Science 9.10.2020

RESEARCH | RESEARCH ARTICLE

Fig. 5. FGF9 sustained expression delays meiosis and promotes XX gonadal masculinization. firmed the association of the FGF9 gene with
(A) Spatiotemporal expression of FGF9 and the meiotic marker SYCP3 [immunostaining, green; the region containing the putative tissue-specific
4′,6-diamidino-2-phenylindole (DAPI), blue]. Insets display zoomed regions from OP. Scale bars, enhancers. We tested one element (fig. S16) in
100 mm. (B) Volcano plot from RNA-seq of XX mutant versus wild-type gonads (E13.5) and GO analysis. mouse LacZ reporter assays. Although this ele-
(C) Hematoxylin and eosin staining of XX gonads of adult mutants and wild-type controls. Cord-like ment displayed enhancer activity in tissues
structures in mutants denote XX-to-XY sex reversal. Inset shows SOX9 expression (immunostaining, such as the eye (fig. S17), we did not observe
green; DAPI, blue). Scale bars, 200 mm. gonadal staining, which could be due to spe-
cific requirements for additional trans-acting
murine locus (Fig. 3D). Adult knock-in mice ined (Fig. 4A and figs. S13 and S14). FGF9 is a factors in the mole. Alternatively, it may reflect
carrying the additional mole enhancer dis- known testis-determining gene that functions known limitations of the reporter assay that
played a three- and twofold up-regulation of in positive feedback with SOX9 and inhibits might be intensified by the interspecies nature
Cyp17a1 expression in females and males, re- the ovary-determining WNT4/b-catenin path- of the experiment (19).
spectively. The increased expression occurs in way (11) (fig. S15). Consequently, loss of Fgf9
the same cell type (steroidogenic cells) as in in XY gonads results in down-regulation of We then explored possible alterations on the
wild-type controls (fig. S12). A similar effect testicular markers and male-to-female sex dimorphic FGF9 expression pattern observed
was observed at embryonic stages (fig. S12), reversal. Comparative analyses against human, in other mammals, which is essential for sup-
thus confirming the increased in vivo activ- representative of the ancestral organization pressing germ cell meiosis (20). In mice, Fg f9
ity of the mole-specific fusion enhancer in at the locus, revealed a large inversion that re- is first expressed in the bipotential gonad of
gonadal tissue. The up-regulation of Cyp17a1 locates a distant genomic region (26 Mb away both sexes and becomes progressively restricted
expression was accompanied by an increase in in the human genome) to the mole FGF9 locus to the testis and turned off in ovaries, allowing
circulating testosterone in females and males (Fig. 4B). Hi-C data showed that the synteny the initiation of female meiosis at embryonic
(two- and threefold respectively) (Fig. 3E). Be- break occurs in the human FGF9 TAD, disrupt- day 13.5 (E13.5). In mole gonads, however, FGF9
cause androgens have potent anabolic effects ing its 3D organization. Thus, the mole locus expression is maintained after sex determina-
in muscle, we performed a grip-strength test is reorganized, with most of the FGF9 TAD tion in both sexes and becomes restricted to the
that revealed a significant increase in muscle remaining conserved but extending beyond OP at later stages (Fig. 4C). Immunostaining
strength in mutants compared with wild-type the synteny breakpoint on the centromeric analyses showed that FGF9 expression persists
controls (Fig. 3F). Thus, the regulatory nature side. This extended interaction domain is de- in the OP across the entire prenatal period and
of the CYP17A1 rearrangement offers a plau- limited by the presence of two CTCF binding becomes confined to a thin rim at postnatal
sible molecular mechanism for the observed sites with divergent orientation, a genomic stages (P7). The spatial reduction in FGF9 ex-
shift in hormone levels and the corresponding signature associated with TAD boundaries pression is concomitant with the initiation of
phenotype. (7) (Fig. 4B). A closer examination of the newly meiosis (Fig. 5A and fig. S18), which is con-
interacting region revealed several elements siderably delayed in female moles until birth,
An inversion at the FGF9 locus is enriched for active epigenetic marks, some an exceptional feature among mammals (21).
associated with delayed meiosis of them specific for the mole ovotestis (fig. Consequently, the observed heterochrony on
and gonadal masculinization S16). This interaction pattern was also val- mole FGF9 expression, compared with mouse,
idated through independent circular chro- and its potential effects are suggestive of a
We identified a rearrangement at the FGF9 mosome conformation capture sequencing contribution to the masculinization of female
locus, which is exclusive to the mole lineage (4C-seq) experiments (Fig. 4B), which con- mole gonads.
and not present in any other mammals exam-
We hypothesized that FGF9 expression during
early XX mole gonadogenesis might prevent
germ cells from entering meiosis in the OP,
allowing the TP to develop further. To test this
hypothesis, we engineered a bacterial artificial
chromosome (BAC) construct to overexpress
Fg f9 in somatic ovarian populations and gen-
erated transgenic mice through PiggyBac trans-
genesis and morula aggregation. Highly chimeric
animals displayed early embryonic lethality,
likely due to Fg f9 misexpression in other or-
gans, an effect that precluded their study in
later stages. Nevertheless, RNA-seq analysis
of E13.5 ovaries demonstrated an inhibition
of the meiotic process, manifested by down-
regulation of meiosis markers (Fig. 5B and
data S10). By contrast, low-chimera individu-
als composed of XX wild-type and XX mutant
cells were viable and displayed Fg f9 expression
during the entire ovarian development. These
animals showed a complete female-to-male
sex reversal, defined by gonadal morphology
and expression of male-specific factors such
as SOX9 (Fig. 5C). These results directly con-
firm the potential of altered FGF9 expression
to induce masculinization in mammalian XX
females.

M. Real et al., Science 370, 208–214 (2020) 9 October 2020 5 of 6

RESEARCH | RESEARCH ARTICLE

Discussion TADs might constitute a mechanism of “evolv- We thank P. Koopman for providing us with the Wt1-BAC vector.
ability” resulting from a modular system with We thank the sequencing and genotyping facility of the Max Planck
Vertebrate sex determination is controlled by vast flexibility and enormous evolutionary Institute of Molecular Cell Biology and Genetics in Dresden for
a limited set of key regulators whose hierarchy helping with the initial long-read sequencing project. We thank the
has evolved dynamically (11). Most of these potential for the origin of novelties. On the Animal Outcome Core Facility of the NeuroCure Center, Charité
genes display pleiotropic effects, controlling University, Berlin, for helping with strength tests. We also thank
regulatory networks in several tissues, often basis of this modularity, genomic rearrange- members of the Lupiáñez and Mundlos labs for fruitful discussions.
making them indispensable for embryonic via- Funding: This research was supported by grants from the
bility (22, 23). It is thus plausible that variations ments can easily change and reconstitute com- Deutsche Forschungsgemeinschaft (grant numbers MU 880/15-1
in sex determination derive from regulatory and MU 880/16-1) and Max Planck Society. D.G.L. was supported
changes that alter gene expression patterns plex expression patterns, thereby contributing by the Fundación Alfonso Martín Escudero and by a Helmholtz
but preserve essential functions, as suggested ERC Recognition Award grant from the Helmholtz-Gemeinschaft
for other evolutionary adaptations (1, 2). These to the saltatory nature of phenotypic innova- (ERC-RA-0033). Work conducted at the E. O. Lawrence Berkeley
genomic changes appear to be linked to the National Laboratory was performed under Department of Energy
evolutionary success of moles and demonstrate tion observed in many phylogenetic lineages. contract DE-AC02-05CH11231, University of California. A.V. and
that regulatory innovation can overcome a M.O. were supported by NIH grant R01HG003988. M.O. was
priori seemingly incompatible situations such We expect that approaches considering these supported by Swiss National Science Foundation grant PCEFP3_186993.
as female fertility in the presence of high andro- O.S. was supported by the Austrian Science Fund FWF grant P32190.
gen levels. But how could an intersex phenotype important aspects will eventually reveal the Author contributions: F.M.R., S.M., and D.G.L. designed the
evolve, and what is its adaptive significance? experimental approach. F.M.R., D.G.L., F.B., and R.J. captured moles and
One potential explanation is the anabolic effects evolutionary basis of many other traits and prepared the samples. F.M.R., M.S., and D.G.L. processed the samples
of androgens on muscle mass. Mole ovotestes for DNA, Hi-C, RNA, ChIP, and ATAC sequencing. S.A.H., B.T., A.M., and
may have evolved to equalize muscular strength substantially enhance the toolbox for unlock- M.V. supervised bioinformatic analyses. R.S., D.H., M.-H.M., and H.K.
among sexes by increasing androgens in fe- assembled the genome. P.F., P.X., and O.S. performed the gene
males. Such an advantage, combined with other ing the secrets of phenotypic variation and annotation and posterior genomic analyses. S.A.H. and A.B. analyzed the
androgen-derived effects such as aggressive be- RNA-seq data. S.A.H. performed the epigenetic analysis and the
havior, might have been key for the adaptation adaptation across the animal kingdom. prediction of synteny breaks. M.H. performed multispecies alignments.
to the demanding requirements of a burrowing R.S., V.H., and T.K. analyzed chromatin interaction datasets. M.F.H. and
underground lifestyle (14). REFERENCES AND NOTES S.A.W. performed hormone measurements. F.M.R., D.G.L., and A.H.
performed the CYP17A1 enhancer transgenic experiment. F.M.R., I.H.,
This study also highlights the evolutionary 1. J. Lopez-Rios et al., Nature 511, 46–51 (2014). and M.O. performed the FGF9 transgenic experiments. L.W. derived the
importance of genomic rearrangements and 2. Y. F. Chan et al., Science 327, 302–305 (2010). XX mESC line and performed blastocyst aggregations. F.M.R., M.S.,
their potential to modulate developmental gene 3. O. Dudchenko et al., Science 356, 92–95 (2017). S.M., and D.G.L. performed ELISA experiments and grip-strength tests.
expression. In most cases, genomic rearrange- 4. J. R. Dixon et al., Nature 485, 376–380 (2012). F.M.R., D.K.N.D., A.V., A.M., S.M., and D.G.L. participated in data
ments would have limited effects, as they pre- 5. E. P. Nora et al., Nature 485, 381–385 (2012). interpretation and discussion. F.M.R., S.M., and D.G.L. wrote the
serve whole regulatory units and do not disrupt 6. N. Harmston et al., Nat. Commun. 8, 441 (2017). manuscript with contributions from all other authors. Competing
enhancer-promoter interactions (6, 24). How- 7. C. Gómez-Marín et al., Proc. Natl. Acad. Sci. U.S.A. 112, interests: The authors declare no competing interests. Data and
ever, as shown here, they can also alter the reg- materials availability: The Whole Genome Shotgun project has been
ulatory potential in the local environment of 7542–7547 (2015). deposited at DDBJ/ENA/GenBank under the accession number
a TAD, as for the CYP17A1 locus, or shuffle 8. M. Spielmann, D. G. Lupiáñez, S. Mundlos, Nat. Rev. Genet. 19, PRJNA494291. All transcriptomes and epigenetic data have been
the functional content of distant TAD units, deposited in GEO under accession number GSE120589. Alignments have
exemplified by the FGF9 locus. Similar effects 453–467 (2018). been deposited at https://bds.mpi-cbg.de/hillerlab/IberianMole.
have been observed in human genetic diseases 9. D. G. Lupiáñez et al., Cell 161, 1012–1025 (2015).
for enhancer duplications (25) and for inver- 10. M. Franke et al., Nature 538, 265–269 (2016). SUPPLEMENTARY MATERIALS
sions (8, 9). Therefore, regulation of genes by 11. B. Capel, Nat. Rev. Genet. 18, 675–689 (2017).
enhancer elements and their organization in 12. F. J. Barrionuevo, F. Zurita, M. Burgos, R. Jiménez, Dev. Biol. science.sciencemag.org/content/370/6513/208/suppl/DC1
Materials and Methods
268, 39–52 (2004). Supplementary Text
13. F. D. Carmona et al., J. Exp. Zool. 310B, 259–266 (2008). Figs. S1 to S20
14. J. Haeck, Neth. J. Zool. 19, 145–248 (1969). Tables S1 to S8
15. C. J. Douady et al., Mol. Phylogenet. Evol. 25, 200–209 (2002). References (26–100)
16. Q. Wang, Y. Lan, E.-S. Cho, K. M. Maltby, R. Jiang, Dev. Biol. MDAR Reproducibility Checklist
Data S1 to S10
288, 582–594 (2005).
17. C. Berthet, E. Aleem, V. Coppola, L. Tessarollo, P. Kaldis, View/request a protocol for this paper from Bio-protocol.

Curr. Biol. 13, 1775–1785 (2003). 27 August 2019; resubmitted 19 April 2020
18. I. Hanukoglu, J. Steroid Biochem. Mol. Biol. 43, 779–804 Accepted 17 August 2020
10.1126/science.aaz2582
(1992).
19. K. Suryamohan, M. S. Halfon, WIREs Dev. Biol. 4, 59–84 (2015).
20. J. Bowles et al., Dev. Cell 19, 440–449 (2010).
21. F. Zurita et al., Sex Dev. 1, 66–76 (2007).
22. J. S. Colvin, A. C. White, S. J. Pratt, D. M. Ornitz,

Development 128, 2095–2106 (2001).
23. W. Bi et al., Proc. Natl. Acad. Sci. U.S.A. 98, 6698–6703

(2001).
24. N. H. Lazar et al., Genome Res. 28, 983–997 (2018).
25. A. J. Will et al., Nat. Genet. 49, 1539–1545 (2017).

ACKNOWLEDGMENTS

We thank A. Stiege, N. Brieske, U. Fisher, K. Macura, J. Fiedler,
K. Zill, C. Franke, and G. Mastrobuoni for technical support.

