Science 2019-12-20 @SciencePDFbooksandmagazines

RESEARCH | RESEARCH ARTICLE

Fig. 4. Treatment with targeted therapies promotes mutagenesis in CRC conditions and normalized to cell viability. NanoLuc signal from MLH1 KO
cells. (A) Schematic representation of the CA-NanoLuc reporter assay. clones was then compared with signal detected in MLH1 wild-type cells (CTR).
(B) MMRd HCT116 and MMRp DiFi, WiDr, and NCIH508 CRC cells were transduced Results represent means ± SD (n = 4). *p < 0.05; **p < 0.01; ***p < 0.001
with the NanoLuc lentivirus. At the indicated time points, NanoLuc signal was (Student’s t test). (D) DiFi, WiDr, and NCIH508 CRC cells were treated
evaluated and normalized to cell viability. Results represent means ± SD (n = 3). as indicated. NanoLuc signal was normalized to cell viability. NanoLuc signal
**p < 0.01; ***p < 0.001 (Student’s t test). NS, not a statistically significant from treated cells was then compared with signal detected in untreated
difference. (C) NanoLuc signal in HT29 MLH1-KO clones (cl. 1 and cl. 2). (NT) cells. Results represent means ± SD (n = 3). *p < 0.05; **p < 0.01;
NanoLuc signal was evaluated after 72 and 96 hours of growth in standard ***p < 0.001 (Student’s t test).

mispairings, incorporation of aberrant DNA 53BP1-positive nuclei upon EGFR and BRAF of reactive oxygen species (ROS) in cancer cells
primer ends, and increased mutagenesis rate blockade (fig. S9, A and B). In direct opposition (36, 37), potentially contributing to DNA dam-
(32, 33). to BRCA1, 53BP1 promotes nonhomologous age during treatment. ROS levels significantly
end joining–mediated DSB repair while pre- increased when CRC cells were exposed to
We therefore investigated whether treat- venting HR through restriction of end resec- EGFR and BRAF inhibitors (Fig. 3C). By con-
ment with targeted therapies leads to genomic tion (35). These data suggest that targeted trast, ROS levels were not increased in per-
damage in cancer cells and if error-prone– therapies trigger a switch from high-fidelity to manently drug-resistant (adapted) cells upon
mediated repair of DNA damage is favored error-prone–mediated repair of DNA damage, drug treatment (fig. S9C).
when CRC cells encounter the hostile environ- thereby potentially increasing the occurrence
ment imposed by targeted therapies. Indeed, of mutations conferring drug resistance. The drug-induced increase in ROS levels
quantification of phosphorylation of H2AX was abrogated when targeted therapies were
at Ser139 (gH2AX), a common marker of DNA We next explored the possible causes of the administered in the presence of the antioxi-
damage (34), revealed a dose- and time- DNA damage observed upon the administra- dant N-acetyl-L-cysteine (NAC) (Fig. 3C). NAC
dependent increase in the number of foci- tion of targeted therapies. Although several administration partially reduced the number
positive nuclei upon drug treatment (Fig. 3, A chemotherapeutic agents directly generate of gH2AX foci-positive nuclei upon EGFR and
and B, and fig. S8, A and B), whereas no fur- DNA damage, drugs interfering with oncogenic BRAF blockade (fig. S10, A and B). However,
ther increase was observed in permanently signaling (such as EGFR or BRAF inhibitors) cotreatment with NAC did not prevent or res-
resistant cells upon drug treatment (fig. S8, C are not directly genotoxic. However, it has been cue down-regulation of DNA-repair genes (fig.
and D). In addition, we observed a dose- and shown that certain targeted therapies, such as S10C). The addition of NAC delayed onset of
time-dependent increase in the number of ABL and BRAF inhibitors, increase the levels relapse to targeted therapies when administered

Russo et al., Science 366, 1473–1480 (2019) 20 December 2019 5 of 8

RESEARCH | RESEARCH ARTICLE

Fig. 5. Adaptive mutability leads to genetic instability in CRC cells in response WiDr cells (parental) treated with cetuximab + DAB for 14 days (persisters) and at
to therapy-induced stress. (A) Percentage of unstable microsatellite regions in resistance. (D) DNA was collected from one vehicle-treated and one cetuximab-
DiFi and WiDr persister and resistant cells compared with their parental counterpart resistant PDX. Percentage of unstable microsatellite regions of the tumor collected
(CTRL). (B) Length distribution of one representative microsatellite region for from the cetuximab-resistant mouse (PDX CTX-R) compared with the vehicle-
drug-resistant DiFi and WiDr cell lines. ***p < 0.001 (c2 test). (C) Number of treated (CTRL) mouse is reported. (E) Length distribution of one representative
unstable microsatellite sites detected by NGS-based high-depth capture panel in microsatellite region. ***p < 0.001 (c2 test).

together with mitogen-activated protein kinase mammalian target of rapamycin (mTOR), of EGFR or KRAS in DiFi cells and of BRAF
(MAPK) pathway inhibitors (fig. S10, D and E) which is a master regulator of mammalian (±EGFR) in WiDr cells led to reduced expres-
(38, 39). cellular stress response (40). Indeed, the mTOR sion of DNA-repair proteins, triggered DNA
effectors pS6K–p70K were down-regulated with damage and mTOR down-regulation (Fig. 3E),
Interfering with oncogenic dependencies kinetics comparable to that of MMR and HR and increased ROS levels (fig. S12). These re-
initiates a stress response in CRCs regulation upon EGFR and BRAF pharma- sults exclude the possibility that drug-induced
cological blockade (Fig. 3D). However, silenc- down-regulation of DNA-repair pathways
To elucidate the mechanistic basis of therapy- ing of mTOR did not affect the expression of could be due to a nonspecific (off-target) effect
induced mutagenesis in cancer cells, we tested DNA-repair proteins or gH2AX (Fig. 3E). It of the anti-EGFR antibody cetuximab or the
whether the adaptive mutability that we ob- is therefore plausible that down-regulation BRAF inhibitor dabrafenib.
served in response to targeted therapies was of mTOR contributes to stress-induced muta-
simply a secondary response to G1 cell-cycle genesis of cancer cells but is not sufficient to Targeted therapies induce adaptive mutability
arrest or DNA damage or if it represented an activate this phenotype. in CRC cells
active stress response. We found that thymidine-
mediated cell-cycle stress (fig. S11, A to C) or The exquisite sensitivity of DiFi and WiDr Next, we tested whether the stress response
direct DNA damage with the alkylating agent cells to EGFR and BRAF blockade reflects cell- induced by targeted therapies translated into
oxaliplatin (fig. S11, D to F) instead promoted specific oncogenic alterations. The EGFR locus increased mutagenesis in CRC cells. We used
the up-regulation of the MMR- and HR-repair is amplified in DiFi cells (2); the WiDr cells a reporter assay in which a dinucleotide CA-
systems (fig. S11, C and F), and G1 cell-cycle carry the BRAF p.V600E oncogenic mutation, repeat microsatellite drives the NanoLuc en-
arrest by nutrient starvation did not lead to but they also become dependent on feedback zyme coding sequence out of frame (Fig. 4A).
modulation of DNA-repair gene expression activation of EGFR when treated with BRAF Random mutations that introduce frameshifts
(fig. S11, G to I). In bacterial cells, both the inhibitors (41). We therefore investigated in this region, in the absence of a functional
DNA damage–activated SOS response and the whether interfering with the oncogenic depen- MMR, would restore the NanoLuc open read-
general stress response appear to be required dency of cancer cells could directly initiate the ing frame, leading to bioluminescence. Analo-
to induce adaptive mutagenesis (14). We there- drug-induced stress phenotype. Indeed, small gous approaches have previously been used to
fore examined the modulation of the kinase interfering RNA (siRNA)–mediated knockdown measure MMR defects in cancer cells (42–44).

Russo et al., Science 366, 1473–1480 (2019) 20 December 2019 6 of 8

RESEARCH | RESEARCH ARTICLE

To validate the assay, we first introduced the to targeted agents (Fig. 5, A and B), as shown by ment the mutational processes operative in
CA-NanoLuc vector into a MMRd human CRC a shift in the length of microsatellite regions, cells (47). In future studies, it will be interest-
cell line (HCT116) and three MMRp human highlighting the impact of targeted therapies ing to establish whether specific mutational
CRC cell lines (DiFi, WiDr, and NCIH508). The on the DNA-repair process and mutagenicity.
NanoLuc signal was significantly higher in To detect the occurrence of microsatellite signatures emerge under targeted therapies.
MMRd cells after 48 hours of standard growth alterations in nonclonal cell populations, we
conditions (Fig. 4B). This difference was further utilized a high-depth capture panel that de- Resolving such processes, which we postulate
increased when HCT116 cells were kept in cul- tects hotspot somatic variants and shifts in
ture for several days, whereas the signal in the the length of microsatellite regions. Indeed, occur transiently in small cell subpopulations,
MMRp lines remained low (Fig. 4B), indicat- such high-sensitivity analysis unveiled a sig-
ing that the CA-NanoLuc assay effectively de- nificant shift in the length of microsatellite is likely to require extensive genomic com-
tects MMR deficiency in cancer cells. regions in both persister and drug-resistant
cells (Fig. 5C and fig. S17). parisons of multiple clones and independent
We next used the CA-NanoLuc system to
measure the impact of ectopic inactivation We next assessed the impact of targeted data points.
of MMR in CRC cells. To this end, we used therapies on the genomic landscape of PDXs
CRISPR-CAS9 to inactivate the MLH1 gene by studying a PDX model (CRC0078) (Fig. 2A These results may have clinical implications.
in the HT29 human CRC cell line. After the and fig. S7D) that was continuously treated
isolation of two independent MLH1 knockout with cetuximab until it developed resistance The knowledge that cancer cells under ther-
(KO) clones (fig. S13, A and B), they were (fig. S18). WES analysis of the cetuximab-
transduced with the CA-NanoLuc vector. MLH1 resistant tumor tissue revealed alterations in apeutic stress down-regulate key effectors of
KO clones exhibited higher levels of NanoLuc microsatellite genomic regions that were not
signal as expected, confirming that the assay present in the PDX tumor collected from the the DNA-repair machinery, such as MMR and
can detect inactivation of DNA MMR (Fig. 4C). corresponding untreated mouse (Fig. 5, D
Next, drug-dependent (transient) MMR down- and E). Overall, these results indicate that CRC HR, exposes a vulnerability that could be clin-
regulation was evaluated. EGFR and BRAF in- cells and a CRC PDX model exposed to tar-
hibition led to time-dependent increases of geted therapies experience loss of replication ically exploited. For example, it will be impor-
bioluminescence (Fig. 4D), paralleling the fidelity in regions of nucleotide repeats.
down-regulation of DNA-repair effectors and tant to assess whether down-regulation of HR
the up-regulation of low-fidelity polymerases. Discussion
We further found that permanently resist- proteins confers sensitivity to poly-ADP-ribose
ant derivatives no longer exhibited adaptive The development of resistance has emerged
mutability in response to targeted therapies as a major limitation of targeted therapies polymerases (PARP) inhibitors as observed in
(fig. S14). directed against oncoproteins such as EGFR, HR-deficient cancers (48–50). Moreover, phar-
BRAF, and ABL (25). macological or genetic interference could be
Genomic alterations in CRC cells upon
treatment with targeted therapies In this study, we tested the hypothesis that deployed to curb the cellular mechanisms that
cancer cells treated with targeted therapies ac-
To determine whether molecular evidence of tivate stress-induced mutagenic mechanisms. initiate drug-driven adaptive mutagenesis with
adaptive mutability was present in the genome We found that persister (drug-tolerant) cancer
of CRC cells treated with EGFR and BRAF cells that survive EGFR and/or BRAF inhibition the goal of reducing the generation of new
inhibitors, we analyzed whole-exome sequenc- exhibit DNA damage, down-regulate mismatch
ing (WES) data from DiFi and WiDr parental, and HR repair proteins, switch from high- variants during treatment. This strategy could
persister, and drug-resistant derivative cells. fidelity to error-prone–mediated repair of DNA
The overall mutational burden (i.e., the num- damage, and transiently increase their muta- potentially increase and prolong the clinical
ber of mutations per megabase) of persisters genic ability.
and the drug-resistant cell population was efficacy of targeted therapies.
only marginally affected (fig. S15, A and B). Stress-induced mutagenesis is a character-
As a control, we assessed whether MMR per- istic trait of unicellular organisms to tran- REFERENCES AND NOTES
manent inactivation induced by MLH1 KO siently accelerate genetic diversity in a fraction
affected the mutational burden of HT29 CRC of the population when encountering a hos- 1. S. E. Luria, M. Delbrück, Genetics 28, 491–511 (1943).
cells and found that it was only marginally tile environment. (16). Indeed, we found that 2. S. Misale et al., Nature 486, 532–536 (2012).
affected (fig. S16A). therapy-induced modulation of DNA repair in 3. L. A. Diaz Jr. et al., Nature 486, 537–540 (2012).
cancer cells is also transitory and reverts back 4. A. Gutierrez et al., Nat. Commun. 4, 1610 (2013).
Given these results, we changed our approach. once a mutational landscape able to restore 5. H. Long et al., Proc. Natl. Acad. Sci. U.S.A. 113, E2498–E2505
Because treatment with targeted therapies led the ability to grow in the presence of the drug
to a transient MMR-deficient phenotype, we is achieved. We postulate that in cells of mul- (2016).
reasoned that MMR status could be more ticellular organisms, stress-induced mutagen- 6. I. Bjedov et al., Science 300, 1404–1409 (2003).
easily detected by examining microsatellite esis is not operational. However, in cancer cells 7. D. M. Fitzgerald, P. J. Hastings, S. M. Rosenberg, Annu. Rev.
regions, where DNA replication slippage errors that have lost tissue-imposed homeostasis—
occur frequently and are ineffectively repaired and in many ways operate like unicellular Cancer Biol. 1, 119–140 (2017).
in the absence of MMR. Indeed, WES analysis organisms—this ancestral program is still avail- 8. P. L. Foster, Crit. Rev. Biochem. Mol. Biol. 42, 373–397
unveiled alterations in microsatellite regions of able and is unleashed by oncoprotein-targeted
HT29 in which the MLH1 gene was genetically drugs. A similar process has also been ob- (2007).
knocked out (fig. S16, B and C). We also de- served in cancer cells undergoing hypoxia- 9. R. S. Galhardo, P. J. Hastings, S. M. Rosenberg, Crit. Rev.
tected increased genetic instability in the micro- driven stress (7, 45, 46).
satellite regions of CRC cells made resistant Biochem. Mol. Biol. 42, 399–435 (2007).
The analysis of mutational signatures has 10. E. Shor, C. A. Fox, J. R. Broach, PLOS Genet. 9, e1003680
emerged as a valuable tool with which to docu-
(2013).
11. E. Heidenreich, Crit. Rev. Biochem. Mol. Biol. 42, 285–311

(2007).
12. D. M. Fitzgerald, S. M. Rosenberg, PLOS Genet. 15, e1007995

(2019).
13. G. P. Rodriguez et al., Proc. Natl. Acad. Sci. U.S.A. 109,

6153–6158 (2012).
14. R. G. Ponder, N. C. Fonville, S. M. Rosenberg, Mol. Cell 19,

791–804 (2005).
15. G. J. McKenzie, S. M. Rosenberg, Curr. Opin. Microbiol. 4,

586–594 (2001).
16. P. L. Foster, BioEssays 22, 1067–1074 (2000).
17. F. Taddei et al., Nature 387, 700–702 (1997).
18. J. Torkelson et al., EMBO J. 16, 3303–3311 (1997).
19. W. A. Rosche, P. L. Foster, Proc. Natl. Acad. Sci. U.S.A. 96,

6862–6867 (1999).
20. G. J. McKenzie, R. S. Harris, P. L. Lee, S. M. Rosenberg,

Proc. Natl. Acad. Sci. U.S.A. 97, 6646–6651 (2000).
21. S. V. Sharma et al., Cell 141, 69–80 (2010).
22. G. R. Oxnard, Nat. Med. 22, 232–234 (2016).
23. M. Ramirez et al., Nat. Commun. 7, 10690 (2016).
24. A. N. Hata et al., Nat. Med. 22, 262–269 (2016).
25. S. Misale, F. Di Nicolantonio, A. Sartore-Bianchi, S. Siena,

A. Bardelli, Cancer Discov. 4, 1269–1280 (2014).
26. S. Kopetz et al., N. Engl. J. Med. 381, 1632–1643 (2019).
27. Z. D. Nagel et al., Proc. Natl. Acad. Sci. U.S.A. 111, E1823–E1832

(2014).
28. A. J. Pierce, R. D. Johnson, L. H. Thompson, M. Jasin,

Genes Dev. 13, 2633–2638 (1999).
29. A. Bertotti et al., Nature 526, 263–267 (2015).
30. C. Isella et al., Nat. Commun. 8, 15107 (2017).
31. A. J. Rattray, J. N. Strathern, Annu. Rev. Genet. 37, 31–66

(2003).

Russo et al., Science 366, 1473–1480 (2019) 20 December 2019 7 of 8

RESEARCH | RESEARCH ARTICLE

32. V. M. Krutiakov, Mol. Biol. (Mosk.) 40, 3–11 (2006). ACKNOWLEDGMENTS DNA-repair capacity. A. Bardelli is a member of the scientific
33. M. F. Goodman, Annu. Rev. Biochem. 71, 17–50 (2002). advisory board of NeoPhore and a shareholder of NeoPhore and
34. L. J. Mah, A. El-Osta, T. C. Karagiannis, Leukemia 24, 679–686 We thank members of Molecular Oncology Laboratory at Candiolo PhoreMost. The other authors declare no competing interests.
Cancer Institute for critically reading the manuscript, A. Cassingena Data and materials availability: The FM-HCR reporter plasmids
(2010). and F. Sassi for their help with experiments, and G. McKenzie are available from Z.D.N. under a material transfer agreement. The
35. S. Panier, S. J. Boulton, Nat. Rev. Mol. Cell Biol. 15, 7–18 (2014). for the NanoLuc plasmid design. Funding: This research was pDRGFP and the pCBASce-I plasmids are available from AddGene
36. G. Cesi, G. Walbrecq, A. Zimmer, S. Kreis, C. Haan, Mol. Cancer supported by Fondazione AIRC under the 5 per Mille 2018 ID 21091 under a material transfer agreement. The NanoLuc-expressing
Program (P.I. A. Bardelli, G.L. S. Siena, G.L. A. Bertotti, G.L. L. Trusolino, plasmid is available from PhoreMost, Ltd. (Cambridge, UK) under
16, 102 (2017). G.L. F. Di Nicolantonio); AIRC 2010 Special Program Molecular a material transfer agreement. The HT29 empty and MLH1 KO
37. M. Koptyra et al., Blood 108, 319–327 (2006). Clinical Oncology 5 per Mille, Project No. 9970 Extension Program cells are available from A. Bardelli (UNITO) under a material transfer
38. M. L. Sagristá, A. E. García, M. A. De Madariaga, M. Mora, (A. Bardelli); AIRC IG no. 16788 (A. Bardelli); AIRC IG 2018–ID 21923 agreement. RNA-sequencing and DNA-sequencing data have been
project (P.I. A. Bardelli); H2020 grant agreement no. 635342-2 deposited in ENA (from EBI; no. PRJEB28674).
Free Radic. Res. 36, 329–340 (2002). MoTriColor (A. Bardelli); Ministero Salute, Ricerca Corrente 2019
39. E. D. Chan, D. W. Riches, C. W. White, Am. J. Respir. Cell Mol. RC2019 (A. Bardelli); H2020 grant agreement no. 754923 SUPPLEMENTARY MATERIALS
COLOSSUS (L.T.). A. Bardelli has also received research funding science.sciencemag.org/content/366/6472/1473/suppl/DC1
Biol. 24, 627–632 (2001). from PhoreMost and NeoPhore. Author contributions: M.R. and Materials and Methods
40. R. A. Saxton, D. M. Sabatini, Cell 168, 960–976 (2017). A. Bardelli conceived the study. M.R., A.S., N.M.R., S.A., S.L., Figs. S1 to S18
41. A. Prahallad et al., Nature 483, 100–103 (2012). V.A., A.M. M.C., M.G., and M.C.L. conducted the experiments and Tables S1 and S2
42. J. Shen et al., Nat. Med. 24, 556–562 (2018). analyzed data. G.C. performed NGS and bioinformatics analysis. References (51–61)
A. Bartolini performed NGS panel-based experiments. L.N. performed
43. N. Chatterjee, Y. Lin, B. A. Santillan, P. Yotnda, J. H. Wilson, RNAseq bioinformatics analysis. Z.D.N. and C.G.P. provided plasmids View/request a protocol for this paper from Bio-protocol.
Proc. Natl. Acad. Sci. U.S.A. 112, 3764–3769 (2015). harboring DNA damage. A. Bertotti and L.T. analyzed and provided
PDX material. I.S. performed and analyzed IHC data; A.A., A.S.-B., 21 September 2018; resubmitted 24 July 2019
44. R. Parsons et al., Cell 75, 1227–1236 (1993). and S.S. provided patient samples. F.D.N. analyzed data. M.R. Accepted 26 October 2019
45. V. T. Mihaylova et al., Mol. Cell. Biol. 23, 3265–3273 and A. Bardelli wrote the manuscript. A. Bardelli supervised the study. Published online 7 November 2019
Competing interests: L.T. is a paid consultant for Eli Lilly, 10.1126/science.aav4474
(2003). AstraZeneca, and Merck KGaA. Z.D.N. is an inventor on a patent
(U.S. 9,938,587) covering methods and kits for determining
46. R. S. Bindra, P. M. Glazer, Cancer Lett. 252, 93–103
(2007).

47. M. Petljak et al., Cell 176, 1282–1294.e20 (2019).
48. L. Li, H. Wang, E. S. Yang, C. L. Arteaga, F. Xia, Cancer Res. 68,

9141–9146 (2008).
49. S. Nowsheen et al., PLOS ONE 6, e24148 (2011).
50. J. Mateo et al., Ann. Oncol. 30, 1437–1447 (2019).

Russo et al., Science 366, 1473–1480 (2019) 20 December 2019 8 of 8

RESEARCH

SUPERFLUIDITY namics of few-vortex clusters in a single shot
and tracked over many minutes as they interact,
Coherent vortex dynamics in a strongly interacting dissipate energy, and annihilate. We find that
superfluid on a silicon chip evaporative heating occurs, where the anni-
hilation of low-energy vortices draws energy
Yauhen P. Sachkou1*, Christopher G. Baker1*, Glen I. Harris1*, Oliver R. Stockdale2, Stefan Forstner1, out of a background flow, causing a net in-
Matthew T. Reeves2, Xin He1, David L. McAuslan1, Ashton S. Bradley3, Matthew J. Davis1,2, Warwick P. Bowen1† crease in free-vortex kinetic energy as the
system evolves.
Quantized vortices are fundamental to the two-dimensional dynamics of superfluids, from quantum
turbulence to phase transitions. However, surface effects have prevented direct observations of coherent Our experiments yield a vortex diffusivity
two-dimensional vortex dynamics in strongly interacting systems. Here, we overcome this challenge five orders of magnitude lower than has been
by confining a thin film of superfluid helium at microscale on the atomically smooth surface of a silicon observed previously for unpinned vortices in
chip. An on-chip optical microcavity allows laser initiation of clusters of quasi–two-dimensional vortices superfluid helium films (34). This verifies that
and nondestructive observation of their decay in a single shot. Coherent dynamics dominate, with the diffusivity can become exceptionally small
thermal vortex diffusion suppressed by five orders of magnitude. This establishes an on-chip platform when operating at temperatures far below the
with which to study emergent phenomena in strongly interacting superfluids and to develop quantum superfluid transition temperature, as conjec-
technologies such as precision inertial sensors. tured from extrapolation of 30-year-old exper-
imental observations (8, 34). Therefore, the
S trongly interacting many-body quantum are generally limited to the regime of weak system operates well within the regime of
systems exhibit rich behaviors of signif- interactions, where the Gross–Pitaevskii equa- coherent vortex dynamics, with the time scale
icance to areas ranging from supercon- tion provides a microscopic model of the dy- for dissipation found to exceed the coherent
ductivity (1) to quantum computation namics of the superfluid. The regime of strong evolution time by more than five orders of
(2, 3), astrophysics (4–6), and even string interactions can be reached by tuning the magnitude. The on-chip system reported here
theory (7). The first example of such a be- atomic scattering length in ultracold gases provides a platform with which to explore the
havior, superfluidity, was discovered >80 years (20–23), although three-body losses can limit dynamics of phase transitions and quantum
ago in cryogenically cooled liquid helium-4. It the system lifetime (20). Superfluid helium, by turbulence in strongly interacting superfluids
was found to persist even in very thin two- contrast, exists naturally in this state, which and to study how such fluids evolve toward
dimensional films, for which the well-known offers the possibility of studying many-vortex thermal equilibrium and dissipate energy.
Mermin–Wagner theorem precludes conden- dynamics over a much greater range of length
sation into a superfluid phase in the thermo- and time scales (24). The strongly interacting Nonequilibrium vortex clusters can be
dynamic limit [see, e.g., discussion in (8)]. regime is relevant for astrophysical superfluid generated with laser light
This apparent contradiction was resolved by phenomena such as pulsar glitches (6) and
Berezinskii, Kosterlitz, and Thouless (BKT), superfluidity of the quark-gluon plasma in the Figure 1A shows a schematic of our experi-
who predicted that quantized vortices allow a early universe (5) and is highly challenging to mental apparatus. A microscale optical ring
topological phase transition into superfluidity treat theoretically (25, 26). The vortex dynam- cavity with laser-reflowed atomically smooth
(9, 10). It is now recognized that quantized ics in this regime are typically predicted using surface, conventionally known as a micro-
vortices dominate much of the two-dimensional phenomenological vortex models. However, toroid (35), is placed inside a sealed sample
dynamics of out-of-equilibrium superfluids questions such as whether the vortices have chamber within a closed-cycle 3He cryostat.
even outside the regime of BKT superfluidity, inertia (27, 28), what is the precise nature of The microtoroid confines light in whispering
including quantum turbulence (11) and anom- the forces they experience from the normal gallery mode resonances, which are excited
alous hydrodynamics (12). component of the fluid (29), and how to treat through an optical nanofiber. The microtoroid
dissipation given the nonlocal nature of the is fabricated from silica, has a radius of R =
Recently, laser control and imaging of vorti- vortex flow fields (8, 30) all remain to be con- 30 mm, and is suspended above a silicon chip
ces in ultracold gases (13, 14) and semicon- clusively answered. Moreover, point-vortex by a pedestal that narrows to a radius of rp ~
ductor exciton-polariton systems (15, 16) have modeling offers limited insight into the pro- 1 mm where it contacts the bottom surface of
provided rich capabilities to study superfluid cesses of vortex creation and annihilation, the microtoroid (Fig. 1A, inset). The sample
dynamics including, for example, the forma- which are crucial to understanding the dy- chamber is filled with 4He gas at room tem-
tion of collective quasi–two-dimensional vortex namics of topological phase transitions. perature. The pressure is chosen so that the
dipoles with negative temperature and large- gas condenses directly into an unsaturated
scale order (17, 18) as predicted by Lars Onsager Here, we report the observation of coherent superfluid film at ~1 K. This film coats the
70 years ago (19). However, these experiments dynamics of quasi–two-dimensional vortices in inside of the chamber, including the optical
a strongly interacting superfluid. We achieve microcavity (31, 36). Vortices and superfluid
1Australian Research Council Centre of Excellence for this by developing a microscale photonic plat- surface waves can coexist in the superfluid
Engineered Quantum Systems, School of Mathematics and form to initialize vortex clusters in a few- film and are geometrically confined to the
Physics, University of Queensland, St. Lucia, Queensland nanometer-thick film of helium-4 on a silicon surface of the microtoroid.
4072, Australia. 2Australian Research Council Centre of chip, confine them, and image their spatial
Excellence in Future Low-Energy Electronics Technologies, distribution over time. Our experiments char- Our experiments operate with a superfluid
School of Mathematics and Physics, University of acterize vortex distributions through their film thickness of d ~ 7.5 nm and a temperature
Queensland, St. Lucia, Queensland 4072, Australia. interactions with resonant superfluid surface of T ~ 500 mK, well below the superfluid
3Department of Physics, Centre for Quantum Science, and waves, taking advantage of ultraprecise sensing transition. In these conditions, the thickness
Dodd-Walls Centre for Photonic and Quantum Technologies, methods from cavity optomechanics (31–33). far exceeds the superfluid healing length, such
University of Otago, Dunedin 9016, New Zealand. Microscale confinement greatly enhances the that the system is outside the regime of BKT
*These authors contributed equally to this work. interactions and enables resolution of the dy- superfluidity. However, from the hydrody-
†Corresponding author. Email: [email protected] namic perspective, the system is quasi–two-
dimensional because both the lowest-energy
vertically propagating phonon modes and the
lowest-energy Kelvin modes of vortices have

Sachkou et al., Science 366, 1480–1485 (2019) 20 December 2019 1 of 6

RESEARCH | RESEARCH ARTICLE

energies comparable to the thermal energy, Interactions with third-sound allow but in this case are isolated from the vorti-
so only a few of these modes are expected to vortex imaging ces because of their physical separation [see
be thermally excited [see section 1.1 of (37)]. The interaction between light and vortices is section 2 of (37)]. We monitor the thermal
extremely weak in superfluid helium because excitations of third-sound modes in real time
We find experimentally that vortex clusters of its exceedingly low refractive index and the through their effect on the height of the super-
can be optically initialized on the surface of the Angström scale of the vortex cores. This pre- fluid film in the optical evanescent field of the
microtoroid in several ways, including pulsing cludes direct optical imaging techniques. Fur- microtoroid (31). This manifests as fluctuations
the intensity of the injected laser to induce su- thermore, the tracer particles used to image of the phase of light confined inside the cavity,
perfluid flow through the “fountain effect” (36) vortex dynamics in bulk three-dimensional which are resolved through balanced homo-
and optomechanical driving of low-frequency helium (24) are too large to be applied in thin dyne detection implemented within a fiber
surface waves by dynamical backaction (31). films. Instead, in our experiments, the vortex interferometer (Fig. 1A). Frequency analysis
Both of these techniques induce flow that ex- dynamics are tracked through their influence of the output photocurrent reveals resonances
ceeds the superfluid critical velocity at the inter- on resonant superfluid third-sound waves with frequencies that are in good agreement
face between the pedestal and the microtoroid. that are also confined to the bottom surface of with expectations for third-sound modes con-
This triggers the generation of vortex pairs the microtoroid (Fig. 2A). Third-sound waves fined to a circular geometry [see sections 1 and
on the bottom surface of the microtoroid in are surface waves analogous to shallow-water 2 of (37)]. The resonances are Bessel modes,
close proximity to the pedestal (Fig. 1B). In the waves but with a restoring force provided by characterized by their radial m and azimuthal
presence of a circular boundary, an ensemble the van der Waals interaction with the sub- n mode numbers (41).
of vortex pairs evolves into a metastable state strate rather than gravity [see, e.g., (39)]. They
characterized at high energies by a large-scale also exist on the top surface of the microtoroid The vortex flow field causes Doppler shifts
negative-temperature Onsager vortex dipole of the frequencies of the third-sound modes,
(17, 19) [see section 3.2 of (37)]. In our case, the
microtoroid pedestal introduces a deep po- A FBS C
tential to which vortices can pin [see section 7 3He cryostat
of (37)], qualitatively modifying the physics. Laser Tapered fiber PD
Vortices of one sign become pinned, creating a 1555 nm
macroscopic circulation, whereas vortices of
the opposite sign evolve into a free-orbiting Microtoroid
metastable cluster similar to the case without
a pinning site [see section 3.2 of (37)]. ΔN

The time scale within which a metastable 20 µm Oscilloscope
state is reached can be estimated from the
characteristic turnover time for internal re- B
arrangement of the free-vortex cluster, t ~ ΔK
rc2/Nk, where N is the number of free vortices,
rc is the radius of the cluster, and k = h/mHe Laser heating Evaporation
is the circulation quantum, with mHe being
the mass of a helium atom and h Planck’s ΔL
constant (38) [see section 3.2 of (37)]. Consid-
ering the case of two free vortices separated Superflow
by the disk radius provides an upper bound
to the turnover time of t ≲ 5 ms. This is sub- 0 pmax
stantially faster than both the dissipation of
the system and the temporal resolution of our Fig. 1. Laser initialization of vortex clusters. (A) Experimental setup. A balanced homodyne detection
measurements, which are ~1 min and 1 s, re- scheme is implemented within a fiber interferometer. FBS, fiber beam splitter; PD, photodetector. Inset shows
spectively. Consequently, the vortex cluster a scanning electron microscope image of the microtoroidal optical cavity used in the experiments. Scale
can be well approximated to exist in a meta- bar, 20 mm. (B) Sketch of the vortex generation process. Laser heating of the microtoroid perimeter
stable state throughout its evolution, with causes superfluid evaporation, followed by superflow (36). The flow exceeds the superfluid critical velocity
this state modified continuously by dissipa- at the top of the microtoroid pedestal, seeding the generation of vortex pairs. (C) Exemplar simulated
tion and in discrete steps by vortex annihila- metastable distributions showing the effects of vortex annihilation and of changes in total kinetic energy K
tion events. and angular momentum L. The color map indicates the free-vortex probability density, with a maximum of
pmax = {0.035, 0.075, 0.023, 0.12} mm−2, respectively, for the top to bottom metastable states. Top metastable
Each possible metastable state is uniquely state: {N,K,L} = {10,0.43 aJ, 120 ag mm2 s–1}; second to top: N → 9; second to bottom: K → 0.41 aJ;
characterized by the number of free vortices, and bottom: L → 44 ag mm2 s–1.
kinetic energy, and angular momentum. Per-
forming point-vortex simulations, we determine
the possible metastable vortex distributions
as a function of these three parameters [see
section 3.2 of (37)]. In their metastable state,
the free vortices exist in an orbiting horseshoe-
shaped cluster separated from the origin,
which, together with the macroscopic circula-
tion, forms a vortex dipole. Figure 1C illustrates
how the distribution of the cluster evolves as
a consequence of changes in vortex number,
kinetic energy, and angular momentum.