M. Real et al., Science 370, 208–214 (2020) 9 October 2020 6 of 6

RESEARCH

◥ which we term the electro-inductive effect, is
that an electrode can modulate nonredox chem-
REPORT ical reactivity. That the electric field induced
by an applied voltage can alter the electronic
ORGANIC CHEMISTRY structure of immobilized molecules on the
electrode is known as the Stark effect (3, 4),
Electro-inductive effect: Electrodes as functional but the electrode was not used as a func-
groups with tunable electronic properties tional group to control chemical reactions
of reactants that were immobilized on the
Joon Heo1,2*, Hojin Ahn1*, Joonghee Won1,2*, Jin Gyeong Son3, Hyun Kyong Shon3, Tae Geol Lee3, surface.
Sang Woo Han1†, Mu-Hyun Baik2,1†
To prove the principle, we explored whether
In place of functional groups that impose different inductive effects, we immobilize the electro-inductive effect could change the
molecules carrying thiol groups on a gold electrode. By applying different voltages, the rate of the base-mediated saponification of
properties of the immobilized molecules can be tuned. The base-catalyzed saponification a benzoic ester. This reaction served as one
of benzoic esters is fully inhibited by applying a mildly negative voltage of –0.25 volt versus of the 39 standard reactions from which the
open circuit potential. Furthermore, the rate of a Suzuki-Miyaura cross-coupling reaction can Hammett parameters were originally derived
be changed by applying a voltage when the arylhalide substrate is immobilized on a gold (1). As highlighted in Fig. 2A, a benzoic ester
electrode. Finally, a two-step carboxylic acid amidation is shown to benefit from a switch carrying a thiol moiety at the para position
in applied voltage between addition of a carbodiimide coupling reagent and introduction could readily be self-assembled as a mono-
of the amine. layer on a gold surface (5). Nucleophilic addi-
tion of a hydroxide ion to the carbonyl of
I n 1937, Hammett reported his ground- voltage (Fig. 1). Applying a negative potential the ester affords the hydrolyzed products:
breaking studies on how functional groups should mimic electron-donating groups, whereas the benzoate anion and the corresponding
affect the rate of chemical reactions (1). a positive potential should reproduce the elec- alcohol. It is easy to understand that electron-
This work led to the broadly accepted and tronic effect of electron-withdrawing groups. withdrawing groups on the substrate make a
widely used concept of the inductive effect In this paradigm, the inductive effect is no carbonyl carbon more electrophilic, leading to
of functional groups to tune the electronic longer discrete, because the applied voltage an acceleration of the saponification reaction,
properties of molecules. Of the countless ap- can be adjusted continuously. Recently, Dawlaty whereas electron-donating groups diminish
plications of this concept in chemistry, the and others showed in a thorough spectroscopic the reaction rate. Our initial studies showed
ability to control the rates of stoichiometric study that such a correlation between Hammett that the saponification of the immobilized ben-
and catalytic reactions by altering the func- parameters and polarization of the molecule zoates containing methyl and ethyl esters was
tional groups that are covalently bound to exists (2). Specifically, the vibrational frequency problematic, which can be attributed to too
reactants or catalysts is perhaps the most shifts of para-substituted mercaptobenzonitrile tightly packed monolayers that are protected
prominent and pervasive use of inductive ef- immobilized on a gold electrode could be di- against chemical attacks (6, 7). Thus, we used
fects. Despite being a cornerstone of chemis- rectly correlated to the applied voltage. Although a sterically bulkier tert-butyl ester, tert-butyl
try at large, this approach to modulating the this study did not explore control of chemical 4-mercaptobenzoate (TBMB), to form a slightly
property of a molecule is neither efficient nor reactions using voltage, the connection between less dense monolayer, allowing the esters to
convenient. Each functional group can only applied voltage and Hammett parameters di- undergo chemical reactions more easily. The
impose an inductive effect that is specific to rectly supports our proposed strategy. If the progress of saponification was conveniently
its identity, necessarily creating a discontin- electrode could act as a functional group with monitored by surface-enhanced Raman spec-
uous series where the range of properties is adjustable inductive effect, it would no longer troscopy (SERS) techniques (8).
limited by what the functional groups offer. be necessary to prepare a series of analogous
Naturally, each functionalized derivative must molecules or catalysts to impose different in- If the electro-inductive effect were substan-
be prepared, which is laborious and limiting, ductive effects on the reactive core. A particu- tial, the saponification of TBMB to generate
because not all possible molecular composi- larly attractive feature of using an electrode in 4-mercaptobenzoate should be accelerated
tions are easily accessible. this way is that the inductive effect applied to when a positive voltage is applied, whereas the
the molecule can be varied over time, thereby reaction should become slower with a nega-
We envisioned a different approach for mod- modulating the electronic structure of a cata- tive voltage, as illustrated in Fig. 2A. The SERS
ulating the electronic properties of a molecule, lyst over the course of a catalytic cycle, for spectra of the reactant and the product show
whereby instead of preparing a series of deri- example. This feature could enable fine con- two common peaks assigned to the breathing
vatives, the molecule of interest may be cova- trol over the progress of chemical reactions by modes of the aromatic ring [v(C–C)ring] with very
lently attached to an electrode and its chemical application of voltage sequences, reminiscent strong intensities at ~1070 (9) and ~1580 cm–1
properties modulated by changing the applied of pulse sequences in nuclear magnetic reso- (Fig. 2B). In addition, other distinct peaks
nance spectroscopy. at 1383, 1706, and 1405 cm–1 correspond to
1Department of Chemistry, Korea Advanced Institute of C(CH3)3, C=O stretching mode [v(C=O)] in the
Science and Technology (KAIST), Daejeon 34141, Republic The idea of covalently attaching molecules reactant, and COO– stretching mode [vs(COO–)]
of Korea. 2Center for Catalytic Hydrocarbon Functionalizations, to an electrode is, of course, not new, but im- in the product, respectively (10–12). As the reac-
Institute for Basic Science (IBS), Daejeon 34141, Republic mobilization efforts thus far have been aimed tion proceeds, we expect a decrease in peak
of Korea. 3Center for Nano-Bio Measurement, Korea Research at reactions where the deposited molecules intensities of the vibrations related to C(CH3)3
Institute of Standards and Science (KRISS), Daejeon 34113, participate in redox reactions. Molecules are and C=O and an increase in peak intensity for
Republic of Korea. typically immobilized on an electrode because vs(COO–). However, because the SERS peaks
*These authors contributed equally to this work. their reactivity hinges on receiving or donat- for C(CH3)3 and vs(COO–) appear in a similar
†Corresponding author. Email: [email protected] ing electrons. The key feature of our approach, region around 1400 cm–1, it proved difficult to
(M.-H.B.); [email protected] (S.W.H.) precisely monitor the reaction progress using
these peaks. Therefore, the moderately strong

Heo et al., Science 370, 214–219 (2020) 9 October 2020 1 of 6

RESEARCH | REPORT

Fig. 1. Proposed design for
exploiting the electro-inductive
effect to control chemical
reactions. (A) Classical approach
for modulating the electronic prop-
erties of a molecule. FG, functional
group; EDG, electron-donating group;
EWG, electron-withdrawing group;
Me, methyl; Et, ethyl. (B) Electronic
tuning of the reaction site with
covalent attachment of a molecule
to an electrode at an applied
potential. WE, working electrode; CE,
counter electrode; d– and d+, partial
negative charge and partial positive
charge.

peak at 1706 cm–1, which is instantly recogniz- v(C–C)ring(1) was found to be 0.183. At OCP— 3 M solution of potassium hydroxide, are highly
able as a C=O vibration in the reactant, was that is, when no voltage was applied—the peak favorable for saponification.
chosen as an indicator for monitoring the reac- at 1706 cm–1 diminished substantially after
tion progress. The intensities of the SERS signals exposure of the surface-bound esters to the There is an alternative explanation for these
are difficult to use as a quantitative measure hydrolysis conditions, as shown in black in experimental observations to consider. When a
of concentration in an absolute sense without Fig. 2C, and the peak ratio was found to be negative potential is applied to the gold elec-
careful calibration; the evolution of these sig- 0.072, indicating a considerable depletion of trode, all anionic species, including the hydrox-
nals over the reaction time of the current ex- esters on the surface. In good agreement with ide reactant, are repelled from the surface,
periment relative to an internal standard is the aforementioned expectation, the reaction whereas cationic species are drawn close to
meaningful and serves as a qualitative indi- could be accelerated by applying a positive the surface. As a result, the local concentra-
cator for the reaction progress (13, 14). voltage. At +0.25 V, the ester peak at 1706 cm–1 tion of anions drops substantially, whereas
dramatically reduced over the reaction time the concentration of cations increases, estab-
Before applying an external voltage, an ap- of 22.5 hours, as shown in blue in Fig. 2C. By lishing the electrochemical double layer. Thus,
propriate potential window was ascertained contrast, the application of a negative voltage the lack of saponification might simply be a
through cyclic voltammetry to confirm the ab- of –0.25 V led to a nearly complete inhibition of result of diminished hydroxide concentration
sence of electron transfer events (see figs. S3 to the reaction, indicated by the SERS spectrum within the double layer. A reasonable way of
S5). Specifically, potentials of +0.25 and –0.25 V shown in red in Fig. 2C that is practically iden- estimating the width of the double layer is to
versus open circuit potential (OCP) were used. tical to the spectrum of the initial state be- use the Gouy-Chapman model (15). Under
Figure 2C shows the SERS spectra of TBMB fore exposure to the saponification conditions, the applied conditions of –0.25 V and a KOH
immobilized on the gold electrode before and which is displayed in green. The relative peak concentration of 3 M, the width of this double
after 22.5 hours at 50°C in 3 M potassium intensity ratio was found to be 0.181 in this layer can be estimated to be in the range of
hydroxide solution with no voltage, +0.25 V, case, very close to the ratio of 0.183 determined ~2 Å. With the sulfide moiety in direct contact
or –0.25 V applied. By comparing the vibra- for the sample before hydrolysis, indicating with the gold surface, the reactive ester func-
tional signature of the ester at 1706 cm–1 to the that most of the esters were still intact on the tionality of the TBMB substrate is calculated
prominent peak at ~1070 cm–1 [v(C–C)ring(1)] surface. These observations suggest that the to be positioned 7 to 9 Å away from the sur-
as an internal reference, the ratios of the peak degree of saponification can be controlled by face, which is well within the diffusion layer
intensities could be determined, as illustrated the applied potential, as proposed. The com- where ionic species, such as the hydroxide ions,
in Fig. 2D. plete inhibition of the ester hydrolysis using should be able to approach the ester group
what may be considered a mildly negative without being affected by the applied poten-
Before the surface-bound substrates were voltage of –0.25 V is notable, given that the tial. Details of these calculations are given in
exposed to the saponification conditions, the reaction conditions, including a concentrated the supplementary materials. Moreover, the
ester-associated peak was clearly visible, and examples discussed below are not affected
the peak intensity ratio relative to the peak of

Heo et al., Science 370, 214–219 (2020) 9 October 2020 2 of 6

RESEARCH | REPORT

Fig. 2. Base-catalyzed
ester hydrolysis reaction.
(A) Schematic representation
for base-catalyzed saponification
of self-assembled monolayer
of TBMB under applied potentials.
Reactions were run in 3 M KOH(aq)
at 50°C for 22.5 hours. (B) SERS
spectra of the reactant (R) TBMB
(green line) and the product (P)
4-mercaptobenzoate (brown line,
deprotonated form). a.u., arbitrary
units; n, stretching mode; ns, sym-
metric stretching mode. (C) Repre-
sentative SERS spectra of base-
catalyzed saponification with
+0.25 V versus OCP (blue line), OCP
(black line), –0.25 V versus OCP (red
line), and R (green line). (D) Relative
intensities of base-catalyzed sapon-
ification under different potentials.
The points and error bars in the
relative intensity graph (D) indicate
the means and standard deviations
of five to seven experiments, respec-
tively. I, intensity.

by the double-layer effect, so this alterna- 1068 cm–1 (18), (ii) aromatic ring vibrations of-flight secondary ion mass spectrometry (TOF-
tive explanation can be excluded with some v(C–C)ring(1) at 1079 cm–1, and (iii) another SIMS) techniques and analyzed the amounts
confidence. aromatic ring vibrational mode v(C–C)ring(2) of the starting material and the product on
at 1560 cm–1 (13, 19). We also immobilized the the surface. As shown in fig. S7, the peaks of
With the initial proof-of-principle study in anticipated product biphenyl-4-thiol on the the reactant ([C6H4S79Br]–) and the product
hand, we sought to test the electro-inductive ([C12H9S]–) are easily distinguished in the
effect in a catalytic reaction. There are two gold electrode to confirm its SERS signature. mass spectrum. In excellent agreement with
possible approaches for carrying out cata- Characteristic v(C–C)ring peaks of the biphenyl the SERS results, the peak intensity ratio of
lytic reactions while taking advantage of the product that were most convenient to monitor the product to the starting material increases
electro-inductive effect: Either the catalyst or could be found at 1586 cm–1 [v(C–C)ring(3)] as the applied potential is set more negative
the reactant could be immobilized on the gold and 1600 cm–1 [v(C–C)ring(4)], as illustrated in (fig. S8).
electrode and their activity changed by apply- Fig. 3B (20). The SERS spectra of the starting
ing different voltages. Immobilizing the cata- material and the spectra taken after 12 hours Finally, we monitored the SERS spectra of
lyst poses several practical challenges such as the reaction over a time course of 3 hours,
low catalyst loading, owing to the relatively under the Suzuki-Miyaura conditions apply- taking measurements in 30-min intervals, as
small size of the gold electrode surface or the ing +0.30 V, +0.15 V, no voltage, and –0.15 V illustrated in Fig. 3E. As expected, the product
synthetic difficulties of installing a thiol linkage are overlaid in Fig. 3C. After 12 hours of re- peak increased most rapidly at –0.15 V, reach-
on the catalyst. Immobilizing the reactant and action time, the product peak of v(C–C)ring(3) ing a saturation plateau after ~90 min, whereas
monitoring its transformation to the product can be identified very clearly, whereas the at other voltages, the reaction was not com-
on the gold surface is more convenient. As an peak of v(C–C)ring(4) has become a shoulder plete after 3 hours. Taken together, these
illustrative example, we chose the Pd-catalyzed feature. The cross-coupling reaction was slowest results demonstrate that the rate of the Suzuki-
Suzuki-Miyaura cross-coupling reaction (16), Miyaura cross-coupling changes notably when
which is one of the most powerful and widely at +0.30 V, indicated by the weakest product the applied voltage is varied. More specifical-
used synthetic tools for C–C bond formation. ly, applying a negative voltage, to mimic an
Mechanistically, it is thought to proceed through peak among the series, but the rate increased electron-donating group, enhances the rate of
three key steps: (i) oxidative addition of an the reaction. The mechanism of the Suzuki-
aryl halide to give a Pd(II)-intermediate, (ii) notably over the series of +0.15 V, OCP, and Miyaura reaction is conceptually well understood,
transmetalation of the arylboron species that –0.15 V, suggesting that negative induction but many details remain unclear. Indeed, the
delivers the second aryl moiety to the Pd cen- increases the rate of the reaction. As stated detailed mechanism of cross-coupling reactions
ter, and (iii) reductive elimination to afford a depends on the reaction conditions, for in-
C–C coupled biaryl product and the Pd(0) above, the SERS signals cannot be used in a stance, the concentration of substrate or cou-
catalyst. pling partner, the catalyst, and the base (21–23).
quantitative sense to precisely determine the Our results are in good agreement with a con-
The SERS spectrum of the immobilized ventional Hammett study that used para-
4-bromobenzenethiol shown in Fig. 3B dis- reaction progress, but by comparing the two substituted arylbromide substrates (24), which
plays three clearly distinguishable features aromatic ring vibrations [v(C–C)ring(1) and showed the same trend of an increase in the
(17): (i) a C–Br stretching mode v(C–Br) at v(C–C)ring(3)], the trend that negative poten-
tial accelerates the reaction can be observed at

least semiquantitatively. Moreover, the con-

sumption of reactants on the surface leads to
a decrease of the v(C–Br) peak against the same
internal reference, as illustrated in Fig. 3D. To

complement the SERS analysis, we used time-

Heo et al., Science 370, 214–219 (2020) 9 October 2020 3 of 6

RESEARCH | REPORT

Fig. 3. Palladium-catalyzed
Suzuki-Miyaura cross-
coupling reaction.
(A) Schematic representation
for Pd-catalyzed Suzuki-
Miyaura cross-coupling of self-
assembled monolayer of
4-bromobenzenethiol under
applied potentials. Reactions
were run with PhB(OH)2
(1.5 mmol), Pd(OAc)2
(0.075 mmol), (2-biphenyl)di-
tert-butylphosphine (0.15 mmol),
KF (4.5 mmol), and tetrabuty-
lammonium hexafluorophos-
phate (0.1 M) in tetrahydrofuran
(3 ml) at 25°C for 12 hours
(26). (B) SERS spectra of the
reactant 4-bromobenzenethiol
(green line, R) and product
biphenyl-4-thiol (brown line, P).
(C) Representative SERS
spectra of Suzuki-Miyaura
cross-coupling with R (green
line), +0.30 V versus OCP
(sky blue line), +0.15 V versus
OCP (royal blue line), OCP
(black line), and –0.15 V versus
OCP (red line). (D) Relative
intensities of Suzuki-Miyaura
cross-coupling under different
potentials. (E) Relative intensities
of P in a time course under
different potentials. The points
and error bars in the relative
intensity graphs [(D) and (E)]
indicate the means and standard
deviations of three to five ex-
periments, respectively.