Sachkou et al., Science 366, 1480–1485 (2019) 20 December 2019 2 of 6

RESEARCH | RESEARCH ARTICLE

A critical flow. We then simultaneously measure
the frequency splittings induced on the (m,n) =
r (1,3), (1,4), (1,5), (1,7), and (1,8) third-sound
modes as the cluster evolves over time, an-
r ticipating vortex-dipole decay as illustrated in
Fig. 3A. As a representative example, Fig. 3B
B min max shows the observed power spectral density
amplitude amplitude and frequency splitting of the (1,7) third-sound
0.25 mode at the start of the measurement run, just
Δf after the vortex cluster has been initialized
0.2 [see section 2 of (37) for the initial power
Δf R2 (Hz mm2) spectral densities of all modes]. Using the
PSD0.15 vortex position–dependent splitting function
D f (r) (Fig. 2B), the frequency shifts expected
CCW CW on each of the third-sound modes are com-
puted for all possible metastable distributions
0.1 and compared with the observed shifts. This
allows us to ascertain both the metastable
0.05 f state that most closely matches the observed
frequency shifts at a given time, as well as the
0 1 range of metastable states for which the shifts
0 0.2 0.4 0.6 0.8 are statistically indistinguishable.
r/R
As shown for the initial metastable vortex
Fig. 2. Interactions between vortices and third-sound on a disk. (A) Top left: Free vortices (blue) and distribution in Fig. 3C (colored circular points),
pinned circulation (red) on the bottom surface of the microtoroid. Bottom left: A third-sound mode on the we find excellent agreement between the ob-
same surface. The vortices and third-sound couple due to the superposition of their flow fields, shown on served frequency splittings and those ob-
the right for the case of a single vortex (blue dot) offset from the disk origin by distance r. Here, the surface color tained with the best-fit metastable distribution.
represents the third-sound mode amplitude profile and the blue lines are vortex streamlines. Confinement Moreover, although the measurements only
within the same microscale domain enhances both the interaction rate between vortices and third-sound and loosely constrain the angular momentum
the resulting frequency splitting between counterpropagating third-sound modes. (B) Normalized splitting [see section 4.2 of (37)], we find that the
per vortex for third-sound modes (m,n) = (1,3), (1,5), and (1,8) calculated by finite-element modeling using the statistical uncertainty in the number of free
techniques detailed in (40) and outlined in section 2.1 of (32), with their respective spatial profiles. The inset vortices and kinetic energy is relatively small.
schematically depicts the vortex-induced splitting Df between clockwise and counterclockwise third-sound The free vortex number in the initial vortex
modes in the presence of a clockwise vortex, which would be observed as a function of frequency f in the power dipole is statistically constrained to a value
spectral density (PSD) of the optically measured superfluid motion. of either 16 or 17. For the case of 16 vortices,
the total kinetic energy and angular momen-
lifting the degeneracy between clockwise and third-sound modes is then equal to the sum tum are constrained to K ¼ 0:8Àþ00::022 aJ and
counterclockwise waves (Fig. 2). The magni- of the splittings generated by each vortex L ¼ 0:2Àþ00::037 fg mm2 s–1, respectively, where-
tude of the frequency splitting induced by a and by the macroscopic circulation pinned as for 17 vortices, they are constrained to
vortex scales inversely with the area in which to the pedestal. We exploit this linearity, com- K ¼ 0:8Àþ00::01 3 aJ and L ¼ 0:3Àþ00::205 fg mm2 s–1.
it is confined (39, 40). As such, the microscale bined with the vortex position–dependent in- We are also able to determine the dipole sep-
confinement provided by our microtoroid teraction and simultaneous measurements of aration, which in the case of 17 vortices is
allows greatly enhanced resolution compared splitting on several third-sound modes, to char- found to be 7:3þÀ01::28 mm.
with previous experiments (39, 42). The split- acterize the spatial distribution of vortex clus-
ting also depends both on the position of the ters in a manner analogous to experiments Continuously monitoring the superfluid
vortex and on the spatial profile of the third- that use multiple cantilever eigenmodes to third-sound modes reveals that their split-
sound mode, as shown for several modes in image the distribution of deposited nano- tings decay over a time scale on the order of
Fig. 2B. Despite the strong interactions be- particles (43). minutes (Fig. 3C). The observed splittings are
tween helium atoms, the phase coherence and well characterized by a metastable vortex di-
incompressibility of the superfluid combine to Vortex clusters evolve coherently pole throughout this entire decay process. By
ensure linearity. Therefore, the total flow field contrast, they are inconsistent with other vor-
of a vortex cluster is given by the linear super- To determine the instantaneous metastable tex dynamics models such as expansion of a
position of the flow of each constituent vortex. vortex distribution from a single continuous single-sign vortex cluster caused by either
The total splitting between counter-rotating measurement, we generate a nonequilibrium vortex-vortex interactions or diffusive hopping
vortex cluster by optically initiating super- between pinning sites on the surface of the
microtoroid [see section 4 of (37)].

The kinetic energy and free-vortex number
of the metastable state are shown as a function
of time for a single shot in Fig. 4, A and B (blue
curves). The total kinetic energy of the dipole
decays continuously with time as a conse-
quence of weak dissipation. This decay occurs
over a period of ~1 min, comparable to pre-
vious nonspatially resolved measurements of
the decay of a persistent current (44). In our

Sachkou et al., Science 366, 1480–1485 (2019) 20 December 2019 3 of 6

RESEARCH | RESEARCH ARTICLE

experiments, the decay is accompanied by a The negativity of the free-vortex energy can reach the pedestal, where its contribution to
reduction in the number of free vortices as be understood by considering the interfer- the total kinetic energy is at a minimum. To do
vortex-vortex interactions and dissipation ence between the flow fields of a free vortex this, it gives up kinetic energy to the remain-
drive vortices into the center of the disk, where and the macroscopic circulation. Although ing free vortices. This process of removing
they can annihilate with quanta of circulation the high flow velocity near the core of a free low-energy vortices has been described as
of opposite sign pinned to the pedestal. These vortex introduces kinetic energy, the vortex evaporative heating (18), in analogy to evap-
dynamics are supported by point-vortex simu- flow field also cancels a component of the orative cooling of ultracold atomic ensembles.
lations (Fig. 4, C and D), which show good background flow. For a sufficiently large cir- However, whereas standard evaporative heat-
quantitative agreement with only the mutual culation, this cancellation effect dominates, ing can explain a per-vortex increase in kinetic
friction coefficient a as a fitting parameter. leading to an overall negative energy cost to energy, its effect is to reduce the net free-
The agreement between experiment and the- introducing the vortex cluster [see section 7.2 vortex kinetic energy (18). The physics is mod-
ory indicates that within experimental uncer- of (37)]. ified here by the presence of a macroscopic
tainties, the vortex dynamics are consistent background flow. The annihilation of a free
with a simple point vortex model including The increase in free-vortex energy over time vortex with a pinned circulation quanta can-
local phenomenological dissipation and with- can be explained by considering the process of cels a component of this flow, reducing its
out the need to introduce surface pinning vortex annihilation in a macroscopic back- kinetic energy while leaving the total kinetic
sites, inertia to the vortex cores (27, 28), or ground flow. To annihilate, a free vortex must
an Iordanskii force between vortices and the
normal component of the fluid (29). A B vortex-induced splitting
C4
The mutual friction coefficient quantifies 102 PSD (a.u.)
the ratio of coherent to dissipative time scales
in the superfluid dynamics and in our system 101 -3.5 0 3.5 7
is expected to be determined by a combina- 100 f - fm (kHz)
tion of temperature-dependent phonon- and
ripplon-scattering processes, as well as surface -7
roughness effects. From the fit, it is found to be
a ~ 2 × 10–6 [see section 3.3 of (37)]. A second 3 (1,3) (1,5) (1,8)
experiment under similar conditions but with
an initial free-vortex number of N ¼ 33Àþ12 is Splitting (kHz) (1,4) (1,7)
broadly consistent, yielding a ~ 3 × 10–6 [see
section 5 of (37)]. These values are similar 2
to both measurements (45, 46) and theoret-
ical predictions [see section 3.4 of (37)] for 1
bulk superfluid helium, suggesting that sur-
face effects are small despite the quasi–two- 0
dimensional geometry. It has generally been 0 50 100 150 200 250 300 350
thought that fast dissipative processes would
preclude the observation of coherent dynam- Time (s)
ics in superfluid helium films. However, the
mutual friction coefficient obtained here shows Fig. 3. Temporal dynamics of third-sound splitting. (A) Vortex-dipole decay process. Red indicates
that this is not the case in general, with co- quantized circulation around the pedestal reduces due to annihilation events. Light blue indicates orbiting
herent dynamics dominating by more than free-vortex cluster spirals toward the origin due to dissipation. Dissipation is exaggerated for clarity.
five orders of magnitude. This is competitive (B) Experimental splitting observed in the PSD of the (m,n) = (1,7) third-sound mode pair immediately
with the best ultracold atom experiments, after vortex-dipole initialization. Gray spectra are the third-sound mode pair without vortices. The residual
which typically achieve a ~ 6 × 10–4 (17). splitting in these unperturbed spectra is caused by irregularities in the circularity of the microtoroid that
break the degeneracy between standing-wave Bessel modes (31) and is accounted for in data processing
Nonequilibrium vortex dynamics is observed in [see section 1.3 of (37)]. f is the frequency of superfluid motion; fm is the mean resonance frequency of
a single shot third-sound mode pair. (C) Temporal decay of splitting of the (m,n) = (1,8), (1,7), (1,5), (1,4), and (1,3) third-
sound modes (top to bottom traces, respectively). These specific modes were chosen because of the high
The pinning of vortices on the microtoroid ped- signal-to-noise ratio of their PSDs. The raw data were recorded on a high-bandwidth, high-memory-depth
estal results in a macroscopic circulation, as dis- oscilloscope. Six continuous measurements were taken, separated by ~10-s data-saving periods. Colored circles
cussed above. The kinetic energy associated at the start of each trace show the theoretical splitting of each third-sound mode pair for the best-fit initial
with the free vortex cluster is given by Kfree = vortex metastable distribution. Insets show spatial amplitude profiles of each third-sound mode.
K – Kpinned, where Kpinned = rd(Nk)2ln(R/rp)/
4p, the kinetic energy of the macroscopic cir-
culation alone, with r = 145 kg/m3 being the
density of superfluid helium. Kfree is shown
for both our experimental data and simu-
lations by the red curves in Fig. 4, A and C.
During the first minute of evolution, it is
negative and, notably, increases with time.
The dynamics are characterized by steps up
in energy during vortex annihilation events
interspersed with a continuous dissipative
decay.

Sachkou et al., Science 366, 1480–1485 (2019) 20 December 2019 4 of 6

RESEARCH | RESEARCH ARTICLE

i ii iii i ii iii

0.9 18 B

A

15

i ii iii

0.45 12

iii

9

06

3

Kinetic energy (aJ) experiment experiment
Vortex number
-0.45 0

0.9 18

CD

15

0.45 12

9
06

-0.45 3 simulation 300
0
simulation 100 200
Time (s)
0
50 100 150 200 250 300 0

Time (s)

Fig. 4. Single-shot evolution of vortex-cluster metastable states. (A) Evolution vortex number. Vertical dashed lines correspond to times in (A). Note that although
of total kinetic energy (blue curve) and free-vortex cluster kinetic energy (red curve). these data display steps in the vortex number, this is a feature of our analysis
Insets show metastable vortex probability densities at times indicated by the that minimizes the root-mean-square uncertainty only over discrete vortex number.
vertical dashed lines. These probability densities have the same color scale as Our experiments approach single-vortex resolution; however, the continuous
those in Fig. 1C, with pmax = {0.0087, 0.015, 0.032} mm−2, respectively, from left to variation of vortex-induced splitting with time precludes direct unambiguous
right. The second and third metastable distributions are taken, respectively, just observation of individual steps in the splitting that result from creation or annihilation
before and just after the 7-to-6 annihilation event. The angular momentum of (40). (C and D) Point-vortex simulations of the system evolution [see section 3.3
each distribution is chosen, within the uncertainty window of the fit, to maximize of (37) for implementation]. (C) Evolution of the total (blue curve) and free-vortex
the entropy of the state and therefore represents the most statistically likely of the (red curve) kinetic energies. (D) Evolution of the free-vortex number. In all traces, the
experimentally plausible distributions. (B) Experimentally determined decay of the shaded area corresponds to a 1-SD uncertainty.

energy essentially unchanged, as can be seen the BKT transition (10) and in the dissipation extrapolation from 30-year-old experimental
by the lack of discrete steps in the blue curves of persistent flow (47). This has motivated sub- observations that the diffusivity could become
of Fig. 4, A and C. Therefore, annihilation stantial research efforts to quantify it both exceedingly small outside the BKT regime at
events increase the kinetic energy of the free- low temperatures (8). Future experiments
vortex cluster by drawing energy out of the near the BKT transition and outside the re- varying the temperature, film thickness, and
background flow. This pushes the cluster gime of BKT superfluidity (8, 34, 44). How- surface properties may further elucidate the
outward to a higher separation, as illustrated ever, achieving a high signal-to-noise ratio has details of dissipation in thin superfluid he-
by the metastable states just before and after generally proved challenging (8), and surface lium films.
an annihilation event, as shown in insets ii pinning is known to have a major influence on
and iii, respectively, in Fig. 4A. measurement outcomes (34, 48). By contrast, Concluding perspectives
our experiments are not dominated by surface
In addition to allowing the observation of pinning [see section 6 of (37)] and achieve The experiments reported here were enabled
evaporative heating in a single continuous near-single-vortex resolution. From them, we by the use of microscale confinement, which
shot, our experiments allow the diffusivity D obtain a value of D = kBTa/rdk ~ 100 nm2 s–1, greatly enhances the interactions between
of vortices to be established in a regime for where kB is the Boltzmann constant. This is vortices, between third-sound and vortices,
superfluid helium films, where it was previ- five orders of magnitude below previous mea- and between third-sound and light. Interac-
ously inaccessible. The vortex diffusivity plays tions with strong pinning sites have previously
an important role in dynamic corrections to surements for which the vortex dynamics are prevented the observation of coherent vortex
not dominated by pinning (34), verifying the

Sachkou et al., Science 366, 1480–1485 (2019) 20 December 2019 5 of 6

RESEARCH | RESEARCH ARTICLE

dynamics in superfluid helium films (42). In REFERENCES AND NOTES 40. S. Forstner et al., New J. Phys. 21, 053029 (2019).
our experiments, vortex-vortex interactions 41. C. G. Baker et al., New J. Phys. 18, 123025 (2016).
dominate because of the increased confinement 1. D. Mitra et al., Nat. Phys. 14, 173–177 (2017). 42. F. M. Ellis, L. Li, Phys. Rev. Lett. 71, 1577–1580 (1993).
and atomically smooth surface of the device. 2. C. Gross, I. Bloch, Science 357, 995–1001 (2017). 43. M. S. Hanay et al., Nat. Nanotechnol. 10, 339–344 (2015).
The smoothness of the surface results in a con- 3. A. D. King et al., Nature 560, 456–460 (2018). 44. D. T. Ekholm, R. B. Hallock, Phys. Rev. Lett. 42, 449–452 (1979).
servative upper bound to the vortex unpinning 4. N. Chamel, J. Astrophys. Astron. 38, 43 (2017). 45. A. I. Golov, P. M. Walmsley, J. Low Temp. Phys. 156, 51–70 (2009).
velocity of 0.2 cm/s [see section S6 of (37)], 5. STAR Collaboration, Nature 548, 62–65 (2017). 46. E. Kozik, B. Svistunov, Phys. Rev. Lett. 100, 195302 (2008).
three orders of magnitude lower than previ- 6. P. W. Anderson, N. Itoh, Nature 256, 25–27 (1975). 47. L. Yu, Phys. Rev. B 25, 198–215 (1982).
ous experiments (42). As such, vortex-vortex 7. P. K. Kovtun, D. T. Son, A. O. Starinets, Phys. Rev. Lett. 94, 48. G. Agnolet, D. F. McQueeney, J. D. Reppy, Phys. Rev. B 39,
interactions dominate even for the smallest
possible clusters containing only two vorti- 111601 (2005). 8934–8958 (1989).
ces. Furthermore, a four-order-of-magnitude 49. S. Serafini et al., Phys. Rev. X 7, 021031 (2017).
enhancement in the interaction strength be- 8. P. W. Adams, W. I. Glaberson, Phys. Rev. B 35, 4633–4652 50. Y. P. Sachkou et al., Experimental and simulation datasets
tween vortices and third-sound compared with (1987).
earlier experiments (39, 42) allows resolution for: Coherent vortex dynamics in a strongly-interacting
approaching the single-vortex level. Together, 9. J. M. Kosterlitz, D. J. Thouless, J. Phys. Chem. 5, L124–L126 superfluid on a silicon chip, Version 1.0, Zenodo (2019);
these capabilities provide a tool for future (1972). https://doi.org/10.5281/zenodo.3524827.
study of the rich dynamics of quasi–two- 51. Y. P. Sachkou et al., Data processing and simulation scripts
dimensional vortices in strongly interacting 10. D. J. Bishop, J. D. Reppy, Phys. Rev. Lett. 40, 1727–1730 for: Coherent vortex dynamics in a strongly-interacting
superfluids on a silicon chip. Vortex dynamics (1978). superfluid on a silicon chip, Version 1.0, Zenodo (2019);
in the regime of BKT superfluidity may also be https://doi.org/10.5281/zenodo.3524872.
accessible by operating with thinner films. 11. T. W. Neely et al., Phys. Rev. Lett. 111, 235301 (2013).
12. P. Wiegmann, A. G. Abanov, Phys. Rev. Lett. 113, 034501 ACKNOWLEDGMENTS
The ability to nondestructively track vortex
dynamics in a single shot opens the prospect (2014). We thank X. Yu, M. Cawte, Y. Sfendla, A. Sadawsky, A. Doherty,
to explore out-of-equilibrium dynamics and 13. D. V. Freilich, D. M. Bianchi, A. M. Kaufman, T. K. Langin, P. Warszawski, and M. Woolley for insightful discussions. Device
stochastic noise-driven processes that are fabrication was performed in part at the Queensland node
challenging to study with other techniques D. S. Hall, Science 329, 1182–1185 (2010). of the Australian National Fabrication Facility. Funding: This work
(16, 49). It also promises to resolve conten- 14. S. Donadello et al., Phys. Rev. Lett. 113, 065302 (2014). was funded by the U.S. Army Research Office through grant no.
tious aspects of vortex dynamics in strongly 15. A. Amo et al., Science 332, 1167–1170 (2011). W911NF17-1-0310. It was also partially supported by the Australian
interacting superfluids, such as dissipation 16. E. Estrecho et al., Nat. Commun. 9, 2944 (2018). Research Council Centre of Excellence for Engineered Quantum
and diffusion models (8, 30), vortex inertia 17. G. Gauthier et al., Science 364, 1264–1267 (2019). Systems (EQUS, project no. CE170100009) and the Australian
(27, 28), and the Iordanskii force (29). Fur- 18. S. P. Johnstone et al., Science 364, 1267–1271 (2019). Research Council Centre of Excellence in Future Low-Energy
thermore, whereas the experiments reported 19. L. Onsager, Nuovo Cim. 6 (S2), 279–287 (1949). Electronics Technologies (FLEET, project no. CE170100039).
here were performed with a relatively small 20. P. Makotyn, C. E. Klauss, D. L. Goldberger, E. A. Cornell, W.P.B. acknowledges Australian Research Council Future
number of vortices, the Angström scale of the Fellowship FT140100650. C.G.B. acknowledges a Fellowship from
vortex core in helium-4 will enable future D. S. Jin, Nat. Phys. 10, 116–119 (2014). the University of Queensland (UQFEL1833877) and a Discovery
research on the dynamics of ensembles of 21. M. W. Zwierlein, J. R. Abo-Shaeer, A. Schirotzek, C. H. Schunck, Early Career Researcher Award from the Australian Research
thousands of vortices, a regime well outside Council (DE190100318). Author contributions: Y.P.S., C.G.B., G.I.H.,
current capabilities with cold atom and exciton- W. Ketterle, Nature 435, 1047–1051 (2005). D.L.M., and X.H. performed the experiments. G.I.H., D.L.M., and
polariton superfluids (38). This could allow 22. C. Eigen et al., Nature 563, 221–224 (2018). W.P.B. designed the experiments. Y.P.S., C.G.B., S.F., O.R.S.,
emergent phenomena in two-dimensional 23. B. Ko, J. W. Park, Y. Shin, Nat. Phys. (2019). M.T.R., M.J.D., and W.P.B. analyzed and interpreted the data. O.R.S.,
turbulence, such as turbulent cascades (38) 24. E. Fonda, D. P. Meichle, N. T. Ouellette, S. Hormoz, M.T.R., M.J.D., and A.S.B. performed the point vortex simulations.
and anomalous hydrodynamics (12), to be Y.P.S., C.G.B., G.I.H., and W.P.B. wrote the paper with contributions
explored. D. P. Lathrop, Proc. Natl. Acad. Sci. U.S.A. 111 (Suppl 1), from the other authors. W.P.B. provided overall leadership for
4707–4710 (2014). the project. Competing interests: The authors declare no
25. H. Hu, B. C. Mulkerin, U. Toniolo, L. He, X.-J. Liu, Phys. Rev. competing interests. Data and materials availability: All data
Lett. 122, 070401 (2019). presented in this work and data processing and simulation scripts
26. F. Ancilotto et al., Int. Rev. Phys. Chem. 36, 621–707 (2017). can be found at Zenodo (50, 51).
27. D. J. Thouless, J. R. Anglin, Phys. Rev. Lett. 99, 105301 (2007).
28. T. Simula, Phys. Rev. A 97, 023609 (2018). SUPPLEMENTARY MATERIALS
29. E. B. Sonin, Phys. Rev. B 55, 485–501 (1997).
30. L. Thompson, P. C. E. Stamp, Phys. Rev. Lett. 108, 184501 science.sciencemag.org/content/366/6472/1480/suppl/DC1
(2012). Materials and Methods
Supplementary Text
31. G. I. Harris et al., Nat. Phys. 12, 788–793 (2016). Figs. S1 to S16
32. A. Kashkanova et al., Nat. Phys. 13, 74–79 (2016). Table S1
33. A. Noury et al., Phys. Rev. Lett. 122, 165301 (2019). References (52–80)
34. K. A. Gillis, S. M. Volz, J. M. Mochel, Phys. Rev. B 40,
6 February 2019; accepted 6 November 2019
6684–6694 (1989). 10.1126/science.aaw9229
35. D. K. Armani, T. J. Kippenberg, S. M. Spillane, K. J. Vahala,

Nature 421, 925–928 (2003).
36. D. L. McAuslan et al., Phys. Rev. X 6, 021012 (2016).
37. Materials and methods are available as supplementary

materials.

38. M. T. Reeves, T. P. Billam, X. Yu, A. S. Bradley, Phys. Rev. Lett.
119, 184502 (2017).

39. F. M. Ellis, H. Luo, Phys. Rev. B 39, 2703–2706 (1989).

Sachkou et al., Science 366, 1480–1485 (2019) 20 December 2019 6 of 6

RESEARCH

NEURODEVELOPMENT the number of morphological synapses (Fig. 1,
C and D). This lack of morphological alter-
Potassium channel dysfunction in human neuronal ations in KO neurons prompted us to investi-
models of Angelman syndrome gate functional changes using whole-cell patch
clamp recordings. By injecting stepped currents
Alfred Xuyang Sun1,2*†, Qiang Yuan3,4*, Masahiro Fukuda4, Weonjin Yu4, Haidun Yan5, into neurons to trigger action potentials (APs),
Grace Gui Yin Lim1, Mui Hoon Nai6, Giuseppe Alessandro D’Agostino7, Hoang-Dai Tran2, we observed that currents higher than 300 pA
Yoko Itahana8, Danlei Wang1‡, Hidayat Lokman4, Koji Itahana8, Stephanie Wai Lin Lim4, resulted in increased AP firing frequencies
Jiong Tang9, Ya Yin Chang1, Menglan Zhang4, Stuart A. Cook7, Owen J. L. Rackham7, in KO neurons (Fig. 1, E to G). There were no
Chwee Teck Lim6, Eng King Tan1,4, Huck Hui Ng2, Kah Leong Lim1,10,11, differences in membrane capacitance, input
Yong-Hui Jiang5§, Hyunsoo Shawn Je4,10† resistance, AP threshold, and AP amplitudes
between WT and KO neurons (fig. S2, B to E).
Disruptions in the ubiquitin protein ligase E3A (UBE3A) gene cause Angelman syndrome (AS). Whereas Therefore, we monitored changes in voltage-
AS model mice have associated synaptic dysfunction and altered plasticity with abnormal behavior, gated sodium (V-Na+) and potassium (V-K+)
whether similar or other mechanisms contribute to network hyperactivity and epilepsy susceptibility channels, which are essential for AP genera-
in AS patients remains unclear. Using human neurons and brain organoids, we demonstrate that tion (13). Current–voltage curves induced by
UBE3A suppresses neuronal hyperexcitability via ubiquitin-mediated degradation of calcium- stepped depolarization were indistinguishable
and voltage-dependent big potassium (BK) channels. We provide evidence that augmented BK between WT and KO neurons (fig. S2F). How-
channel activity manifests as increased intrinsic excitability in individual neurons and subsequent ever, when we examined the fast components
network synchronization. BK antagonists normalized neuronal excitability in both human and mouse of after-hyperpolarization (fAHPs) (fig. S2G),
neurons and ameliorated seizure susceptibility in an AS mouse model. Our findings suggest that which can affect neuronal excitability in a
BK channelopathy underlies epilepsy in AS and support the use of human cells to model human bidirectional manner (14–16), we observed that
developmental diseases. fAHPs were significantly augmented in UBE3A
KO neurons [WT: −8.6 ± 1.5 mV (mean ± SEM);
A ngelman syndrome (AS) is a neuro- using neurons differentiated from AS patient– KO: −14.8 ± 1.1 mV; P = 0.0014] (Fig. 1, H and I).
developmental disorder characterized derived induced pluripotent stem cells (iPSCs) The changes in AP firing and fAHP observed
have suggested disrupted maturation (9, 10), in KO neurons were restored by the ectopic
by delayed development, intellectual dis- but neither the pathological mechanisms un- expression of WT UBE3A (fig. S2, H to L), con-
derlying seizures in AS nor the biological sub- firming that the lack of UBE3A was responsi-
ability, and seizures (1). Approximately strate(s) of UBE3A has been established. ble for these functional changes. To verify that
90% of AS cases are caused by a loss-of- these differences are not specific to this parti-
Altered excitability and fAHP in UBE3A-null cular hESC line (H9), we generated independ-
function mutation of the ubiquitin protein human neurons ent pairs of isogenic UBE3A WT and KO clones
in another hESC line (H1) (fig. S2M) and were
ligase E3A (UBE3A) gene, which encodes a To understand the cellular and functional able to reproduce similar functional changes
HECT E3 ubiquitin ligase (2). Loss of the UBE3A consequences of the loss of UBE3A in human (fig. S2, N to S). In humans, AS is associated
protein could result in accumulation of AS- neurons, we used the CRISPR-Cas9 system with loss of the maternal UBE3A allele, and
to generate UBE3A knockout (KO) cells in the this could not be recapitulated in our homo-
relevant substrate proteins and contribute to human embryonic stem cell line (hESC) H9 zygous KO clones. Thus, we sought to determine
(Fig. 1A). One clone (KO) harbored a homozy- whether neurons from AS patient–derived iPSCs
disease pathogenesis (3). Studies using an AS gous 5-bp deletion in exon 6 that resulted in a exhibit a similar increase in excitability with
mouse model have demonstrated impaired frame shift and early translational termination augmented fAHP. To this end, we generated
(Fig. 1A). This clone lacked the UBE3A protein neurons from an AS iPSC line that carried a
synaptic connectivity, an imbalance between (Fig. 1A). We found no off-target effects from microdeletion in chromosome 15q11.2-q13.1
CRISPR-Cas9 on the top 10 predicted off-target (which includes the UBE3A gene) and confirmed
network excitation and inhibition, and delayed sites of our single guide RNA sequence (table that the UBE3A protein was not observed in
S1). UBE3A KO hESCs were karyotypically nor- induced neurons (Fig. 1J). The introduction of
neurodevelopmental processes (4–8). Studies mal (fig. S1A), expressed key pluripotency mark- WT UBE3A into these AS iPSC–derived neu-
ers, and proliferated normally (fig. S1, B and C). rons similarly rescued functional changes in
1National Neuroscience Institute, 11 Jalan Tan Tock Seng, AP firing and fAHPs (Fig. 1, K to N). Taken
Singapore 308433, Singapore. 2Genome Institute of Singapore, Next, we induced neuronal differentiation together, our results established that the loss
60 Biopolis Street, Singapore 138672, Singapore. 3Graduate from these isogenic hESCs by ectopically expres- of UBE3A in human neurons results in altered
School for Integrative Sciences and Engineering, National sing Ngn2 (11). This protocol generates homog- neuronal excitability associated with enhanced
University of Singapore, 28 Medical Drive, Singapore 117456, enous populations of electrically mature cortical fAHP.
Singapore. 4Signature Program in Neuroscience and Behavioral neurons in a time frame shorter than that
Disorders, Duke-NUS Medical School, 8 College Road, achieved with morphogen-guided differentia- BK augmentation underlies increased fAHP
Singapore 169857, Singapore. 5Department of Neurobiology, tion protocols (12). Both UBE3A wild-type (WT)
Duke University School of Medicine, Durham, NC 27710, USA. and KO hESCs were converted to neurons with fAHPs are primarily mediated by calcium- and
6Department of Biomedical Engineering, National University similar efficiency (fig. S2A). At 35 days post- voltage-dependent big potassium (BK) channels
of Singapore, Singapore 117576, Singapore. 7Program in induction, we observed morphologically mature in neurons (17). To determine whether the en-
Cardiovascular and Metabolic Disorders, Duke-NUS Medical neurons with dendritic arborizations decorated hanced fAHPs observed in KO neurons resulted
School, 8 College Road, Singapore 169857, Singapore. with synaptic markers (Fig. 1B). There were no from increased BK channel activity, we pharma-
8Program in Cancer and Stem Cell Biology, Duke-NUS Medical significant differences between WT and KO cologically isolated BK currents (fig. S3A). Com-
School, 8 College Road, 169857, Singapore. 9Singapore neurons in terms of dendritic complexity or pared with WT-derived neurons, UBE3A KO
Bioimaging Consortium, Agency for Science Technology neurons exhibited larger BK currents (Fig. 2A
and Research (A*STAR), Singapore 138667, Singapore.
10Department of Physiology, Yong Loo Lin School of Medicine,
National University of Singapore, Singapore 117597, Singapore.
11Lee Kong Chian School of Medicine, Nanyang Technological
University, 11 Mandalay Road, Singapore 308232, Singapore.
*These authors contributed equally to this work.
†Corresponding author. Email: [email protected] (A.X.S.);
[email protected] (H.S.J.)
‡Present address: Department of Neurobiology and Anatomy,
University of Utah, Salt Lake City, UT 84112, USA. §Present address:
Department of Genetics, Yale University School of Medicine,
New Haven, CT 06520, USA.

Sun et al., Science 366, 1486–1492 (2019) 20 December 2019 1 of 7

RESEARCH | RESEARCH ARTICLE

and fig. S3B). Although a previous report in- confirmed this finding by employing single- the presence of BK channels at that location.
dicated that SK2 (small-conductance Ca2+- molecule, atomic force microscopy (AFM) to The results revealed that the BK channel den-
activated K+ channel 2) is a substrate of UBE3A assess the densities of BK proteins on cellular sity was higher on the membranes of KO cells
(18), we did not observe any significant differ- membranes by analyzing the physical inter- (Fig. 2, C and D). If increased BK levels lead to
ence in SK2-mediated medium AHPs (fig. S3, actions between BK proteins and BK-specific enhanced fAHP and elevated excitability in
antibodies conjugated to the AFM tip (Fig. 2B KO neurons, we hypothesized that treatment
C and D). The higher BK currents observed in and fig. S4) (19). Sequential AFM force mapping with a BK antagonist would normalize these
was conducted by spatially moving the func- differences. Indeed, application of BK antago-
KO neurons might simply be due to increased tionalized tip on the soma membranes of nists (either 5 mM paxilline or 100 nM IBTX)
neurons. Figure 2C shows an example of a (Fig. 2, E to H, and fig. S5, A to C), but not of an
BK protein expression. Indeed, Western blot heatmap in which each colored block indicates SK antagonist, apamin (200 nM) (fig. S5D),

analyses showed that the level of the oblig-
atory subunit of BK (BKa) was higher in KO
neurons (fig. S3, E and F). We independently

A hUBE3A exon 6 exon 14 B Analyses C D

hESCs+Ngn2 Replate 35 20 MAP2 SYNAPSIN
H9 hESC CRISPR KO No. of Intersections WT
WT CACCAGTTAACTGAGGG Day-1 0 5
KO CACCA - - - - - CTGAGGG WT KO
10
WT KO MAP2 SYNAPSIN Synapsin N.S.
WT (25) density 20
UBE3A 100 kDa KO (25) (per 10 µm)
0 10
Actin 42 kDa 0 25 50 75 100
µm from soma 15 18

0
WT KO

Spike frequency (Hz)E WT (21) F G * H I **
KO (25)
40 WT 30 21 25 WT -20
30 20 KO
-65 mV Max. spike freq. (Hz) 10 WT KO 10 mV fAHP amplitude(mV)
2 ms
20 0 * -10
KO fAHP
L 19 20
10 -65 mV M

0 20 mV AS+ 21 25
0 0.1 0.2 0.3 0.1 s UBE3A
Current (nA) AS
0.36 nA 500 ms 0
10 mV WT KO
2 ms
JK N **

-20
MAP2 UBE3A Spike frequency (Hz) 40 Max. spike freq. (Hz) 40 fAHP amplitude(mV)
AS+ UBEA3S+A 30 UBEA3S+A
UBE3A 30 20
AS 10 AS
AS 20 AS+ -10
UBE3A 0
10 AS 19 20

0 0.1 0.2 0.3 0
0 Current (nA)

Fig. 1. Altered functional properties of UBE3A-deficient human neurons. injections in WT and KO induced neurons. (F and G) Representative traces
(A to D) Generation of human induced neurons from WT and UBE3A KO and quantification of maximal spike frequencies by current injection in
hESCs. (A) Schematic illustrating the CRISPR-Cas9-mediated gene editing induced WT and KO neurons. (H and I) Representative traces and
approach used to knock out UBE3A in hESCs (top) and immunoblot showing quantification of spike fAHP amplitudes of induced WT and KO neurons.
the absence of UBE3A protein in UBE3A KO hESCs (bottom). (B) Schematic (J to N) Induced neurons derived from AS iPSCs showed reproduced
illustrating the protocol used to generate the human neurons used in this excitability and fAHP changes. (J) Immunostaining for UBE3A in AS neurons,
study (top) and representative morphological images of day-35 WT and with and without ectopic expression of UBE3A. Scale bar, 10 mm. (K) F-I curves
KO neurons immunostained for MAP2 (blue) and synapsin (magenta) showing spike frequencies versus current injection in induced neurons derived
(bottom). Scale bar, 20 mm. (C) Sholl analysis of WT and KO induced neurons from AS iPSCs. (L) Quantification of maximal spike frequencies in the current
at day 35. (D) Representative images and quantification of synaptic injections. (M and N) Representative traces and amplitude quantification of
puncta densities, calculated from neurons immunostained for synapsin. Scale spike fAHP. Data represent means ± SEM. The two-tailed unpaired Student’s
bar, 1 mm. (E to I) Altered excitability in KO neurons (H9 derived). (E) t test was used for all analyses. The numerals in all bars indicate the number of
Frequency–current (F-I) curves showing spike frequency versus current analyzed neurons. **P < 0.01; *P < 0.05; N.S., not significant.

Sun et al., Science 366, 1486–1492 (2019) 20 December 2019 2 of 7

RESEARCH | RESEARCH ARTICLE

Fig. 2. UBE3A deletions A BC WT D WT Norm. BK density (a.u.) *
increase BK channel KO
function in human BK current AFM scanning 80pN 200 2

neurons. (A) Representative 0.5 nA Cantilever No. of BK binding events 150
traces and quantification 0.1 s 1
of BK currents isolated BK current 0
from WT and KO neurons density (pA/pF) 15 100 23 21
treated with paxilline y y KO
(5 mM). (B) Diagrams 10 x 0
illustrating the detection x BK probe 50 WT KO
of BK channels using a 5
functionalized probe y 0
with AFM. (C) Representative 12 10 x 0 10 20 30 40 50
heatmaps of specific Unbinding force (pN)
0
WT KO

BK probe binding events. E DMSO DMSO F G WT+DMSO H
Force–distance curves KO+DMSO
(10 data points by 10 data WT -20 40 WT+Paxilline *
-65 mV KO+Paxilline
points) were obtained * 50
over 1-mm2 areas. Colors KO 40 N.S.
fAHP amplitude(mV) Spike frequency (Hz) Max. spike freq. (Hz)
indicate the measured -65 mV N.S. 30 30
force of specific binding Paxilline Paxilline -10

events. (D) Unbinding WT 20 20
force distribution and BK
channel density on the -67 mV 10 10 20 17 16 16
surface of WT and KO
neurons. (E to H) Pharma- KO 20 17 16 16 0 0
cological rescue by the 0 0.1 0.2 0.3 DMSOPaxilline
BK antagonist paxilline. -64 mV 20 mV 0 Current (nA)
0.1 s DMSOPaxilline
20 mV

0.36nA 500 ms10 ms

(E and F) Representative

traces and quantification of fAHP with and without paxilline (5 mM). DMSO, dimethyl sulfoxide. (G and H) F-I curves showing spike frequencies versus

current injections and related quantification in induced neurons. Data represent means ± SEM. In all bars, the values indicate the number of analyzed neurons.

The two-tailed unpaired Student’s t test was used. *P < 0.05; N.S., not significant.

normalized the differences in fAHP amplitude responding decline in the BK protein level, and slight perturbation in the transcriptome over-
and AP firing frequency between KO and WT this decrease in the BK level was abolished by all: in total, 129 differentially expressed genes
neurons, demonstrating that BK channels me- the proteasome inhibitor MG312 (fig. S6, F to (adjusted P < 0.05) (figs. S7D and S8A and
diate the augmented fAHP observed in KO H). Collectively, these data demonstrate that table S2) were identified, and all of these genes
neurons. BK is one of the targeted substrates of UBE3A- showed minor changes, as indicated by the
mediated ubiquitination and proteasomal de- small range of log2(fold change) values. Con-
UBE3A mediates ubiquitination and gradation. A small subset of people with AS sistent with our quantitative PCR results, we
proteasomal degradation of BK do not carry genetic deletions in UBE3A but found no significant differences in the expres-
instead possess missense mutations in UBE3A sion levels of BK channel genes (fig. S8B and
Next, we sought to determine how the loss of (2). Ectopically expressing AS-associated UBE3A table S2). Taken together, our results support
UBE3A led to BK augmentation. As UBE3A is missense mutants (Glu550→Leu, Leu502→Pro) the notion that, in human neurons, UBE3A pri-
an E3 ligase, we hypothesized that BK is one (20), but not an autism-associated mutant marily regulates BK levels via a posttranslational
of the substrates of UBE3A-mediated ubiquiti- (Thr485→Ala) (21), failed to down-regulate BK mechanism.
nation and proteasomal degradation (3). To protein levels (fig. S6I), indicating that in-
test this hypothesis, we first performed in vitro creased neuronal BK levels may also be a A BK antagonist restores neuronal excitability
coimmunoprecipitation (IP) in heterologous pathophysiological phenotype in AS patients and network activity in 3D organoids
cells and confirmed that there is an interaction with missense mutations in UBE3A. Although
between UBE3A and BK (fig. S6A). Second, our data thus far suggest that UBE3A affects One advantage of using our single-step neu-
when FLAG-tagged BK was coexpressed with BK protein abundance via its E3 ligase activity, ronal induction protocol instead of morphogen-
hemagglutinin (HA)–tagged ubiquitin, we ob- UBE3A could regulate BK levels via transcrip- guided differentiation protocols is that our
served more BK ubiquitination when WT UBE3A tional modulation (22). However, quantitative protocol can produce electrically mature neu-
was present than when a catalytically dead real-time reverse transcription polymerase chain rons (11). However, neurons obtained using this
mutant (MT; C885A) was used (fig. S6B). BK reaction (PCR) analysis revealed that there protocol do not follow a normal developmental
ubiquitination by UBE3A was further con- was no significant difference in BKa subunit trajectory, in that they bypass the progenitor
firmed in an in vitro ubiquitination assay using expression between WT and KO neurons (fig. stage, raising the question of whether these
recombinant proteins (fig. S6C) and by assess- S7A). To identify which transcripts are altered functional deficits are relevant to AS disease
ing endogenous BK ubiquitination using mouse by UBE3A deficiency at the genome-wide level, progression. To address this issue, we gener-
cortical neurons derived from either WT or we performed mRNA sequencing on WT and ated three-dimensional (3D) cortical organoids
Ube3am-/p+ mice (fig. S6, D and E). Further- KO neurons (fig. S7, B to D, and tables S2 to and used this system to investigate functional
more, coexpressing BK cDNA with increasing S4). Differential expression analysis showed a changes in UBE3A-deficient neurons that
amounts of UBE3A cDNA resulted in a cor- follow a normal developmental maturation

Sun et al., Science 366, 1486–1492 (2019) 20 December 2019 3 of 7

RESEARCH | RESEARCH ARTICLE

A BRN2 CUX1 B WT C WT D -20 ***
KO
Brightfield CTIP2 30 40 ** -15
WT -10
30
-66 20 20 -5 26 15
mV KO 10 26 15 0

Spike frequency (Hz) 0 WT KO
WT KO
10 Max. spike freq. (Hz)
fAHP amplitude (mV)
-68 0
mV WT+Paxilline 0 10 20 30

KO -65 30 Paxilline Paxilline Paxilline
KO mV KO+Paxilline 40 -20
20
-65 20 -15
mV 20 mV 10
11 11 -10
0.1 s 0 -5 11 11
0 10 20 30 0
30 pA 500 ms WT KO 0
Current injection (pA) WT KO

E WT KO F GH ***
WT
Cell 1 1 0.6 ** 2
Cell 2 Cell 1 Calcium amp.(ΔF/F)
Cell 3 Cell 2 Cumulative IEI 0.5 Calcium transient *** 1 Ca. amp.(ΔF/F) Ca. transient
Cell 4 Cell 3 freq. (x10-2 Hz)
Cell 5 Cell 4 0 0.4 freq. (x10-2 Hz)0
Cell 6 Cell 5 0 20 4 Base.
Cell 7 Paxilline***
Cell 8 Cell 6 KO
Cell 9 Cell 7 0.6
Cell 10 Cell 8 10 s
Cell 9 +Paxilline 2 0.2 0.4
I Cell 10 0.2
10 s
Cell 1 0 WT KO 0
Cell 2
Cell 3 1ΔF/F 40 60 80 0
Cell 4 IEI (s) WT KO

Cell 1 SI:0.12 20s SI:0.57 K WT KO
Cell 2 1
Cell 3 Cell # J Cell # 0.6 ***
Cell 4
1ΔF/F Cell 1 1ΔF/F Sync.Ind.0.4
Cell 2 0.2
10 s SI:0.09 Cell 3 SI:0.22 ***
+Paxilline Cell 4 0 0

Cell # Cell 1 0.5ΔF/F Cell # Cell #
Cell 2
0.5ΔF/F Cell 3 Base.
Cell 4 Paxilline

Cell #
10 s

Fig. 3. Altered functional properties of neurons and enhanced network respectively, for WT and KO organoids. (G) Quantification of the amplitudes of
activity in UBE3A-deficient organoids. (A) Bright-field images of organoids calcium transients recorded in WT and KO organoids. (H) Quantification of the
and immunostaining of cortical layer markers in WT and KO organoids at frequencies and amplitudes of calcium transients before and after paxilline
day 120. Scale bars, 5 mm (bright-field images) or 50 mm (immunostaining). treatment (10 mM). N = 10 and 11 organoids for WT and KO, respectively. (I and J)
(B to D) Altered electrophysiological properties in neurons from KO organoids, as Representative traces of calcium transients in individual neurons (left) and
illustrated with representative traces (B), F-I curves evoked by current injection correlation heatmaps (right) obtained from WT (I) and KO (J) organoids. SI,
(C), and quantifications (D), with and without paxilline (10 mM). a.u., arbitrary synchronization index. (K) Summary of the synchronization index recorded in WT
units. (E to K) Two-photon (2P) live calcium imaging of WT and KO organoids. and KO organoids upon paxilline treatment (10 mM). N = 10 and 11 organoids for
(E) Calcium transient traces extracted from individual neurons of WT and KO WT and KO, respectively. Data represent means ± SEM. Numerals in bars indicate
organoids. (F) Cumulative distribution of interevent intervals (IEI) in time bins for the number of analyzed neurons. The two-tailed paired Student’s t test was used
calcium transients recorded in WT and KO organoids. The inserted bar graph for analysis of data in (H) and (K); the two-tailed unpaired Student’s t test was used
shows the quantification. n = 131 and 236 neurons from N = 12 and 17 organoids, for analysis of data in all other panels. **P < 0.01; ***P < 0.001.

sequence (23). These brain organoids have uishable in size. They were similarly composed tinued to grow over the next 100 days to ~1
been shown to better recapitulate the devel- of polarized, proliferating neuroepithelial cells to 2 mm (Fig. 3A). After 120 days, immuno-
opmental timelines of human brains (24). After that expressed canonical neuronal progenitor staining revealed the presence of a substan-
20 days of differentiation, both WT and KO markers, including PAX6, SOX2, and NESTIN tial number of NeuN-positive (NeuN+) neurons.
cells produced spheroids that were indisting- (fig. S9, A and B). These brain organoids con- Consistent with previous reports, multiple

Sun et al., Science 366, 1486–1492 (2019) 20 December 2019 4 of 7

RESEARCH | RESEARCH ARTICLE A Flurothyl induced seizure B Picrotoxin induced seizure

Fig. 4. BK modulation ameliorates seizure 150 ** N.S. 6 N.S.
susceptibility in a mouse model of AS.
(A) Latency to myoclonic seizure induced by Normalized latency to **
flurothyl in WT and KO (Ube3am−/p+) mice. myoclonic seizure (100%)
WT without paxilline, N = 23; KO without 5
paxilline, N = 17; WT with paxilline, N = 16;
KO with paxilline, N = 12. (B) Epilepsy grade 100 4 Epilepsy grade
induced by picrotoxin in WT and KO mice.
WT without paxilline, N = 16; KO without 3
paxilline, N = 18; WT with paxilline, N = 16;
KO with paxilline, N = 16. (C) Samples 50 2
of spectrogram and averaged delta power
data for LFP from BIC of WT and KO mice with 1
sound stimuli. Delta rhythmicity below the
dashed line was averaged for analysis. 0 WT KO 0
(D) Summary of power spectral density WT KO Paxilline WT KO WT KO
(PSD) of LFP recorded from BIC of WT and Paxilline
KO mice with sound stimuli. N = 8 mice
from two batches. Normalization was C 20 WT D E *
performed to remove the baseline difference
due to batch effect. A two-way analysis 15Frequency (Hz) Delta 200
of variance (ANOVA) with the Bonferroni 10 Mean delta power (a.u.)
post hoc test was used for frequencies 250 **
of 1 to 4 Hz to analyze differences 5
between WT and KO mice. (E) Mean Delta
delta (1 to 4 Hz) power of WT and KO mice 0 20 Normalized LFP power (a.u.) 200 100
LFPs. N = 8 mice from two batches. 20 0 150
(F) Sample spectrogram and averaged delta -20 100 88
power of LFPs in BIC of WT and KO mice KO
with paxilline treatment (0.35 mg/kg; Delta (a.u.) 15 WT
four times, at 1-hour intervals). (G) Summary 10 KO 0
of PSD of LFPs in BIC recorded in WT and
KO mice with paxilline treatment. Two-way 5
ANOVA with the Bonferroni post hoc test was 0
used for frequencies of 1 to 4 Hz to analyze
differences between WT and KO mice. 20 Delta 1-4 Hz power
(H) Mean delta power (1 to 4 Hz) of WT and
KO mice LFPs with paxilline treatment. 10 WT WT
N = 8 mice from two batches. Numerals in KO
bars indicate the number of animals 0 KO
analyzed. Data represent means ± SEM. 0 1234 5 50
The two-tailed unpaired Student’s t test was Time (min) 0 5 10 15 20
used to analyze data for all panels except Frequency (Hz)
for (D) and (G). *P < 0.05; **P < 0.01;
N.S., not significant. F Paxilline G Paxilline H

20 250 Paxilline
15 WT
Normalized LFP power (a.u.) 200
10 N.S.
Frequency (Hz) Mean delta power (a.u.)
5
N.S.

0 200

20 20 100
15 KO 0
-20 88
10 150
WT 0
5 5 KO

0 100 WT
KO
Delta (a.u.) 15 Delta 1-4 Hz power
10 W
1234 T
5 Time (min)
0 KO

0 50 5 10 15 20
0 Frequency (Hz)

cortical layer markers were also observed, in- tive of mature astrocytes, were also observed exhibited augmented excitability and elevated
cluding BRN2/CTIP2/CUX1, and these segre- in both WT and KO organoids (fig. S9D). Col- fAHPs (Fig. 3, B to D), whereas other electro-
gated the neurons into upper and deep layers lectively, these data demonstrated that both the physiological properties remained the same,
(Fig. 3A). Quantification analyses showed that WT and KO organoids contain similar com- indicating that the overall developmental matu-
there was no significant difference in the positions of mature cortical cell types when ration is similar between WT and KO organ-
number of NeuN+ neurons or the expression grown in long-term cultures. oids (fig. S10, A to C). Immunoblots of BK
of the cortical layer markers between WT and proteins confirmed that BK levels were higher
KO organoids (fig. S9C). We detected a few Next, we performed whole-cell patch-clamp in KO organoids (fig. S10D). Paxilline treat-
g-aminobutyric acid–positive (GABA+) neu- recordings to evaluate the intrinsic properties ment normalized the augmented neuronal
rons, indicating the presence of inhibitory of neurons in WT and KO organoids after 120 excitability and fAHP changes observed in
neurons, within the organoids. Glial fibrillary to 150 days in culture. Similar to the results neurons of KO organoids (Fig. 3, B to D). We
acidic protein–positive cells, which are indica- obtained in 2D-induced neurons, pyramidal- obtained similar results with cortical organoids
shaped neurons within UBE3A KO organoids

Sun et al., Science 366, 1486–1492 (2019) 20 December 2019 5 of 7

RESEARCH | RESEARCH ARTICLE

generated from AS iPSCs (fig. S10, E to G). ed BK channel activities in neurons derived (Fig. 4, F to H). Taken together, we conclude
Thus, our recordings performed in long-term from both isogenic UBE3A KO hESCs and AS that BK augmentation similar to that observed
cultured 3D human cortical organoids repro- iPSCs. Because none of these phenotypes in human UBE3A KO neurons is observed in
duced the phenotypes observed in 2D human neurons of Ube3am−/p+ mice, and paxilline
neurons. had been previously reported in AS mouse treatment ameliorates the seizure threshold
as well as the high delta oscillation observed
Because 3D organoids are capable of as- models, we sought to determine whether in these mice.
sembling more sophisticated neuronal networks
than can be achieved in 2D neuronal cul- they are also present in neurons obtained Outlook
tures, we sought to use calcium (Ca2+) imaging from mice with a maternal Ube3a deletion
to monitor differences in neuronal network (Ube3am−/p+) (fig. S11A), which has been shown By characterizing the functional properties of
dynamics in large populations of cells in WT to recapitulate many of the symptoms asso- 2D human neurons and 3D cortical organoids
and KO cortical organoids. Incubating cortical ciated with AS in humans (29). At postnatal derived from UBE3A KO hESCs and AS iPSCs,
organoids with Fluo-4AM (2 mM) resulted in day 28, the BK currents were significantly we demonstrated an evolutionally conserved
labeled cells located up to 170 mm from the channelopathy that contributes to network
surface (movie S1). In WT organoids, bath ap- larger in hippocampal neurons obtained from dysfunction and hyperactivity in AS. Our re-
plication of TTX (2 mM) abolished spontaneous Ube3am−/p+ mice than in those obtained from sults suggest that BK channels are substrates
calcium transients (fig. S10H and movie S2). WT mice (fig. S11B). Similar to human neu- for UBE3A and that BK might thus be a
Whereas neurons in day-120 to day-150 WT therapeutic target for treatment of patients
organoids displayed prominent spontaneous rons, changes in excitability and comparatively with AS.
Ca2+ transients, at the network level synchro-
nization of transients was rare, and most neu- higher fAHP were observed in neurons ob- REFERENCES AND NOTES
rons fired in a stochastic manner (Fig. 3E and tained from Ube3am−/p+ mice (fig. S11, C and D)
movie S3) (25). In contrast, neurons in UBE3A (fAHP; WT: −4.4 ± 1.4 mV, n = 26; Ube3am−/p+: 1. K. Buiting, C. Williams, B. Horsthemke, Nat. Rev. Neurol. 12,
KO organoids exhibited more frequent firing −6.4 ± 1.2 mV, n = 29; p < 0.001). Paxilline 584–593 (2016).
with augmented dynamic ranges of Fluo-4 normalized these changes (fig. S11, C and D).
fluorescence, and neurons in some KO organ- 2. A. M. Mabb, M. C. Judson, M. J. Zylka, B. D. Philpot, Trends
oids seemed to discharge regularly and synchro- These results indicate the presence of evolu- Neurosci. 34, 293–303 (2011).
nously (Fig. 3, E to G, and movie S3). Soma
calcium transients were correlated with intrin- tionally conserved BK augmentation in mouse 3. G. L. Sell, S. S. Margolis, Front. Neurosci. 9, 322 (2015).
sic neuronal activity, as indicated by the finding 4. Y. H. Jiang et al., Neuron 21, 799–811 (1998).
that the intensity of somatic Fluo-4 fluores- neurons that lack Ube3a, and they prompted 5. M. L. Wallace, A. C. Burette, R. J. Weinberg, B. D. Philpot,
cence gradually increased along with graded
electrical train stimulations in individual neu- us to test whether the enhanced seizure sus- Neuron 74, 793–800 (2012).
rons (fig. S10I), demonstrating that regular ceptibility observed in Ube3am−/p+ mice (7) 6. M. L. Wallace, G. M. van Woerden, Y. Elgersma, S. L. Smith,
fluorescence fluctuations were related to intrin- could be normalized upon BK antagonist treat-
sic neuronal bursting, as shown in previous B. D. Philpot, J. Neurophysiol. 118, 634–646 (2017).
reports (26, 27). In addition, perforated patch ment. First, we exposed mice to the putative 7. M. C. Judson et al., Neuron 90, 56–69 (2016).
recordings revealed that neurons exhibit spon- 8. M. C. Judson et al., J. Neurosci. 37, 7347–7361 (2017).
taneous burst firing in UBE3A KO organoids GABA type A receptor (GABAAR) antagonist 9. S. J. Chamberlain et al., Proc. Natl. Acad. Sci. U.S.A. 107,
(fig. S10J). Paxilline treatment completely con- flurothyl (100 ml/min), and we observed that
verted spontaneous burst firing into single AP Ube3am−/p+ mice exhibited a significant reduc- 17668–17673 (2010).
firing in these neurons (fig. S10, J and K) and tion in the latency to myoclonic seizure, and 10. J. J. Fink et al., Nat. Commun. 8, 15038 (2017).
decreased both the frequencies and the am- 11. Y. Zhang et al., Neuron 78, 785–798 (2013).
plitudes of calcium transients in the organoids this effect could be normalized by paxilline 12. F. Yi et al., Science 352, aaf2669 (2016).
(Fig. 3H), suggesting that augmented BK ac- 13. B. P. Bean, Nat. Rev. Neurosci. 8, 451–465 (2007).
tivity promoted burst firing in individual neu- pretreatment [0.35 mg/kg, intraperitoneal (i.p.) 14. N. Gu, K. Vervaeke, J. F. Storm, J. Physiol. 580, 859–882 (2007).
rons and subsequent hyperactivity in the entire 15. J. R. Montgomery, A. L. Meredith, Proc. Natl. Acad. Sci. U.S.A.
network. Finally, we investigated network-level injection] (Fig. 4A). Similarly, another GABAAR
synchronization in cortical organoids. We found antagonist, picrotoxin (3 mg/kg, i.p.), induced 109, 18997–19002 (2012).
that KO organoids and AS iPSC–derived or- 16. T. Kimm, Z. M. Khaliq, B. P. Bean, J. Neurosci. 35,
ganoids exhibited higher levels of synchroni- a grade 1 to 2 seizure phenotype based on
zation and that this effect was abolished by Racine’s scale (30) in WT mice, whereas the 16404–16417 (2015).
treatment with paxilline (Fig. 3, I to K, and same treatment induced a much more severe 17. J. F. Storm, J. Physiol. 385, 733–759 (1987).
movies S3 and S4). Taken together, our data seizure phenotype (grade 5) in Ube3am−/p+ 18. J. Sun et al., Cell Rep. 12, 449–461 (2015).
suggest that aberrant intrinsic firing of UBE3A- mice (Fig. 4B). Paxilline pretreatment was able 19. J. L. Maciaszek, H. Soh, R. S. Walikonis, A. V. Tzingounis,
null neurons collectively elicits network- to normalize the seizure grade of Ube3am−/p+
level hyperactivity and synchronization that mice to a level comparable to that observed G. Lykotrafitis, J. Neurosci. 32, 11435–11440 (2012).
resemble the seizure activity observed in AS 20. B. Sadikovic et al., Hum. Mutat. 35, 1407–1417 (2014).
patients (28). in WT mice (Fig. 4B). Next, to investigate the 21. J. J. Yi et al., J. Biol. Chem. 292, 12503–12515 (2017).
22. S. Ramamoorthy, Z. Nawaz, Nucl. Recept. Signal. 6, e006 (2008).
A BK antagonist ameliorates epileptic AS-like electroencephalogram (EEG) abnor- 23. A. M. Paşca et al., Nat. Methods 12, 671–678 (2015).
susceptibility in a mouse model of AS 24. I. Kelava, M. A. Lancaster, Cell Stem Cell 18, 736–748 (2016).
malities observed in these mice, we recorded 25. P. Kirwan et al., Development 142, 3178–3187 (2015).
Our data thus far demonstrate that changes in 26. J. N. Kerr, D. Greenberg, F. Helmchen, Proc. Natl. Acad. Sci. U.S.A.
intrinsic excitability are mediated by augment- resting-state local field potentials (LFPs), which
102, 14063–14068 (2005).
are analogous to intracortical EEG changes, in 27. V. Emiliani, A. E. Cohen, K. Deisseroth, M. Häusser, J. Neurosci.
freely moving mice (fig. S12A) (31). As previ-
ously reported, compared to the WT mice, 35, 13917–13926 (2015).
Ube3am−/p+ mice exhibited a strong trend 28. K. Pelc, S. G. Boyd, G. Cheron, B. Dan, Seizure 17, 211–217 (2008).
toward total spectral power being increased in 29. Y. Jiang, T. F. Tsai, J. Bressler, A. L. Beaudet, Curr. Opin. Genet. Dev.

the delta rhythmicity (1 to 4 Hz) (Fig. 4, C to E, 8, 334–342 (1998).
P = 0.0044) (31). We also observed increased 30. R. J. Racine, Electroencephalogr. Clin. Neurophysiol. 32,
power in other bands, including theta (4 to
281–294 (1972).
8 Hz), beta (12 to 30 Hz), and gamma (30 to 31. M. S. Sidorov et al., J. Neurodev. Disord. 9, 17 (2017).
80 Hz) bands, in Ube3am−/p+ mice, but these
differences were not statistically significant ACKNOWLEDGMENTS

(fig. S12B). Because increased delta rhythmic- We thank E. Chua and N. Meenubharathi for their assistance in cell
culture, molecular cloning, and Western blotting and N. Harmston
ity has been linked to seizure susceptibility of for help with RNA sequencing (RNA-seq) data processing. We
Ube3am−/p+ mice (7), we sought to determine thank C. Yong (ZJU) for sharing the BK plasmid. We thank
whether paxilline treatment could normalize G. Augustine, J. Lee, and W.-K. Ho for valuable discussions. We
thank S.T. Li (SJTU) for help with animal epilepsy experiments. We
the delta rhythmicity. Indeed, paxilline treat- thank the Advanced Bioimaging Core of Duke-NUS/SingHealth
for their technical support with multiphoton microscopy.
ment normalized the enhancement observed Funding: This work was supported by the Singapore Ministry of
in the delta band in Ube3am−/p+ mice (P = 0.85) Education (MOE) Academic Research Fund (MOE2014-T2-2-071
to H.S.J.), a National Medical Research Council Collaborative

Sun et al., Science 366, 1486–1492 (2019) 20 December 2019 6 of 7

RESEARCH | RESEARCH ARTICLE

Research Grant (NMRC/OFIRG/0050/2017 to A.X.S. and H.S.J.), W.Y., organoid electrophysiology. H.Y. and Y.-H.J., mouse SUPPLEMENTARY MATERIALS
a KHOO Bridge funding (Duke-NUS-KBrFA/2017/0004 to H.S.J.), electrophysiology. G.G.Y.L. and K.L.L., ubiquitin assays. M.H.N. and science.sciencemag.org/content/366/6472/1486/suppl/DC1
a Duke-NUS Signature Research Program block grant (to H.S.J. C.T.L., AFM. G.A.D., S.A.C., and O.J.L.R., RNA-seq. H.-D.T., CRISPR- Materials and Methods
and K.I.), a Khoo Postdoctoral Fellowship Award (Duke-NUS-KPFA/ Cas9 genome editing. Y.I. and K.I., co-IP and biochemistry. D.W. Figs. S1 to S13
2016/0007 to M.F.), the National Parkinson’s Disease and H.L., cell culture and gene expression. S.W.L.L., cell culture Tables S1 to S5
Translational Clinical Research Program fund (NMRC/TCR/ and biochemistry. J.T., biochemistry. Y.Y.C and M.Z., imaging. References (32–47)
013-NNI/2014 to E.K.T.), a Singapore Translational Research E.K.T. and H.H.N., research support and data discussion. A.X.S., Movies S1 to S4
Investigator Award (NMRC/STaR/014/2013 to E.K.T.), and an Q.Y., and H.S.J. conceived and designed the study and wrote
NMRC Young Investigator Research Grant (NMRC/OFYIRG/ the manuscript, with input from all members. Competing View/request a protocol for this paper from Bio-protocol.
0022/2016) to O.J.L.R. Author contributions: A.X.S., cell interests: The authors declare no competing interests. Data
culture, biochemistry, and molecular biology experiments. and materials availability: RNA sequencing raw data and counts 26 September 2018; resubmitted 29 September 2019
Q.Y., electrophysiology, biochemistry, and molecular biology were deposited in the Gene Expression Omnibus with accession Accepted 13 November 2019
experiments and behavioral tests. M.F., organoid Ca2+ imaging. number GSE120225. 10.1126/science.aav5386