rate with electron-donating groups. Whereas carboxylic acid is enhanced. Next, the carbonyl acid, as shown in Fig. 4B. Similarly, applying
the mechanistic interpretation of these re- carbon in the O-acylurea may act as an elec- a voltage of +0.40 V for 1 hour had no effect,
sults for the Suzuki-Miyaura reaction requires trophile to react with an amine to afford the
additional work, we establish that catalytic re- desired amidated product. In this second step, which is expected because making the benzoic
action rates can be modulated by the electro- higher electrophilicity of the carbon should acid less electron-rich should inhibit the urea
inductive effect. increase the reaction rate. Thus, the EDC-
mediated amidation should be facilitated by formation. Applying a negative potential gives
Encouraged by these two successful dem- first applying a negative potential and then rise to a new peak at 1772 cm–1, which is attri-
onstrations of the electro-inductive effect, we switching to a positive potential, as concep-
sought to test the idea of first enhancing the tualized in Fig. 4A. buted to the carbonyl stretching mode of the
nucleophilicity of the immobilized molecule O-acylurea, as is discussed in greater detail
and then increasing electrophilicity by invert- The SERS spectrum of the reactant 4-
ing the applied voltage. For this purpose, an mercaptobenzoic acid is shown in Fig. 4B and in the supplementary materials. The SERS
amidation mediated by a coupling reagent was reveals representative peaks for vs(COO–) and spectrum acquired after applying a voltage of
chosen (25). As outlined in Fig. 4, a carboxylic v(C=O) at ~1400 and ~1700 cm–1, respectively. –0.70 V for 1 hour clearly shows this new car-
acid such as benzoic acid first attacks 1-ethyl- First, we confirmed that under the given con- bonyl peak (Fig. 4C). Applying the –0.70 V for
3-(3-dimethylaminopropyl)carbodiimide ditions with all reagents present in solution, only 40 min and acquiring the spectrum after
(EDC) at the carbodiimide carbon to give an no reaction occurs if no voltage is applied, 20 min of applying no voltage gives a strong
O-acylurea. This first step should be accel- indicated by a stable and unchanged SERS
erated if the oxygen nucleophilicity of the spectrum of the reactant 4-mercaptobenzoic signal as well, and we can see a second feature
at 2228 cm–1 as a tiny peak. This is a charac-
teristic signature of a C≡N stretching mode
[v(C≡N)] (11), which is a molecular tag that
we had intentionally installed on the amine

Heo et al., Science 370, 214–219 (2020) 9 October 2020 4 of 6

RESEARCH | REPORT

Fig. 4. Carbodiimide-mediated amidation reaction. (A) Schematic condition (bottom graph). (E) Relative intensities of Int and P under different
representation for EDC-mediated amidation of 4-mercaptobenzoic acid. potentials. The prominent peak assigned to the vibration of the aromatic
After immobilizing the 4-mercaptobenzoic acid on the gold electrode, ring at ~1070 cm–1 was used as a reference peak. *Run at –0.70 V for 40 min,
EDC (0.3 M), 4-(aminomethyl)benzonitrile (2 mM), and NaClO4 (1 M) were then at OCP for 20 min. †Run at –0.70 V for 40 min, then at +0.40 V for
added to MES-buffered (0.1 M) aqueous solution (pH 5.5), and reactions were 20 min. ‡Ten cycles were conducted, where each cycle consisted of –0.70 V
run at 25°C for 1 hour. DG‡, Gibbs free-energy barrier. (B) Representative for 3 min, then +0.40 V for 3 min (total reaction time of 1 hour). The
spectra of 4-mercaptobenzoic acid (top graph, green line, R), EDC-mediated representative peaks of R, Int, and P are highlighted in yellow, pink, and cyan,
amidation with +0.40 V versus OCP (middle graph), and OCP (bottom graph). respectively. Insets show an enlarged SERS spectrum of the highlighted
(C) Representative SERS spectra of EDC-mediated amidation with –0.70 V regions. Scale bars in (C) and (D) indicate 0.05 relative intensity. The
(top graph) and a sequence of –0.70 V to OCP (bottom graph). Int, points and error bars in the relative intensity graph (E) indicate the means
intermediate. (D) Representative SERS spectra of EDC-mediated amidation and standard deviations of four to seven experiments, respectively. MES,
after a sequence of –0.70 V to +0.40 V (top graph) and pulsed potential 2-(N-morpholino)ethanesulfonic acid; R1, –C2H5; R2, –C3H6(NH(CH3)2)+.

Heo et al., Science 370, 214–219 (2020) 9 October 2020 5 of 6

RESEARCH | REPORT

coupling partner 4-(aminomethyl)benzonitrile technique. Another promising application is in designate these peaks as v(C–Br) and v(C–C)ring for
for monitoring purposes. The emergence of catalysis, where the electronics of the catalyst convenience.
this peak is the first indication that the two- or substrate can be changed easily. Whereas 19. G. Varsanyi, Vibrational Spectra of Benzene Derivatives
step reaction we envisioned has taken place. the gold electrode was convenient for the proof- (Elsevier, ed. 1, 1969).
Next, we applied –0.70 V for 40 min and then of-principle studies presented above, the rela- 20. F. Benz et al., Nano Lett. 15, 669–674 (2015).
switched to +0.40 V for 20 min, which sub- tively small surface area limits the effective 21. C. Amatore, A. Jutand, G. Le Duc, Chem. Eur. J. 17, 2492–2503
stantially increased the intensity of the peak concentration of the immobilized material (2011).
at 2228 cm–1 (Fig. 4D), suggesting that the to a picomolar range, which is not useful for 22. C. Amatore, A. Jutand, G. Le Duc, Angew. Chem. Int. Ed. 51,
positive potential accelerated the second step a large-scale application. Carbon-based elec- 1379–1382 (2012).
of the reaction, as anticipated. Finally, a pulse trodes offer a promising solution, and future 23. S. B. Kedia, M. B. Mitchell, Org. Process Res. Dev. 13,
mode was tested, where –0.70 V was applied efforts will concentrate on using such elec- 420–428 (2009).
for 3 min followed by 3 min of +0.40 V and trodes to increase the scale of the reactions 24. J. F. Hartwig, Inorg. Chem. 46, 1936–1947 (2007).
repeated 10 times during a total reaction time that can be accelerated or inhibited by the 25. E. Valeur, M. Bradley, Chem. Soc. Rev. 38, 606–631
of 1 hour. Both the carbonyl and cyanide peaks electro-inductive effect. (2009).
are clearly visible in both operational modes, 26. J. P. Wolfe, S. L. Buchwald, Angew. Chem. Int. Ed. 38,
as illustrated in Fig. 4D. The relative inten- REFERENCES AND NOTES 2413–2416 (1999).
sities of the two signals observed at differ- 27. J. Heo et al., Data for the paper: “Electro-inductive effect:
ent potentials are summarized in Fig. 4E 1. L. P. Hammett, J. Am. Chem. Soc. 59, 96–103 (1937). electrodes as functional groups with tunable electronic
and show that the application of a negative 2. S. Sarkar, J. G. Patrow, M. J. Voegtle, A. K. Pennathur, properties”. Zenodo (2020); http://doi.org/10.5281/zenodo.
potential leads to the formation of the O- 3970448.
acylurea. If the negative voltage is switched J. M. Dawlaty, J. Phys. Chem. C 123, 4926–4937 (2019).
off, these intermediates can react slowly with 3. S. G. Boxer, J. Phys. Chem. B 113, 2972–2983 (2009). ACKNOWLEDGMENTS
the amine in solution to afford the final prod- 4. S. D. Fried, S. Bagchi, S. G. Boxer, Science 346,
uct, albeit in very small amounts. The sec- Funding: This research was supported by the Institute for
ond step can be accelerated by applying a 1510–1514 (2014). Basic Science (IBS-R010-A1) in Korea. This work was also
positive voltage, which lowers the barrier by 5. C. Vericat, M. E. Vela, G. Benitez, P. Carro, R. C. Salvarezza, supported by the Basic Science Research Program
increasing the electrophilicity of the immo- (2015R1A3A2033469) through the National Research
bilized intermediate. Chem. Soc. Rev. 39, 1805–1834 (2010). Foundation of Korea (NRF) funded by the Ministry of Science
6. B. Vaidya, J. Chen, M. D. Porter, R. J. Angelici, Langmuir 17, and ICT. Author contributions: J.H., H.A., J.W., S.W.H., and
Our studies demonstrate that electrodes can M.-H.B. conceived and designed the project and wrote the
be used as functional groups with adjustable 6569–6576 (2001). manuscript; J.H., H.A., and J.W. carried out the experiments;
inductive effects to change the chemical re- 7. J. Lahann et al., Science 299, 371–374 (2003). J.H. and J.W. performed computational studies; J.G.S.,
activity of molecules that are attached to them, 8. A. Campion, P. Kambhampati, Chem. Soc. Rev. 27, H.K.S., and T.G.L. conducted TOF-SIMS analyses; M.-H.B.
closely mimicking the inductive effect of con- organized the research; and all authors analyzed the data,
ventional functional groups. This constitutes 241–250 (1998). discussed the results, and commented on the manuscript.
a potentially powerful new way of control- 9. The peak at ~1070 cm–1 is assigned to an in-plane ring Competing interests: The authors declare no competing
ling chemical reactions, offering an alternative financial interests. Data and materials availability: The
to preparing derivatives to install electron- breathing mode coupled with C–S stretching. We designate supplementary materials contain microscope images,
donating or electron-withdrawing functional this peak as v(C–C)ring for convenience. cyclic voltammetry studies, density functional theory
groups. Classical Hammett studies that aim to 10. A. Michota, J. Bukowska, J. Raman Spectrosc. 34, 21–25 (2003). calculations, SERS spectra, TOF-SIMS studies, and nuclear
gain mechanistic insight about the electronic 11. G. Socrates, Infrared and Raman Characteristic Group magnetic resonance spectra. The raw data of the base-catalyzed
demand of the rate-determining step can easily Frequencies: Tables and Charts (Wiley, 2004). hydrolysis reaction before the background correction
be imagined to be replicated through this new 12. H. Wackerbarth, L. Gundrum, C. Salb, K. Christou, W. Viöl, can be found at Zenodo (27).
Appl. Opt. 49, 4367–4371 (2010).
13. M. J. Pelletier, Appl. Spectrosc. 57, 20A–42A (2003). SUPPLEMENTARY MATERIALS
14. S. E. J. Bell et al., Angew. Chem. Int. Ed. 59, 5454–5462
(2020). science.sciencemag.org/content/370/6513/214/suppl/DC1
15. A. J. Bard, L. R. Faulkner, Electrochemical Methods: Materials and Methods
Fundamentals and Applications (Wiley, ed. 2, 2000). Supplementary Text
16. N. Miyaura, K. Yamada, A. Suzuki, Tetrahedron Lett. 20, Figs. S1 to S17
3437–3440 (1979). Tables S1 and S2
17. Y. Zhao, L. Du, H. Li, W. Xie, J. Chen, J. Phys. Chem. Lett. 10, References (28–45)
1286–1291 (2019).
18. The peaks at 1068 and 1079 cm–1 are assigned to 8 March 2020; accepted 21 August 2020
in-plane ring breathing modes coupled with C–Br 10.1126/science.abb6375
stretching and C–S stretching, respectively. We