Sun et al., Science 366, 1486–1492 (2019) 20 December 2019 7 of 7

RESEARCH

◥ ing because the six tensor components must
be calculated by detecting multiple diffraction
REPORT spots per grain. Additionally, stress tensor
measurement in grains with plastic strains is
MATERIALS SCIENCE difficult because small changes in the x-ray
scattering directions caused by shear stress
Intragranular three-dimensional stress tensor fields components become buried in the more dom-
in plastically deformed polycrystals inant changes caused by crystallographic
rotations.
Yujiro Hayashi*, Daigo Setoyama, Yoshiharu Hirose, Tomoyuki Yoshida, Hidehiko Kimura
Obtaining 3D type III stress measurements
The failure of polycrystalline materials used in infrastructure and transportation can be catastrophic. for nondeformed bulk polycrystals requires
Multiscale modeling, which requires multiscale measurements of internal stress fields, is the key to scanning 3DXRD, where the intragranular
predicting the deformation and failure of alloys. We determined the three-dimensional intragranular local lattice parameters in deeply embedded
stress tensor fields in plastically deformed bulk steel using a high-energy x-ray microbeam. We grains are determined from multiple diffrac-
observed intragranular local stresses that deviated greatly from the grain-averaged stresses and tion spots per grain, detected using a focused
exceeded the macroscopic tensile strength. Even under deformation smaller than the uniform monochromatic high-energy x-ray beam (29).
elongation, the intragranular stress fields were in highly triaxial stress states, which cannot be The six stress tensor components in nonde-
determined from the grain-averaged stresses. The ability to determine intragranular stress tensor formed polycrystals with small intragranular
fields can facilitate the understanding and prediction of the deformation and failure of materials misorientations are obtained by determining
through multiscale modeling. six lattice parameters (19). The six stress tensor
components are derived using the differences
P olycrystalline materials such as steel are Generally, 3DXRD allows for grain-resolved in the lattice parameters between strained
ubiquitous in infrastructure and trans- 3D measurements of stress tensors for em- and unstrained grains. However, type III stress
portation industries. The failure of such bedded grains in bulk polycrystals using measurement for plastically deformed bulk
materials caused by the accumulation of synchrotron-based, high-energy x-rays. Grain- alloys is still difficult because the six lattice
deformation can be catastrophic. Plastic averaged stress tensors, called type II stresses, parameters of strained grains cannot be accu-
deformation of single crystals is predictable are obtained by determining the lattice pa- rately determined because of large, intragran-
because the stress tensors acting along slip rameters of individual grains from multiple ular misorientations.
planes can be determined from macroscopic diffraction spots per grain (19). 3DXRD-based
loads. For polycrystalline materials, multiscale type II stress evaluation is a useful method for We overcame this obstacle by determining
modeling is required to predict the deforma- materials science (20, 21), modeling (22, 23), elastic strains directly from the difference in
tion of a huge number of grains and multiscale and engineering (24–26). One example is the lattice spacing between strained and unstrained
microstructures (1). Establishing a multiscale study from Sedmak et al., which determined grains, without needing to accurately determine
model requires nondestructive, multiscale, ob- the macroscopic (type I) stress tensor fields in a the six lattice parameters of strained grains. We
servational data for the three-dimensional (3D) tensioned polycrystalline alloy using the type II determined six stress tensor components from
fields of the stresses and plastic strains in bulk stress tensors they measured to verify martens- the elastic strains in various directions, which
polycrystals and their evolution produced by ite transformation models (20). Their work also we obtained by detecting multiple diffraction
plastic deformation (2). Nondestructive, multi- showed the deviation of type II stresses from spots per grain. Our analysis using the multi-
scale 3D observations of the plastic strains and the type I stress fields. Abdolvand et al. used ple diffraction spots also allows us to measure
microstructures in bulk polycrystals have be- type II stresses that they measured to verify intragranular local stresses. The difference
come possible using a variety of synchrotron- crystal plasticity models and found a valid in the position of a diffraction spot caused by
based x-ray diffraction and imaging techniques, model for predicting type II stresses based on an intragranular local stress is small because
including x-ray computed tomography (CT) (3), the orientations of neighboring grains (23). the focused x-ray beam penetrates the grain.
scanning x-ray diffraction CT (4), scanning However, 3DXRD cannot be used to mea- The sensitivity to the local stress is increased
Laue diffraction microscopy (5), diffraction sure intragranular local stress tensors, called by penetrating the region with the local stress
contrast tomography (DCT) (6), 3D x-ray dif- type III stresses. The measurement of these from various directions and detecting multi-
fraction microscopy (3DXRD) (7–11), scan- stresses, which reveal the stress tensor fields in ple diffraction spots. We also detected non-
ning 3DXRD (12), differential-aperture x-ray individual grains, is the key to verifying state- overlapping diffraction spots from plastically
microscopy (DAXM) (13–15), dark-field x-ray of-the-art multiscale models. deformed, millimeter-sized bulk steel with a
microscopy (DFXM) (16), and Bragg coher- large number of grains using a conical slit.
ent diffraction imaging (17, 18). Nevertheless, Using DAXM allows for the nondestructive Thus, we achieved the 3D visualization of the
advances in 3D multiscale measurements of measurement of type III stresses by scanning a difference between type II and type III stress
the stress tensors in bulk polycrystals are still focused polychromatic x-ray beam and detect- tensor fields (Fig. 1) (30). Although we still
needed because stresses cause only small ing multiple Laue diffraction spots per grain need to overcome some remaining limitations,
changes in the scattering directions of dif- (27). However, DAXM is limited to the mea- such as intragranular averaging and artifacts
fracted x-rays compared with the changes surement of near-surface grains because the for quantitative reconstruction (30), this dem-
caused by plastic strain–induced crystallo- polychromatic x-ray illumination of numerous onstration reveals the existence of type III
graphic rotations. grains causes massive overlap of diffraction stress tensor fields with complex distributions
spots. DFXM imaging of type III elastic strain and type III stresses that deviated greatly from
Toyota Central R&D Laboratories, Nagakute, Aichi 480-1192, fields in embedded single grains in bulk poly- type II stresses. These stresses are important
Japan. crystals can be accomplished by magnifying for materials modeling.
*Corresponding author. Email: [email protected] individual diffraction spots (28). However, de-
termining a type III stress tensor is challeng- We conducted the scanning 3DXRD mea-
surements at beamline BL33XU (31) of SPring-8
for low-carbon steel with a mean grain size of
20 mm, a yield strength of sYS = 230 MPa, a

Hayashi et al., Science 366, 1492–1496 (2019) 20 December 2019 1 of 5

RESEARCH | REPORT

tensile strength of sTS = 330 MPa, a uniform A
elongation of 10.5%, and a total elongation of
DE F
25% (figs. S2 and S3). We illuminated a low-
carbon steel sample with a 1 mm–by–1 mm x-ray B
microbeam. We detected diffracted beams
GH I
through the conical slit as diffraction spots on
C JK L
an area detector (Fig. 1B). We continuously
MN O
acquired diffraction images every 0.6° for 180°
Fig. 1. 3D type II and type III stress tensor fields in plastically deformed bulk steel. (A) In situ stress–
with an exposure time of 38 ms while we ro- strain curve. After 5.1% elongation, the scanning 3DXRD measurement was conducted under an applied
tated the sample around its z axis. We con- tensile load of 290 MPa, indicated by the red arrow and red dot. The tensile direction was parallel to the
ducted this scan at an interval of 1.2 mm in the z direction. Eng., engineering. (B) Schematic diagram of scanning 3DXRD setup with a conical slit. (C) 3D
y direction for 20 min. We repeated the 2D orientation map. Orientation is represented by the inverse pole figure for the tensile direction. (D to I) 3D grain-
scan at an interval of 1.2 mm in the z direction averaged (type II) maps of (D) sx, (E) sy, (F) sz, (G) txy, (H) tyz, and (I) tzx, where sx is a normal stress in
for 12 hours. We mapped the orientations of the x direction and txy is a shear stress on the plane normal to the x axis acting in the y direction. (J to O) 3D
aFe and elastic strains for 373 voxels with a intragranular (type III) fields of (J) sx, (K) sy, (L) sz, (M) txy, (N) tyz, and (O) tzx.
voxel size of 1.2 mm by 1.2 mm by 1.2 mm using
a cluster computer from 3.5 × 105 diffraction AB

images acquired by the 3D scan. This task took Volume fraction

1 week to complete. Volume fraction
The sample was elongated by 5.1% in the z
Fig. 2. Statistical comparison of type II and type III stress tensors. (A) Histograms of the volume
direction (Fig. 1A). After stress relaxation from fractions of elastic strain tensors with a bin width of 1 × 10−4. (B) Histograms of the volume fractions of

302 to 290 MPa, we obtained the 3D maps of type II and type III stress tensors with a bin width of 25 MPa.

the elastic strains. The average of the type III
longitudinal elastic strains, ezz, after the 5.1%
elongation agreed with the macroscopic strain
of 1 × 10−3 at the macroscopic elastic limit. The

averages of the type III transverse elastic
strains, exx and eyy, were −2 × 10−4 and −3 ×
10−4, respectively, and they were consistent

with the Poisson effect. Though the intragran-

ular averaging remains, we observed type III
strains of up to 2 × 10−3 (Fig. 2A). We estimated

the lower limits of the spatial resolutions to

be two and four voxels for the type III long-

itudinal and transverse strains, respectively,
using a reconstruction simulation (30). Similar
high intragranular local elastic strains have
been found using high–angular resolution elec-
tron backscatter diffraction (HR-EBSD) (32).
The HR-EBSD method can detect higher local
elastic strains (33) because of its higher spa-
tial resolution compared with that of scanning

3DXRD. However, HR-EBSD is limited to sur-

face strain measurements.

We converted the elastic strain tensors into

stress tensors using the elastic stiffness con-
stants of aFe and the grain orientations, thereby
obtaining the 3D type III stress tensor fields

(2D, see fig. S40 and movie S3; 3D, see fig. S42

and movie S4). We then compared the type II

and type III stress tensors (Fig. 2B). The type III
longitudinal stresses (sz) reached 500 MPa be-
yond sTS, whereas the type II sz did not exceed
sTS. The type I sz (average for all voxels) was
240 MPa, which approximately agrees with

the applied load. Even though no loads were
applied in the transverse directions (x and
y), we observed type III transverse stresses
(sx and sy) comparable to sTS. These high
type III stresses were not observed before the

5.1% elongation (2D, see fig. S36 and movie S1;

3D, see fig. S38 and movie S2).

In continuum mechanics and failure the-

ories, many stress criteria have been proposed

Hayashi et al., Science 366, 1492–1496 (2019) 20 December 2019 2 of 5

RESEARCH | REPORT

AB for predicting yielding and failure in ductile
alloys. The von Mises (or equivalent) stress, se,
C F GH calculated using six stress tensor components,
is most commonly used to predict yielding. An
D I JK important stress component, or factor, for pre-
dicting ductile failure is hydrostatic (or mean)
E LM stress, sm, or stress triaxiality, sm/se (34). High
sm/se limits plastic deformation and leads to
N QRS the growth and coalescence of voids that orig-
inate at inclusions. We produced maps of se,
O TUV sm, and sm/se in the xy plane from the type III
stress tensors obtained before and after 5.1%
P WX Y elongation (2D, see Fig. 3, figs. S37 and S41,
and movie S5; 3D, see figs. S39 and S43 and
Fig. 3. Type III stress tensor fields obtained before and after plastic deformation. (A) In situ stress– movie S6). Even under the small 5.1% elonga-
strain curve. Orientation maps and stress tensor fields were obtained at S1 and S2 (stress–strain states 1 and tion, the highest type III sm/se was approx-
2). (B) Histograms of the volume fractions of sz with a bin width of 25 MPa at S1 and S2. Histograms imately three times as high as the average
for the other strain and stress components are shown in figs. S44 and S45. 2D maps (z = 4.8 mm) of (0.35) and the macroscopic sm/se (0.33) for
(C) orientation, (D) intragranular misorientation (IGM), and (E) kernel-averaged misorientation (KAM) uniaxial uniform elongation (fig. S45). This
obtained at S1, where KAM was evaluated in individual grains. (F to K) 2D fields of (F) sx, (G) sy, (H) sz, raises the question of how low-carbon steel
(I) txy, (J) tyz, and (K) tzx obtained at S1. (L and M) 2D maps of (L) se and (M) sm obtained at S1, where can endure such high stress triaxiality until
se is the equivalent stress and sm is the mean stress. (N to P) 2D maps of (N) orientation, (O) IGM, the total elongation at fracture. In general,
and (P) KAM obtained at S2. (Q to V) 2D fields of (Q) sx, (R) sy, (S) sz, (T) txy, (U) tyz, and (V) tzx obtained at plastic strain is coupled with stress triaxiality
S2. (W to Y) 2D maps of (W) se, (X) sm, and (Y) sm/se obtained at S2, where sm/se is the stress triaxiality. as a failure criterion. Although plastic strain
cannot be directly derived, the increase in
misorientation implies the production of plas-
tic strain, because plastic deformation in
low-carbon steel is caused by the slipping of
crystal planes, which leads to local crystallo-
graphic rotations. We observed deformation-
induced local misorientations, as shown in
the kernel-averaged misorientation (KAM)
maps obtained before and after the 5.1% elon-
gation (Fig. 3, E and P), where KAM was eval-
uated in individual grains. We found that the
positions of high KAM do not necessarily
agree with those of high stress triaxiality, as
shown in grains G1 and G2 (Fig. 3, P and Y).
We speculate that low-carbon steel can en-
dure high local stress triaxiality under the
5.1% elongation because of the nonuniform
distribution of plastic strain, where the small
local plastic strain in the high-triaxiality re-
gion is compensated by the large plastic strain
in the adjacent region. The local plastic strain
grows with further elongation and eventu-
ally reaches the plastic failure strain in high-
triaxiality regions.

In addition to high type III sm/se, we ob-
served the coexistence of high and low sm/se
in single grains, which resulted in moderate
type II sm/se, as shown in G1 (Fig. 3Y). We
compared some type III and type II stresses as
a function of position in the xy plane (Fig. 4).
In G1, the type III sz ranges from 20 ± 100 to
470 ± 170 MPa, and the type II sz is 280 MPa
(Fig. 4D), where the measurement errors of
the type III stresses were evaluated from the
deviations of forces from equilibrium (30). The
type III sz reaches 520 ± 160 MPa inside G1
(Fig. 4E) and 560 ± 100 MPa near the grain
boundary in grain G3 (Fig. 4H). In G3, the
type III sx and sy range from 10 ± 93 to 390 ±
94 MPa and from 10 ± 130 to 430 ± 140 MPa,

Hayashi et al., Science 366, 1492–1496 (2019) 20 December 2019 3 of 5

RESEARCH | REPORT B Fig. 4. Type III stresses that deviated
markedly from type II stresses as a function
A of position. Type III (A) sx and (B) sy at z =
1.2 mm and sz at (C) z = 1.2 mm, (D) z = 6.0 mm,
C and (E) z = 12.0 mm, as a function of position
in grain G1. Type III (F) sx and (G) sy at
E D z = 18.0 mm and (H) sz at z = 20.4 mm, as a
H function of position in grain G3. The type III
G stresses were extracted from the unsmoothed
F stress maps shown in fig. S29 and movie S7.
Type II stresses are represented by solid lines,
which are independent of position in each
grain. Error bars indicate the measurement
errors of the type III stresses, evaluated from
the deviations of forces from equilibrium.

respectively, and the type II sx and sy are 170 3D stress tensor fields with complex distribu- 10. H. F. Poulsen, Three-Dimensional X-ray Diffraction
and 90 MPa, respectively (Fig. 4, F and G). The tions in individual grains for elongated bulk Microscopy: Mapping Polycrystals and Their Dynamics
type III sx and sy reach compressive maximum steel. The intragranular stresses we observed (Springer, 2004).
values of −190 ± 93 and −360 ± 180 MPa, respec- deviated by a value as high as the yield strength
tively, near the boundary in G1 (Fig. 4, A and B). from the grain-averaged stresses. The stress 11. H. F. Poulsen, J. Appl. Cryst. 45, 1084–1097 (2012).
fields were in highly triaxial stress states under 12. Y. Hayashi, Y. Hirose, Y. Seno, J. Appl. Cryst. 48, 1094–1101
The macroscopic sYS increases with decreas- plastic deformation smaller than the uniform
ing grain size and satisfies the Hall-Petch equa- elongation. The ability to nondestructively map (2015).
intragranular 3D stress tensors voxel by voxel 13. B. C. Larson, W. Yang, G. E. Ice, J. D. Budai, J. Z. Tischler,
tion as a function of grain size. The Hall-Petch for multiple grains will facilitate the develop-
equation for low-carbon steel predicts that sYS ment of materials modeling for predicting Nature 415, 887–890 (2002).
will increase from 234 MPa for a grain size of the observed evolution of local stresses, stress 14. L. E. Levine et al., Nat. Mater. 5, 619–622 (2006).
20 mm to 700 MPa for a grain size of 1 mm. The triaxiality, and plastic strains as yielding and 15. G. E. Ice, J. D. Budai, J. W. Pang, Science 334, 1234–1239
strong dependence of sYS on grain size means failure criteria. Advancement in multiscale
that the plastic deformation resistance (pin- materials modeling, which connects material (2011).
responses at the nanoscale to the macroscopic 16. H. Simons et al., Nat. Commun. 6, 6098 (2015).
ning of dislocation propagation) is localized mechanical behavior, is crucial for the quan- 17. A. Yau, W. Cha, M. W. Kanan, G. B. Stephenson, A. Ulvestad,
titative prediction of key properties concern-
near grain boundaries. Therefore, local regions ing the plastic deformation, creep, fatigue, and Science 356, 739–742 (2017).
fracture of engineering alloys. 18. M. J. Cherukara et al., Nat. Commun. 9, 3776 (2018).
with high strength, including inclusions inside 19. J. Oddershede et al., J. Appl. Cryst. 43, 539–549 (2010).
REFERENCES AND NOTES 20. P. Sedmák et al., Science 353, 559–562 (2016).
grains, may exist even in coarse-grained steel. 21. L. Renversade, A. Borbély, J. Appl. Cryst. 50, 1144–1157
1. R. Pippan, P. Gumbsch, Eds., Multiscale Modelling of
Although the reconstruction became difficult Plasticity and Fracture by Means of Dislocation Mechanic (2017).
(Springer, 2010). 22. T. Turner et al., Metall. Mater. Trans. A 48, 627–647 (2017).
for voxels near grain boundaries, the number 23. H. Abdolvand, J. Wright, A. J. Wilkinson, Nat. Commun. 9, 171
of voxels with sz > 450 MPa was 0.3% of the 2. R. Suter, Science 356, 704–705 (2017).
total number of reconstructed voxels after plas- 3. E. Maire, P. J. Withers, Int. Mater. Rev. 59, 1–43 (2014). (2018).
4. P. Bleuet et al., Nat. Mater. 7, 468–472 (2008). 24. J. Oddershede et al., Acta Mater. 60, 3570–3580 (2012).
tic deformation. This means that 5% elongated, 5. F. Hofmann et al., Nat. Commun. 4, 2774 (2013). 25. J. C. Schuren et al., Curr. Opin. Solid State Mater. Sci. 19,
6. A. King, G. Johnson, D. Engelberg, W. Ludwig, J. Marrow,
coarse-grained low-carbon steel with a grain 235–244 (2015).
size of 20 mm contains tensile longitudinal Science 321, 382–385 (2008). 26. D. Naragani et al., Acta Mater. 137, 71–84 (2017).
stress corresponding to the sYS of fine-grained 7. L. Margulies, G. Winther, H. F. Poulsen, Science 291,
low-carbon steel with a grain size of 3 mm at a 27. Y. Guo et al., Acta Mater. 96, 229–236 (2015).
volume ratio of at least 0.3%. 2392–2394 (2001). 28. H. Simons et al., Nat. Mater. 17, 814–819 (2018).
8. S. Schmidt et al., Science 305, 229–232 (2004). 29. Y. Hayashi, D. Setoyama, Y. Seno, Mater. Sci. Forum 905,
We achieved intragranular 3D stress tensor 9. B. Jakobsen et al., Science 312, 889–892 (2006).
157–164 (2017).
measurements for plastically deformed bulk 30. Materials and methods are available as supplementary materials.
31. T. Nonaka et al., AIP Conf. Proc. 1741, 030043 (2016).
polycrystalline materials by overcoming grain 32. A. J. Wilkinson, G. Meaden, D. J. Dingley, Mater. Sci. Technol.

averaging. Although the intragranular aver- 22, 1271–1278 (2006).
33. M. Ojima, Y. Adachi, S. Suzuki, Y. Tomota, Acta Mater. 59,
aging of stress fields remains, we determined
4177–4185 (2011).
34. G. Mirone, Eng. Fract. Mech. 74, 1203–1221 (2007).

ACKNOWLEDGMENTS

The authors thank H. Ohashi and H. Yumoto for the x-ray optics design,
Y. Seno and Y. Takatani for assistance in the experiments, and

Hayashi et al., Science 366, 1492–1496 (2019) 20 December 2019 4 of 5

RESEARCH | REPORT

T. Ohsuna for comments on the manuscript. Funding: This research at SPring-8. The derived data are available from https://github.com/ Supplementary Text
received no specific grants from any funding agency. Author Scanning3DXRDpublish/2017B7002. Figs. S1 to S45
contributions: Y.Ha., D.S., and Y.Hi. designed the research. Y.Ha. References (35–89)
and T.Y. performed the experiments. Y.Ha. and H.K. wrote the SUPPLEMENTARY MATERIALS Movies S1 to S8
manuscript. All authors discussed the results and commented on the science.sciencemag.org/content/366/6472/1492/suppl/DC1
manuscript. Competing interests: The authors declare no conflicts Materials and Methods 4 May 2019; accepted 13 November 2019
of interest. Data and materials availability: Raw data were generated 10.1126/science.aax9167

Hayashi et al., Science 366, 1492–1496 (2019) 20 December 2019 5 of 5

RESEARCH

QUANTUM GASES phase shift of the intracavity field across the
cavity resonance: The light field scattered
Dissipation-induced structural instability and chiral from the pump into the cavity acquires a

dynamics in a quantum gas phase shift fk that effectively mixes the ortho-
gonal quadratures, giving rise to a dissipative

coupling between the DM and the SM (Fig. 1,

Nishant Dogra*, Manuele Landini†, Katrin Kroeger, Lorenz Hruby, Tobias Donner‡, Tilman Esslinger A and B).
The strengths lD and lS of the coherent
Dissipative and unitary processes define the evolution of a many-body system. Their interplay gives rise
to dynamical phase transitions and can lead to instabilities. In this study, we observe a nonstationary couplings between the BEC and the DM or
state of chiral nature in a synthetic many-body system with independently controllable unitary and the SM, respectively, are tuned by the lattice
dissipative couplings. Our experiment is based on a spinor Bose gas interacting with an optical depth VTP and polarization angle ϕ of the trans-
resonator. Orthogonal quadratures of the resonator field coherently couple the Bose-Einstein condensate verse pump (20). These couplings soften the
to two different atomic spatial modes, whereas the dispersive effect of the resonator losses mediates effective excitation frequencies wD and wS of
a dissipative coupling between these modes. In a regime of dominant dissipative coupling, we observe both modes, such that at a critical lattice
the chiral evolution and relate it to a positional instability. depth the frequency of the more strongly
coupled mode vanishes. This mode can then
be macroscopically occupied, and the system

undergoes a self-organization phase transi-

I n a many-body system, unitary processes couplings are mediated via photons scattered tion (21), breaking a spatial Z(2) symmetry.
generally give rise to coherent evolu- by the atomic system from a standing-wave Simultaneously, the corresponding quadra-
tion, whereas dissipative processes lead transverse pump laser field into a high-finesse ture of the cavity mode is coherently pop-
to stationary states (1). The interplay of optical cavity mode (Fig. 1B). The DM and the ulated, which we detect with a heterodyne
these two types of processes in a driven- SM interact with orthogonal quadratures detection system analyzing the light field leak-

dissipative setting can influence a many-body of the cavity mode. We engineer a dissipa- ing from the cavity (22).

system in profound ways. Examples are dis- tive coupling in this system, exploiting the The cavity-induced phase shift fk and hence

sipative phase transitions (2–9), the emergence finite cavity decay rate k and the associated the dissipative coupling strength K2 between

of new universality classes (10, 11), dissipation-

induced topological effects (12), complex dy-

namics (13), and the splitting of multicritical A BEC B
points (14). In this study, we observe a pheno-

menon in which chiral nonstationary dynam-

ics emerges if the energy scales of dissipative HD

and unitary processes are similar. In our ex-

periment, we use a quantum gas to create a DM SM
driven many-body system with controllable

unitary and dissipative couplings. This allows y B
us to explore the system’s macroscopic behav- zx
ior at the boundary between stationary and C
nonstationary states. We gain a conceptual

understanding of the observed dynamics by D

considering dissipation as a structure-dependent DSI
force, analogous to mechanical nonconserva-

tive positional forces (15–17). In addition, our

observations permit us to draw connections DM SM
to limit cycles (18, 19).

Our experiment consists of a spinor Bose-

Einstein condensate (BEC) of two different 0

Zeeman states that is coherently coupled to

two different spatial atomic configurations Fig. 1. Engineered dissipative coupling. (A) A spinor BEC containing a mixture of two Zeeman

(20): One coupling induces a density mode states (depicted by up and down arrows) is coherently coupled with rates lD and lS to a DM and
(DM) in which the density of the gas shows a an SM, respectively, via the two quadratures of a cavity field with annihilation operator ^a. A dissipative

checkerboard modulation, whereas the other coupling (orange arrow) between DM and SM is induced by the phase response of the cavity.

coupling favors a spin mode (SM) in which Occupation of the DM (SM) leads to a density (spin) modulation, displayed as atoms occupying the

the gas instead develops a checkerboard mod- same (opposite) checkerboard lattice sites of an exemplary green (blue) two-by-two square unit

ulation of the spin degree of freedom (unit cell of dimensions l by l, where l is the wavelength of the transverse pump field (see below).

cells are shown in Fig. 1A). These coherent (B) The spinor BEC is coupled with a cavity mode and irradiated by a standing-wave transverse pump

in the z direction with linear polarization (black arrow) at angle ϕ with respect to the y direction.

The magnetic field B is oriented along the z direction. Light leaking out of the cavity at decay

Institute for Quantum Electronics, ETH Zurich, CH-8093 rate k is analyzed via a heterodyne detector (HD). (Inset) Phase fk and amplitude A response of
Zurich, Switzerland. the cavity as a function of detuning Dc. (C) Dissipation acts like a chiral force (orange); here,

*Present address: Cavendish Laboratory, University of Cambridge, xD (xS) represents the amplitude of the density (spin) modulation. (D) Schematic representation of
Cambridge CB3 0HE, UK. †Present address: Institut für three different regimes as a function of dissipative coupling strength K2 and detuning between the
Experimentalphysik und Zentrum für Quantenphysik, Universität
Innsbruck, 6020 Innsbruck, Austria. frequencies wD and wS of the DM and SM, respectively. The green (blue) region represents the
‡Corresponding author. Email: [email protected]
DM (SM) being dominantly coupled to the spinor BEC.

Dogra et al., Science 366, 1496–1499 (2019) 20 December 2019 1 of 4

RESEARCH | REPORT

the two modes can be controlled by the detuning strength of the resulting force field increases of the intracavity light field for two different
Dc between the cavity resonance and the fre- sets of parameters. We find two qualitatively
quency of the transverse pump (23). The effect with the strength of the dissipative coupling, different behaviors: Above a critical pump
of this dissipative coupling can be understood such that the instability occurs also for in- power, the cavity field has a nonzero ampli-
in the xD-xS plane, where xD and xS repre- tude and either a well-defined (Fig. 2A) or a
sent the amplitudes of the density and spin creasingly nondegenerate modes, as qualita- monotonically changing phase (modulo 2p;
modulation [see below and (23)] caused by tively shown by the broadening of the DSI Fig. 2B). A well-defined phase indicates that
the occupation of DM and SM, respectively only one quadrature of the cavity field is ex-
(Fig. 1C). In this plane, the dissipative coupling region in the schematic in Fig. 1D. cited, corresponding to either the SM or the
acts as a force field that favors a rotation of the We prepare the 87Rb spinor BEC with N = DM being populated, which is determined by
system’s state around the origin: If the DM the prevailing coherent coupling. In contrast,
and the SM are degenerate, even an infini- (23 ± 2) × 103 atoms in each of the different a monotonically changing phase is observed
tesimally small dissipative coupling leads to a Zeeman states jF ¼ 1; mF ¼ T1i, where F and when the dissipative coupling is dominant and
dissipation-induced structural instability (DSI) mF denote the total angular momentum and signals that the system is continuously evolving
in which the system rotates with fixed chirality the corresponding magnetic quantum number, through the different spatial modes linked with
between the different atomic modes. The the two quadratures of the cavity field.
respectively. We linearly ramp up the lattice
depth of the transverse pump in 50 ms and The many-body Hamiltonian of the system
is given by
analyze the cavity output with our heterodyne

setup. In Fig. 2, A and B, we show the mean
photon number nph and phase f (modulo 2p)

A 60 (rad) nph 10 VTP (ER) B4 (rad) nph 15 VTP (ER) H^ ¼ ÀℏDca^†a^ þ ℏw0ðJ^z;þ þ J^z;ÀÞ

30 5 2 10 ½þ pℏffiffiffiffi lDða^† þ a^ÞðJ^x;þ þ J^x;ÀÞ
0 0
π 0 π 5 N
10 15 ŠÀ ilSða^† À a^ÞðJ^x;þ À J^x;ÀÞ
ð1Þ

05 VTP (ER) 0 10 VTP (ER) where a^ða^†Þ is the annihilation (creation)

-π 0 -π 5 operator corresponding to the y-polarized
3 5 7 9 11 13 10 20
Time (ms) 12 14 16 18 cavity mode in the frame of the transverse
Time (ms)
xS pump frequency. The atomic system is rep-

xS xS resented by an ensemble of N effective “spins”

xD xD xD in each of the two Zeeman states (labeled
with ±), with J^x;T; J^y;T; and J^zT being the cor-

responding angular momentum operators. This

pseudo-spin is constructed from the macro-

VTP (ER) VTP (ER) VTP (ER) scopically occupied zero momentum state j0i
13 19 26 13 19 26 13 19 26
06 32 0 6 32 0 6 32 of the BEC and an excited state jki with a
102
15 C D E symmetric superposition of four momentum

7 100 states jpx ¼ Tℏk; pz ¼ Tℏki, which represent
0 one recoil momentum ℏk each in the cavity
−7 10−2
ν (kHz) and the transverse pump direction (21). The
nph
wave number of the transverse laser field with

wavelength l is given by k ¼ 2p=l. For zero

−15 10 20 30 40 50 0 10 20 30 40 50 0 10− 4 lattice depth VTP of the transverse pump, the
0 Time (ms) Time (ms) 10 20 30 40 50
energy difference between the states j0i and
Time (ms) jki is ℏw0 ¼ 2ER ¼ ℏ2k2=m, where m is the
mass of a 87Rb atom and ℏ is Planck’s con-

stant divided by 2p. The first two terms in

Fig. 2. Detection of the DSI. (A and B) Time evolution of the mean number of photons nph (black) in Eq. 1 describe the energies of the bare pho-
the cavity and the corresponding phase ϕ modulo 2p (orange) for (A) ϕ ¼ 58:3ð1:0Þ°, D~c=2p ¼ À5:5ð1ÞMHz
and (B) ϕ ¼ 63:3ð1:0Þ°, D~c=2p ¼ À5:5ð1ÞMHz. The detuning D~c includes the effect of the atomic dispersive tonic and atomic modes, respectively, whereas
shift on the detuning Dc (23). The linear ramp of the lattice depth VTP of the transverse pump is shown
as dashed blue line. Gray and light orange lines correspond to zero occupation of the cavity. (C to E) the third term captures the couplings of
Spectrograms showing the mean number of photons ~nph as a function of frequency n and time for (C)
ϕ ¼ 58:3ð1:0Þ°, D~c=2p ¼ À5:4ð2Þ MHz; (D) ϕ ¼ 63:3ð1:0Þ°, D~c=2p ¼ À2:5ð2Þ MHz; and (E) ϕ ¼ 58:3ð1:0Þ°, the DM and the SM to the respective quad-
D~c=2p ¼ À2:4ð2Þ MHz. A window size of 5 ms is used to construct the spectrograms, and each data set is
averaged over 20 experimental realizations. Frequency n is defined relative to the transverse pump. Corresponding ratures of the cavity field (23). In this frame-

insets show the evolution of the atomic state in the xD-xS plane extracted from the spectrograms work, the amplitudes of the DM and the
in a duration of 500 ms at the position of the gray lines. In the inset to (C), we plot the data and its mirror
SM are given by xD ¼ 1 ð hJ^x;þi þ hJ^x;Ài Þ and
about xS to illustrate the Z(2) symmetry breaking of the self-organization phase transition. Dashed lines in the ðhJ^ x;þ i À hJ^ N respectively.
insets to (D) and (E) are the qualitative predictions of the noninteracting theory for vanishing gain (Eq. 3). xS ¼ 1 x;ÀiÞ,
N
All insets are rescaled with respect to each other and the theoretical lines for better visibility. Linecuts through The damping of the cavity mode can be

spectrograms for similar data are presented in (23). modeled by a decay term Àka^† in the equation
of motion of the operator a^†. It causes a phase

shift fk ¼ tanÀ1ðÀk=DcÞ of the field scattered

into the cavity by the atomic system. We

adiabatically eliminate the cavity field and

write the linearized equations of motion

near the ground state of the noninteracting

Dogra et al., Science 366, 1496–1499 (2019) 20 December 2019 2 of 4

RESEARCH | REPORT

˜Δ c/ 2π (MHz)A0 DSI VTPsin2fkcosdf can thus be enhanced by either phase of the cavity field, as shown in Fig. 2A.
DM SM increasing the cavity-induced phase shift or We identify this as the formation of a static
−4 ν (kHz) g (kHz) making the two modes degenerate. This dis- checkerboard density pattern that coherently
−8 0.5 sipative coupling generates a chiral force scatters the pump field into the cavity (20, 23).
−12 0.0 orthogonal to the current position vector of The corresponding steady state of the system
the system in the xD-xS plane (Fig. 1C) and pro- is displayed in the inset to Fig. 2C.
B vides an example of a positional force (17, 24).
Such positional forces are known in mecha- In contrast, Figs. 2, D and E, depict red (n <
6 nical systems such as a rotating shaft subject 0) and blue (n > 0) detuned sidebands with the
to friction caused by an incompressible vis- peak frequency being a function of the lattice
4 cous fluid in a bearing. The incompressibility depth of the transverse pump. Observation of
of the fluid leads to unequal frictional forces only a red sideband, as in Fig. 2D, is equivalent
2 −0.5 on opposite sides of the shaft, resulting in a to a linearly running phase. For small lattice
positional force orthogonal to the direction depths (VTP < 6ER), the sideband frequency is
0 30 60 90 of the displacement (15, 16). These positional expected to be close to the root mean square
ϕ (deg) forces represent the only possible linear mechan- we of theptwffiffiffioffiffiffiffiffimffiffiffiffiffioffiffidffiffiffieffiffiffiffiffifffirffi equencies—that is,
ical forces besides conservative trapping forces, w ≈ we ¼ ðwD2 þ wS2Þ=2 . Evolution at this
Fig. 3. Boundary of the instability-dominated centripetal forces, and frictional forces (23). In intermediate frequency reflects a synchroni-
regime. (A) Critical detuning, including the our system, when the two atomic modes are zation process (26) between the two spatial
dispersive shift (23), for the onset of the degenerate, this positional force cannot be modes arising from the dissipative coupling.
dissipation-induced instability as a function of counteracted by the restoring harmonic force For large lattice depths, the sideband fre-
polarization angle ϕ. The black data points indicate pointing toward the origin, thus leading to a quency depends on the dissipative coupling
the smallest detuning, where the strength of the dissipation-induced structural or positional in- strength: w ≈ K 2=2jwej (23). In this limit, the
sidebands becomes larger than the signal at stability (17, 24, 25). two mode frequencies become imaginary, cor-
the transverse pump frequency in the spectrograms responding to the self-organization phase tran-
for a fixed lattice depth of 21ER. The white diamonds Mode degeneracy (i.e., lD ¼ lS such that sition in the absence of dissipative coupling.
indicate the parameters for the spectrograms shown df ¼ 0) is reached at the critical polarization
in Fig. 2, C to E. Error bars are given by the discretized angle ϕc ¼ 65:1° for the chosen wavelength The relative strength R of the blue with re-
detuning interval in which the transition was of the transverse pump of l = 784.7 nm. To spect to the red sideband increases toward 1 as
detected. The polarization angle determination has explain our observations, we analyze the solu- ϕ deviates from the critical angle ϕc. The pres-
an error of 1°. The green and blue regions tions of Eq. 2 and the corresponding intra- ence of the blue sideband is connected to
correspond to DM and SM being respectively cavity light field for polarization angles close nonzero df (Eq. 3) and leads to an elliptical
more strongly coupled to the spinor BEC. In the to ϕc—that is, for a small df (23) evolution in the xD-xS plane. The relative
orange region, the system is dominated by the strength R, and hence the ellipticity of the
dissipation-induced instability. The boundary wt þ Dc df chiral solution, can be influenced via Dc or
between different regions is obtained from a k jdfj . Microscopically, the blue sideband is
noninteracting theory [see main text and (23)]. ½xDc ; xScŠ ¼ A coswt; sin egt connected to the motion of a different number
(B) Green and blue solid (dashed) lines show the of atoms in each Zeeman state. The insets to
frequency n of the two eigenmodes of the system as ha^ † iºðxDc À ixSc Þ À df ðxDc þ ixSc Þ Fig. 2, D and E, show data of the time-varying
a function of the polarization angle in the presence 2 trajectory of the system together with solu-
(absence) of dissipative coupling. Without dissipation, tions obtained from Eq. 2 for g = 0, illustrating
the two modes are the DM (green) and the SM º A eÀiwt À df 1 þ i Dc eiwt egt the nonstationary chiral state. When prepar-
(blue). Dissipation leads to level attraction between 2k ð3Þ ing the system in the DSI at a fixed lattice
the two modes and hence changes their frequencies. depth, we observe a narrow red sideband that
In the instability-dominated regime, the two modes In the limit df ¼ 0, this time-dependent sol- persists over tens of milliseconds (23), corre-
are synchronized (orange). Correspondingly, the gains ution implies that the system is rotating in the sponding to a uniform circular motion in the
g of the amplified mode (black) and the damped xD-xS plane with fixed chirality at frequency w xD-xS plane. This observation hints at the pres-
mode (gray) are plotted. The lines are drawn and amplification rate g (additionally, t is time ence of a limit cycle in our system (18, 19).
for D~c=2p ¼ À2:7MHz or Dc=2p ¼ À4MHz and and A is the amplitude at t=0) (23). Microscop-
VTP ¼ 1:5ER. ically, this rotation is associated with the Our mean-field description of the system in
atomic spins moving from one l-periodic spa- terms of a positional instability does not de-
spinor BEC for the average amplitudes xD tial pattern to another (Fig. 1C). Because the scribe all aspects of the observed dynamics.
and xS (23) two atomic modes are connected to different For example, the number of photons in the
quadratures of the cavity, the phase of the case of the DSI (Fig. 2, D and E) is much
d2 xD ÀwD2 cavity field evolves monotonically, as observed smaller than during self-organization (Fig. 2C),
dt2 xS K2 ÀK 2 xD in Fig. 2B and shown in Eq. 3. although the instability is associated with a
¼ Àw2S xS ð2Þ finite amplification rate g. Even the limit cycle–
The frequency spectrum of the light leaking like behavior persisted for a long time without
where w2D À wS2ºsindf with df ¼ 2tanÀ1ðlS= from the cavity is also accessible with our het- much increase in the corresponding amplitude.
lDÞ À p=2 is a measure of the relative coupling erodyne setup (22). Figure 2, C to E, shows Moreover, there is a finite lattice depth thresh-
spectrograms for three different sets of param- old for the onset of the DSI (Fig. 2B), although
strength of the BEC to the SM and the DM. eters of data similar to Fig. 2, A and B, but g is positive for infinitesimal lattice depths. We
The strength of the dissipative coupling K 2º averaged over 20 repetitions. Figure 2C shows attribute this, as well as the observed pulsing
a spectrogram where the signal is located at behavior in the number of photons (Fig. 2B),
zero frequency (n = 0). It corresponds to the to collisional interactions between the atoms
frequency of the transverse pump and is iden- (23). Such a behavior of collisional interac-
tical to the observation of a constant time tions can possibly be modeled as an atomic

Dogra et al., Science 366, 1496–1499 (2019) 20 December 2019 3 of 4

RESEARCH | REPORT

dissipation channel that competes with the couplings are independently tunable. Our ob- Conference Publications 2133, NASA, 1982), vol. 2250,
cavity-induced instability. Similar competi- pp. 274–283.
tions between multiple heat baths are pre- servation of a dissipation-induced instability 26. A. Pikovsky, M. Rosenblum, J. Kurths, Synchronization:
dicted to give rise to dissipative frustration A Universal Concept in Nonlinear Science (Cambridge Univ.
(27, 28). represents a form of quantum many-body dy- Press, 2003).
27. H. Kohler, F. Sols, New J. Phys. 8, 149 (2006).
We experimentally map out the boundary namics that shares characteristics with but is 28. D. Maile, S. Andergassen, W. Belzig, G. Rastelli, Phys. Rev. B
of the DSI region by ramping up the trans- distinct from limit cycles (18, 19, 34) and time 97, 155427 (2018).
verse pump lattice to 25ER in 50 ms for various crystals (35, 36). 29. I. Rotter, J. Phys. A Math. Theor. 42, 153001 (2009).
polarization angles ϕ and detunings Dc (Fig. 30. R. El-Ganainy et al., Nat. Phys. 14, 11–19 (2018).
3A). For a given polarization angle, we define REFERENCES AND NOTES 31. N. R. Bernier, L. D. Tóth, A. K. Feofanov, T. J. Kippenberg,
the onset of the DSI by the minimum detuning Phys. Rev. A 98, 023841 (2018).
where the strength of the sidebands exceeds 1. H.-P. Breuer, F. Petruccione, The Theory of Open Quantum 32. W. D. Heiss, J. Phys. A Math. Theor. 45, 444016 (2012).
the signal at the transverse pump frequency. Systems (Oxford Univ. Press, 2002). 33. M.-A. Miri, A. Alù, Science 363, eaar7709 (2019).
We theoretically obtain the boundary of the 34. M. Gutzwiller, Chaos in Classical and Quantum Mechanics
DSI (orange region in Fig. 3A) from Eq. 2 as 2. N. Syassen et al., Science 320, 1329–1331 (2008). (Springer, 1990).
jw2D À wS2j ¼ 2K 2 , which results in a critical 3. G. Barontini et al., Phys. Rev. Lett. 110, 035302 (2013). 35. J. Zhang et al., Nature 543, 217–220 (2017).
detuning Dc;b ¼ Àk=tanjdfj . A study of the 4. F. Brennecke et al., Proc. Natl. Acad. Sci. U.S.A. 110, 36. S. Choi et al., Nature 543, 221–225 (2017).
extent of the DSI as a function of polariza- 37. N. Dogra, M. Landini, K. Kroeger, L. Hruby, T. Donner,
tion angle and transverse pump lattice depth 11763–11767 (2013). T. Esslinger, “Dissipation Induced Structural Instability
is presented in (23). 5. R. Labouvie, B. Santra, S. Heun, H. Ott, Phys. Rev. Lett. 116, and Chiral Dynamics in a Quantum Gas,” ETH Zurich
Research Collection (2019); https://doi.org/10.3929/
Physically, the boundary of the DSI can be 235302 (2016). ethz-b-000366051.
understood as the onset of synchronization 6. T. Tomita, S. Nakajima, I. Danshita, Y. Takasu, Y. Takahashi,
between DM and SM. Such a synchronization ACKNOWLEDGMENTS
process is the result of a dissipation-induced Sci. Adv. 3, e1701513 (2017). We thank B. Buca, E. Rodriguez Chiacchio, F. Ferri, D. Jaksch,
level attraction, a hallmark of non-Hermitian 7. I. Carusotto, C. Ciuti, Rev. Mod. Phys. 85, 299–366 (2013). A. Nunnenkamp, and J. Tindall for discussions. Funding:
systems (29, 30). Figure 3B shows the frequen- 8. T. Fink, A. Schade, S. Höfling, C. Schneider, A. Imamoglu, We acknowledge funding from SNF [project nos. 182650 and
cies of the DM and SM in the absence (dashed 175329 (NAQUAS QuantERA) and NCCR QSIT], EU Horizon
lines) and presence (solid lines) of dissipation. Nat. Phys. 14, 365–369 (2018). 2020 [ERC Advanced Grant TransQ (project no. 742579)], and
In the vicinity of the critical angle, this level 9. J. T. Young, A. V. Gorshkov, M. Foss-Feig, M. F. Maghrebi, SBFI (QUIC, contract no. 15.0019). Author contributions:
attraction leads to the emergence of two de- N.D., M.L., K.K., and L.H. prepared the experiment and collected
generate modes with opposite chirality. Whereas arXiv:1903.02569 [cond-mat.quant-gas] (6 March 2019). the data; N.D., M.L., and K.K. evaluated the data; N.D. and
one of these modes is damped, the other is 10. S. Diehl, A. Tomadin, A. Micheli, R. Fazio, P. Zoller, Phys. Rev. M.L. developed the theory; T.D. and T.E. supervised the project;
amplified, which gives rise to the DSI. Similar and all authors contributed to discussions and writing of
forms of level attraction leading to instability Lett. 105, 015702 (2010). the manuscript. Competing interests: The authors declare
are observed in optomechanical systems (31) 11. D. Nagy, G. Szirmai, P. Domokos, Phys. Rev. A 84, 043637 no competing interests. Data and materials availability:
and, in general, in non-Hermitian photonic Data and scripts are available in the ETH Zurich Research
systems, where they can be associated with (2011). Collection repository (37).
parity–time symmetry-breaking phase transi- 12. S. Diehl, E. Rico, M. A. Baranov, P. Zoller, Nat. Phys. 7, 971–977
tions (30) and exceptional points (32, 33). We SUPPLEMENTARY MATERIALS
have experimentally studied a many-body sys- (2011). science.sciencemag.org/content/366/6472/1496/suppl/DC1
tem in which both coherent and dissipative 13. B. Buča, J. Tindall, D. Jaksch, Nat. Commun. 10, 1730 Materials and Methods
Supplementary Text
(2019). Figs. S1 to S5
14. M. Soriente, T. Donner, R. Chitra, O. Zilberberg, Phys. Rev. Lett. References (38–41)

120, 183603 (2018). 20 December 2018; accepted 6 November 2019
15. B. L. Newkirk, H. Taylor, Gen. Electr. Rev. 28, 559–568 10.1126/science.aaw4465

(1925).
16. P. L. Kapitsa, Zh. Tekh. Fiz. 9, 124–147 (1939).
17. D. R. Merkin, Introduction to the Theory of Stability (Springer,

1996).
18. F. Piazza, H. Ritsch, Phys. Rev. Lett. 115, 163601 (2015).
19. E. I. Rodríguez Chiacchio, A. Nunnenkamp, Phys. Rev. Lett. 122,

193605 (2019).
20. M. Landini et al., Phys. Rev. Lett. 120, 223602 (2018).
21. K. Baumann, C. Guerlin, F. Brennecke, T. Esslinger, Nature 464,

1301–1306 (2010).
22. R. Landig, F. Brennecke, R. Mottl, T. Donner, T. Esslinger,

Nat. Commun. 6, 7046 (2015).
23. Supplementary materials.
24. R. Krechetnikov, J. E. Marsden, Rev. Mod. Phys. 79, 519–553

(2007).
25. S. H. Crandall, in “Rotordynamic Instability Problems

in High-Performance Turbomachinery” (NASA

Dogra et al., Science 366, 1496–1499 (2019) 20 December 2019 4 of 4

RESEARCH

MECHANOCHEMISTRY ful and attractive strategy to reduce the en-

Redox reactions of small organic molecules using ball vironmental impact of chemical processes;
milling and piezoelectric materials photoredox approaches, similarly to other con-

Koji Kubota1,2*, Yadong Pang2, Akira Miura2, Hajime Ito1,2* ventional organic reactions, often require com-
plicated reaction setups, substantial amounts
Over the past decade, photoredox catalysis has harnessed light energy to accelerate bond-forming reactions.
We postulated that a complementary method for the redox-activation of small organic molecules in of dry and degassed organic solvents, and an
response to applied mechanical energy could be developed through the piezoelectric effect. Here, we inert gas atmosphere (1–5).
report that agitation of piezoelectric materials via ball milling reduces aryl diazonium salts. This mechanoredox
system can be applied to arylation and borylation reactions under mechanochemical conditions. For a proof-of-concept study, we selected
commercially available, inexpensive, and easy-
V isible light photoredox catalysis repre- These particles might act as strong reductants
sents a recent key development in con- to transfer electrons to small organic mole- to-handle BaTiO3 nanoparticles as the piezo-
temporary organic synthesis (1–5). In cules, followed by oxidative quenching of a electric material (Fig. 1B). This choice was
these transformations, the photoexcited donor, thereby inducing the selective forma-
catalyst can act as a potent single-electron tion of bonds in a manner analogous to pho- motivated by pioneering studies, in which ultra-
oxidant, transferring an electron to an accep- toredox catalysis (Fig. 1B). Such a mechanoredox sonic agitation of BaTiO3 produces a suitable
tor, after which single-electron oxidation of a approach could potentially represent a power- electrochemical potential to overcome the water
donor affords the product under concom- splitting potential (1.23 V) (28, 29) and reduce a
itant regeneration of the ground-state catalyst N,N,N′,N′,N″,N″-hexamethyl[tris(aminoethyl)
(Fig. 1A). The broad success of photoredox amine] (Me6TREN)–ligated CuBr2 complex,
catalysis hinges on the susceptibility of the which has a reduction potential of −0.30 V
coupling partners to redox activation and en- [versus saturated calomel electrode (SCE)]
suing bond-forming reactions with high levels (29, 36). These studies suggested that mechanical
of efficiency and selectivity. agitation of BaTiO3 in a ball mill could gen-
erate an electrochemical potential suitable for
There has also been substantial parallel
progress in mechanochemical organic trans-
formations using ball milling (6–18). Since the
term mechanochemistry was introduced by
Ostwald in 1887, mechanochemical synthesis
has been extensively exploited in materials
science (6–8), polymer chemistry (9), and in-
organic synthesis (8), but its application to
organic synthesis is more recent (10–18). Ad-
vantages of mechanochemical synthesis using
ball milling include the avoidance of poten-
tially harmful organic solvents and external
heating, shorter reaction times, and simpler
operational handling. Additionally, mechano-
chemical reactions are particularly useful for
substrates that are poorly soluble in com-
mon organic solvents.

Inspired by the unique profile of photoredox
systems based on light irradiation and the
utility of ball milling in mechanochemistry,
we hypothesized that redox activation of small
organic molecules could be achieved through
a mechanistically distinct approach using
mechanical energy (19–22). Specifically, we
envisioned that the agitation of piezoelectric
materials (23–35) via ball milling could gen-
erate temporarily highly polarized particles.

1Institute for Chemical Reaction Design and Discovery Fig. 1. Working hypothesis for a mechanoredox system for the activation of small organic molecules in
(WPI-ICReDD), Hokkaido University, Sapporo, Hokkaido organic synthesis. (A) A commonly employed photoredox catalyst and a generic photoredox oxidative quenching
060-8628, Japan. 2Division of Applied Chemistry, Graduate cycle. A, acceptor; D, donor. (B) Proposed mechanoredox paradigm using a ball mill and a piezoelectric material.
School of Engineering, Hokkaido University, Sapporo, (C) Proposed mechanism for the mechanoredox arylation and borylation using BaTiO3 and ball milling. tBu, tertiary
Hokkaido 060-8628, Japan. butyl; ppy, (2-pyridinyl)phenyl; dtbpy, 4,4′-di-tert-butyl-2,2′-bipyridine; pin, pinacolato; R, substituent.

*Corresponding author. Email: [email protected] (H.I.);

[email protected] (K.K.)

Kubota et al., Science 366, 1500–1504 (2019) 20 December 2019 1 of 5

RESEARCH | REPORT

the activation of redox-active, small organic Fig. 2. Mechanoredox arylation and borylation using mechanical force. (A) Optimization of the
mechanoredox arylation and borylation reactions. (B) Reaction mixture of the mechanoredox arylation
molecules in organic synthesis. of 2a after grinding in a ball mill. A stainless steel milling jar (1.5 ml) and a stainless steel ball (diameter,
The photoredox systems that activate aryl 5 mm) were used. See supplementary materials for details. equiv, equivalent; h, hour.

diazonium salts for coupling with heteroar- This result is consistent with our hypothesis yield of 5a, whereas other common solvents,
enes (37) or borylation (38) have been reported that the required piezoelectric potential is gen- such as N,N-dimethylformamide (DMF), DMSO,
by König and Yan, respectively. The key step in erated by mechanical force provided by the toluene, or hexane, led to little or no improve-
these transformations is the photochemical ball milling of BaTiO3. Using other piezoelec- ment. We also found that prolonging the re-
tric materials, such as LiNbO3 and ZnO, also action time led to a higher yield of 3a (89%)
reduction of aryl diazonium salts to aryl rad- afforded 3a, albeit in lower yield (24% and when MeCN was used as the LAG additive.
ical species. Andrieux and Pinson have reported 15%, respectively). Using a bigger jar (5.0 ml) Using a bigger jar (5.0 ml) and ball (7.5 mm)
a reduction potential of −0.16 V (versus SCE) and ball (7.5 mm) provided 3a in high yield did not substantially affect the transformation
for phenyl diazonium tetrafluoroborate (39), (82%, table S2). Product 3a was isolated by (86%, table S3). The reaction without BaTiO3
which was within range for feasible redox ac- filtration of the obtained crude solid mixture did not proceed, supporting our mechanistic
tivation using BaTiO3. In our postulated mech- (Fig. 2B), followed by column chromatography hypothesis. Compared with the analogous pho-
anism (Fig. 1C), the agitation of BaTiO3 via ball on silica gel. toredox reaction, the present mechanoredox
milling generates a temporary electrochemical borylation exhibited much faster reaction ki-
potential in response to mechanical impact. We then investigated the possibility of mech- netics and a better product yield (38).
anoredox borylation using the same setup
According to the aforementioned inorganic (Fig. 2A). We found that the borylation of 1a Subsequently, we explored the scope of the
studies (28, 29), the temporary polarization with 4 in the presence of BaTiO3 afforded mechanoredox arylation reaction with various
should be sufficiently persistent to reduce an the desired arylboronate (5a) in 21% yield. aryl diazonium salts using a 5-ml stainless steel
aryl diazonium salt (1) via a single-electron Next, we attempted to improve the reactivity milling jar with a 7.5-mm diameter stainless
transfer (SET) mechanism analogous to the by using liquid-assisted grinding (LAG), in steel ball (Fig. 3A). Electron-deficient aryl dia-
photoredox reaction, to furnish the corre- which a substoichiometric amount of liquid zonium salts (1a to 1h) were converted into the
sponding aryl radical I. The addition of I to is added (41). In all LAG reactions, the ratio desired products (3a to 3h) in good yield under
heteroarene 2 would afford radical addition of liquid additive (microliters) to reactant the optimized conditions. Simple aromatic
intermediate II, which would be subsequently (milligrams) was 0.12. Use of acetonitrile substrates (1i to 1k) also reacted to give the
oxidized by the hole in the agitated BaTiO3 to (MeCN) as the LAG additive improved the corresponding products (3i to 3k) in good
form carbocation intermediate III (37). Final-
ly, deprotonation of III would lead to aryla-
tion product 3. In the borylation (38), the
generated radical I reacts with bis(pinacolate)
diboron (4), to cleave the B−B bond, result-
ing in the formation of the boryl substitution
product 5 as well as radical anion intermed-
iate IV. Subsequently, oxidation of IV by the
agitated BaTiO3 could form F−B(pin) (V) as a
by-product.

To explore this mechanistic hypothesis, we
first attempted the proposed mechanoredox
C−H arylation with 1a and furan (2a) in the
presence of commercially available BaTiO3,
using a Retch MM400 mixer mill (1.5-ml stain-
less steel milling jar with 5-mm diameter stain-

less steel ball) (Fig. 2A). The corresponding
arylation product (3a) was obtained in 40%
yield after milling at 20 Hz for 1 hour under
air. The reaction did not proceed in the ab-

sence of BaTiO3, suggesting that the mechani-
cal energy provided by ball milling generated
the piezoelectric potential to reduce 1a. In
contrast, even when ultrasound irradiation
was applied to the mixture with BaTiO3 in
dimethyl sulfoxide (DMSO) under a nitrogen
atmosphere, the formation of the product was
not observed. A small amount of 3a was ob-
tained when the reaction was carried out in

the presence of SrTiO3, which exhibits piezo-
electric properties upon applying in-plain
strain (40). When using non-piezoelectric ce-
ramic materials, such as TiO2, BaCO3, or Al2O3,
the reaction did not proceed, which suggests
that piezoelectric materials are essential for this

arylation reaction. We found that conducting
the reaction at higher ball-milling frequency

(30 Hz) substantially improved the yield (81%).

Kubota et al., Science 366, 1500–1504 (2019) 20 December 2019 2 of 5

RESEARCH | REPORT

Fig. 3. Scope of the
mechanoredox arylation
and borylation reactions
using aryl diazonium
salts. Data for each entry
(in bold) are reported
as isolated yield percentages.
Proton NMR integrated
yields are shown in parenthe-
ses. (A) Substrate scope
of the mechanoredox aryla-
tion of heteroaromatic
compounds. (B) Substrate
scope of the mechanoredox
borylation of aryl diazonium
salts. Reactions were per-
formed at 0.3 mmol scale
using a stainless steel milling
jar (5 ml) and stainless steel
ball (diameter, 7.5 mm).
Arylation conditions:
1 (0.3 mmol), 2 (4.5 mmol),
and BaTiO3 (1.5 mmol).
Borylation conditions:
1 (0.3 mmol), 4 (0.3 mmol),
BaTiO3 (1.5 mmol), and
MeCN (0.12 mL/mg). See
supplementary materials
for details. Me, methyl;
Boc, tertiary butoxycarbonyl.

yield. However, in the case of an electron-rich, tions with thiophene, whereas the photoredox ried out on an 8-mmol scale in a 25-ml stain-
methoxy-substituted diazonium salt (1l), the system developed by König and co-workers,
product (3l) was obtained in relatively low using eosin Y, produced 3ma and 3pa as less steel ball-milling jar with one 15-mm
yield. This was probably due to the relatively single isomers (37). We also confirmed that the diameter stainless steel ball, affording 3o in
high reduction potential of 1l. Other hetero- developed mechanoredox borylation conditions 71% yield (Fig. 4A). After separation from the
arenes, namely, thiophene and pyrrole, suc- were applicable to a variety of aryl diazonium
salts (Fig. 3B). crude reaction mixture and washing, BaTiO3
cessfully reacted to form the desired products could be reused for the mechanoredox aryla-
(3m and 3n). Arylation reactions using thienyl To demonstrate the practical utility of this tion of furan (2a) under the same reaction
diazonium salts afforded the corresponding protocol, we investigated a gram-scale synthe- conditions at least three times before the yield
heterobiaryls (3o and 3p), which are typical sis of heterobiaryls under the developed mech- of 3k declined substantially (Fig. 4B).
structural motifs for organic semiconductors anoredox conditions, as well as the recycling
(42). Minor regioisomers (3mb and 3pb) were of BaTiO3 (Fig. 4, A and B). The mechanoredox Direct C−H arylation of polycyclic aromatic
observed as competing products in the reac- C−H arylation of furan (2a) with 1o was car- hydrocarbons (PAHs) has attracted considerable
interest (43, 44) on account of their role in organic
light-emitting diodes, organic photovoltaics,

Kubota et al., Science 366, 1500–1504 (2019) 20 December 2019 3 of 5

RESEARCH | REPORT

Fig. 4. Exploration of a
gram-scale synthesis, cat-
alyst recycling, and the
arylation of polyaromatic
hydrocarbon compounds.
(A) Mechanoredox arylation
of 2a with 1o on the gram
scale using a 25-ml ball-
milling jar (pictured on the
right). (B) BaTiO3 recycling
experiments using 1k and
2a. (C) Mechanoredox aryl-
ation of the polyaromatic
hydrocarbons 6 with 1f.
See supplementary
materials for details.
rt, room temperature.

organic semiconductors, and organic thin- photoredox reactions. These results demon- action mixture containing 1a, furan (2a), and
strate the promising potential of the mecha- BaTiO3 halted the arylation process, and the
film transistors. With this in mind, we con- noredox arylation as an operationally simple TEMPO-trapped intermediates 9 and 10 were
and mild route to synthesize functionalized detected (Fig. 5A). Compound 10 might have
ducted a preliminary investigation of the PAHs from poorly soluble substrates that are been formed through the reaction of TEMPO
feasibility of the C−H arylation of PAHs using incompatible with photoredox conditions. (8) with the intermediate II (Fig. 1C), followed
our mechanoredox approach (Fig. 4C). The re- by oxidative aromatization by BaTiO3 or at-
action of 1f with pyrene (6a) in the presence of We postulated that the mechanoredox C−H mospheric oxygen. Overall, the identified
BaTiO3 and a small amount of MeCN as the arylation and borylation were proceeding via
LAG additive afforded the desired C−H aryla- SET events, in which an aryl radical is gener- compounds suggest that the mechanoredox
tion product (7aa) in good yield with high ated through piezoelectric reduction (Fig. 1C);
regioselectivity (7aa:7ab = 92:8). Furthermore, this assumption is supported by the results of activation with BaTiO3 proceeds via a radical
coronene (6b) was also arylated in moderate preliminary mechanistic investigations (Fig. 5). pathway.
yield under the mechanoredox conditions. When 2,2,6,6-tetramethylpiperidinoxyl (TEMPO)
When the arylation of coronene (6b) was at- (8) was treated with aryl diazonium salt 1a in We used a scanning electron microscopy
tempted via König’s photoredox procedure using the absence of furan (2a), the TEMPO-trapped
eosin Y, no reaction occurred (37). This was intermediate 9 was obtained (Fig. 5A). Fur- (SEM) analysis to confirm that the mechanical
probably due to the low solubility of coronene thermore, the addition of TEMPO (8) to a re-
(6b) in the polar solvents required for such stimulus provided by ball milling is transferred

onto the BaTiO3 particles under the applied
conditions (Fig. 5B) (24). The SEM image of
commercially available BaTiO3 powder shows

Kubota et al., Science 366, 1500–1504 (2019) 20 December 2019 4 of 5

RESEARCH | REPORT

Fig. 5. Mechanistic studies. (A) Radical scavenger experiments. (B) SEM images of BaTiO3 particles (i) before 9. R. Boulatov, Ed., Polymer Mechanochemistry (Springer, 2015).
and (ii) after ball milling (30 Hz; 60 min). (C) Thermographic image of the milling jar after the mechanoredox 10. S. L. James et al., Chem. Soc. Rev. 41, 413–447 (2012).
arylation of 2a with 1a. See supplementary materials for details. GC-MS, gas chromatography–mass spectrometry. 11. G.-W. Wang, Chem. Soc. Rev. 42, 7668–7700 (2013).
12. J.-L. Do, T. Friščić, ACS Cent. Sci. 3, 13–19 (2017).
a regular shape of the BaTiO3 particles before wrapped in a piece of a weighing paper and 13. J. G. Hernández, C. Bolm, J. Org. Chem. 82, 4007–4019
the reaction (approximate size, <75 mm) [Fig. placed in a zipper-locking plastic bag, followed
5B (i)]. Subsequently, the powder was sub- by striking with a hammer over 200 times. (2017).
jected for 60 min to ball milling at 30 Hz and Even under these crude conditions, the mech- 14. D. Tan, T. Friščić, Eur. J. Org. Chem. 2018, 18–33 (2018).
then analyzed by SEM. The resulting image anoredox borylation product 5i was obtained 15. J. L. Howard, Q. Cao, D. L. Browne, Chem. Sci. 9, 3080–3094
clearly shows a pronounced distortion of the in 43% yield, as assessed by nuclear magnetic
shape and a decrease in the size of the BaTiO3 resonance (NMR) integration. (2018).
particles [Fig. 5B (ii)]. These results suggest 16. J. Andersen, J. Mack, Green Chem. 20, 1435–1443 (2018).
that the mechanical stimulus provided by ball The present mechanoredox reactions can be 17. C. Bolm, J. G. Hernández, Angew. Chem. Int. Ed. 58,
milling is efficiently transferred onto the BaTiO3 carried out on gram scale without the use of
particles, which would result in the generation large amounts of dry and degassed organic 3285–3299 (2019).
of localized electrochemical potentials on the solvents in air, and the reactions do not re- 18. F. Gomollón-Bel, Chem. Int. 41, 12–17 (2019).
surface of the BaTiO3 particles that can be used quire special operating conditions. This oper- 19. J. C. Robertson, M. L. Coote, A. C. Bissember, Nat. Rev. Chem.
for the activation of the aryl diazonium salts. ational simplicity suggests that the present
approach may complement existing photo- 3, 290–304 (2019).
To investigate whether the friction during redox transformations in a practical and en- 20. J. Liang, J. M. Fernández, J. Am. Chem. Soc. 133, 3528–3534
ball milling generates a thermal effect, the tem- vironmentally friendly manner. Beyond the
perature inside the milling jar during the mech- immediate benefits of this protocol, our strat- (2011).
anoredox arylation of 2a with 1a was measured egy could be applicable to light-sensitive or 21. P. Dopieralski et al., Nat. Chem. 5, 685–691 (2013).
using thermography immediately after opening light-absorbing substrates that cannot be sub- 22. H. Yan et al., Nature 554, 505–510 (2018).
the jar (Fig. 5C). The crude mixtures were pre- jected to conventional photoredox systems. 23. S. Ikeda et al., Phys. Chem. Chem. Phys. 1, 4485–4491 (1999).
pared under optimized conditions. The obtained 24. M. Hara et al., Appl. Catal. A Gen. 190, 35–42 (2000).
image showed that the temperature after aryla- REFERENCES AND NOTES 25. Z. L. Wang, J. Song, Science 312, 242–246 (2006).
tion in the ball mill was around 30°C, which 26. X. Wang, J. Song, J. Liu, Z. L. Wang, Science 316, 102–105
discounts the possibility of thermal activa- 1. C. Stephenson, T. Yoon, D. W. C. MacMillan, Visible Light
tion of the aryldiazonium salts generating aryl Photocatalysis in Organic Chemistry (Wiley, ed. 1, 2018). (2007).
radical species by the heat provided from ball 27. Y. Qin, X. Wang, Z. L. Wang, Nature 451, 809–813 (2008).
milling. 2. C. K. Prier, D. A. Rankic, D. W. C. MacMillan, Chem. Rev. 113, 28. K.-S. Hong, H. Xu, H. Konishi, X. Li, J. Phys. Chem. Lett. 1,
5322–5363 (2013).
To demonstrate the robustness of the mech- 997–1002 (2010).
anoredox transformations, we conducted the 3. N. A. Romero, D. A. Nicewicz, Chem. Rev. 116, 10075–10166 (2016). 29. H. Mohapatra, M. Kleiman, A. P. Esser-Kahn, Nat. Chem. 9,
borylation of 1k with BaTiO3 in air using a 4. C.-S. Wang, P. H. Dixneuf, J.-F. Soulé, Chem. Rev. 118,
hammer (fig. S8 and movie S1). First, the re- 135–139 (2017).
action mixture was prepared by gentle grind- 7532–7585 (2018). 30. M. B. Starr, J. Shi, X. Wang, Angew. Chem. Int. Ed. 51,
ing in a mortar. Subsequently, the mixture was 5. K. L. Skubi, T. R. Blum, T. P. Yoon, Chem. Rev. 116,
5962–5966 (2012).
10035–10074 (2016). 31. M. B. Starr, X. Wang, Sci. Rep. 3, 2160 (2013).
6. T. Friščić, J. Mater. Chem. 20, 7599–7605 (2010). 32. H. Lin et al., Appl. Phys. Lett. 104, 162907 (2014).
7. S.-E. Zhu, F. Li, G.-W. Wang, Chem. Soc. Rev. 42, 7535–7570 33. W. Lv et al., J. Chem. Technol. Biotechnol. 92, 152–156 (2017).
34. M. B. Starr, X. Wang, Nano Energy 14, 296–311 (2015).
(2013). 35. Z. Wang et al., ACS Macro Lett. 8, 161–165 (2019).
8. P. Baláž, Mechanochemistry in Nanoscience and Minerals 36. K. Matyjaszewski, Macromolecules 45, 4015–4039 (2012).
37. D. P. Hari, P. Schroll, B. König, J. Am. Chem. Soc. 134,
Engineering (Springer, 2008).
2958–2961 (2012).
38. J. Yu, L. Zhang, G. Yan, Adv. Synth. Catal. 354, 2625–2628 (2012).
39. C. P. Andrieux, J. Pinson, J. Am. Chem. Soc. 125, 14801–14806

(2003).
40. J. H. Haeni et al., Nature 430, 758–761 (2004).
41. T. Friščić, S. L. Childs, S. A. A. Rizvi, W. Jones, CrystEngComm

11, 418–426 (2009).
42. M. E. Cinar, T. Ozturk, Chem. Rev. 115, 3036–3140 (2015).
43. Y. Segawa, H. Ito, K. Itami, Nat. Rev. Mater. 1, 15002 (2016).
44. Y. Segawa, T. Maekawa, K. Itami, Angew. Chem. Int. Ed. 54,

66–81 (2015).