Heo et al., Science 370, 214–219 (2020) 9 October 2020 6 of 6

RESEARCH

NOVEL COMMUNITIES nals of compositional change: (i) rapid change
to (ii) an unprecedented state. We quantified
Increased extinction in the emergence of novel these by decomposing an ecological time se-
ecological communities ries within a site into two sets of dissimilarity
scores (using the Jaccard index): a metric of
John M. Pandolfi1*†, Timothy L. Staples1†, Wolfgang Kiessling2 community change through time (rate of
change, called “instantaneous dissimilarity”)
Environmental change is transforming ecological assemblages into new configurations, resulting in novel (Fig. 1, A and D) and the smallest dissim-
communities. We developed a robust methodology to detect novel communities, examine patterns of ilarity to any past composition (community
emergence, and quantify probabilities of local demographic turnover in transitions to and from novel novelness, called “cumulative dissimilarity”)
communities. Using a global dataset of Cenozoic marine plankton communities, we found that the (Fig. 1, A and C). We classified communities
probability of local extinction, origination, and emigration during transitions to a novel community as novel only where these observed dissim-
increased two to four times that of background community changes. Although rare, novel communities ilarities exceeded expectations, generated
were five times more likely than chance to shift into another novel state. For marine plankton for each time series using parametric spline-
communities at a 100,000-year time grain, novel communities were sensitive to further extinctions based models (20). These models estimated
and substantial community change. average dissimilarities (Fig. 1B, dashed lines)
that could increase or decrease along the
P rofound changes in the biodiversity of tribution of novel climates (8, 16), including time series, meaning that novelty was assessed
global ecosystems (1, 2) are leading to under various Representative Concentration relative to local windows of time. Novelty oc-
the formation of novel communities, Pathway scenarios (18, 19). curred where observed dissimilarities exceeded
where species composition and diver- 95% predictive intervals of expectations (Fig.
sity are transformed into new, nonhis- We build on these previous approaches to 1B, gray shading). This resulted in four possible
torical configurations with altered ecosystem provide a new novelty-detection framework community states: faster-than-expected com-
functions (3–6). Key factors driving novel com- (20) that accords with modern definitions munity turnover (“instantaneous novelty,” I), a
munity emergence include the rapid pace of novel ecosystems (3) and enables compar- community state more distinct than expected
of global climate change (7–9), breakdown of ative ecosystem approaches to understand from any prior state (“cumulative novelty,”
biogeographic barriers, species invasions, and general trends, causes, and consequences of C), coincident instantaneous and cumulative
ecosystem degradation (10–12). Here, we in- novelty on a global scale. Our framework de- novelty (“novel community,” N), and “back-
troduce a reproducible, objective, and quanti- fines novel communities as having two sig- ground communities” (B) that lacked novelty
tative approach to detect novel communities (Fig. 1B) (20).
from time series compositional data, apply
this approach to investigate the frequency Fig. 1. Description of novel community
of emergence and transition probabilities of detection, showing the time-ordered,
community novelty over long temporal scales, space-restricted nature of our framework.
and explore the demographic processes that (A) Calculation of instantaneous (pairwise
influence the rise and fall of novel commu- dissimilarities; red) and cumulative (smallest
nity states. dissimilarity to any past state; blue)
dissimilarity in a compositional time series.
The conceptual basis of novel ecosystems (B) Mean expected dissimilarities (dashed
spans multiple disciplines. For conservation lines) obtained from generalized additive
biologists, novel ecosystems contain historically models for a single time series. Observed
unprecedented combinations of species that are dissimilarities (red and blue lines) were calculated
driven by human agency, often with altered eco- for 100,000-year sampling bins. Gray shading
logical functions (3). The no-analog community indicates the upper and lower 95% predictive
concept from plant paleoecology focuses on boundary. Bins that exceeded the upper
the taxonomic composition or environmental boundary were classified as either instanta-
framework of past communities that are com- neous novelty (I, in red) or cumulative
positionally unlike any found today (13). A va- novelty (C, in blue; y axis is inverted); bins
riety of quantitative analytical approaches have classified as both were considered to be
been used to understand how and why past true novel communities (N, in orange); and bins
(13–16) or even future (4) vegetation dynam- not classified as I, C, or N are background
ics in focal ecosystems differ from the present. states (B, in gray). (C and D) Examples of (C)
Early emphasis on the role of novel climates cumulative and (D) instantaneous novelty test
in ecological change (13, 17) has prompted projected on ordination axes. Points are time
more recent robust quantitative investigation series sampling bins (in direction of arrows);
of the geographic variation and temporal dis- square points indicate two example target
compositions, but novelty framework tests were
1Australian Research Council Centre of Excellence for Coral applied to all time series points. Target points are
Reef Studies, School of Biological Sciences, The University of colored where observed dissimilarities (Cobs and
Queensland, St Lucia, Queensland 4072, Australia. Iobs, respectively) exceed 95% predictive intervals
2GeoZentrum Nordbayern, Friedrich-Alexander Universität (radii of gray circles and Ccrit and Icrit, respec-
(FAU) Erlangen‐Nürnberg, 91054 Erlangen, Germany. tively). Gray circle radii are equivalent to the
*Corresponding author. Email: [email protected] upper edge of grey predictive regions in (B).

†These authors contributed equally to this work.

Pandolfi et al., Science 370, 220–222 (2020) 9 October 2020 1 of 3

RESEARCH | REPORT

We used our novelty-detection framework tion between instantaneous and cumulative Fig. 3. Expected probability of transitions
to investigate the emergence probability and dissimilarity scores [correlation coefficient between each novelty classification, and the
demographic drivers of community novelty (r) = 0.637; 95% confidence interval (CIs), ratio of observed to expected transition
in the Cenozoic marine plankton record, using 0.628 to 0.646; n = 17,719 sampling bin probabilities. Points are halved and dual-colored,
the Neptune Sandbox (NSB), a global set of comparisons; P ≤ 0.001]. Thus, subsequent representing the classification of the preceding and
microfossil data from deep sea drilling cores analyses built on particular sets of novel com- succeeding community. y axis points are ratios.
(21, 22). Incorporating updated taxonomy and munities are unlikely to be an artifact of our Faded points are transitions where the observed-to-
age models allowed us to build community arbitrary a cut-off. expected ratio was not different from one. Both
data for groups of taxa across geological time. axes are on a natural log scale. The result summary
Each time series in our analyses consisted of We also compared the observed pair-wise is available in table S5.
single or multiple cores aggregated into Long- probabilities of community state transitions
hurst biogeographical provinces (23), with spe- with nonautocorrelated estimates calculated tion of new taxa, likely through invasion or
cies presence and absence grouped separately from the occurrence probabilities of each state speciation.
for four groups of marine plankton (calcareous in the transition multiplied together (20). Most
nannoplankton, foraminifers, radiolarians, observed transitions were from background In contrast to the emergence of novelty,
and diatoms) every 100,000 years over the past to background states (Fig. 3, right). Transi- taxonomic turnover played a relatively minor
66 million years (fig. S1). tions between the three types of novelty and role in the subsequent transition from nov-
a background state were much rarer (Fig. 3, elty (I, C, or N) to background states (Fig. 4,
We found emergence probabilities of nov- middle), and many occurred less often than A and B). However, when novel communities
elty to be consistent across the four marine expected by chance. Transitions from one transition to a subsequent, different novel
planktonic groups. These probabilities esti- novelty state to another were rarer still and, state, these transitions were accompanied by
mated the frequency that novel planktonic with the exception of one transition, led to increased probabilities of demographic changes,
communities emerged over 100,000-year time subsequent novelty between 2 and 10 times including local extinction (Fig. 4A) and local
grains and under natural marine conditions, more often than expected (Fig. 3, left). Thus, origination (Fig. 4B). Pairing this result with
in the absence of human impacts (Fig. 2). These despite the infrequent occurrence of novelty the increased probability of novel commun-
probabilities were sensitive to the a cut-off throughout our time series, transitions be- ities emerging after existing novel commun-
threshold we used to detect novelty, which was tween novelty states were disproportionately ities (Fig. 3, left), naturally occurring novel
set to 0.05 (fig. S2A), but not choice of dissim- followed by subsequent novelty states, dem- communities recorded from the deep-sea
ilarity index (fig. S3). However, the proportional onstrating a heightened propensity for nov- marine record had the potential to cascade
overlap among the three novelty classifica- elty to beget further novelty. through multiple novelty transitions, each
tions was largely stable (fig. S2, B and C), and associated with increased rates of extinction
there was a strong positive Pearson correla- We evaluated the role of four demographic and origination.
processes in driving transitions between one
Fig. 2. Novelty emergence probabilities for community state and another (20). Local ex- To evaluate the robustness of our results, we
four marine planktonic groups. (A) Venn diagram tinction (the permanent loss of a species from computed sample completeness (figs. S5D and
with overall mean and 95% CI probabilities of a time series) and emigration (a transient loss S6); plotted species accumulation curves (fig.
novelty emergence (table S1). (B to D) Novelty with the species reappearing later in the time S7); undertook simulations (figs. S8 and S9
emergence probabilities and raw proportions for series) lead to species loss, whereas local orig- and table S12); and accounted for position
individual planktonic groups (tables S2 to S4). ination (the first occurrence of a species in a within time series (figs. S10 and S11), potential
Colored points indicate raw proportion of sampling time series) and immigration (a species sub- reworking (figs. S12 to S14), and species rich-
bins in each time series categorized as each novelty sequently reappearing after a temporary dis- ness (figs. S5C and S15). We also conducted
classification. White outlined points are mean appearance) lead to species gain. Transitions additional comparative analyses, with novelty
probabilities for each planktonic group (D, diatoms; from background to background states acted assessed at the scale of individual cores (figs.
F, foraminifers; N, calcareous nannoplankton; as a reference baseline for comparing demo- S16 to S18), with low richness communities ex-
and R, radiolarians). Lines indicating 95% CIs graphic probabilities. cluded (figs. S19 to S21), with varying sampling
are obscured behind mean estimates. bin widths (200,000 to 500,000 years) (figs. S22
Species turnover was a major feature in to S24), and analyzing taxa separately (figs. S25
the transition to novelty (fig. S4). On average, and S26) and through time (fig. S5, A and B)
taxa in the preceding community had a 10 to (20). Our results were largely consistent across
14% probability of going locally extinct in the these comparative scenarios.
transition to a novel community, which is more
than twice as high as transitions to background
communities (Fig. 4A). Similarly, taxa present
in the newly emerged novel community had
an 18 to 27% probability on average of being
entirely new to the time series (Fig. 4B, local
origination), which was more than three times
higher than transitions to background states.
Species gain in the transition to instantaneous
novelty (I) was driven more by species re-
colonization (Fig. 4B, immigration) than by the
origination of new taxa, which is consistent
with instantaneous novelty being a large shift
to a state similar to past states. By contrast,
species gain in transitions to new, previously
unseen community states (C) and novel com-
munities (N) came about through the addi-

Pandolfi et al., Science 370, 220–222 (2020) 9 October 2020 2 of 3

RESEARCH | REPORT

Fig. 4. Mean probabilities of taxonomic loss and fossil communities relative to modern com- system states that are linked with heightened
gain in the transition between two communities munities. However, short-term fluctuations in extinction risk.
along a time series. (A) Probability of local extinction community structure would have been aver-
plotted against probability of emigration. (B) Proba- aged out over the 100,000-year time bins REFERENCES AND NOTES
bility of local origination plotted against probability of analyzed in our fossil time series data, poten-
immigration. Points are halved and dual-colored, tially leading to lower estimates of past novelty 1. M. Dornelas et al., Science 344, 296–299 (2014).
representing the classification of the preceding and than in modern communities sampled over 2. S. A. Blowes et al., Science 366, 339–345 (2019).
succeeding community. The dashed line represents a much shorter time intervals (24). These and 3. R. J. Hobbs, E. Higgs, J. A. Harris, Trends Ecol. Evol. 24,
1:1 ratio; error bars are 95% CIs. Model summaries other factors will need to be assessed before
are available in tables S6 to S11. fossil assemblages can be used as an ecolog- 599–605 (2009).
ical baseline for the frequency of novelty in 4. V. C. Radeloff et al., Ecol. Appl. 25, 2051–2068
Our approach provides a quantitative tool living ecosystems. Nevertheless, our analyses
with which to examine the frequency and provide a foundation for understanding the (2015).
drivers of novelty in ecological communities past frequency and demographic drivers of nov- 5. A. E. Magurran, Science 351, 448–449 (2016).
that can be broadly applied at any temporal elty in natural settings over an extended time 6. R. J. Hobbs, L. E. Valentine, R. J. Standish, S. T. Jackson,
scale, for any taxonomic group, and any eco- frame and in the absence of human impacts.
system. However, we caution that additional Trends Ecol. Evol. 33, 116–128 (2018).
factors such as generation times and the time Of particular concern to conservation biol- 7. D. Stralberg et al., PLOS ONE 4, e6825–e6825 (2009).
frame under study must be considered in com- ogists is the fate of modern communities that 8. A. Ordonez, J. W. Williams, J. C. Svenning, Nat. Clim. Chang. 6,
parative analyses of the frequency or drivers of have transitioned into novel states (25): Will
community novelty. For example, long-term these new configurations of species persist, or 1104–1109 (2016).
studies that connect ecological and evolution- are they likely to lead to further community 9. B. R. Scheffers et al., Science 354, aaf7671 (2016).
ary time scales need to consider the role of change or even ecosystem collapse? Novelty 10. A. E. Lugo, T. A. Carlo, J. M. Wunderle Jr., Anim. Conserv. 15,
speciation (and extinction) in ecosystems from in our fossil marine planktonic communities
the deep past when comparing the frequen- most often transitioned to a background state, 233–241 (2012).
cy of novelty with present ecosystems. Our suggesting that many shifts to novel states 11. A. Traveset et al., Proc. Biol. Sci. 280, 20123040
analyses of novelty in marine plankton com- may persist through time. Where functional
munities across 66 million years inevitably redundancy exists, these changes in species (2013).
included an evolutionary component that composition may not reduce ecosystem func- 12. A. Anton et al., Nat. Ecol. Evol. 3, 787–800 (2019).
could lead to higher estimates of novelty in tioning per se. However, deep time transitions 13. J. W. Williams, B. N. Shuman, T. Webb III, Ecology 82,
to novel communities in marine plankton ex-
hibited both species loss and species gain but 3346–3362 (2001).
ultimately resulted in compositional shifts with 14. W. Finsinger, T. Giesecke, S. Brewer, M. Leydet, Ecol. Lett. 20,
at least twice the probability of local extinction
(Fig. 4A). We also observed that novel com- 336–346 (2017).
munities were five times more likely to give rise 15. V. Iglesias, C. Whitlock, T. R. Krause, R. G. Baker, J. Biogeogr.
to other novel communities than expected (Fig.
3, left). Although there are fundamental differ- 45, 1768–1780 (2018).
ences in ecological and environmental factors 16. K. D. Burke et al., Philos. Trans. R. Soc. Lond. B Biol. Sci. 374,
governing species composition over different
time scales, our results from fossil communities 20190218 (2019).
raise the intriguing hypothesis that modern 17. J. W. Williams, S. T. Jackson, J. E. Kutzbach, Proc. Natl. Acad.
novel ecological communities could also be
more likely than expected to experience tran- Sci. U.S.A. 104, 5738–5742 (2007).
sitions to subsequent novel community states, 18. M. C. Fitzpatrick et al., Glob. Change Biol. 24, 3575–3586
driven in part by further extinctions. Our nov-
elty framework is well suited to further study (2018).
of modern time series to test this directly. 19. C. R. Mahony, A. J. Cannon, T. Wang, S. N. Aitken,

Although we are aware of the dangers in Glob. Change Biol. 23, 3934–3955 (2017).
direct comparison of ecological dynamics over 20. Materials and methods and comparative analyses are available
vastly different time scales, our results raise
the possibility that efforts to reduce extinction as supplementary materials.
risk are consistent with the active management 21. D. Lazarus, Math. Geol. 26, 817–832 (1994).
and conservation of novelty in modern ecolog- 22. C. Spencer-Cervato, Palaeontol. Electronica 1999, 2
ical communities (26–28). Under the influence
of human impacts, failing to improve the con- (1999).
ditions that brought about the transition to 23. A. R. Longhurst, Ecological Geography of the Sea (Academic
novelty may facilitate further novelty accom-
panied by additional species extinction. Our Press, ed. 2., 2007).
results contravene any notion that a fixed 24. D. B. Kemp, K. Eichenseer, W. Kiessling, Nat. Commun. 6, 8890
historical baseline of a community can or
should be the only conservation goal for nov- (2015).
el communities (3). Rather, ecosystem man- 25. E. Marris, Nature 460, 450–453 (2009).
agement should also focus on how to prevent 26. A. M. Truitt et al., Environ. Manage. 55, 1217–1226
transitions to additional previously unseen eco-
(2015).
27. S. J. Capon, G. J. Palmer, Solutions 9, 3 (2018).
28. S. Clement, R. J. Standish, J. Environ. Manage. 208, 36–45

(2018).
29. T. Staples, TimothyStaples/novelty-cenozoic-microplankton:

Primary release of code used for data analysis and modeling.
Zenodo (2020); doi:10.5281/zenodo.4031861.

ACKNOWLEDGMENTS

We are grateful for discussion about this work with S. Jackson,
D. Lazarus, J. Levine, and R. Norris. Funding: This work was
funded by the Australian Research Council’s Centre of Excellence
for Coral Reef Studies (CE140100020) and by the Deutsche
Forschungsgemeinschaft (DFG; German Research Foundation,
KI 806/16-1). Author contributions: The ideas for the paper were
conceived by J.M.P. and W.K. and enhanced by T.L.S.; T.L.S.
undertook all analyses and developed analytical approaches, with
input from J.M.P. and W.K.; and J.M.P. and T.L.S. wrote the
paper, with W.K. contributing to writing and editing. Competing
interests: The authors declare no competing interests. Data
and materials availability: Data used in this study are available
from the Neptune Sandbox microfossil database (http://nsb-mfn-
berlin.de/search). R code to conduct all analyses and produce
all tables and figures are available at Zenodo (29).

SUPPLEMENTARY MATERIALS

science.sciencemag.org/content/370/6513/220/suppl/DC1
Materials and Methods
Figs. S1 to S26
Tables S1 to S12
References (30–43)
MDAR Reproducibility Checklist

21 February 2020; accepted 20 August 2020
10.1126/science.abb3996

Pandolfi et al., Science 370, 220–222 (2020) 9 October 2020 3 of 3

RESEARCH

ELECTRON MICROSCOPY To explain this movement, we suggest a
physical model, based on considering the
Cryo-EM with sub–1 Å specimen movement stresses that are created in the specimen
during vitrification and the pseudo-diffusion of
Katerina Naydenova1, Peipei Jia1,2*, Christopher J. Russo1† the water molecules under electron irradiation
(9) (Fig. 2 and supplementary text). In typical
Most information loss in cryogenic electron microscopy (cryo-EM) stems from particle movement cryo-EM specimens, the ice deforms twice, once
during imaging, which remains poorly understood. We show that this movement is caused by during freezing and again during irradiation
buckling and subsequent deformation of the suspended ice, with a threshold that depends with the electron beam. Vitreous water (low-
directly on the shape of the frozen water layer set by the support foil. We describe a density amorphous ice) has 6% lower density
specimen support design that eliminates buckling and reduces electron beam–induced particle than liquid water at 4°C (10, 11). During cryo-
movement to less than 1 angstrom. The design allows precise foil tracking during imaging plunging, freezing occurs over a time interval
with high-speed detectors, thereby lessening demands on cryostage precision and stability. of ≤10–4 s (10). Within this time, the water
It includes a maximal density of holes, which increases throughput in automated cryo-EM density change is most rapid near the homoge-
without degrading data quality. Movement-free imaging allows extrapolation to a neous nucleation temperature, ~235 K (12, 13).
three-dimensional map of the specimen at zero electron exposure, before the onset of As the water solidifies, the rapid cooling does
radiation damage. not allow sufficient time for structural rear-
rangements of the water molecules and causes
D espite the current success of cryo-EM the particles by 150 to 250 Å (4, 5). More re- the buildup of compressive strain within the
in determining the structure of numer- cent tracking methods yield similar types of thin film (Fig. 2A). Once the compression ex-
ous macromolecular complexes that trajectories for various smaller protein speci- ceeds a critical value, the film buckles, thus
were previously intractable, several out- mens (6, 7). Gold nanoparticles embedded in momentarily relieving the radial stress in the
standing problems remain: Specimen vitreous ice provide a high-contrast fiducial layer. Buckling only occurs if the stress ex-
movement at the beginning of electron beam marker to measure the movement of the ice ceeds a critical point, which is determined by
irradiation degrades image quality and reduces during a typical high-resolution, low-dose the dimensions of the ice film, its elastic moduli,
information about the undamaged structure micrograph (Fig. 1A and fig. S1). The types of specific volume change relative to the support,
(1). Thus, all current structures determined movement that occur during cryo-EM imag- and the constraints at the edge of the hole
by cryo-EM have varying degrees of radiation ing, ordered by magnitude, are (i) stage drift, (supplementary text and figs. S13 to S15). As
damage incorporated into the resulting atomic (ii) bending of the support foil, (iii) ice bending, the vitreous film continues to cool to the tem-
models. (iv) Brownian movement during irradiation, perature of the surrounding cryogen (typically
and (v) molecular vibrations (supplementary liquid ethane at 90 to 93 K), more stress builds
The throughput of modern synchrotron text). To isolate the movement of the ice layer, up as a result of further changes in relative den-
crystallography beamlines vastly exceeds that we eliminate the movement of the specimen sity. This stress is stored in the film indefinitely
of current state-of-the-art electron microscopes; support by using ultrastable gold foils that at 77 K (liquid nitrogen temperature).
hence, high-resolution structure determina- do not move during irradiation (8) and in-
tion by cryo-EM is feasible but throughput is clude the edge of the gold hole in the image High-energy electron irradiation increases
comparatively limited in practice (2). Rapid for accurate (<0.1 Å) stage drift correction. the effective diffusivity of water molecules by
structure determination by cryo-EM is fur- What remains is then the movement of the ice 46 orders of magnitude (Fig. 2B) (14), causing
ther hampered by structural heterogeneity within the hole and the individual particles the ice in the electron beam to behave as an
in many samples, as well as by irreproduci- within the ice. ultraviscous fluid. This change in the mechanical
ble and destructive interactions with surfaces properties of the film results in unbalanced
during specimen preparation (3). Here, we From measurements of the movement of bending moments in it; these cause a correlated
present a physical theory of the causes of move- ~15,000 particles from hundreds of holes in and rapid displacement of the particles in the
ment of cryo-EM specimens during imag- ultrastable gold foils (8), with hole diameters same direction as the preexisting buckling (Fig.
ing and use this work to design and implement from the typical 1 to 2 mm in standard specimen 1D), apparent in cryomicrographs. The move-
a specimen support that eliminates move- support foils down to 200 nm, a pattern of ment is enabled and accompanied by random
ment and helps to address some of these movement becomes clear (Fig. 1, B and C, and movement due to the continuous pseudo-
limitations. figs. S2 to S8). There is an abrupt, spatially diffusion of the surrounding water and only
correlated displacement of the particles at occurs during electron irradiation. The pre-
Proteins rapidly frozen for cryo-EM in the onset of irradiation (i.e., during the first dicted critical aspect ratio (diameter:thickness),
amorphous ice are inherently low-contrast, ~4 e–/Å2). This is followed by a less correlated above which the film is expected to buckle
making them difficult to track and image. movement that continues with reduced speed during freezing and become unstable during
Previous work using virus particle tracking on to the end of the exposure. The initial movement irradiation, is ~11:1 (supplementary text). When
amorphous carbon supports showed a “drum- is particularly apparent in the tilted micro- the aspect ratio of the vitreous ice film is below
like motion” that included deformation of the graphs of the 2-mm holes, where the vertical the critical value (i.e., holes 330 nm or 180 nm
support foil and the ice layer, which together (out of plane) displacement of the ice is 10 to in diameter for 300-Å-thick ice in Fig. 1, B and
cause a rotation and vertical displacement of 50 Å in a 30 e–/Å2 exposure (fig. S4). In con- C), the ice never builds up enough stress to
trast, the movement of the particles in 300- and buckle during freezing or to deform under ir-
1MRC Laboratory of Molecular Biology, Cambridge CB2 0QH, 200-nm holes is isotropic (the same in the radiation. Such films remain stable throughout
UK. 2ARC Centre of Excellence for Nanoscale BioPhotonics plane of the support and perpendicular to it), irradiation, and only diffusive movement occurs
(CNBP), Institute for Photonics and Advanced Sensing, is spatially uncorrelated, and scales with the (fig. S16); this is limited to <1 Å RMS (root mean
School of Physical Sciences, University of Adelaide, Adelaide square root of the incident electron fluence, as squared) in 30 e–/Å2.
5005, Australia. would be the case for diffusional motion (figs.
*Present address: Shenzhen Topmembranes Technology Co. Ltd., S2 and S9 to S12). On the basis of the above theory, we propose
Shenzhen 518000, China. a simple method for choosing the dimensions
†Corresponding author. Email: [email protected] of a support foil to prevent buckling for a given

Naydenova et al., Science 370, 223–226 (2020) 9 October 2020 1 of 4

RESEARCH | REPORT

specimen: The optimal foil thickness is set by small holes, the thickness of the suspended ice of gold less than ~400 Å thick are not stable
the desired ice thickness, and the optimal foil is largely controlled by the foil thickness; uni-
hole diameter is ≤11 times the ice thickness form and stable thin films of vitreous water because of their polycrystalline grain structure
(fig. S17). For example, for a 200-Å-thick particle can be formed. Unfortunately, for most single- (15). We propose a specimen support design,
in 300-Å-thick ice, the optimal hole diameter particle specimens, the desired foil hole size “HexAuFoil” (Fig. 3)—which entirely elimi-
is below 330 nm. For 100-Å-thick ice, poten- is less than the practical limit of conventional nates the buckling of the suspended amorphous
tially desirable for sub–100-kDa specimens, photolithography techniques currently used to ice during irradiation—and provide a scale-
the optimal diameter is below 110 nm. In such make specimen supports, and evaporated foils able method for manufacturing these supports

based on phase interference lithography [Talbot

A B 20 Ce– 20 Ee–

hole ø 2.0
RMS displacement (Å)
RMS displacement (Å)15 0o tilt15 30o tilt
Diameter (µm)

11:11as5:p1ecta rsaptieoct ratio
10 hole ø 10 unstable ratios
1.5 (buckling)

5 5

0 0 1.0
0 10 20 30
0 10 20 30 Fluence (e-/Å2) 0.5 stable ratios
buckling if N > N0 DPS (no buckling)
Fluence (e-/Å2) (2a/h > 11)

D Au particle h N

vitreous ice

Au foil 0.0
0
gold particle N 500 1000 1500 2000
displacement Thickness (Å)
2a

Fig. 1. Movement of gold nanoparticles in vitreous ice in a range of foil denote SEM. (D) Thin films of ice used in cryo-EM buckle during vitrification if the
hole sizes. (A) Typical drift-corrected electron micrograph used for tracking gold compressive stress (N) exceeds a critical value (N0), determined by the aspect
particles in vitreous water on all-gold supports; scale bar, 0.5 mm. The inset ratio (2a/h) of the film. Electron irradiation causes the film to move in response
(20 nm × 20 nm) shows an overlay of the positions of a gold nanoparticle to additional stresses in it, as is evident from the correlated particle movement at
at the beginning of irradiation and after a fluence of 60 e–/Å2. (B and C) RMS the beginning of irradiation. (E) The range of stable hole diameters can be
displacements of 200 to 2000 particles from 10 to 50 movies of holes of determined for a given ice thickness (blue shaded region). The two critical aspect
different diameters (yellow, UltrAuFoil R2/2, 1.9-mm holes; green, UltrAuFoil ratios (dashed and dotted lines) are calculated according to two theoretical
R1.2/1.3, 1.2-mm holes; pink, UltrAuFoil R 0.6/1, 0.8-mm holes; purple and blue, models of buckling during vitrification (supplementary text). The black and red
custom-made grids with 0.3-mm and 0.2-mm holes, respectively) are plotted as a data points show the hole sizes and ice thicknesses for the gold particle and
function of cumulative electron fluence for 0° tilt (B) and 30° tilt (C). Error bars DPS datasets, respectively. Error bars denote SD.

Fig. 2. Model of stress accumulation A 77 84 91 147 ~238 277K B 77 147
in thin films of amorphous ice during (4°C)
cryoplunging, and their response to 1.01 nitrogen boiling point glass liquid 10+10 liquid
electron irradiation. (A) The density of microscope stage aqueous
1.00 liquid ethane plunge specimen
diffusivity 1 Å2/s
liquid and amorphous water is plotted 0.99 1
(black markers) as a function of tem-
cryoplunge
perature, as reported in (12, 13), with a Density (g/mL) 0.98 homogeneous Diffusivity (Å2/s) 10-10 electron irradiation electron flux 0.1 to 10 e–/Å2/s
black line to guide the eye. During 0.97 nucleation 10-20
cryoplunging into liquid ethane, water is 0.96 10-30
rapidly cooled, typically from 277 K to 0.95 critical stress buildup
91 K (black arrow). The largest specific
volume change experienced by water thin film
buckling

below its homogeneous nucleation point 0.94 10-40 Stable cryo regime
(red) is (DV/V)max ≈ 6%. The thin film
can only withstand compression of up to 0.93 10–46
(DV/V)critical before it buckles (purple
range corresponds to a 300-Å-thick 0.92 10-50 diffusivity
layer in a 1-mm hole). (B) The diffusivity 0.91 at 84K
of water molecules in liquid and amor- Stable cryo < 10–4 seconds
phous ice is plotted (black crosses) as a regime 10-60
100 200 300 0 100 200 300
0 Temperature (K)
Temperature (K)

function of temperature, as reported in
(14). The black line is a fit to these values, as proposed in (14). The extrapolated diffusivity in amorphous ice at 84 K is vanishingly low, ~10–46 Å2/s. The blue shaded

region indicates the range of diffusivity in amorphous ice at temperatures in the 0 to 100 K range, where it is stable indefinitely. During imaging with 300-keV
electrons, water molecules move pseudo-diffusively by 1 Å2/(e–/Å2) (9). At typical imaging fluxes of 0.1 to 10 (e–/Å2)/s, this is equivalent to 0.1 to 10 Å2/s (orange)

and corresponds to an instantaneous local temperature of 147 K.