ACKNOWLEDGMENTS

Funding: This work was supported by the Japan Society for the
Promotion of Science (JSPS) through KAKENHI grants 18H03907,
17H06370, and 19K15547; by the JST CREST grant number
JPMJCR19R1; and by the Institute for Chemical Reaction Design and
Discovery (ICReDD), which was established by the World Premier
International Research Initiative (WPI), MEXT, Japan. Y.P. thanks the
Otsuka Toshimi Scholarship Foundation for a scholarship. We thank
D. F. Toste and T. Shimada for advice on the preparation of this
manuscript. We thank Nippon Chemical Industrial Co., Ltd. for the
gift of piezoelectric materials. Author contributions: H.I. came up with
the original idea; K.K. and H.I. directed the project; K.K. and H.I.
designed the experiments; K.K. and Y.P. performed the experiments;
A.M. prepared ceramic materials used in this study; and K.K. and H.I.
wrote the manuscript with feedback from the other authors.
Competing interests: H.I. and K.K. are inventors on patent application
JP, 2019-163323 submitted by Hokkaido University that covers
mechanoredox reactions using piezoelectric materials and
manufacturing methods using the reaction setup with the ball mill.
The authors declare no other competing interests. Data and materials
availability: All data are available in the supplementary materials.

SUPPLEMENTARY MATERIALS

science.sciencemag.org/content/366/6472/1500/suppl/DC1
Materials and Methods
Figs. S1 to S9
Tables S1 to S5
NMR Spectra
References (45–58)
Movie S1

22 July 2019; accepted 30 October 2019
10.1126/science.aay8224

Kubota et al., Science 366, 1500–1504 (2019) 20 December 2019 5 of 5

RESEARCH

SOLID-STATE PHYSICS the nodes of the triangular pore lattice con-
nected to three links. The islands shrink, and
Intermediate bosonic metallic state in the the links become more resistive, with increas-
superconductor-insulator transition ing etching time (Fig. 1C). The behavior of the
measured RS(T) curves, where RS is the sheet
Chao Yang1*, Yi Liu2,3*, Yang Wang1, Liu Feng1, Qianmei He1, Jian Sun2,3, Yue Tang2,3, Chunchun Wu1, resistance and T is the temperature, of the
Jie Xiong1†, Wanli Zhang1, Xi Lin2,3,4,5, Hong Yao3,6,7, Haiwen Liu8, Gustavo Fernandes9, Jimmy Xu9,10, nanopatterned YBCO thin films is consistent
James M. Valles Jr.10†, Jian Wang2,3,4,5,7†, Yanrong Li1 with a quantum phase transition (QPT) (Fig.
1D). The YBCO thin film subjected to the
Whether a metallic ground state exists in a two-dimensional system beyond Anderson localization minimal RIE time remains superconducting
remains an unresolved question. We studied how quantum phase coherence evolves across (Fig. 1D). The superconducting onset critical
superconductor–metal–insulator transitions through magnetoconductance quantum oscillations temperature Tconset is defined as the temper-
in nanopatterned high-temperature superconducting films. We tuned the degree of phase ature at which the sheet resistance RS(T) curve
coherence by varying the etching time of our films. Between the superconducting and insulating deviates from the linear extrapolation of the
regimes, we detected a robust intervening anomalous metallic state characterized by saturating normal state (Fig. 1D and fig. S3), and the
resistance and oscillation amplitude at low temperatures. Our measurements suggest that the normal state sheet resistance is defined as
anomalous metallic state is bosonic and that the saturation of phase coherence plays a prominent RN ¼ RSðTconsetÞ. With increasing etching time,
role in its formation. the superconducting transition region be-
comes wider, and the zero-resistance state
T he scaling theory of localization predicts phase coherence could be key to understand- fades away. We define this state as the anom-
that metallic states are absent at zero ing the 2D metallic state. If the phase co- alous metal (AM) state because there is a
herence length of electrons saturates at low plateau of residual resistance at lower tem-
temperature in two-dimensional (2D) temperatures, owing to the weakening of con- peratures. The sheet resistance of a repre-
structive quantum interference, the diffusion sentative film (AM1) is shown in Fig. 1D. For
disordered systems because of quantum should not be absolutely prohibited, result- films with the normal resistance larger than
interference (1, 2). Consistently, when ing in a possible metallic state in a low- the critical value RS ≈ 13 kilohms, RS(T) has a
electrons form Cooper pairs in 2D disordered dimensional electron system (27–29). Thus, it positive slope (dRS/dT > 0) below Tconset that
is crucial to investigate the phase coherence of turns negative (dRS/dT > 0) at lower temper-
systems, the ground state should be a super- Cooper pairs in the anomalous metallic state atures, indicating an insulating transition;
of 2D superconductors. we define this as a transitional state (TS). A
conductor or an insulator. Nevertheless, ap- typical TS curve is marked in Fig. 1D. The
Here, we present a systematic study of samples with more etching exhibit a more
proaching zero temperature, an anomalous Cooper pair phase coherence through the pronounced insulating upturn emerging at
superconductor–anomalous metal phase transi- higher temperatures. With further etching,
metallic state has been experimentally ob- tion followed by the anomalous metal-insulator RS(T) increases monotonically (dRS/dT > 0)
served (3–15) that shows a residual resistance phase transition in high-temperature (Tc) toward infinity at zero temperature, indicat-
far below the quantum resistance at low tem- superconductor (HTS) films patterned with ing an insulating property over the whole
an array of holes. temperature range. An example labeled as
peratures. The existence and origin of this un- INS (insulating state) is shown in Fig. 1D.
We used reactive ion etching (RIE) to etch a More details of electrical transport of all
expected metallic state are intensely debated high-quality, 12-nm-thick YBa2Cu3O7-x (YBCO) films can be found in fig. S3.
(16–25). thin film through a contact mask with a tri-
angular array of holes (fig. S1). Each mask is The characteristics of the anomalous metal
The resistance saturation as a characteristic an anodized aluminum oxide (AAO) mem- state in HTS films with an array of nanopores
brane formed on and then separated from an are shown in Fig. 2. The Arrhenius plot of the
of an anomalous metallic state in 2D super- aluminum base (Fig. 1A). The approach offers anomalous metallic–state films and supercon-
conductors (3–15) has been experimentally ob- the advantages of minimal molecular residuals ducting films are presented in Fig. 2A. For
served in the ultralow temperature regime. and processing contamination (30–32). superconducting films, the resistance below
Tconset drops to zero within the measurement
However, extrinsic effects such as noise and The RIE transfers the pattern, which has resolution. For metallic-state films, in the low-
~70-nm-diameter holes with center-to-center temperature regime, the resistance drops and
carrier overheating may also disrupt phase hole spacing of ~103 nm, onto the HTS film saturates to a finite value that can be up to five
(Fig. 1B). In the process, the ion bombardment orders of magnitude smaller than that of the
coherence in superconductors, which casts impacts the sidewalls of the resultant pores to normal-state resistance. In the smallest saturat-
create a less crystalline and strained YBCO ing resistance film, the saturation of resistance
doubt on the existence of 2D metallic ground shell, which becomes thicker with increasing starts at ~5 K, which is much higher than that
states in superconducting systems (26). The etching time (32, 33). This was confirmed in of conventional superconductors (4–8, 10–15).
a detailed high-resolution transmission elec- To exclude external noise effects, we measured
1State Key Laboratory of Electronic Thin Films and Integrated tron microscopy (HRTEM) analysis, which a metallic-state film using well-filtered electri-
Devices, University of Electronic Science and Technology of showed a high density of structural defects cal leads in a dilution refrigerator cryostat
China, Chengdu 610054, China. 2International Center for around the nanopatterned holes (34). Thus, down to 50 mK (Fig. 2B). The current-voltage
Quantum Materials, School of Physics, Peking University, the normal state resistance grows with in- relation (I-V) curves are linear below 100 nA,
Beijing 100871, China. 3Collaborative Innovation Center of creasing etching time (fig. S2). We deduced indicating that the measurements are in the
Quantum Matter, Beijing 100871, China. 4CAS Center for from the transport data presented below that ohmic regime (Fig. 2B, inset). The resistance
Excellence in Topological Quantum Computation, University this process forms superconducting islands at saturation remains almost the same with or
of Chinese Academy of Sciences, Beijing 100190, China.
5Beijing Academy of Quantum Information Sciences, Beijing
100193, China. 6Institute for Advanced Study, Tsinghua
University, Beijing 100084, China. 7State Key Laboratory of
Low Dimensional Quantum Physics, Tsinghua University,
Beijing 100084, China. 8Center for Advanced Quantum
Studies, Department of Physics, Beijing Normal University,
Beijing 100875, China. 9School of Engineering, Brown
University, 182 Hope Street, Providence, RI 02912, USA.
10Department of Physics, Brown University, 182 Hope Street,
Providence, RI 02912, USA.
*These authors contributed equally to this work.
†Corresponding author. E-mail: [email protected]
(J.W.); [email protected] (J.M.V.); [email protected].
cn (J.X.)

Yang et al., Science 366, 1505–1509 (2019) 20 December 2019 1 of 5

RESEARCH | REPORT

without the resistor-capacitor (RC) filters, dem- sentative metallic-state film (Fig. 2C, inset). netoresistance is simultaneously observed as
Shown in Fig. 2C is the temperature depen-
onstrating that the metallic state is intrinsic. dence of Rxy, which becomes zero within the another important feature of this anomalous
measurement resolution at a relatively high metallic state (Fig. 2D) (13). The observed zero
To further explore the nature of the anom- temperature of ~50 K, whereas Rxx is still Hall resistance and giant positive magneto-
finite. The absence of Hall resistance at low tem-
alous metallic state, the Hall resistance (Rxy) peratures reveals the emergence of particle- resistance can be further confirmed in another
and longitudinal resistance (Rxx) were mea- hole symmetry in the anomalous metal. Fur- representative anomalous metallic–state film
sured simultaneously. Below Tconset, the slopes thermore, a giant positive longitudinal mag- (AM10) (fig. S4).
of Rxy(H) curves, where H is the magnetic field
strength, are largely suppressed and drop to Below Tconset, the perpendicular magneto-
conductance of all the YBCO nanopore thin
zero with decreasing temperature for a repre-

Fig. 1. Superconductor–anomalous
metal–insulator transition in
nanopore-modulated YBCO thin films.
(A) Illustration of the fabricating process
of nanopore-modulated YBCO thin films
by means of reactive ion etching patterned
by AAO (41). (B) Scanning electron microscopy
image of etched nanopore YBCO films along with
the AAO pattern. (C) Illustration of etched nanopore
YBCO films. The gray halos indicate the parts
weakened by the ion bombardment and etching;
blue triangles indicate the superconducting
islands. Cooper pairs are indicated with purple
sphere pairs. (D) Temperature dependence
of resistance for the etched nanopore YBCO films
for different etching times, which increase from
the bottom to the top curve, ranging from 20
to 160 s. The nanopore YBCO films undergo a
QPT by tuning the etching time. The resistance
of four representative films marked as
superconducting (SC), anomalous metallic
(AM1), transitional (TS), and insulating (INS)
are shown in black.

Fig. 2. The characteristics of anomalous metallic
states. (A) Arrhenius plot of all the anomalous
metallic–state films (AM1-AM7) and superconducting
films. (B) Arrhenius plot of longitudinal resistance
(Rxx) down to 50 mK, with and without filters of an
anomalous metallic–state film (AM8). (Inset) I-V
curve of the anomalous metallic–state film. The good
linearity below 100 nA indicates that the measure-
ments are within the ohmic region. (C) Arrhenius
plot of Hall resistance (Rxy) and longitudinal
resistance (Rxx) of a representative anomalous
metallic–state film (AM9). The Hall resistance (Rxy)
goes to zero at ~50 K. (Inset) Hall resistance
(Rxy) of a representative anomalous metallic–state
film. (D) Giant positive magnetoresistance of a
representative anomalous metallic–state film (AM9).

Yang et al., Science 366, 1505–1509 (2019) 20 December 2019 2 of 5

RESEARCH | REPORT

films in Fig. 1D oscillates at low fields (Fig. 3, contrast, the oscillation amplitude saturates effects in the multiply connected geometry
in the anomalous metallic region below 5 K imposed by the array structure (36, 37). These
A to C) for three representative films. We (Fig. 3D), which is reminiscent of the residual effects enter through the Josephson coupling
resistance plateau of the anomalous metallic
present the data as the negative change of state at low temperatures (Fig. 2A). For in- energy between nodes of the array, as given
magnetoconductance –DG = G0 – G(H), where sulating films (INS films), Gosc peaks and then by EJ = –Jcos(fi – fj –Aij), where fi – fj is
G0 is the conductance at zero magnetic field. drops with decreasing temperature. With in- the gauge invariant phase difference of the
The magnetoconductance of all films is sym- creasing disorder, Gosc varies by at least nine order parameter between islands i and j, and
orders of magnitude from the supercon- Aij ¼ 2e=ħ∫ij A Á dl is the line integral of the
metric, and the magnetoconductance oscil- ducting films to most insulating films in the vector potential between neighboring nodes
zero-temperature limit, whereas the normal (ħ is Planck’s constant h divided by 2p). J is
lations appear on a monotonically rising resistance only changes by two orders of mag- proportional to the order parameter amplitude
nitude. This finding demonstrates that the on the islands. Josephson coupling energy EJ
background (full data are shown in fig. S5). Cooper pair coherence can be controlled over oscillates with increasing magnetic field be-
a wide range by tuning the link resistance of
Up to four magnetoconductance oscillations nanopatterned YBCO films. cause of the requirement that the line inte-
were detected with a constant period of H0 ~
2.25 kOe. This period is consistent with one The electrical transport and the magneto- gral of the phase around a closed loop be a
superconducting flux quantum F0 = h/2e conductance oscillations in YBCO nanopore multiple of 2p. The dimensions of the array
threading an area of one unit cell of the thin films can be qualitatively understood by unit dictate the period of the oscillations. The
nanopattern (around 9200 nm2), where h is modeling them as resistively shunted Josephson
Planck’s constant and e is the electron charge. junction arrays. With increasing etching time, magnetoconductance oscillates on the super-
The appearance of the charge 2e quantum the link resistances grow, and the supercon- conducting side because EJ is the barrier to
oscillations just below Tconset demonstrate ducting islands shrink (35), which drives the thermally activated vortex motion (37); on the
that Cooper pairs participate in the transport system through a superconductor–anomalous insulating and anomalous metal side, the mag-
in all the states through the superconductor– metal–insulator transition. For all films, Tconset netoconductance oscillates because EJ is pro-
metal–insulator transitions. varies by no more than 15% from the super- portional to the Cooper pair tunneling rate.
conducting film to the insulating film, which
To highlight the temperature dependence indicates that the strength of local supercon- We now discuss the evolution of the tem-
ducting pairing remains stable through the
of the oscillations, we define the magneto- QPT, whereas Cooper pair phase coherence perature dependence of the oscillations across
conductance oscillation amplitude as Gosc ¼ can be tuned by increasing the link resistance.
jGð1=2H0Þ À G0j after subtracting the rising the QPT within this array model. For all the
background (fig. S6). The temperature de- The magnetoconductance oscillations orig-
pendence of Gosc is shown in Fig. 3D on a inate from Cooper pair quantum interference YBCO films, the oscillations appear just below
logarithm scale for the films that exhibit super- Tconset , indicating that local phase coherence
develops as soon as the amplitude of the
conducting, anomalous metallic, and insulat-
order parameter forms on the islands. At
ing states. For the films in the superconducting
lower temperatures, the oscillation ampli-
state, the oscillation amplitude monotonically
tude is approximately proportional to the
grows to a giant value beyond our measure-
phase coherence as the growth of the order
ment limit with decreasing temperature. By

Fig. 3. The evolution of
Cooper pairs coherence
through the QPT. (A to
C) Perpendicular negative
change of magnetoconduc-
tance –DG/GQ of three repre-
sentative YBCO nanopore
films at various temperatures.
Shown are the data from a
(A) superconducting-state film
(SC), (B) anomalous metallic–
state film (AM1), and (C)
insulating-state film (INS). Up
to four magnetoconductance
oscillations are detected
with a constant period of
H0 ~ 2.25 kOe, corresponding
to one superconducting
flux quantum per unit
cell of the nanopattern.
(D) Temperature dependence
of magnetoconductance
oscillations amplitude of all the
YBCO films on a logarithmic
scale. For the anomalous
metallic–state films, Gosc saturates with decreasing temperature at ~5 K. By contrast, Gosc diverges at low temperatures for superconducting–state films, and
Gosc peaks and drops at low temperatures for insulating–state film. Three representative YBCO films are shown in black. (E) Deduced phase coherence length of
insulating and some metallic-state films with Gosc smaller or comparable with quantum conductance. In the anomalous metallic regime, the phase coherence
length Lf saturates at lower temperatures; in the insulating regime, the phase coherence length Lf decreases with decreasing temperature.

Yang et al., Science 366, 1505–1509 (2019) 20 December 2019 3 of 5

RESEARCH | REPORT

Fig. 4. Phase diagram of disorder tuned QPT in
nanopatterned high-temperature superconductors.
(A and B) Color-scaled maps of the (A) resistance
and (B) differential resistance as a function of
temperature and the normal-state resistance. The
normal state and other states are separated by Tconset
(top dashed line). Below Tconset, with increasing the normal-
state resistance, the system is initially in the SC phase
(zero resistance state within the instrument resolution), then
AM phase (residual resistance plateau at low temperatures),
and last, CPI phase (Cooper pair localized insulator).
When approaching zero temperature, with increasing normal
resistance, the system evolves from AM to CPI, separated
by a quantum critical point around RN ~ 13 kilohms.

parameter on the islands ceases. For super- related to the etching time (fig. S2) and in- REFERENCES AND NOTES

conducting films, the oscillation amplitude creases with increasing disorder in the nano- 1. E. Abrahams, P. W. Anderson, D. C. Licciardello,
patterned films (35). The normal state and T. V. Ramakrishnan, Phys. Rev. Lett. 42, 673–676
monotonically grows beyond our measurement other states are separated by Tconset . At the (1979).
boundary from dRs/dT > 0 to dRs/dT = 0, the
range with decreasing temperature, indicating array resistance drops below Tconset to zero 2. B. L. Altshuler, A. G. Aronov, in Electron–Electron Interactions in
(SC) or a plateau of residual resistance (AM) Disordered Systems A. L. Efros, M. Pollak, Eds. (North-Holland,
that the phase coherence length diverges in 1985).
at the lowest temperatures, as shown in the
the zero-temperature limit. By contrast, the os- 3. H. M. Jaeger, D. B. Haviland, B. G. Orr, A. M. Goldman,
bottom-left part of the phase diagram (Fig. 4). Phys. Rev. B 40, 182–196 (1989).
cillation amplitude of the anomalous metallic–
The short dashed line in Fig. 4 represents the 4. D. Ephron, A. Yazdani, A. Kapitulnik, M. R. Beasley, Phys. Rev. Lett.
state films saturates below ~5 K after a steep 76, 1529–1532 (1996).
crossover between the AM state and the re-
growth. The saturating temperature depen- 5. N. Mason, A. Kapitulnik, Phys. Rev. Lett. 82, 5341–5344
gion where resistance drops rapidly. Below (1999).
dence of the oscillation amplitude is consistent Tconset , the onset of insulating behavior is vis-
ible in the lower right region in Fig. 4, where 6. C. Christiansen, L. M. Hernandez, A. M. Goldman, Phys. Rev.
with that of the resistance, indicating that dRs/dT is negative. We label this region as the Lett. 88, 037004 (2002).
Cooper pair insulator (CPI). Approaching zero
phase coherence saturation should be crucial temperature, with increasing RN, there is a 7. S. Eley, S. Gopalakrishnan, P. M. Goldbart, N. Mason, Nat. Phys.
regime where dRs/dT changes from zero to a 8, 59–62 (2012).
to the formation of the anomalous metallic negative value, which can be interpreted as
8. W. Liu et al., Phys. Rev. Lett. 111, 067003 (2013).
state. For the insulating films, as the link re- the critical regime of a quantum phase tran- 9. J. Garcia-Barriocanal et al., Phys. Rev. B 87, 024509
sistance rises beyond RQ = h/4e2 with decreas-
sition. The phase diagram shows the evolution (2013).
ing temperature, the inter-island capacitance 10. Z. Han et al., Nat. Phys. 10, 380–386 (2014).
of an anomalous metallic state to a Cooper 11. Y. Saito, Y. Kasahara, J. Ye, Y. Iwasa, T. Nojima, Science 350,
and self-capacitances of the islands can pro-
pair insulating state separated by a QCP around 409–413 (2015).
duce a Coulomb blockade to Cooper pair trans- RN ~ 13 kilohms. 12. N. P. Breznay, A. Kapitulnik, Sci. Adv. 3, e1700612 (2017).
13. A. Kapitulnik, S. A. Kivelson, B. Spivak, Rev. Mod. Phys. 91,
port (38). Thus, Gosc and the phase coherence To date, the nature of the anomalous metal-
length decrease as the Cooper pairs become 011002 (2019).
lic state in 2D systems remains a mystery. 14. C. G. L. Bøttcher et al., Nat. Phys. 14, 1138–1144 (2018).
localized. In order to quantify phase coherence 15. Z. Chen et al., Nat. Commun. 9, 4008 (2018).
Candidate theories invoke coupling to a dis- 16. E. Shimshoni, A. Auerbach, A. Kapitulnik, Phys. Rev. Lett. 80,
through the QPT, we estimated the phase co- sipative bath (16–18), a Bose metal phase (19–21),
order parameter fluctuations (22–24), and a 3352–3355 (1998).
herence length Lf by u sing the formula composite Fermi liquid phase (25). In our HTS 17. A. Kapitulnik, N. Mason, S. A. Kivelson, S. Chakravarty,
1:5 quasi-particle system, this anomalous metallic state is clearly
Gosc ¼ 4e2 exp À pr for a bosonic state demonstrated by the h/2e Phys. Rev. B 63, 125322 (2001).
h Lf Lf periodic quantum oscillations and zero Hall 18. V. M. Galitski, G. Refael, M. P. A. Fisher, T. Senthil, Phys. Rev. Lett.

pr resistance, which appears to rule out the sce- 95, 077002 (2005).
narios that are based on charge e carriers. The 19. D. H. Lee, S. Kivelson, S. C. Zhang, Phys. Rev. Lett. 67,
quantum interference (29, 39), where r is half phase coherence saturation is argued to be
3302–3305 (1991).
of the center-to-center hole spacing ~50 nm. the key to understanding the low-dimensional 20. D. Das, S. Doniach, Phys. Rev. B 60, 1261–1275 (1999).
metallic state at low temperatures (27–29, 40). 21. P. Phillips, D. Dalidovich, Science 302, 243–247 (2003).
The above formula can only be applied to in- Investigations of Cooper pair phase coherence 22. M. Feigel’man, A. Larkin, Chem. Phys. 235, 107–114

sulating or some metallic-state films with rela- may be the right approach for revealing the (1998).
23. B. Spivak, A. Zyuzin, M. Hruska, Phys. Rev. B 64, 132502
tively large residual resistance when Gosc is nature of this anomalous metallic state. Fur-
smaller or comparable with quantum con- (2001).
thermore, our work on this HTS also paves 24. B. Spivak, P. Oreto, S. A. Kivelson, Phys. Rev. B 77, 214523
ductance. The temperature dependence of Lf
are presented in Fig. 3E, indicating a satura- the way for investigations on the anomalous (2008).
25. M. Mulligan, S. Raghu, Phys. Rev. B 93, 205116 (2016).
tion of quantum coherence for anomalous state of matter by using finite frequency 26. I. Tamir et al., Sci. Adv. 5, eaau3826 (2019).
27. P. Mohanty, E. M. Q. Jariwala, R. A. Webb, Phys. Rev. Lett. 78,
metallic states in the low-temperature regime. probes, such as terahertz spectroscopy, that
3366–3369 (1997).
The magnetoconductance oscillation ampli- are more difficult to apply at dilution refrig- 28. J. J. Lin, J. P. Bird, J. Phys. Condens. Matter 14, 201

tude observed in the anomalous metallic states erator temperatures. (2002).
29. F. Pierre et al., Phys. Rev. B 68, 085413 (2003).
with small residual resistance is giant and 30. J. Liang, H. Chik, A. Yin, J. Xu, J. Appl. Phys. 91, 2544–2546

much larger than the quantum conductance, (2002).

which is difficult to explain by using only quasi- 31. H. Chik, J. M. Xu, Mater. Sci. Eng. Rep. 43, 103–138
(2004).
particle quantum interference, leaving this an
32. S. G. Cloutier, P. A. Kossyrev, J. Xu, Nat. Mater. 4, 887–891
open question for further investigation. (2005).

The phase diagrams of resistance and dif- 33. S. A. Cybart et al., Nat. Nanotechnol. 10, 598–602 (2015).
34. S. G. Cloutier, C. H. Hsu, P. A. Kossyrev, J. Xu, Adv. Mater. 18,
ferential resistance as a function of temper-
841–844 (2006).
ature and normal resistance RN in Fig. 4 35. J. M. Valles Jr. et al., Phys. Rev. B 39, 11599–11602 (1989).
provide a summary of the results. We use RN
as the tuning parameter because it is directly

Yang et al., Science 366, 1505–1509 (2019) 20 December 2019 4 of 5

RESEARCH | REPORT

36. M. D. Stewart Jr., A. Yin, J. M. Xu, J. M. Valles Jr., Science 318, Y. Xing for the fruitful discussions. Funding: This work was supported instruction from J.M.X. and G.F.; H.L. and H.Y. gave the theoretical
1273–1275 (2007). by the National Natural Science Foundation of China (grants analysis; C.Y., Y.Liu, J.W., J.X., and J.M.V. analyzed the data, with
11888101, 51722204, 11774008, 91421110, 11474175, 11674028, and the contribution of H.L.; C.Y., Y.Liu, J.W., J.X., J.M.X., and J.M.V.
37. I. Sochnikov, A. Shaulov, Y. Yeshurun, G. Logvenov, I. Bozović, 11825404), the National Key Research and Development Program wrote the manuscript, with input from all authors. Competing
Nat. Nanotechnol. 5, 516–519 (2010). of China (grants 2018YFA0305604, 2015CB358600, interests: The authors declare no competing interests. Data and
2017YFA0303300, 2016YFA0301001, and 2017YFA0304600), materials availability: All experimental data shown in the main
38. Y. Takahide, H. Miyazaki, A. Kanda, Y. Ootuka, J. Phys. Soc. Fundamental Research Funds for the Central Universities text and in the supplementary materials are accessible at the
Jpn. 72, 96–99 (2003). (ZYGX2016Z004), the Strategic Priority Research Program of Chinese Zenodo repository (42).
Academy of Sciences (grant XDB28000000), the Open Research
39. E. Akkermans, G. Montambaux, Mesoscopic Physics of Fund Program of the State Key Laboratory of Low-Dimensional SUPPLEMENTARY MATERIALS
Electrons and Photons (Cambridge Univ. Press, 2007). Quantum Physics, Tsinghua University (grant KF201703), Beijing
Natural Science Foundation (Z180010), Beijing Municipal Science science.sciencemag.org/content/366/6472/1505/suppl/DC1
40. S. Washburn, R. A. Webb, Adv. Phys. 35, 375–422 (1986). and Technology Commission (Z181100004218001), and China Materials and Methods
41. Materials and methods are available as supplementary Postdoctoral Science Foundation (grant 2019M650290). Author Supplementary Text
contributions: J.W., J.X., and J.M.V. conceived and instructed the Figs. S1 to S7
materials. research; C.Y. and Y.Liu performed the transport measurements,
42. C. Yang et al., Raw data for “Intermediate bosonic metallic with the assistance of Y.W., Q.H., C.W., J.S., Y.T., X.L., and J.W.; 4 April 2019; accepted 1 November 2019
C.Y., L.F., J.X., W.Z., and Y.Li fabricated the samples with the Published online 14 November 2019
state in the superconductor-insulator transition”. Zenodo 10.1126/science.aax5798
(2019); doi:10.5281/zenodo.3517821.