Naydenova et al., Science 370, 223–226 (2020) 9 October 2020 2 of 4

RESEARCH | REPORT

displacement (16, 17)] and low-temperature thick gold foil with 260-nm holes (Fig. 3F calculated from a single frame corresponding
evaporation (figs. S18 and S19). The holes are and figs. S20 to S22). The average resolution, to a total fluence of 1 e–/Å2 (supplementary text
arranged in a hexagonal pattern in a foil whose from an initial reconstruction from ~9 hours and fig. S24), which is less than the typical dose
thickness (280 Å for the example in Fig. 3F) of automated data collection on a modern limit used in x-ray crystallography [10 MGy
can be matched to the specimen; the holes still 300-keV microscope, easily reached <2 Å; the (25)]. Second, the complex structure factors
remain round (fig. S19B). Interestingly, the total particle displacement was 0.86 Å RMS in at each pixel in Fourier space can be fit with
nanoscale dimensions of the array imply that 35 e–/Å2 of irradiation (Fig. 4A, fig. S23, and exponential functions decaying with dose, which
the foil has plasmon resonances in the visible supplementary text). The absence of buck- extrapolate back to an initial “undamaged” val-
range, which causes the foil to appear yellow ling also ensured that no tilting of the par- ue at zero exposure (Fig. 4, B and C). A sim-
on reflection with white light but blue on ticles occurred during imaging. In contrast ilar method for amplitudes only was previously
transmission (Fig. 3, A and B, and fig. S18G)—a to all previous sub–2 Å resolution single- proposed but has not been widely used in x-ray
property that might be used in the future for particle cryo-EM datasets to date (supplemen- crystallography (26). Maps produced by both
characterizing the specimen before imaging tary text), maps reconstructed from each frame of these methods show improved densities for
with electrons. The foil is suspended across a showed that the first frame (1 e–/Å2 or 3 MGy) radiation-
3-mm hexagonal mesh grid; together these contained the most structural information (Fig. sensitive side chains, ordered water molecules,
hexagonal arrays increase the usable area by 4A) and that the quality (B-factor) of sequential and other complexed ions; high-resolution,
a factor of 10 relative to a standard cryo-EM frames decayed linearly with dose/fluence. A <10 MGy reconstructions avoid potential prob-
grid. More than 800 images can be acquired linear decay in B-factor with dose is expected lems in accurately modeling radiation-sensitive
from a single stage position, enabling a faster from studies of radiation damage in x-ray and parts of biological molecules and make the per-
rate of automated data collection (18, 19), electron crystallography (21, 22) but had never atom B-factors directly interpretable as move-
and more than 5000 individual holes can be been observed for single-particle cryo-EM be- ment. Further, the progressive effects of radiation
imaged in a single 25-mm-wide hexagon (Fig. cause of movement of the specimen at the onset are directly evident in the sequential maps and
3C). Having the edge of the gold foil in each of irradiation. can be used to improve atomic modeling tech-
image (Fig. 3, E and F, and fig. S20) has the niques and inform understanding of radiation
additional benefit of allowing rapid and ac- Current cryo-EM data processing and recon- damage (Fig. 4D). After fully removing move-
curate drift tracking independent of the sig- struction algorithms derive the final recon- ment, it is also possible to separate out the
nal from the specimen within the holes, which structed map by summing the information in effects of other factors that reduce resolution,
reduces the demands on stage precision and each frame, down-weighted to account for such as particle heterogeneity and deviation
stability. movement and damage (23, 24). In contrast, from symmetry (fig. S22A). Elimination of
decoupling specimen movement from radia- beam-induced foil and ice movement pro-
To demonstrate the use of movement-free tion damage affords new approaches to recon- vides an overall B-factor improvement of 54 ±
specimen supports for high-resolution cryo-EM, struction based on the physical theory of how 10 Å2 over motion-corrected micrographs for
we determined the structure of the 223-kDa the structure factors decay with exposure due this specimen in the sub–4 Å resolution range
DNA protection during starvation (DPS) pro- to mass loss and radiation damage only (sup- (fig. S22B).
tein (20), plunge-frozen on grids with 280-Å- plementary text). First, a reconstruction can be

AC DF

200 µm 230 nm
600 nm
B
E

200 nm

10 µm 50 nm

Fig. 3. HexAuFoil is an all-gold specimen support designed for 5000 holes in a regular pattern. The green circle encloses more than
movement-free cryo-EM imaging. (A and B) Optical micrographs showing 800 holes, which can all be imaged at high magnification without moving
reflected (A) and transmitted (B) unpolarized white illumination of the the stage during high-speed data collection (fig. S21). (D) Transmission
patterned gold foil (hexagonal array of 200-nm-diameter holes with electron micrograph of the holey gold foil. The arrows show the pitch
600-nm pitch) on a 600-mesh thin-bar gold grid. The scale and the of the regular hexagonal pattern. (E) Transmission electron micrograph of a
corresponding area are the same for (A) and (B). The blue color of the single hole in the nanocrystalline foil. The roundness of the 200-nm hole is
foil in transmitted light is due to a strong red absorption enhancement by improved by reducing the gold grain size to 10 nm or less, which can be
the periodic hole pattern (fig. S18G). (C) Transmission electron micrograph achieved by deposition onto a cooled template (fig. S19). (F) Low-dose
of a single hexagonal grid opening on one of these grids. A 3-mm grid transmission electron micrograph, at 1.5-mm defocus, of the protein DPS
contains ~800 of these hexagons, each of which includes more than (223 kDa) vitrified on a HexAuFoil grid with 260-nm holes.

Naydenova et al., Science 370, 223–226 (2020) 9 October 2020 3 of 4

RESEARCH | REPORT

Fig. 4. Structure of DPS determined at <2 Å A - 5.0 ± 0.3 Å 2/(e -/Å 2) EM map struct. factor phases (rad)CEM map struct. factor amplitudes (arb)1 × 10-2
resolution by extrapolation to zero dose with Relative B factor (Å2) π 0 -π
the use of a 260-nm hole support. (A) Plot of -200 -150 -100 -50 0 4.7 Å
Mean squared displacement (Å2) 3.1 Å
12 0.862 0.72 0.52 0
the mean squared displacement during 5 × 10 -3 2.3 Å

irradiation (red) for all DPS particles used in the 0 0.02 Å2 /(e- /Å2 ) 30 4.7 Å
reconstruction, and the relative B-factor for each
frame with respect to the first (black) with linear B 10 20 0 4.7 Å
fits to both. The displacement of the particles Fluence (e-/Å2) 4.7 Å
corresponds to diffusion with a constant of 3.1 Å
0.02 Å2/(e–/Å2) (red line). The B-factor decay 0 10 20 2.3 Å
agrees with the expected slope from radiation Fluence (e-/Å2) 30
damage alone (32). (B) The real (triangles) and
imaginary (squares) parts of selected Fourier EM map struct. factors (arb) 5 × 10 -3 D 0 e-/Å2 1 e-/Å2 10 e-/Å2 20 e-/Å2 30 e-/Å2
pixels are plotted as a function of fluence. 0 MGy 3 MGy 30 MGy 60 MGy 90 MGy
(C) The phases (triangles) and amplitudes
(squares) of selected Fourier pixels (at 2.3 Å, - 5 × 10 -3 0 Im 4.5 Å His51
3.1 Å, and two different pixels at 4.7 Å resolution) Re 7.0 Å Glu82
are plotted as a function of fluence. The real Re 3.1 Å
and imaginary parts (B) or phases and amplitudes Re 2.2 Å
Im 2.2 Å
Im 3.1 Å
Im 7.0 Å
Re 4.5 Å

(C) are extrapolated to their values before Asp156
the onset of irradiation, corresponding to H2O

the undamaged structure (solid symbols at 0 10 20 30
0 fluence). The lines in (B) and (C) are Fluence (e-/Å2)

exponential fits (to the real parts, imaginary parts,

and amplitudes) or linear fits (to the phases). (D) Selected side chains and a water molecule from zero-dose extrapolated and per-frame DPS reconstructions show the

progression of radiation damage. The residues from the refined model are colored by atom (gray, C; blue, N; red, O), and the contoured density map is shown as a mesh.

The specimen support described here re- 3. R. M. Glaeser, Curr. Opin. Colloid Interface Sci. 34, 1–8 ACKNOWLEDGMENTS
duces particle movement in a cryomicro- (2018).
graph to the limit set by pseudo-diffusion, We thank A. Howie for discussions and advice throughout this
which is on average less than the diameter of 4. R. Henderson et al., J. Mol. Biol. 413, 1028–1046 project; M. J. Peet, R. Henderson, G. McMullan, T. Nakane, S. Scheres,
a hydrogen atom. This allows reconstruction (2011). R. A. Crowther, A. Leslie, J. Dickerson, L. A. Passmore, and P. A. Midgley
of a complete map of an undamaged struc- for helpful discussions; Y. Lee and P. C. Edwards for the DPS
ture at 1.9 Å resolution. We note that inter- 5. A. F. Brilot et al., J. Struct. Biol. 177, 630–637 (2012). specimen; S. Chen, G. Cannone, G. Sharov, and A. Yeates of the
action of the specimen with the air/water 6. S. Q. Zheng et al., Nat. Methods 14, 331–332 (2017). Laboratory of Molecular Biology Electron Microscopy Facility; K. Sader,
interface remains a limitation for cryo-EM spe- 7. J. Zivanov, T. Nakane, S. H. W. Scheres, IUCrJ 6, 5–17 P. Qian, and P. Castro-Hartmann (Thermo Fisher Scientific) for
cimen preparation (3) and can be the domi- assistance with data collection on the Cambridge Pharma Consortium
nant factor determining the success of a (2019). microscope housed in the Department of Materials Science and
project (fig. S22D). Several approaches have 8. C. J. Russo, L. A. Passmore, Science 346, 1377–1380 Metallurgy, University of Cambridge; and J. Grimmett and T. Darling
been used to address this, including addition (MRC Laboratory of Molecular Biology Scientific Computing) and the
of a surface such as functionalized graphene (2014). MRC Laboratory of Molecular Biology Mechanical and Electronics
(27–30) and decreasing the time of interac- 9. G. McMullan, K. R. Vinothkumar, R. Henderson, workshops for technical assistance. Funding: Supported by a Vice-
tion with surfaces before freezing (31). Both Chancellor’s Award (Cambridge Commonwealth, European and
are compatible with the described support, Ultramicroscopy 158, 26–32 (2015). International Trust) and a Bradfield scholarship (K.N.), an Australian
and current work is focused on integrating 10. J. Dubochet et al., Q. Rev. Biophys. 21, 129–228 (1988). Nanotechnology Network overseas travel fellowship (P.J.), and Medical
the small aspect ratio and hexagonal pat- 11. T. Loerting et al., Phys. Chem. Chem. Phys. 13, 8783–8794 Research Council grant MC_UP_120117. Author contributions:
tern of holes with other existing technolo- K.N., P.J., and C.J.R. designed and performed the templated
gies and manufacturing the devices at scale. (2011). small-hole grid fabrication experiments; K.N. and C.J.R. collected
Movement-free imaging will also allow the 12. F. Mallamace et al., Proc. Natl. Acad. Sci. U.S.A. 104, and analyzed the cryo-EM data, performed other movement,
investigation of lower-temperature cryomi- density, and low-temperature physical vapor deposition–related
croscopy (closer to 0 K) where the secondary 18387–18391 (2007). experiments, developed the theory of ice movement, and wrote the
effects of radiation damage may be further 13. V. Holten, M. A. Anisimov, Sci. Rep. 2, 713 (2012). paper. Competing interests: The authors are inventors on a patent
reduced from those at liquid nitrogen tem- 14. R. S. Smith, Z. Dohnalek, G. A. Kimmel, K. P. Stevenson, application GB2004272.7 on the design of the specimen support,
peratures, thus affording more contrast per filed by the Medical Research Council as part of United Kingdom
image. With these improvements, cryo-EM B. D. Kay, Chem. Phys. 258, 291–305 (2000). Research and Innovation. Data and materials availability: The
will continue to rapidly expand our under- 15. C. J. Russo, L. A. Passmore, J. Struct. Biol. 193, 33–44 entire DPS dataset is publicly available in the Electron Microscopy
standing of the structures and functions of Public Image Archive (EMPIAR-10445). The final reconstructed
biological molecules. (2016). maps from each frame and the weighted sum are deposited in the
16. H. H. Solak, C. Dais, F. Clube, Opt. Express 19, 10686–10691 Electron Microscopy Data Bank (EMD-11210), and the refined
REFERENCES AND NOTES atomic model in the Protein Data Bank (PDB-6ZGL). All code for
1. R. Henderson, C. J. Russo, Microsc. Microanal. 25 (suppl. 2), (2011). zero-dose data processing is freely available from the author’s
17. K. Jefimovs et al., in Advances in Patterning Materials and website, www.mrc-lmb.cam.ac.uk/crusso/, and is deposited in Zenodo
4–5 (2019). at https://dx.doi.org/10.5281/zenodo.3944689.
2. M. Saur et al., Drug Discov. Today 25, 485–490 (2020). Processes XXXIV, C. K. Hohle, Ed. (SPIE, 2017), pp. 140–146.
18. C. Suloway et al., J. Struct. Biol. 151, 41–60 (2005). SUPPLEMENTARY MATERIALS
19. M. Schorb, I. Haberbosch, W. J. H. Hagen, Y. Schwab,
science.sciencemag.org/content/370/6513/223/suppl/DC1
D. N. Mastronarde, Nat. Methods 16, 471–477 (2019). Materials and Methods
20. R. A. Grant, D. J. Filman, S. E. Finkel, R. Kolter, J. M. Hogle, Supplementary Text
Figs. S1 to S24
Nat. Struct. Biol. 5, 294–303 (1998). Table S1
21. M. Warkentin, J. B. Hopkins, J. B. Haber, G. Blaha, R. E. Thorne, References (33–82)

Acta Crystallogr. D 70, 2890–2896 (2014). 24 March 2020; accepted 15 July 2020
22. J. Hattne et al., Structure 26, 759–766.e4 (2018). 10.1126/science.abb7927
23. S. H. W. Scheres, eLife 3, e03665 (2014).
24. T. Grant, N. Grigorieff, eLife 4, e06980 (2015).
25. R. Henderson, Proc. R. Soc. B 241, 6–8 (1990).
26. K. Diederichs, S. McSweeney, R. B. Ravelli, Acta Crystallogr. D

59, 903–909 (2003).
27. K. Naydenova, M. J. Peet, C. J. Russo, Proc. Natl. Acad. Sci. U.S.A.

116, 11718–11724 (2019).
28. X. Fan et al., Nat. Commun. 10, 2386 (2019).
29. E. D’Imprima et al., eLife 8, e42747 (2019).
30. F. Wang et al., J. Struct. Biol. 209, 107437 (2020).
31. A. J. Noble et al., eLife 7, e34257 (2018).
32. M. J. Peet, R. Henderson, C. J. Russo, Ultramicroscopy 203,

125–131 (2019).

Naydenova et al., Science 370, 223–226 (2020) 9 October 2020 4 of 4

RESEARCH

PLANT SCIENCE for the detection of CMV. Although the virus
was directly inoculated into stem cells in the
WUSCHEL triggers innate antiviral immunity in plant L1 cell layer and the organizing center (OC) in
stem cells the L2 cell layer, where the WUS protein is
located, we observed that the CMV only spread
Haijun Wu1, Xiaoya Qu1, Zhicheng Dong2, Linjie Luo1, Chen Shao1, Joachim Forner3*, under the L2 layer (Fig. 1, J and K), suggesting
Jan U. Lohmann3, Meng Su1, Mengchu Xu1, Xiaobin Liu2, Lei Zhu4, Jian Zeng1, Sumei Liu1, that the WUS protein–containing regions, in-
Zhaoxia Tian1†, Zhong Zhao1† cluding stem cells and the OC, are resistant to
CMV invasion.
Stem cells in plants constantly supply daughter cells to form new organs and are expected to safeguard
the integrity of the cells from biological invasion. Here, we show how stem cells of the Arabidopsis To test whether WUS proteins are involved
shoot apical meristem and their nascent daughter cells suppress infection by cucumber mosaic virus in the host defense against viruses, we exam-
(CMV). The stem cell regulator WUSCHEL responds to CMV infection and represses virus accumulation ined WUS protein distribution patterns during
in the meristem central and peripheral zones. WUSCHEL inhibits viral protein synthesis by repressing the CMV infection at 9 dpi using wus/pCLV3::
expression of plant S-adenosyl-L-methionine–dependent methyltransferases, which are involved in mCherry-NLS; pWUS::WUS-GFP rescue plants.
ribosomal RNA processing and ribosome stability. Our results reveal a conserved strategy in plants to In virus-free plants, WUS protein expression
protect stem cells against viral intrusion and provide a molecular basis for WUSCHEL-mediated broad- overlapped with CLV3 transcript expression
spectrum innate antiviral immunity in plants. in stem cells (Fig. 2, A to D). Upon CMV inva-
sion, WUS protein expression (Fig. 2, E, G, and
P lant stem cells and their newly differ- blocks virus movement or if WUS inhibits virus H, and fig. S4), but not transcript expression
entiated daughter cells in the shoot apical accumulation. We generated an mp clv3-7 (fig. S5), increased. WUS protein was ectopi-
meristem (SAM) provide cells for all above- double mutant with an enlarged SAM, which cally expressed in the entire PZ and young
ground organs during postembryonic floral primordia in a pattern complementary
development (1). This tissue is free from allowed us to inoculate CMV directly into and to the CMV distribution in the shoot apex
viral invasion. Indeed, meristem tip culture below the WUS expression domain (Fig. 1I). (Fig. 1D).
eliminates viruses from infected plants and We performed a serial sectioning–based mis-
is widely used to develop virus-free cultivars match assay, with one section for the detection In plants in which WUS was inducible with
(2). Here, we analyzed the broad-spectrum of WUS expression and the adjacent section dexamethasone (DEX) treatment [pUBQ10::
antiviral mechanism that safeguards pluri- WUS-GR plants (9)], we induced WUS 1 day
potent cells in the SAM from viral invasion. after inoculating with CMV. Without induction,
89% of the plants were infected by viruses.
To test the hypothesis that the top of the
SAM is virus free (3), we followed virus dis-
tribution in the Arabidopsis SAM after inocu-
lating the leaves with cucumber mosaic virus
(CMV) (Fig. 1A). At 9 days postinoculation
(dpi), we observed CMV accumulations in the
shoot and the rib zone of the SAM, but not in
the central zone (CZ), peripheral zone (PZ), or
floral primordia (Fig. 1, B to D, and figs. S1 and
S2). We observed CMV accumulation just below
the WUSCHEL (WUS) (4, 5) expression domain
(Fig. 1, E and F). In the clv3-7 mutant that
ectopically expressed WUS in the L2 cell layer
(6–8) (Fig. 1G), CMV moved upward and re-
mained beneath the WUS protein–containing
domain (Fig. 1H and fig. S3). We then inves-
tigated whether the WUS expression domain

1Hefei National Laboratory for Physical Sciences at the Fig. 1. Plant stem cells and their progeny are immune to CMV infection. (A) CMV was inoculated in
Microscale, CAS Center for Excellence in Molecular Plant leaves to trigger systemic infection in Arabidopsis. (B to D) CMV distributions in mock (B), 4 dpi (C), and
Sciences, School of Life Sciences, Division of Life 9 dpi (D) plants. Red box indicates the SAM. Col-0, Columbia-0. (E and F) CMV (E) and WUS (F) distributions
Sciences and Medicine, University of Science and in the SAM at 9 dpi. (G and H) WUS and CMV accumulation patterns in the clv3-7 mutant. (I to K) WUS (J)
Technology of China, Hefei 230027, China. 2School of and CMV (K) distributions in the mp clv3-7 double mutant with viral infection (I). Red lines indicate the
Life Sciences, Guangzhou University, Guangzhou Higher surgical incision in the SAM. Red arrows indicate the location of CMV inoculation. Scale bars, 100 mm in (B) to
Education Mega Center, Guangzhou 510006, China. (D) and (I) to (K) and 50 mm in (E) to (H).
3Department of Stem Cell Biology, Centre for Organismal
Studies, Heidelberg University, Heidelberg D-69120,
Germany. 4State Key Laboratory of Biotherapy and
Cancer Center, National Clinical Research Center for
Geriatrics, West China Hospital, Sichuan University,
Chengdu 610041, China.

*Present address: Max Planck Institute of Molecular Plant

Physiology, Organelle Biology, Potsdam D-14476, Germany.

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

[email protected] (Z.T.)

Wu et al., Science 370, 227–231 (2020) 9 October 2020 1 of 5

RESEARCH | REPORT

With DEX-induced WUS activity, 90% of the (9). Ethanol induction reduced the WUS- The role of WUS in antiviral immunity did
plants were free from CMV invasion (Fig. 2, I linker–green fluorescent protein (GFP) signal not depend on RNA silencing (figs. S7 to S9),
to L). To determine whether the WUS protein in the SAM. In such plants, CMV invaded the autophagy (fig. S10), or phytohormone (fig. S11)
is required for the antiviral capacity of stem SAM CZ and PZ (Fig. 2, M to T, and fig. S6). pathways, but WUS did lead to protein silenc-
cells, we used wus/pWUS::WUS-linker-GFP/ Therefore, the stem cell regulator WUS func- ing in tobacco leaves (fig. S7). To determine
pCLV3::AlcR/pAlcA::NSlmb-vhhGFP4 plants to tions in inhibition of CMV infection and sup- whether WUS acts as a transcription factor in
inducibly degrade WUS protein in stem cells ports antiviral immunity in the SAM. that context, we coexpressed 35S::WUS-GR and
35S::GFP in tobacco leaves and observed that
Fig. 2. WUS protein responds to CMV and inhibits viral infection. (A to H) WUS protein distribution GFP was silenced only if the translocation of
and CLV3 expression patterns in the wus/pWUS::WUS-GFP; pCLV3::mCherry-NLS rescue plant with [(E) to WUS was induced (fig. S12A). We also found
(H)] or without [(A) to (D)] CMV infection at 9 dpi (n = 15). (I and J) CMV accumulations at 8 dpi in pUBQ10:: that the efficiency of transgene silencing was
WUS-GR plants with (J) or without (I) DEX induction for 7 days. (K) Percentage of plants infected by viruses dependent on the dosage of WUS-GR (fig. S12B).
with (n = 159) or without (n = 117) DEX treatment at 8 dpi. (L) WUS inhibited CMV genomic RNA Mutations in the WUS-box (m1) and EAR-like
accumulation in the SAM, as determined by RNA gel blot at 8 dpi. (M to T) WUS-linker–GFP signals and motif (m3), but not the acidic region motif (m2),
CMV distributions in wus/pWUS::WUS-linker-GFP/pCLV3::AlcR/pAlcA::NSlmb-vhhGFP4 lines with or abolished WUS functions in GFP silencing (fig.
without 1% ethanol treatment for 48 hours with CMV infection at 0 dpi [(Q) to (T)] and 8 dpi [(M) to (P)]. S13, A to F, I, and J). Moreover, we fused the C
Scale bars, 50 mm in (A) to (H) and (M) to (T) and 100 mm in (I) and (J). terminus of WUS-m1 with the SRDX repression
domain or the VP16 activation domain and ob-
served recovery of GFP silencing in WUS-m1-
SRDX, but not in WUS-m1-VP16 (fig. S13, G to
J), suggesting that WUS acts as a transcrip-
tional repressor to inhibit GFP accumulation.

To investigate whether WUS inhibits protein
accumulation by accelerating protein degrada-
tion or by repressing protein synthesis, we used
the proteasome inhibitor MG132. GFP degra-
dation was repressed by MG132 treatment in
tobacco leaves, allowing GFP accumulation.
However, with the addition of WUS, GFP did not
accumulate (fig. S14A). Similarly, in Arabidopsis,
MG132 treatment did not affect WUS-mediated
GFP degradation in pUBQ10::WUS-GR; 35S::
GFP plants (fig. S14, B and C). Thus, WUS in-
hibits protein synthesis rather than promoting
protein degradation.

To examine the effect of WUS on viral pro-
tein synthesis, we fused GFP to the C terminus
of each of the CMV proteins 2a, 3a, CP, and 2b
(10). We coinfiltrated these constructs with
35S::WUS in tobacco leaves. Expression of 3a-
GFP, 2a-GFP, CP-GFP, and 2b-GFP was reduced
by coinfiltration with 35S::WUS [as shown by
fluorescence and Western blot analysis (Fig. 3,
A to C)]. To determine whether WUS represses
CMV protein accumulation in the Arabidopsis
SAM, we inoculated pUBQ10::WUS-GR plants
with CMV and then induced WUS with DEX
treatment. At 8 dpi, 2b protein abundance in
the SAM decreased (Fig. 3D). To elucidate
whether WUS-mediated protein inhibition
is virus specific or if it has global effects on
protein synthesis, we followed incorporation
of O-propargyl-puromycin to assess nascent
protein levels in pUBQ10::WUS-GR plants
(11, 12). Three days after DEX induction, we
observed reduced total nascent protein con-
tent in pUBQ10::WUS-GR plants (Fig. 3, E
and F), but not control plants (fig. S15), con-
sistent with the cycloheximide treatment
(Fig. 3, E and F). Thus, in both tobacco and
Arabidopsis, WUS represses CMV protein accu-
mulation and inhibits global protein synthesis.

To study how WUS inhibits protein synthe-
sis, we profiled gene expression in apices of

Wu et al., Science 370, 227–231 (2020) 9 October 2020 2 of 5

RESEARCH | REPORT

Fig. 3. WUS represses CMV protein accumulation in plants. (A) The CMV in the SAM of Arabidopsis at 8 dpi. (E and F) WUS globally inhibits nascent protein
proteins 3a (35S::3a-GFP), 2a (35S::2a-GFP), CP (35S::CP-GFP), and 2b (35S::2b-GFP) synthesis in Arabidopsis (E), and the relative fluorescence intensity is quantified in
were repressed in tobacco leaves 4 days after coinfiltration of 35S::WUS. (F). (n > 7). 35S::GUS was used as a negative control. CHX, cycloheximide; DIPA,
(B) Quantification of the relative fluorescence intensity of GFP in (A) (n = 5). (C) Western 4′,6-diamidino-2-phenylindole; OP-P, O-propargyl-puromycin. Scale bars, 50 mm in (A)
blots of CP and 2b proteins in (A). (D) WUS represses CMV 2b protein accumulation and 100 mm in (E). **P < 0.01, ***P < 0.001, unpaired, two-tailed Student’s t test.

DEX-induced pUBQ10::WUS-GR plants with sion was repressed by CMV infection (fig. S19). processing and increased rRNA intermedi-
or without CMV infection. We identified 11 ates (fig. S20), consistent with observations of
genes involved in the regulation of protein syn- Because NOP2A and its homolog, NOP2B, WUS-overaccumulating plants (fig. S21). With
thesis, including 7 S-adenosyl-L-methionine– polysome profiling, we observed reduced lev-
dependent methyltransferases (MTases) that were predicted to be MTases responsible for els of 80S and 60S ribosomes in plants after
are predicted to regulate RNA methylation m5C methylation at C2860 in 25S ribosomal activating WUS (Fig. 4E and fig. S22). Because
(fig. S16). We examined these (fig. S17A) and RNA (rRNA) (13–15), we assessed total m5C NOP2A was repressed by WUS in stem cells,
12 other known MTases (fig. S17B) using levels. We observed decreases in nop2a/NOP2B we observed less nascent protein synthesis in
quantitative reverse transcription polymerase RNA interference (RNAi) transgenic plants stem cells in an area coincident with the WUS
chain reaction (qRT-PCR). Six of the genes protein domain (Fig. 4F). Although the cata-
were repressed by WUS and up-regulated (Fig. 4A), which we confirmed with liquid lytic activities of AT3G54150, AT1G69523, and
in the wus-6 mutant (fig. S17, A to C). Five of AT1g55450 remain unclear, 80S and 60S ribo-
these MTases contained a WUS DNA-binding chromatography–tandem mass spectrometry somes levels were also reduced in these three
motif. Indeed, we found with chromatin im- (LC-MS/MS) quantification (Fig. 4B). RNA mutants (fig. S23). Thus, WUS represses ribo-
munoprecipitation and electrophoretic mo- immunoprecipitation showed m5C-modified some assembly at least in part by negatively
bility shift assays that WUS associates with regulating MTases, resulting in a global decrease
the MTase promoter both in vivo and in 25S rRNAs enriched around C2860 in the in protein synthesis.
vitro (fig. S17, D to I), suggesting that it wild-type plants and less so in nop2a/NOP2B
represses the transcription of MTases in RNAi plants (Fig. 4C), demonstrating that Given that five MTase genes are directly
the SAM. NOP2A and NOP2B act redundantly in m5C repressed by WUS, we analyzed MTases in
methylation of 25S rRNAs in vivo. Using RNA the context of a WUS-mediated antiviral
Among these five MTase genes, only NUCLEOLAR bisulfite sequencing, we quantified the m5C mechanism. We coexpressed p35S::WUS-GR
PROTEIN 2A (NOP2A, AT5G55920) showed with MTase genes in tobacco leaves and ob-
expression in the PZ (fig. S18), and its expres- levels of 25S rRNA and observed a 64% increase served that WUS-mediated GFP silencing was
of unmethylated C2860 in nop2a/NOP2B RNAi
plants (Fig. 4D).