ACKNOWLEDGMENTS

We thank S. A. Kivelson, J. M. Kosterlitz, D. Oller, D. He, J. H. Kim,
D. J. Lee, K. Liu, J. Qi, Yudong Liu, Q. Li, Yanzhao Liu, and

Yang et al., Science 366, 1505–1509 (2019) 20 December 2019 5 of 5

RESEARCH

SOLAR CELLS As confirmed by x-ray photoelectron spec-
troscopy (XPS), the surface of the as-fabricated
Constructive molecular configurations for perovskite thin film synthesized by a two-step
surface-defect passivation of perovskite photovoltaics method was Pb-rich (fig. S1), and we focused
on the (100) surface with PbI2 termination in a
Rui Wang1*, Jingjing Xue1*†, Kai-Li Wang2*, Zhao-Kui Wang1,2†, Yanqi Luo3, David Fenning3, Pb-rich condition. The types of surface defects
Guangwei Xu1, Selbi Nuryyeva1,4, Tianyi Huang1, Yepin Zhao1, Jonathan Lee Yang5, Jiahui Zhu1, studied and their corresponding top-layer
Minhuan Wang1, Shaun Tan1, Ilhan Yavuz6†, Kendall N. Houk4†, Yang Yang1,7† view of atomic structures are shown in Fig. 1A.
Using the Dispersion Correction 3 (DFT-D3)
Surface trap–mediated nonradiative charge recombination is a major limit to achieving high-efficiency method, we calculated the defect formation
metal-halide perovskite photovoltaics. The ionic character of perovskite lattice has enabled molecular energies (DFEs) (table S1) of VPb, VI, PbI, and
defect passivation approaches through interaction between functional groups and defects. However, IPb on the surface to be 3.20, 0.51, 0.57, and
a lack of in-depth understanding of how the molecular configuration influences the passivation 3.15 eV, respectively. Compared with the values
effectiveness is a challenge to rational molecule design. Here, the chemical environment of a functional reported in bulk perovskite, VPb, VI, and IPb
group that is activated for defect passivation was systematically investigated with theophylline, caffeine, defects show similar DFEs (24), whereas the
and theobromine. When N-H and C=O were in an optimal configuration in the molecule, hydrogen-bond PbI antisite defect exhibited particularly lower
formation between N-H and I (iodine) assisted the primary C=O binding with the antisite Pb (lead) defect to formation energy than that in the bulk. Thus,
maximize surface-defect binding. A stabilized power conversion efficiency of 22.6% of photovoltaic the PbI antisite defect should form more read-
device was demonstrated with theophylline treatment. ily and predominate on the surface. We did
not consider VI further despite its DFE being
D efect passivation to reduce unproduc- density functional theory (DFT) calculations as low as that of PbI, because the interaction
tive charge recombination is an effective to compare the formation energies of selected of molecules with the VI turned out to be not
strategy to increase the power conver- native defects on the perovskite surface. Par- energy favorable (fig. S2).
sion efficiency (PCE) of polycrystalline ticularly taken into consideration were Pb-
metal-halide perovskite thin-film pho- and I-involving point defects, Pb vacancy (VPb), On the basis of these results, we focused on
tovoltaics (PVs) (1–6). The ionic nature of the I vacancy (VI), and Pb-I antisite (PbI and IPb, the interaction between the surface PbI anti-
perovskite lattice enables molecular passiva- corresponding to I site substitution by Pb and site defect and candidate molecules for defect
tion through coordinate binding based on Pb site substitution by I, respectively) because passivation. A set of small molecules sharing
Lewis acid-base chemistry (7–10). Organic mo- the band edges of perovskite were reported to the identical functional groups but with stra-
lecules containing various functional groups, be composed of Pb and I orbitals (22, 23). tegically varying chemical structure were in-
such as carbonyl groups, can passivate de- vestigated, namely theophylline, caffeine, and
fects (11–17). The selection of molecules with theobromine, interacting with the defects
optimal binding configurations for defect (Fig. 1B). These molecules are found in natural
passivation would benefit from molecular de-
sign rules.

We demonstrate high efficiencies for
(FAPbI3)x(MAPbBr3)1−x (where FA is for-
mamidinium and MA is methylammonium;
x is 0.92 in the precursor) perovskite PV de-
vices through defect identification followed
by rational design and comprehensive investi-
gation of the chemical environment around the
active functional group for defect passivation
(18). In high-quality perovskite polycrystalline
thin films that have monolayered grains (19–21),
interior defects of perovskite are negligible
compared with the surface defects. We used

1Department of Materials Science and Engineering and Fig. 1. Surface-defect identification and constructive configuration of the C=O group in three
different chemical environments. (A) Top view of the various types of surface defects. (B) Theoretical
California NanoSystems Institute, University of California Los models of perovskite with molecular surface passivation of PbI antisite with theophylline, caffeine, and
Angeles, CA 90095, USA. 2Institute of Functional Nano & theobromine. (C) J-V curves of perovskite solar cells with or without small-molecules treatment under reverse
scan direction.
Soft Materials (FUNSOM), Jiangsu Key Laboratory for

Carbon-Based Functional Materials & Devices, Soochow
University, Suzhou 215123, China. 3Department of

Nanoengineering, University of California, San Diego, La Jolla,
CA 92093, USA. 4Department of Chemistry and

Biochemistry, University of California Los Angeles, CA
90095, USA. 5College of Chemistry, University of California
Berkeley, CA 94720, USA. 6Department of Physics, Marmara
University, Ziverbey, Istanbul 34722, Turkey. 7School of

Engineering, Westlake University, Hangzhou 310024, China.

*These authors contributed equally to this work.

†Corresponding author. Email: [email protected] (J.X.);

[email protected] (Z.-K.W.); [email protected] (I.Y.);

[email protected] (K.N.H.); [email protected] (Y.Y.)

Wang et al., Science 366, 1509–1513 (2019) 20 December 2019 1 of 5

RESEARCH | REPORT

products (tea, coffee, and chocolate, respectively) devices with ITO/SnO2/perovskite/Spiro- I of PbI62− octahedron through a hydrogen
and are readily accessible. In these molecules, OMeTAD/Ag structure under reverse scan di- bond (H-bond), which strengthened the ab-
the conjugated structure as well as the dipoles
induced by the hetero atoms tend to increase rection [where ITO is indium tin oxide, SnO2 sorption of theophylline onto the PbI defect,
the intermolecular interaction. This renders is tin oxide, and Spiro-OMeTAD is 2,2′7,7′- resulting in an interaction energy (Eint, defined
them nonvolatile in nature, which is key to the tetrakis-(N,N-di-p-methoxyphenyl amine)-9,9′- as Emolecule-perovskite − Eperovskite − Emolecule) as
investigation of their interactions with defects spirobifluorene]. Current density-voltage (J-V) strong as −1.7 eV.
in perovskite and long-term stability of the curves of the PV devices with and without
devices. The xanthine core also helps maintain This observation suggested that the neigh-
the coplanarity of the carbonyl group and the theophylline treatment are compared in Fig.
N–H. Unlike other small molecules with flex- boring H-bond between the xanthine mole-
ible alkyl chains, this rigidity allows us to de- 1C and table S2. The control device showed an cule and the PbI62− octahedron can contribute
fine the configuration and distance between open-circuit voltage (VOC) of 1.164 V, a short- to the defect passivation. A methyl group was
the carbonyl group and N–H when they are in- circuit current (JSC) of 24.78 mA cm−2, and a fill
teracting with the defects, as a result of which factor (FF) of 72.88%, whereas the target device added to the N on the imidazole ring of the-
the constructive molecular configuration for showed a VOC of 1.191 V, a JSC of 25.24 mA cm−2,
defect passivation can be unraveled. and an FF of 78.11%. The enhancement in the ophylline (resulting in caffeine) to eliminate
VOC was attributed to the surface passivation the effect from H-bonding between the N–H
We incorporated theophylline onto the sur- by theophylline through the Lewis base-acid and I, leaving just the interaction with surface
face of perovskite thin film using a post-
treatment method, and a PCE enhancement interaction between C=O group and the antisite PbI defects (Fig. 1B). The missing H-bond be-
from 21.02% (stabilized 20.36%) to 23.48% tween N–H and PbI62− octahedron resulted in
(stabilized 22.64%) was observed in the PV Pb. As shown in the surface structure model a weakened interaction and a less favorable Eint
of −1.3 eV. Compared with the theophylline-
of perovskite with theophylline (Fig. 1B), the treated device, a caffeine-treated perovskite PV

C==O group on theophylline strongly interacted device had a lower PCE of 22.32% along with
with the antisite Pb. The neighboring N–H a lower VOC of 1.178 V, JSC of 25.04 mA cm−2,
on the imidazole ring also interacted with the and FF of 75.76%.

Fig. 2. Investigation of the interactions between surface defects and the small molecules. FTIR spectra of (A) pure theophylline and theophylline-PbI2 films,
(B) pure caffeine and caffeine-PbI2 films, and (C) pure theobromine and theobromine-PbI2 films. (D) PL spectra of perovskite films without and with small-molecules treatment.
(E) tDOS in perovskite solar cells with or without small-molecules treatment. (F) Nyquist plots of perovskite solar cells with or without small-molecules treatment measured
in the dark and at corresponding open-circuit voltages. a.u., arbitrary units; C, junction capacitance; Rrec, recombination resistance; Rs, Series Resistance.

Wang et al., Science 366, 1509–1513 (2019) 20 December 2019 2 of 5

RESEARCH | REPORT

Fig. 3. Characterization of perovskite films and interfaces with theophylline treatment. (A) XPS data for Pb 4f 7/2 and Pb 4f 5/2 core-level spectra
in perovskite films with or without theophylline treatment. (B) UPS spectra of perovskite films with or without theophylline treatment. (C) AFM and
KPFM images of perovskite films with (right) or without (left) theophylline treatment. (D) Time-resolved PL spectra of perovskite films before and after
depositing Spiro-OMeTAD without and with theophylline treatment. (E) Cross-section SEM images and the corresponding EBIC images and line profile
of the perovskite solar cells with (right) or without (left) theophylline treatment.

When the N–H group was located next to copy (FTIR). The C==O in pure theophylline shift of C==O stretching vibration frequency
the C==O group on the same six-membered showed a typical stretching vibration mode at from 1655 to 1620 cm−1 and the short distance
ring, producing a shorter distance between 1660 cm−1 that it shifted to 1630 cm−1 upon between O and Pb (Fig. 2C). However, this
the C=O and the N–H, in theobromine, the binding to PbI2 (Fig. 2A). The downward shift strong interaction was enabled by the free
spatially effective interaction between the of 30 cm−1 of the C==O stretching vibration rotation of PbI2, which resulted in a different
N–H and I was disabled as C==O was bound frequency resulted from the electron delocal- configuration than that in theophylline and
to antisite Pb, resulting in an even weaker ization in C==O when a Lewis base-acid adduct caffeine. Hence, when the configuration of
Eint of −1.1 eV (Fig. 1B). Although C==O and was formed, demonstrating a strong interac- PbI2 was fixed and had a 90° angle between
N–H are both present on the molecule, the tion between PbI2 and C==O in theophylline. Pb and I atom, like that on perovskite surface
lack of appropriate coordination of I to the The atomic distance between the O in C==O (the PbI62− octahedron), the N–H was in a po-
molecule led to a spatially destructive molec- and the Pb in PbI2, on the basis of theoretical sition that led to an unfavorable interaction
ular configuration. The theobromine-treated modeling, was as low as 2.28 Å. with I. This configuration would cause either
devices showed a decrease in PCE to 20.24% weakened interaction between the molecule
with a lower VOC of 1.163 V, JSC of 24.27 mA cm−2, When the H atom was replaced by a methyl and the perovskite surface or distorted PbI62−
and FF of 71.58% compared with the refer- group on the N of imidazole to eliminate the octahedron, resulting in the ineffectiveness of
ence device. This result emphasizes the im- effect of an H bond, the vibration frequency defect passivation and perhaps causing even
portance of the constructive configuration of of C==O in caffeine shifted only 10 cm−1 upon more defects through lattice distortion (fig. S3).
N–H and C==O groups that enable the coop- addition of PbI2, indicating a weakened inter-
erative multisite interaction and synergistic action between the C==O and PbI2 (Fig. 2B). The surface passivation effects of the three
passivation effect. The atomic distance between the correspond- molecules with different configuration were
ing O and Pb also increased to 2.32 Å. In the further studied by photoluminescence (PL).
We studied the variation in the C==O and the case of theobromine, when the N–H was in a The PL intensity increased noticeably with
PbI2-terminated perovskite surface interac- closer position to C=O, the interaction between the treatment by theophylline (Fig. 2D), im-
tion with different molecular configurations the molecule and PbI2 became comparable to plying the suppressed nonradiative charge-
using Fourier-transform infrared spectros- that in theophylline, as evidenced by the large recombination sites from defects (15). With

Wang et al., Science 366, 1509–1513 (2019) 20 December 2019 3 of 5

RESEARCH | REPORT

Fig. 4. Enhanced photovoltaic performance and long-term stability of perovskite solar cells with or without theophylline treatment, as shown in
perovskite solar cells with theophylline treatment. (A) J-V curves (A), recorded under simulated 1-sun AM1.5G illumination. (D) PCE distribution of
of perovskite solar cells with or without theophylline treatment. (B) EQE perovskite solar cells with or without theophylline treatment. (E) Evolution of the
curves of perovskite solar cells with or without theophylline treatment. PCEs measured from the encapsulated perovskite solar cells with or without
(C) Stabilized maximum power output and the photocurrent density at theophylline treatment exposed to continuous light (90 ± 10 mW cm−2) under
maximum power point as a function of time for the best-performing open-circuit condition.

the caffeine treatment, enhanced PL intensity skite. According to Fig. 2F, the device with difference indicated a less n-type surface after
was also observed, but not as strong as that theophylline surface treatment has the smallest theophylline treatment, which could improve
with the theophylline, suggesting a less effec- impedance, signifying a substantially sup- the hole extraction in devices.
tive passivation effect. For theobromine, how- pressed charge recombination at the interface,
ever, a decrease in PL intensity was observed which originated from the reduced surface- Atomic force microscopy (AFM) combined
compared with the reference material, which defect states. A larger impedance was observed with Kelvin probe force microscopy (KPFM)
can be attributed to the destructive molec- in caffeine-treated device, and an even larger was further applied to probe the effect of the-
ular configuration of the passivation agents impedance was measured in theobromine- ophylline on the surface morphology and sur-
causing more charge recombination sites. treated device. face potential. The theophylline-treated surface
exhibited a higher electronic chemical poten-
The trap density of states (tDOS) of the as- Further characterizations were performed tial than that of reference film, while keeping
fabricated devices were also deduced from the to better understand the perovskite interface the surface morphology unchanged (Fig. 3C).
angular frequency-dependent capacitance. with theophylline. High-resolution XPS pat- The transient PL of the perovskite films with
As shown in Fig. 2E, the tDOS as a function terns of the Pb 4f for the theophylline-treated hole-transporting layer (HTL) was compared
of the defect energy demonstrated a reduction film showed two main peaks located at 138.48 in Fig. 3D to delineate the carrier dynamics of
in trap states for theophylline- and caffeine- and 143.38 eV, corresponding to the Pb 4f 7/2 the devices. The perovskite film with theophyl-
treated perovskite compared with the reference and Pb 4f 5/2, respectively (Fig. 3A), whereas line treatment showed a slightly longer carrier
device. By contrast, theobromine treatment the reference film showed two main peaks at lifetime than the reference film, whereas a
induced more trap states, consistent with the 138.27 and 143.13 eV. The peaks from Pb 4f faster decay profile was observed when add-
decrease in PCE. The change in tDOS with shifted to higher binding energies in the film ing the HTL on top of the perovskite film. This
different surface treatments was also con- with theophylline surface treatment, indicat- result demonstrated a better hole extraction
firmed by theoretical modeling (fig. S4). In ing the interaction between the theophylline with theophylline treatment (20), most likely
addition, electrochemical impedance spectros- and the Pb on perovskite surface. We used arising from lesser recombination sites at the
copy (EIS) characterization was performed to ultraviolet photoelectron spectroscopy (UPS) interface and the slightly shallower work func-
demonstrate the carrier transport processes to measure the surface band structure with tion of the perovskite film with theophylline.
under illumination at the interface. The mid- and without the theophylline surface treat-
dle frequency zone of the EIS semicircle should ment. The work function was determined to The improved carrier dynamics originat-
be dominated by junction capacitance and be −4.77 and −4.96 eV with the valance band ing from the effective surface passivation by
recombination resistance related to the inter- maximum of −5.66 and −5.73 eV for reference theophylline was further characterized by
faces between transport materials and perov- and theophylline, respectively (Fig. 3B). This cross-sectional electron-beam–induced current
(EBIC) measurement. In EBIC measurement,

Wang et al., Science 366, 1509–1513 (2019) 20 December 2019 4 of 5

RESEARCH | REPORT

the electron-beam–excited carriers were col- (Fig. 4C). The histogram of PV efficiencies for 18. W. J. Yin, T. Shi, Y. Yan, Appl. Phys. Lett. 104, 063903 (2014).
lected on the basis of the collection probability 40 devices is shown in Fig. 4D (the detailed 19. W. S. Yang et al., Science 356, 1376–1379 (2017).
CP (x, Ld), where x is the distance between parameters are shown in table S2), which 20. N. J. Jeon et al., Nat. Energy 3, 682–689 (2018).
junction and incident beam position, and Ld confirms good reproducibility of the perform- 21. E. H. Jung et al., Nature 567, 511–515 (2019).
is the diffusion length of the carriers (fig. S5). ance improvement with theophylline (11.1% 22. M. A. Green, A. Ho-Baillie, H. J. Snaith, Nat. Photonics 8,
The device with theophylline treatment ex- improvement in an average PCE from 20.36 ±
hibited higher EBIC current compared with 0.53% to 22.61 ± 0.58% with the incorporation 506–514 (2014).
the reference device (Fig. 3E). The average of the theophylline). 23. Z. Xiao, Z. Song, Y. Yan, Adv. Mater. 31, e1803792 (2019).
intensity extracted from these EBIC maps 24. N. Liu, C. Yam, Phys. Chem. Chem. Phys. 20, 6800–6804
demonstrated a general increase in the EBIC The changes in PCE of the encapsulated
signal after treatment with theophylline (fig. devices at a relative humidity of 30 to 40% and (2018).
S6), indicating an enhanced carrier collection temperature of 40°C were tracked over time to 25. E. Edri et al., Nat. Commun. 5, 3461 (2014).
efficiency (25). Specifically, in Fig. 3E, a rep- test the long-term operational stability (Fig. 26. R. Wang et al., Adv. Funct. Mater. 29, 1808843 (2019).
resentative EBIC line profile of the reference 4E). The reference device (initial PCE 19.34%) 27. J. A. Christians et al., Nat. Energy 3, 68–74 (2018).
device showed a current decay from the HTL- degraded by more than 80% in 500 hours, 28. Y. Hou et al., Science 358, 1192–1197 (2017).
perovskite to the SnO2-perovskite interface. whereas the target device maintained >90%
The decay indicates that carrier collection was of its initial efficiency (21.32%) during this ACKNOWLEDGEMENTS
limited by the hole-diffusion length as the beam time. Also, as shown in fig. S7, the shelf stab-
position moved away from the HTL-perovskite ility of the device based on theophylline treat- Funding: Y.Y. acknowledges the Office of Naval Research (ONR)
interface. By contrast, the device with theophyl- ment was noticeably enhanced, maintaining (N00014-17-1-2,484) for their financial support. Part of this
line treatment displays minimal decay in the >95% of its original PCE (22.78%) when stored material is based upon work supported by the U.S. Department
perovskite layer in the EBIC line profile. This under ambient conditions with 20 to 30% of Energy’s Office of Energy Efficiency and Renewable Energy
difference suggests that a longer diffusion humidity at 25°C for 60 days. By contrast, the (EERE) under the Solar Energy Technologies Office Award Number
length of holes was present in theophylline- reference device lost >35% of its initial effi- DE- EE0008751. Z.-K.W. acknowledges the Natural Science
treated sample and balanced electron and ciency (20.67%). The strong interaction be- Foundation of China (no. 91733301). This project was also supported
that hole charge transport and collection was tween the theophylline and the surface defects by the Collaborative Innovation Center of Suzhou Nano Science and
achieved, which is likely the result of fewer likely suppressed deleterious ion migration Technology. Y.L. and D.F. are grateful for the financial support of
surface recombination sites (Fig. 3E). (26–28). a California Energy Commission Advance Breakthrough award
(EPC-16-050). This work was performed in part at the San Diego
Further assessment of the performance of REFERENCES AND NOTES Nanotechnology Infrastructure (SDNI) of UCSD supported by the
the PV devices based on the theophylline sur- National Science Foundation (grant ECCS-1542148). Part of the
face passivation was performed. The devices 1. J. Tong et al., Science 364, 475–479 (2019). computations were performed in the SIMULAB of Marmara University,
showed a negligible hysteresis (4.1%) (Fig. 4A) 2. H. Tan et al., Science 355, 722–726 (2017). Physics Department and in the UHEM cluster of Turkey. K.N.H. and
because of the balanced charge collection orig- 3. X. Zheng et al., Nat. Energy 2, 17102 (2017). S.N. are grateful to the National Science Foundation (CHE‐1764328)
inating from the effective surface passivation, 4. J. J. Yoo et al., Energy Environ. Sci. 12, 2192–2199 (2019). for financial support of this research. Computer time was provided
whereas the reference device showed a large 5. Q. Jiang et al., Nat. Photonics 13, 460–466 (2019). by the UCLA Institute for Digital Research and Education (IDRE).
hysteresis (up to 7.6%) (table S3). External 6. N. Li et al., Nat. Energy 4, 408–415 (2019). Author contributions: R.W., J.X., and Y.Y. conceived the idea for
quantum efficiency (EQE) spectra of the de- 7. J. S. Manser, J. A. Christians, P. V. Kamat, Chem. Rev. 116, the study. R.W. and J.X. fabricated the solar cell devices and
vices were compared in Fig. 4B. An integrated designed the experiments. K.-L.W. performed the film and device
JSC of 24.42 mA cm−2 from the target device 12956–13008 (2016). characterizations under the supervision of Z.-K.W. Y.L. and D.F.
matched well with the value measured from 8. H. Zhang et al., ACS Appl. Mater. Interfaces 10, 42436–42443 performed the EBIC measurement. G.X. carried out the tDOS
the J-V scan (<5% discrepancy), whereas the measurement. S.N., I.Y., and K.N.H. performed the DFT calculation.
control device showed an integrated JSC of (2018). T.H. carried out the TPC and TPV measurement. Y.Z., J.L.Y.,
23.56 mA cm−2. A stabilized PCE of 22.64% 9. J. W. Lee, H. S. Kim, N. G. Park, Acc. Chem. Res. 49, 311–319 J.Z., M.W., and S.T. assisted with the device fabrication and
was achieved with the target device when characterizations. R.W., J.X., and Y.Y. wrote the manuscript. All
biased at 1.00 V, whereas that of the control (2016). authors discussed the results and commented on the manuscript.
device was 20.36% when biased at 0.98 V 10. Y. Zong et al., Chem 4, 1404–1415 (2018). Y.Y. supervised the project. Competing interests: None declared.
11. D. Bi et al., Nat. Energy 1, 16142 (2016). Data and materials availability: All (other) data needed to
12. B. Chen, P. N. Rudd, S. Yang, Y. Yuan, J. Huang, evaluate the conclusions in the paper are present in the paper
or the supplementary materials.
Chem. Soc. Rev. 48, 3842–3867 (2019).
13. T. Wu et al., Adv. Energy Mater. 9, 1803766 (2019). SUPPLEMENTARY MATERIALS
14. T. Niu et al., Adv. Mater. 30, e1706576 (2018).
15. R. Wang et al., Joule 3, 1464–1477 (2019). science.sciencemag.org/content/366/6472/1509/suppl/DC1
16. W. Xu et al., Nat. Photonics 13, 418–424 (2019). Materials and Methods
17. H. Zhang, M. K. Nazeeruddin, W. C. H. Choy, Adv. Mater. 31, Supplementary Text
Figs. S1 to S13
e1805702 (2019). Tables S1 to S3
References (29–41)

1 August 2019; accepted 8 November 2019
10.1126/science.aay9698

Wang et al., Science 366, 1509–1513 (2019) 20 December 2019 5 of 5

RESEARCH

ORGANIC CHEMISTRY Over several decades, numerous companies
all over the world, including BASF, Dupont,
Direct synthesis of adipic acid esters via palladium- Shell, Dow, Kuraray, and Sinopec, investigated
catalyzed carbonylation of 1,3-dienes the prospect of accessing adipate esters via
butadiene dicarbonylation. However, despite
Ji Yang1, Jiawang Liu1, Helfried Neumann1, Robert Franke2,3, Ralf Jackstell1, Matthias Beller1* extensive explorations, no such industrially
viable transformation was developed (23–38).
The direct carbonylation of 1,3-butadiene offers the potential for a more cost-efficient and Some pilot tests were implemented (23–36),
environmentally benign route to industrially important adipic acid derivatives. However, owing to but those processes all involved multistep re-
the complex reaction network of regioisomeric carbonylation and isomerization pathways, a actions with insufficient selectivity (~60 to
selective practical catalyst for this process has thus far proven elusive. Here, we report the design 80%) for the desired linear diester.
of a pyridyl-substituted bidentate phosphine ligand (HeMaRaphos) that, upon coordination to
palladium, catalyzes adipate diester formation from 1,3-butadiene, carbon monoxide, and butanol Here, we present a palladium-catalyzed dicar-
with 97% selectivity and 100% atom-economy under industrially viable and scalable conditions bonylation of 1,3-butadiene that provides dialkyl
(turnover number > 60,000). This catalyst system also affords access to a variety of other adipates in ≥95% yield and ≥97% selectivity.
di- and triesters from 1,2- and 1,3-dienes. Key to success was the ligand design. Recently,
we developed bidentate phosphine ligands
C arbonylation reactions are among the the building blocks of polyamides and poly- for palladium-catalyzed alkoxycarbonylation
most important applications of indus- esters currently produced on a multimillion– reactions in which basic pyridyl substituents
trial catalysis (1–5): Using carbon mon- metric ton scale (16, 17). More specifically, on phosphorus proved essential for high ac-
oxide (CO) as a highly versatile C1 adipate diesters are used for plasticizers, per- tivity (39). On the basis of that work and our
building block with olefins, more than fumes, lubricants, solvents, several active phar- long-standing interest in carbonylation reac-
10 million metric tons of various carbonyl maceutical ingredients, and, with respect to tions (40), we proceeded to investigate the di-
compounds (aldehydes, acids, and esters) are scale, most importantly for the production carbonylation reaction of 1,3-butadiene with
produced annually for numerous consumer of nylons. Now, the main industrial route to butanol as a model for the direct synthesis of
products. CO is a central intermediate in the produce adipate diesters involves oxidation of adipate diesters.
chemical industry that can be easily produced a mixture of cyclohexanol and cyclohexanone
either from fossil-based resources (coal or by an excess of nitric acid, followed by es- As shown in Fig. 1, there are multiple chal-
gas) or from renewables (CO2 or biowaste). terification with the corresponding alcohols lenges associated with this catalytic process:
Despite the initial discovery of homogeneous- (18–20). This process requires special equip- (i) The catalyst must promote two different
ly catalyzed carbonylation processes nearly ment owing to the acid’s corrosiveness and carbonylation reactions on the diene substrate
80 years ago (6–15), several unattained objec- produces stoichiometric amounts of nitrous (which could not be achieved previously); (ii)
tives remain, perhaps most saliently the direct oxide (N2O) (21), which is a major scavenger of the linear dicarbonylation product must be
dicarbonylation of 1,3-dienes. This reaction stratospheric ozone and has nearly 300 times
would enable more environmentally benign, the atmospheric heat-trapping capacity of 1Leibniz-Institut für Katalyse e.V. an der Universität Rostock,
atom-economical production of adipate esters, CO2 (22). Albert-Einstein Str. 29a, D-18059 Rostock, Germany. 2Evonik
Performance Materials GmbH, Paul-Baumann-Str. 1, 45772
Marl, Germany. 3Theoretical Chemistry, Ruhr-University
Bochum, 44780 Bochum, Germany.
*Corresponding author. Email: [email protected]

Fig. 1. Reaction network involved in synthesis of adipates from 1,3-butadienes. The green outlines indicate the starting materials (1,3-butadiene, carbon
monoxide, and alcohol) and desired product adipic diester.

Yang et al., Science 366, 1514–1517 (2019) 20 December 2019 1 of 4

RESEARCH | REPORT

formed selectively, despite the fact that isom- tural fragments of L1 and L4: The bulky and carbonylation occurred to selectively produce
erization of the initially formed monocarbon- electron-rich di-tert-butylphosphino fragment n-butyl pent-3-enoate 3a. This intermediate
ylated intermediate to the terminal olefin is could promote fast isomerization of carbon- continuously accumulated to reach a maximum
thermodynamically particularly unfavorable; carbon double bonds (43–45) while the tert- yield of about 50% after 90 min. Stopping the
and (iii) other side reactions such as telomer- butyl-2-pyridyl phosphino group facilitated reaction at this time allowed isolation of 3a
ization (41), hydroalkoxylation, and (co)poly- formation of the active palladium hydride com- from the reaction mixture. Meanwhile, the
merization must be suppressed. plex and accelerated the final alcoholysis step. active catalyst also promoted olefin isomer-
Mixing L5 and Pd(0) bis(dibenzylideneacetone) ization. The terminal olefin n-butyl pent-4-
We sought to realize the selective dicarbon- [Pd(dba)2] in the presence of HCl resulted in enoate 3c could not be detected, which we
ylation of 1,3-butadiene by using base-modified the formation of a bright yellow palladium attribute to its fast conversion into the linear
derivatives of the 1,2-bis[(di-tert-butylphosphino) complex, which was suitable for x-ray crystal- adipate diester.
methyl]benzene ligand (L1, dtbpx), which is lography (fig. S1). Although no coordination of
used for the bulk production of methyl met- the pyridine-N atom to the palladium center Next, detailed optimization studies on the
hacrylate (42). Initial optimization studies was observed in solid state, the durability of effect of palladium precursor, acid, temper-
with this ligand (Fig. 2) showed slight ac- the catalyst in solution might be enhanced ature, and pressure were performed to further
tivity but good selectivity to give the linear by such hemilabile binding. To our delight, the improve the practicality of the process (tables S2
di-n-butyl adipate 4a at 120°C and 40 bar CO dicarbonylation of 1,3-butadiene proceeded in to S7). In particular, excellent catalyst turnover
with p-toluenesulfonic acid as a cocatalyst. the presence of HeMaRaphos and Pd(II) tri- numbers (>60,000) were obtained using the
To improve the catalyst performance, dtbpx fluoroacetate [Pd(TFA)2] to adipate diester presynthesized Pd(II)-HeMaRaphos complex
derivatives L2 and L3 were tested. However, with a yield of 85% and a linear selectivity of under optimal conditions (table S9). Scaled-
no increase of activity was observed. Accord- 97%. We benchmark the distinct behavior of up reactions of 1,3-butadiene with methanol
ing to our hypothesis above, the incorpora- L5 in comparison with more than 70 other and n-butanol were then carried out at low
tion of suitable basic groups on this specific ligands, including well-known mono- and bi- catalyst loading [<0.5 mole % (mol %)] (table
ligand backbone should increase the activity dentate phosphines, in table S1. S8). The resulting esters were smoothly ob-
of the corresponding palladium catalyst sys- tained in 88 to 95% yield and >97% linear
tem in alkoxycarbonylation reactions. Indeed, To understand the performance of the pal- selectivity. As an example, a reaction with-
using L4 considerably increased activity and ladium catalyst with HeMaRaphos, we con- out additional solvent could be performed
yield of diesters (77%) but at the expense of ducted kinetic monitoring experiments (fig. on a >200-g scale with a Pd loading of only
insufficient selectivity (48%). Considering the S3). In the first half hour, formation of the ac- 0.05 mol % (fig. S2 and table S8).
appropriate reactivity of L4 and the suitable tive palladium hydride complex was observed
selectivity of L1, we designed the ligand L5 in the in situ mixture of Pd(TFA)2, L5, and Beyond the specific importance of the di-
(HeMaRaphos), which combines the two struc- p-toluenesulfonic acid (PTSA). Then, alkoxy- carbonylation of 1,3-butadiene in the chem-
ical industry, this methodology also offers

Fig. 2. Ligand optimization for palladium-catalyzed dicarbonylation of 1,3-butadiene. Reaction conditions for ligand optimization: butadiene (1.0 mmol, solution
in toluene), Pd(TFA)2 (0.005 mmol, 0.5 mol %), ligand (0.01 mmol, 1.0 mol %), PTSA•H2O (2.0 mol %), nBuOH (2.0 ml), CO (40 bar), 120°C, and 24 hours; the ratio

of products and yields were determined by gas chromatography analysis with mesitylene as the internal standard. Reactivity represents the yield of the diester.
Selectivity represents the ratio of linear diester to all diesters. nBu, n-butyl; tBu, tert-butyl.

Yang et al., Science 366, 1514–1517 (2019) 20 December 2019 2 of 4

RESEARCH | REPORT

Fig. 3. Palladium-catalyzed dicarbonylation of 1,2- and 1,3-dienes. Me, methyl; Et, ethyl; nPr, n-propyl; iPr, isopropyl.

prospects for valorization of other dienes for more difficult to control, afforded excellent natural products [(L)-menthol, (L)-borneol,
fine chemical production (46–48). To show- 1,4-site selectivity. Using myrcene 1m as an and cholesterol] yielded the desired products
case the generality of the catalyst system, exemplary inexpensive, widely available natural with high regioselectivity.
15 different dienes and more than 30 alcohols diterpene, tri-alkoxycarbonylation was achieved
were converted in high yield and selectivity smoothly. In addition to conjugated 1,3-dienes, REFERENCES AND NOTES
to the corresponding diesters (Fig. 3). For 1,2-dienes 1n and 1o were converted to the
example, several linear conjugated dienes 1a corresponding linear diesters with similar 1. W. Bertleff, M. Roeper, X. Sava, in Ullmann’s Encyclopedia of
to 1f showed excellent reactivity and regiose- activity and regioselectivity. Industrial Chemistry (Wiley‐VCH, 2007).
lectivity. Even for internal conjugated double
bonds (1f), isomerization reactions led prefer- Numerous aliphatic alcohols (products 4p 2. R. Franke, D. Selent, A. Börner, Chem. Rev. 112, 5675–5732
entially to the terminal products. Increased to 4z and 5a to 5h) were also tolerated by (2012).
steric hindrance on the carbon chain, including the catalyst system and gave diesters in high
a tetra-substituted conjugated diene, influenced yields and selectivity (>93%). Various func- 3. J.-B. Peng, H.-Q. Geng, X.-F. Wu, Chem 5, 526–552 (2019).
the reactivity; however, the regioselectivity to tional groups (products 5m to 5p)—including 4. D. U. Nielsen, X.-M. Hu, K. Daasbjerg, T. Skrydstrup, Nat. Catal.
the corresponding linear diesters remained electron-withdrawing fluorine, chlorine, and
very high (1g to 1k). 1-Phenyl–substituted 1,3- trifluoromethyl, as well as electron-donating 1, 244–254 (2018).
diene 1l, for which the regioselectivity might be methoxy—were compatible. Furthermore, a 5. H. Alper, J. Organomet. Chem. 300, 1–6 (1986).
heterocyclic derivative (5q) and hydroxylated 6. R. Otto, German Patent 849548 (1938).
7. H. Adkins, G. Krsek, J. Am. Chem. Soc. 71, 3051–3055

(1949).
8. M. E. Piotti, H. Alper, J. Am. Chem. Soc. 118, 111–116

(1996).
9. P. B. Webb et al., J. Am. Chem. Soc. 125, 15577–15588

(2003).

Yang et al., Science 366, 1514–1517 (2019) 20 December 2019 3 of 4

RESEARCH | REPORT

10. R. A. Periana, O. Mironov, D. Taube, G. Bhalla, C. J. Jones, 29. T. Omatsu, Y. Tokito, Japanese Patent 04208247 A (1992). ACKNOWLEDGMENTS
Science 301, 814–818 (2003). 30. E. M. Atadan, H. S. Bruner, U.S. Patent 5292944 A (1994).
31. A. Matsuda, T. Fujii, Japanese Patent 06192175 A (1994). We thank the analytical staff of the Leibniz-Institute for Catalysis,
11. Q. Liu et al., Angew. Chem. Int. Ed. 49, 3371–3374 (2010). 32. P. Denis, C. Patois, R. Perron, European Patent 0648731 A1 Rostock, for their excellent service. Funding: We gratefully
12. D. Willcox et al., Science 354, 851–857 (2016). acknowledge the support of Federal Ministry of Education and
13. X. Ren et al., Angew. Chem. Int. Ed. 56, 310–313 (2017). (1995). Research (BMBF), the State of Mecklenburg-Vorpommern, Evonik
14. J. Y. Wang, A. E. Strom, J. F. Hartwig, J. Am. Chem. Soc. 140, 33. O. E. Sielcken, H. Hovenkamp, Patent WO 9506027 A1 (1995). Industries, and the European Research Council. J.Y. thanks China
34. A. D. Poole, J. G. Sunley, UK Patent 2327420 A (1999). CSC Scholarship for financial support (201704910879). Author
7979–7993 (2018). 35. E. Drent, W. W. Jager, Patent WO 00056695 A1 (2000). contributions: J.Y. and J. L. contributed equally. J.Y., M.B., R.F.,
15. C. You et al., J. Am. Chem. Soc. 140, 4977–4981 (2018). 36. O. E. Sielcken et al., Patent WO 0121569 A1 (2001). and R.J. planned the project. J.Y., H.N., and M.B. developed and
16. E. M. Aizenshtein, Fibre Chem. 42, 341–349 (2011). 37. J. Mormul et al., ACS Catal. 6, 2802–2810 (2016). prepared the catalysts. J.Y., J.L., and R.J. designed and performed
17. M. T. Musser, in Ullmann’s Encyclopedia of Industrial Chemistry 38. A. Tortajada, R. Ninokata, R. Martin, J. Am. Chem. Soc. 140, all the experiments. J.Y., M.B., and R.F. wrote the paper.
Competing interests: A patent has been filed on the catalyst
(Wiley‐VCH, 2000). 2050–2053 (2018). system under EP3121184 A2. Data and materials availability:
18. A. Castellan, J. C. J. Bart, S. Cavallaro, Catal. Today 9, 237–254 39. K. Dong et al., Nat. Commun. 8, 14117–14127 (2017). X-ray data are available free of charge from the Cambridge
40. X. Fang, H. Li, R. Jackstell, M. Beller, Angew. Chem. Int. Ed. 53, Crystallographic Data Centre under CCDC 1941490; all other data
(1991). are in the supplementary materials.
19. K. Sato, M. Aoki, R. Noyori, Science 281, 1646–1647 (1998). 9030–9034 (2014).
20. K. C. Hwang, A. Sagadevan, Science 346, 1495–1498 (2014). 41. R. Jackstell et al., Angew. Chem. Int. Ed. 41, 986–989 (2002). SUPPLEMENTARY MATERIALS
21. M. H. Thiemens, W. C. Trogler, Science 251, 932–934 (1991). 42. W. Clegg et al., Chem. Commun. 1999, 1877–1878 (1999).
22. P. Forster et al., in Climate Change 2007: The Physical Science 43. Y. Liu, K. Dong, M. Beller, S. Mecking, ACS Catal. 8, 9232–9237 science.sciencemag.org/content/366/6472/1514/suppl/DC1
Materials and Methods
Basis. Contribution of Working Group I to the Fourth (2018). Figs. S1 to S4
Assessment Report of the Intergovernmental Panel on Climate 44. P. B. Webb, T. E. Kunene, D. J. Cole-Hamilton, Green Chem. 7, Tables S1 to S9
Change, S. Solomon, D. Qin, Eds. (Cambridge Univ. Press, NMR and HRMS Spectra
2007), pp. 130–234. 373–379 (2005). References (49–52)
23. J. B. Zachry, C. L. Aldridge, U.S. Patent 3161672 (1964). 45. J. Liu, J. Yang, F. Ferretti, R. Jackstell, M. Beller, Angew. Chem.
24. K. N. Von, U.S. Patent 3876695 A (1975). 14 August 2019; accepted 7 November 2019
25. Y. Isogai, M. Hosokawa, T. Ookawa, N. Wakui, T. Watanabe, Int. Ed. 58, 4690–4694 (2019). 10.1126/science.aaz1293
Japanese Patent 5788147 A (1982). 46. N. Herrmann, D. Vogelsang, A. Behr, T. Seidensticker,
26. C. K. Hsu, F. Dobinson, U.S. Patent 4575562 A (1986).
27. R. Maerkl et al., U.S. Patent 4777284 A (1988). ChemCatChem 10, 5342–5365 (2018).
28. E. Drent, J. Van Gogh, U.S. Patent 4861912 A (1989). 47. J. Tsuji, Acc. Chem. Res. 6, 8–15 (1973).
48. J. R. Zbieg, E. Yamaguchi, E. L. McInturff, M. J. Krische,

Science 336, 324–327 (2012).

Yang et al., Science 366, 1514–1517 (2019) 20 December 2019 4 of 4

RESEARCH

EMOTION AND LANGUAGE Biologically evolved physiology should pro-
vide universal structure to emotion seman-
Emotion semantics show both cultural variation and tics, but the exact sources of this structure
universal structure are not clear. Constructionist models of emo-
tion emphasize the roles of valence—the he-
Joshua Conrad Jackson1*, Joseph Watts2,3,4,5†, Teague R. Henry1†, Johann-Mattis List2, donic pleasantness versus unpleasantness of
Robert Forkel2, Peter J. Mucha6,7, Simon J. Greenhill2,8, Russell D. Gray2,9, Kristen A. Lindquist1* emotions—and activation—the physiological
arousal associated with experiencing emotions
Many human languages have words for emotions such as “anger” and “fear,” yet it is not clear whether (8–10). According to these models, valence and
these emotions have similar meanings across languages, or why their meanings might vary. We activation reflect basic neurophysiological
estimate emotion semantics across a sample of 2474 spoken languages using “colexification”—a processes that signal when the body shifts
phenomenon in which languages name semantically related concepts with the same word. Analyses away from homeostasis (9), and the universal
show significant variation in networks of emotion concept colexification, which is predicted by importance of these processes may lead all
the geographic proximity of language families. We also find evidence of universal structure in emotion languages to differentiate emotions primar-
colexification networks, with all families differentiating emotions primarily on the basis of hedonic ily on the basis of their degree of valence and
valence and physiological activation. Our findings contribute to debates about universality and activation. Other accounts, however, suggest
diversity in how humans understand and experience emotion. that factors such as dominance, certainty, so-
ciality, and approach-avoidance may also rep-
M any human languages have rich vo- primary emotions (e.g., anger, sadness) that resent universal dimensions of variance in
cabularies devoted to communicating evolved in mammalian brains (4). In turn, emotion semantics (12–15).
many languages may develop words for pri-
emotions. Although not all emotion mary emotion concepts such as “anger” and Predictions about the influence of culture
words are common—the German word “sadness” because these concepts name ex- and biology on emotion have long been exam-
Sehnsucht refers to a strong desire for periences derived from universal biological ined and debated, yet findings from past studies
an alternative life and has no direct trans- structures that are shared by all humans (2–4). are mixed. An early study found that human
lation in English—there are many words that These theories do allow for cultural and lin- subjects from remote Papua New Guinea
appear to name similar emotional states across guistic variation in emotion, but tend not to matched posed facial expressions to emotional
the world’s spoken languages. Translation dic- model or predict this variation. situations at similar rates to North Americans
tionaries, for example, suggest that the English (16), whereas recent field studies among other
word love can be equated with the Turkish There is a growing recognition, however, small-scale societies have found considerably
word sevgi and the Hungarian word szerelem. that emotions can vary systematically in their more cultural variability in people’s concep-
But does this mean that the concept of “love” is meaning and experience across culture and tualization of emotion (17). These mixed results
the same in English, Turkish, and Hungarian? language (5–7). Constructionist models of emo- may be due to methodological limitations of
tion in particular claim that concepts such as past research. Owing to logistical challenges,
Here, we explore this question by examining “anger” and “sadness” do not derive from dedi- the vast majority of cross-cultural studies have
cated brain structures (8), but occur when hu- been two-group comparisons (17), and the few
the meaning of emotion concepts in a sample mans make socially learned inferences about multigroup studies on emotion have sampled
the meaning of basic physiological processes predominantly from industrial and globalized
of 2474 languages from 20 major language linked to maintaining the body’s homeostasis nations (18, 19). Moreover, human subject–
(9, 10). The meaning of emotion concepts (i.e., based studies seldom present emotions as they
families. Using a new method from compara- “emotion semantics”) should thus draw from naturally occur, instead using posed facial ex-
both culturally evolved conceptualizations as pressions, fictional vignettes, and exaggerated
tive linguistics, we examine sources of varia- well as biologically evolved physiology. vocalizations as test stimuli. Finally, human
subject–based studies may be susceptible to
tion and structure in emotion semantics across If cultural evolutionary processes shape the demand characteristics and researcher bias:
meaning of emotion concepts, the historical Studies with imposed training phases and
this global sample of languages. relationships between language groups should forced choice paradigms have found evidence
predict which languages have the most similar for universal recognition of emotion (16),
Early theories of emotion, drawing from emotion semantics. Language groups in closer whereas studies with fewer constraints have
Darwin (1), suggested that there are a discrete geographic proximity are the most likely to found more cultural variability (17).
number of universal emotions from which all engage in borrowing (the sharing of concepts,
other emotions are derived (2–4). Many of norms, etc.) and also tend to share more re- As an alternative to human subjects–based
these theories claimed that, just as there are cent common ancestors than geographically research, analyses of naturally occurring lan-
distant groups (11). We thus hypothesize that guage can have high ecological validity and
primary colors (e.g., yellow, red), there may be emotion semantics are associated with a lan- do not rely on human subject recruitment.
guage group’s geographic location: Language Language may be an imprecise metric of ex-
1Department of Psychology and Neuroscience, University of groups in close geographic proximity may have perience, but analyzing how people use words
North Carolina at Chapel Hill, Chapel Hill, NC, USA. more similar emotion semantics than distant can reveal how they experience emotions as
2Department of Linguistic and Cultural Evolution, Max Planck groups. Although cultural variation in emotion similar or different. Several linguistic studies
Institute for the Science of Human History, Jena, Germany. is plausible under many models of emotion, a have conducted these analyses by qualitatively
3Religion Programme, University of Otago, Dunedin, New link between geographic distance and emotion comparing the meaning of emotion words by
Zealand. 4Center for Research on Evolution, Belief, and semantics would support constructionism’s searching for semantic primitives that have
Behaviour, University of Otago, Dunedin, New Zealand. claim that emotions are conceptualized using similar meanings across many languages (20).
5Social and Evolutionary Neuroscience Research Group, social learning. Yet few studies have quantitatively compared
Department of Experimental Psychology, University of the meaning of emotion words because the
Oxford, Oxford, UK. 6Carolina Center for Interdisciplinary field lacks metrics that quantify the semantic
Applied Mathematics, Department of Mathematics, University
of North Carolina at Chapel Hill, Chapel Hill, NC, USA.
7Department of Applied Physical Sciences, University of
North Carolina at Chapel Hill, Chapel Hill, NC, USA. 8ARC
Centre of Excellence for the Dynamics of Language, ANU
College of Asia and the Pacific, Australian National
University, Canberra, Australia. 9School of Psychology,
University of Auckland, Auckland, New Zealand.
*Corresponding author. Email: [email protected] (J.C.J.);
[email protected] (K.A.L.) †These authors contributed
equally to this work.

Jackson et al., Science 366, 1517–1522 (2019) 20 December 2019 1 of 5

RESEARCH | REPORT

distance between words such as the English dialect of Dargwa uses the word-form dard For instance, languages were more likely to
love and the Turkish sevgi (21). to express both the concepts of “grief” and colexify concepts such as “water” and “sea,”
“anxiety.” Persian speakers may therefore than concepts such as “sun” and “water,” im-
To overcome this challenge, we take a new understand “grief” as an emotion more sim- plying that speakers of these languages viewed
quantitative approach to estimate variabil- ilar to “regret,” whereas Dargwa speakers “water” and “sea” as semantically similar con-
ity and structure in emotion semantics. Our may understand “grief” as more similar to cepts and “sun” and “water” as distinct. We use
approach examines cases of colexification, “anxiety.” a similar approach to estimate the variation
instances in which multiple concepts are co- and structure of emotion semantics across
expressed by the same word form within a Past research has used colexification pat- language families.
language. Colexifications are useful for ad-
dressing questions about semantic structure terns across languages to examine the seman- To gather a high-powered sample, we com-
because they often arise when two concepts putationally aggregated colexifications into
are perceived as conceptually similar (22, 23) tic structure of non-emotion concepts. Youn a database of cross-linguistic colexifications
(see fig. S5). Persian, for instance, uses the (CLICS) featuring 2474 languages and 2439
word-form ænduh to express both the con- and colleagues coded dictionaries from 81 lan- distinct concepts—including 24 emotion con-
cepts of “grief” and “regret,” whereas the Sirkhi guages to show that concepts such as “sun,” cepts. We then used a random walk proba-
“river,” “mountain,” and “hill” had universal bility procedure to generate colexification
patterns of colexification that reflected con- networks (24). In these networks, nodes rep-
cepts’ material and functional properties (21). resented emotion concepts, and edges repre-
sented colexifications between these concepts,
Fig. 1. Colexification of emotion concepts across all languages (top left) and the largest language weighted by the number of languages that
families. Nodes are emotion concepts, and node size represents the number of colexifications involving the possessed a particular colexification. We used
concept. Edges represent colexifications, and edge thickness represents the number of colexifications this procedure to construct a network for all
between two emotion concepts. Node color designates community. languages in our database, and then for 20 in-
dividual language families whose colexification
networks had a significant level of modularity
(ps < 0.001). Although nodes in each language
family network were labeled with the same
emotion concepts (“anger”), comparing pat-
terns of colexification across language families
allowed us to test whether these nodes ac-
tually showed universal semantic equivalence
or whether their patterns of association ref-
lected semantic variation (see supplementary
text for more details).

A key step in these network comparisons
involved identifying communities: clusters of
emotion concepts that are more tightly col-
exified with one another than with emotion
concepts outside of the community. For each
network, we computed community structure
using the Cluster Optimal algorithm (25).
Figure 1 displays the global colexification net-
work and the five largest language family–
specific networks, and fig. S1 displays the
remaining language families. Family-specific
colexification networks allowed us to estimate
global variability in emotion semantics and to
predict variation and structure in emotion
semantics across language families.

We estimated global variation in emotion
semantics by comparing the community struc-
tures of language family networks. We quanti-
fied agreement in community structure using
adjusted Rand indices (ARIs), which indicate
the similarity of two networks’ community
structures (26). Negative ARI values indicate
that two networks’ community partitions vary
more than would be expected by chance, ARI
values of 0 indicate that two networks’ com-
munity partitions vary at a level that would be
expected at chance, and ARI values approach-
ing 1 reflect high agreement in community
structure between two networks. The distrib-
ution of raw ARIs indicated high variability

Jackson et al., Science 366, 1517–1522 (2019) 20 December 2019 2 of 5

RESEARCH | REPORT

in community structure across language fam- olded set of community comparisons. Through permutation-robust community comparisons.
ilies, with a mean ARI of 0.09 (SD = 0.11). Be- a series of permutation tests, we identified With this more conservative set of compar-
cause ARIs can be artificially low in networks pairs of communities that were more similar isons, the mean ARI was 0.22 (SD = 0.09), still
with few edges owing to isolated nodes, we than would be expected by chance and then reflecting high variability in emotion seman-
also examined the ARI values for a thresh- thresholded our sample to only include these tics across language families.

Fig. 2. The distributions of all pairwise language family ARI values for emotion concepts (in orange) To test whether variation in emotion co-
and color concepts (in light blue), and the distributions of permutation-robust ARI values for emo- lexification patterns merely arose from meth-
tion concepts (in red) and color concepts (in dark blue). Emotion concepts had significantly lower ARI odological factors, such as the way that
values than color concepts, showing more semantic variability. concepts were glossed in our database, we
next compared the ARI values from our emo-
Fig. 3. The relation- tion concept comparisons to ARI values for
ship between geo- colexification networks involving color con-
graphic proximity and cepts. Color concepts have also been studied
pairwise ARI values. cross-linguistically (27) and are frequently
Point size illustrates compared to emotion concepts (4), making
the number of them an appropriate sample of comparison
languages in a com- concepts. In the full sample of comparisons,
parison. In the key, the color concepts had a mean ARI of 0.35 (SD =
nodes denoting point 0.17), significantly higher than the full sample
size are not colored of emotion concept comparisons, t(390) =
because they apply to 18.51, p < 0.001. In the permutation-robust
both red and orange sample of comparisons, color concepts had a
points. The red mean ARI of 0.41 (SD = 0.15), again showing
points display the more universality than the permutation-
permutation-robust robust sample of emotion concept compar-
ARI values (r = −0.29), isons, t(158) = 11.44, p < 0.001 (Fig. 2). This
and the orange points difference also replicated when equating the
display the remaining number of color and emotion concepts, t(334) =
ARI values (r = −0.26). 15.52, p < 0.001 (see materials and methods
The regression line for more details). Emotion semantics thus vary
is fitted to all cases, widely across language families, and their var-
and the shading rep- iation is significantly greater than variation in
resents standard error. color semantics.

Our next analysis investigated whether geo-
graphic proximity predicted the pattern of
variation in emotion semantics across lan-
guage families. We tested this hypothesis by
correlating the geographic proximity of lan-
guage families (via the latitude and longitude
coordinates of their languages) with their pair-
wise ARI values. As predicted, language fami-
lies with higher pairwise ARI values were in
closer geographic proximity, both in the full
sample of our ARI comparisons, r(188) = −0.26,
p < 0.001, and in the smaller permutation-
robust sample, r(55) = −0.29, p = 0.03 (Fig. 3).
These associations suggest that emotion se-
mantics do not vary randomly; their variation
is tied to the cultural evolutionary relationship
between language families.

Finally, we tested whether any psychophys-
iological dimensions could predict the se-
mantic structure of emotion across language
families. We examined the explanatory power
of six dimensions (valence, activation, dom-
inance, certainty, approach-avoidance, and
sociality) by testing whether they predicted
the community membership of emotion con-
cepts across colexification networks. Using
ratings of 200 online participants (90 female,
110 male; Mage = 34.11, SDage = 10.52), we first
classified our emotion concepts on these di-
mensions using a 1-10 Likert-type scale. We
also classified a set of five “neutral” concepts

Jackson et al., Science 366, 1517–1522 (2019) 20 December 2019 3 of 5

RESEARCH | REPORT

(ordinary, nondescript, indifferent, neutral, was removed from the model. By contrast, with the highest AIC fit decrements (MAIC =
and impartial). Using a multilevel structural removing nonpredictive dimensions would 323.50) for the all-family network and for 13
equation model in which participants’ ratings have less of an impact on the model’s AIC fit.
of emotion concepts on these dimensions pre- We ran this analysis for all language families of the 19 language families in our analysis.
dicted the community membership of emo- except the Nuclear Macro-Je, for which models
tion concepts, we were then able to test how did not converge because only a single com- Activation was the most predictive dimension
well each dimension differentiated emotion munity contained multiple emotion concepts. for the remaining six language families (MAIC =
communities from our set of neutral words. If 208.76). Approach (MAIC = 35.82), certainty
a dimension was highly predictive, the model’s The results of this leave-one-out analysis re- (MAIC = 30.26), dominance (MAIC = 26.18), and
Akaike information criteria (AIC) fit would vealed higher predictive power for valence and sociality (MAIC = 7.41) had far less predictive
show a large decrement when the dimension activation than for other dimensions (Fig. 4). power than valence and activation, and com-
Valence was the most predictive dimension,
paring the distributions of fit decrements

across language families revealed that both

Fig. 4. Results from a leave-one-out analysis examining relative decrements in model fit following the removal of each dimension. The top panel represents
the AIC fit decrements associated with removing dimensions from a predictive model of emotion community membership. Higher decrements indicate that the
dimension was more predictive. The bottom panel shows the distribution of AIC fit decrements for each dimension. Valence and activation had significantly higher
average decrements than other dimensions.

Jackson et al., Science 366, 1517–1522 (2019) 20 December 2019 4 of 5

RESEARCH | REPORT

valence (ps < 0.001) and activation (ps < 0.001) although there were notable exceptions to this 17. M. Gendron, C. Crivelli, L. F. Barrett, Curr. Dir. Psychol. Sci. 27,
had significantly higher decrements (i.e., ex- pattern. For example, some Austronesian lan- 211–219 (2018).
plained more variance) than these other dimen- guages colexified the concepts of “pity” and
sions, and that valence had a higher average “love,” which implies that these languages 18. K. R. Scherer, R. Banse, H. G. Wallbott, J. Cross Cult. Psychol.
fit decrement than activation, t(19) = 2.70, p = may conceptualize “pity” as a more positive 32, 76–92 (2001).
0.01. These findings suggest that languages (or “love” as a more negative) concept than
around the world primarily differentiate emo- other languages. The ability of valence and 19. D. Matsumoto, S. H. Yoo, S. Nakagawa; 37 member of the
tions on the basis of valence and activation activation to consistently predict structure in Multinational Study of Cultural Display Rules, J. Pers. Soc. Psychol.
(see materials and methods for further analy- emotion semantics across language families 94, 925–937 (2008).
ses and discussion). suggests that these are common psychophys-
iological dimensions shared by all humans. 20. A. Wierzbicka, Cogn. Sci. 16, 539–581 (1992).
Our findings reveal wide variation in emo- 21. H. Youn et al., Proc. Natl. Acad. Sci. U.S.A. 113, 1766–1771
tion semantics across 20 of the world’s lan- Questions about the meaning of human
guage families. Emotion concepts had different emotions are age-old, and debate about the (2016).
patterns of association in different language nature of emotion persists in scientific litera- 22. A. François, “Semantic maps and the typology of colexification:
families. For example, “anxiety” was closely ture. The colexification approach that we take
related to “fear” among Tai-Kadai languages, here provides a new method and a set of Intertwining polysemous networks across languages” in From
but was more related to “grief” and “regret” metrics to answer these questions by creating Polysemy to Semantic Change: Towards a Typology of Lexical
amongst Austroasiatic languages. By contrast, vast networks of how people use words to Semantic Associations (John Benjamin, Philadelphia, PA,
“anger” was related to “envy” among Nakh- name experiences. Analyzing these networks 2008), pp. 163–215.
Daghestanian languages, but was more re- sheds light on the cultural and biological 23. J. M. List et al., Linguist. Typol. 22, 277–306 (2018).
lated to “hate,” “bad,” and “proud” among evolutionary mechanisms underlying how 24. P. Pons, M. Latapy, Computing communities in large
Austronesian languages. We interpret these emotions are ascribed meaning in languages networks using random walks, in International Symposium
findings to mean that emotion words vary around the world. Although debates about the on Computer and Information Sciences (Springer, 2005),
in meaning across languages, even if they relationship between language and conscious 284–293.
are often equated in translation dictionaries. experience are notoriously difficult to resolve 25. U. Brandes et al., IEEE Trans. Knowl. Data Eng. 20, 172–188
The supplementary materials contain an ex- (28), our findings also raise the intriguing pos- (2008).
tended discussion of why other technical and sibility that emotion experiences vary system- 26. D. Steinley, Psychol. Methods 9, 386–396 (2004).
sampling artifacts are unlikely to account atically across cultural groups. More broadly, 27. B. Berlin, P. Kay, Basic Color Terms: Their Universality and
for the variation that we observed in emotion our study shows the value of combining large Evolution (Univ. of California Press, Berkeley, CA, 1991).
semantics. comparative linguistic databases with quan- 28. F. Jackson, Philos. Q. 32, 127–136 (1982).
titative network methods. Analyzing the di-
Geography partly explained variation in verse ways that people use language promises ACKNOWLEDGMENTS
emotion semantics, such that geographically to yield insights into human cognition on an We acknowledge the many linguists who provided the word
closer language families tended to colexify unprecedented scale. lists necessary to detect and analyze colexifications across
emotion concepts in more similar ways than languages. We also acknowledge the feedback of our editor and
distant language families. Geographically prox- REFERENCES AND NOTES six anonymous reviewers; K. Gray, K. Payne, E. McCormick,
imal societies often have more opportunities J. Leshin, and N. Caluori; and the research assistance of
for contact through trade, conquest, and mi- 1. C. Darwin, Expression of the Emotions in Man and Animals R. Drabble, I. Khismatova, and A. Veeragandham. Funding:
gration and share more recent common an- (Univ. of Chicago Press, Chicago, 1965 [1872]). This study was supported by a National Science Foundation
cestry than distant groups (11). This suggests Graduate Research Fellowship and a Thomas S. and Caroline
that historical patterns of contact and com- 2. P. Ekman, Cogn. Emotion 6, 169–200 (1992). H. Royster Fellowship to J.C.J.. The compilation of the CLICS data
mon ancestry may have shaped cross-cultural 3. S. S. Tomkins, R. McCarter, Percept. Mot. Skills 18, 119–158 and software used in this study was funded by the Max Planck
variation in how people conceptualize emo- Society (as part of the CLLD project, https://clld.org), the Max
tions. We encourage future research to examine (1964). Planck Institute for the Science of Human History and the Royal
the specific vertical and horizontal transmis- 4. R. Plutchik, Am. Sci. 89, 344–350 (2001). Society of New Zealand (Marsden Fund grant 13-UOA-121 and
sion processes that give rise to geographic 5. R. E. Jack, O. G. B. Garrod, H. Yu, R. Caldara, P. G. Schyns, GlottoBank project, https://glottobank.org), the DFG research
variation in emotion semantics. fellowship grant 261553824 and the ERC Starting Grant 715618
Proc. Natl. Acad. Sci. U.S.A. 109, 7241–7244 (2012). (both awarded to J.M.L.), and the ARC’s Discovery Project
Despite this variation, we find evidence for a 6. B. Mesquita, M. Boiger, J. De Leersnyder, Curr. Opin. Psychol. 8, DE 120101954 and the ARC Center of Excellence CE140100041
common underlying structure in the meaning (both awarded to S.J.G.). J.W. is supported by funds from the
of emotion concepts across languages. Valence 31–36 (2016). Templeton Religious Trust (TRT0153). No funding agency was
and physiological activation—which are linked 7. C. Heyes, Behav. Brain Sci. 42, e169 (2019). involved in the conceptualization, design, data collection, analysis,
to neurophysiological systems that maintain 8. K. A. Lindquist, T. D. Wager, H. Kober, E. Bliss-Moreau, decision to publish, or preparation of this manuscript, and the views
homeostasis (9)—served as universal constraints expressed in this manuscript do not necessarily reflect the views
to variability in emotion semantics. Positively L. F. Barrett, Behav. Brain Sci. 35, 121–143 (2012). of our funding agencies. Author contributions: J.C.J., J.W.,
and negatively valenced emotions seldom be- 9. L. F. Barrett, Soc. Cogn. Affect. Neurosci. 12, 1–23 (2017). and K.L. conceptualized and designed the study. J.C.J., J.W., T.H.,
longed to the same colexification communities, 10. J. A. Russell, Psychol. Rev. 110, 145–172 (2003). J.M.L., and P.J.M. acquired and analyzed the data. J.M.L., R.F.,
11. E. A. Eff, Struct. Dyn. 3, 208–222 (2008). and S.J.G. contributed software and data used in our analyses. J.C.
12. J. R. Fontaine, K. R. Scherer, E. B. Roesch, P. C. Ellsworth, J., J.W., T.H., P.J.M., and K.L. interpreted the analysis. J.C.J.,
J.W., K.L., J.M.L., and R.D.G. wrote the manuscript. All authors
Psychol. Sci. 18, 1050–1057 (2007). approved the submitted manuscript. Competing interests: The
13. M. Kossowska, M. Bukowski, A. Van Hiel, Pers. Individ. Dif. 45, authors have no competing interests to declare. Data and
materials availability: All data, scripts, and materials are available
744–749 (2008). at https://osf.io/d9tr5/.
14. A. S. Cowen, D. Keltner, Proc. Natl. Acad. Sci. U.S.A. 114,
SUPPLEMENTARY MATERIALS
E7900–E7909 (2017).
15. A. A. Marsh, N. Ambady, R. E. Kleck, Emotion 5, 119–124 (2005). science.sciencemag.org/content/366/6472/1517/suppl/DC1
16. P. Ekman, W. V. Friesen, J. Pers. Soc. Psychol. 17, 124–129 Materials and Methods
Supplementary Text
(1971). Figs. S1 to S5
Tables S1 to S6
References (29–62)

27 January 2019; accepted 20 November 2019
10.1126/science.aaw8160

Jackson et al., Science 366, 1517–1522 (2019) 20 December 2019 5 of 5