rRNA methylation stabilizes rRNA process-

ing and ribosome structure (16). In MTase
mutants, we observed disturbed pre-rRNA

Wu et al., Science 370, 227–231 (2020) 9 October 2020 3 of 5

RESEARCH | REPORT

Fig. 4. MTase mediates
WUS-triggered innate
antiviral immunity.
(A) Dot-blot assay showing
that total m5C levels were
decreased in nop2a/NOP2B
RNAi plants. (B) LC–MS/
MS quantifications of the
total m5C levels of wild-type
(WT) (n = 7), nop2a (n = 8),
and independent nop2a/
NOP2B RNAi T1 (n = 17)
plants. (C) m5C-RNA
immunoprecipitation
qRT-PCR of 4-week-old WT
(n = 6), nop2a (n = 6), and
independent nop2a/NOP2B
RNAi T1 plants (n = 8). The
region of 25S#2 contains
the predicted m5C target
C2860 of 25S rRNA. The
regions of 25S#1 and 18S#1
were used as the negative
controls. (D) Quantification
of the 25S rRNA m5C levels
of C2860 in WT (n = 2),
nop2a (n = 2), and nop2a/
NOP2B RNAi (n = 4) plants
using RNA bisulfite
sequencing. Black box indi-
cates the C2860 site.
(E) Polysome profile of
10-day-old pUBQ10::WUS-
GR seedlings treated with
DEX for 2 days (n = 2).
(F) Nascent protein synthe-
sis in the SAM. Inflorescen-
ces without puromycin
(Puro) incorporation or with
Puro and CHX treatment
were used as the negative
controls. Puro-incorporated
nascent proteins were
mainly observed in the PZ
and primordia in areas
complementary to the WUS
protein domain. Scale bars,
50 mm. (G) Overexpression
of MTases blocks WUS-
mediated antiviral immunity. The shoot apices of MTases overexpression lines in the UBQ10::WUS-GR background with CMV infection at 8 dpi and DEX induction
for 7 days were used for detection of the 2b protein with an anti-2b antibody and of the viral genomic RNA with the CMV RNA3 probe. *P < 0.05, **P < 0.01,
***P < 0.001, Student’s t test. NS, no significant difference.

partially blocked by overexpression of MTases against a range of viral infections, which and have conserved functions in shoot and
(fig. S24). We tested the genetic interaction of suggests a conserved and broad-spectrum root meristems (17, 18), WUS/WOX–mediated
WUS and MTases in Arabidopsis during viral antiviral mechanism in the SAM. We there- antiviral immunity might be a conserved mech-
infection by overexpressing six MTases in pUBQ10:: fore tested three additional viruses that trigger anism in plants.
WUS-GR plants and observed increases in viral systemic infection in Arabidopsis, including
protein and genomic RNA accumulation upon turnip crinkle virus (TCV), tobacco rattle virus As a parasitic strategy, viruses hijack the
WUS induction (Fig. 4G). Thus, WUS-mediated (TRV), and turnip mosaic virus (TuMV) (fig. S25). host protein synthesis machinery to complete
transcriptional repression of MTases is essen- We observed that WUS repressed multiple viral their life cycle. Plants in turn have evolved
tial for innate antiviral immunity in plants. infections in the SAM of Arabidopsis (fig. S26). various mechanisms to protect them against
Given that WUSCHEL-related homeobox (WOX) viral infection. Here, we show how WUS-
Meristem tip culture has been used to genes are widespread in the plant kingdom triggered antiviral immunity safeguards stem
produce virus-free plants of many species cells and their nascent daughter cells from

Wu et al., Science 370, 227–231 (2020) 9 October 2020 4 of 5

RESEARCH | REPORT

viral invasion. In stem cells, WUS directly 13. U. Fujikura, G. Horiguchi, M. R. Ponce, J. L. Micol, H. Tsukaya, performed the Western blot and qRT-PCRs. Z.D., X.L., L.Z., Z.T.,
represses MTase transcription, impairing Plant J. 59, 499–508 (2009). and Z.J. conducted RNA-seq and small RNA-seq and analyzed the
global protein synthesis and limiting the rep- obtained data. L.L., C.S., and M.X. performed the EMSA
lication and spread of the virus (fig. S27). 14. A. L. Burgess, R. David, I. R. Searle, BMC Plant Biol. 15, 199 experiments. J.F. and J.U.L. generated the pWUS::WUS-GFP;
(2015). pCLV3::mCherry-NLS and pUBQ10::WUS-GR transgenic lines. S.L.
REFERENCES AND NOTES performed vector construction and plant transformation.
15. A. Pavlopoulou, S. Kossida, Genomics 93, 350–357 (2009). Competing interests: The authors declare no competing
1. E. Aichinger, N. Kornet, T. Friedrich, T. Laux, Annu. Rev. Plant 16. A. Gigova, S. Duggimpudi, T. Pollex, M. Schaefer, M. Koš, interests. Data and materials availability: Raw sequence
Biol. 63, 615–636 (2012). data generated during this study are available in the Gene
RNA 20, 1632–1644 (2014). Expression Omnibus (GEO) repository under accession
2. G. Morel, CR Acad. Sci. Paris 235, 1324–1325 (1952). 17. K. Mukherjee, L. Brocchieri, T. R. Bürglin, Mol. Biol. Evol. 26, no. GSE128778 for RNA-seq and no. GSE128863 for small
3. P. J. Wang, C. Y. Hu, “Regeneration of virus-free plants through RNA-seq. Requests for materials should be addressed to Z.Z.
2775–2794 (2009). All data presented in this study are available in the main text or
in vitro culture,” in Advances in Biomedical Engineering, 18. J. Nardmann, P. Reisewitz, W. Werr, Mol. Biol. Evol. 26, the supplementary materials.
A. Fiechter, Ed. (Springer, 1980), vol. 18, pp. 61–99.
4. R. K. Yadav et al., Genes Dev. 25, 2025–2030 (2011). 1745–1755 (2009). SUPPLEMENTARY MATERIALS
5. K. F. X. Mayer et al., Cell 95, 805–815 (1998).
6. J. C. Fletcher, U. Brand, M. P. Running, R. Simon, ACKNOWLEDGMENTS science.sciencemag.org/content/370/6513/227/suppl/DC1
E. M. Meyerowitz, Science 283, 1911–1914 (1999). Materials and Methods
7. U. Brand, J. C. Fletcher, M. Hobe, E. M. Meyerowitz, R. Simon, We thank J. Wang, X. Zhang, F. Qu, X. Wang, Y. Xiong, C. Xiang, Figs. S1 to S27
Science 289, 617–619 (2000). I.R. Searle, M. Sun, T. Laux, D. Yu, X. Yuan, J. Gao, and Y. Liu for Table S1
8. H. Schoof et al., Cell 100, 635–644 (2000). sharing plant and virus materials. Funding: This work was References (19–45)
9. Y. Ma et al., Nat. Commun. 10, 5093 (2019). supported by the Strategic Priority Research Program of the MDAR Reproducibility Checklist
10. S.-W. Ding, B. J. Anderson, H. R. Haase, R. H. Symons, Virology Chinese Academy of Sciences (XDB27030105 to Z.Z.), the National
198, 593–601 (1994). Natural Science Foundation of China (31870264 and 31570273 View/request a protocol for this paper from Bio-protocol.
11. S. Blanco et al., Nature 534, 335–340 (2016). to Z.Z., 31400215 to H.W., 31871289 to Z.D.), the Ministry of
12. J. Liu, Y. Xu, D. Stoleru, A. Salic, Proc. Natl. Acad. Sci. U.S.A. Science and Technology of China (2013CB967300 to Z.Z.), the 15 March 2020; resubmitted 28 July 2020
109, 413–418 (2012). China Postdoctoral Science Foundation (2015M582004 to H.W.), Accepted 21 August 2020
and the Deutsche Forschungsgemeinschaft (SFB1101 to J.U.L). 10.1126/science.abb7360
Author contributions: Z.Z., H.W., and Z.T. designed the
experiments, analyzed the data, and wrote the paper. H.W.
performed ChIP, RNA in situ hybridization, RNA blot, polysome
profiling, nascent protein assay, and LC-MS/MS. X.Q. and M.S.

Wu et al., Science 370, 227–231 (2020) 9 October 2020 5 of 5

RESEARCH

SUPERCONDUCTIVITY Fig. 1A for the case of 2H-TaS2, intercalation
of organic molecules (15) [(Py)0.5-TaS2, where
Clean 2D superconductivity in a bulk Py is pyridine] and incommensurate spacer
van der Waals superlattice layers (16) [(PbS)1.13TaS2] has been used to
drive these materials toward the 2D limit and
A. Devarakonda1, H. Inoue1*, S. Fang2†, C. Ozsoy-Keskinbora3‡, T. Suzuki1, M. Kriener4, L. Fu1,
E. Kaxiras2,3, D. C. Bell3,5, J. G. Checkelsky1§ reduce the inversion-symmetric coupling be-
tween adjacent layers (17). These have led to
Advances in low-dimensional superconductivity are often realized through improvements in material important advances, such as establishing that
quality. Apart from a small group of organic materials, there is a near absence of clean-limit
two-dimensional (2D) superconductors, which presents an impediment to the pursuit of numerous superconductivity is a property of the 2D planes
long-standing predictions for exotic superconductivity with fragile pairing symmetries. We developed a
bulk superlattice consisting of the transition metal dichalcogenide (TMD) superconductor 2H-niobium in (Py)0.5-TaS2. More recently, developments
disulfide (2H-NbS2) and a commensurate block layer that yields enhanced two-dimensionality, high in the exfoliation of van der Waals (vdW) lay-
electronic quality, and clean-limit inorganic 2D superconductivity. The structure of this material may
naturally be extended to generate a distinct family of 2D superconductors, topological insulators, and ered materials have made atomically thin 2D
excitonic systems based on TMDs with improved material properties.
superconductors more readily accessible (e.g.,
S uperconducting states with real- or groundbreaking studies that observed the ML-TaS2) (18–21). Subsequent studies have re-
momentum-space nodes in the gap func- vortex-antivortex unbinding transition of Bere- vealed a form of superconductivity characterized
zinskii-Kosterlitz-Thouless (BKT) that had been
tion are acutely sensitive to disorder. predicted for 2D superconductors (5). Shortly by strong Ising spin-orbit coupling equivalent to
Such nodal pairing states are weakened thereafter, studies of similar films revealed
that the disorder induced a superconductor- applying magnetic fields on the order of 100 T
by the momentum-space averaging of insulator transition in two dimensions (6). (19–21). However, flakes exfoliated from the
the gap function caused by disorder-induced Further developments in thin-film growth have bulk are often subject to degradation and re-
led to the observation of the BKT transition
scattering of Cooper pairs across the Fermi in crystalline films of Nb (c-Nb) (7) and, more duction in quality during the fabrication pro-
recently, to high-quality 2D superconductors cess (22). Here, we show that high-quality
surface. Conventionally, this disorder aver- formed at epitaxial interfaces in LaAlO3/SrTiO3 H-NbS2 monolayers with electronic mobilities
aging can be avoided when Cooper pairs have (8) and in d-doped Nb-SrTiO3 (9). more than three orders of magnitude larger
well-defined crystal momentum k within the than in bulk 2H-NbS2 can be realized in a bulk
Pippard coherence length x0; this regime is In parallel, the study of anisotropic bulk single-crystal superlattice formed with a com-
realized when x0 is smaller than the electro- superconductors has also yielded substan-
nic mean free path ‘ (i.e., x0/‘ ≪ 1). Two- tial insight into the nature of the superconduct- mensurate block layer. Correspondingly, we
dimensional superconductors in this so-called ing state. These materials can also be framed
clean limit play a central role in proposals in- in the context of 2D superconductivity (see find that this material is a clean-limit 2D super-
Fig. 1A); prototypical examples include cup- conductor exhibiting a BKT transition at TBKT =
cluding archetypal platforms for finite-momentum rates, pnictides, organic charge-transfer salts, 0.82 K and prominent 2D Shubnikov-de Haas
Cooper pairing (1, 2) and recent constructions graphite intercalation compounds, and transi-
for unconventional superconducting phases tion metal dichalcogenides (TMDs). Consid- (SdH) quantum oscillations.
that leverage normal-state spin textures (3, 4). erable effort has been invested in pursuing
At present, there is a paucity of materials that tunable dimensionality in these systems. For The fundamental structural unit in hexago-
allow access to this regime. example, the search for new cuprates in the nal TMDs is the H-MX2 layer, where M and X
Ruddlesden-Popper series and their oxygen- are a transition metal and chalcogen, respec-
Figure 1A presents a historical survey of deficient variants (10) has led to a diverse set
superconducting materials sorted according of materials with variable dimensionality and tively. As shown in Fig. 1B, this structure [point
a rich variety of exotic behavior (11). Discovery group symmetry 6 m2 (D3h)] breaks inversion
to their dimensionality and cleanliness, the of the structurally related layered perovskite symmetry in the layer plane owing to the tri-
former characterized by Hcc2=Hca2b (the ratio of Sr2RuO4 realized a 2D stoichiometric super- gonal prismatic coordination of X around M
the upper critical magnetic fields perpendicu- conductor well within the clean limit, in prin- (the missing inversion partners are shown as
lar and parallel to the 2D layer) and the latter ciple satisfying the stability requirements for
by x0/‘. Early experimental work in granular fragile superconducting gap functions that would dashed circles), and as a result yields an out-
Al (g-Al) and amorphous Bi (a-Bi) films otherwise be disrupted by nonmagnetic dis- of-plane (Ising) spin texture (18–21). For thin
demonstrated 2D superconductivity through order (12). Despite a great deal of interest in such flakes deposited on substrates, the substrate-
precise control of the superconducting layer phases, this very clean and highly 2D regime
has largely been limited to organic systems (13). flake interface breaks mirror symmetry (Fig. 1C)
thickness; these systems were later used in
Substantial efforts have gone toward achiev- and yields an in-plane (Rashba) spin texture
1Department of Physics, Massachusetts Institute of Technology, ing 2D superconductivity in TMDs, which, (23). Taken together, for TMD flakes on sub-
Cambridge, MA 02139, USA. 2Department of Physics, Harvard given their intrinsically strong spin-orbit cou- strates, the simultaneous breaking of mirror
University, Cambridge, MA 02138, USA. 3John A. Paulson School pling and inversion symmetry breaking, are
of Engineering and Applied Sciences, Harvard University, expected to yield exotic forms of supercon- and inversion symmetry leads to a mixed spin
Cambridge, MA 02138, USA. 4RIKEN Center for Emergent Matter ductivity in the clean limit (14). As shown in
Science (CEMS), Wako 351-0198, Japan. 5Center for Nanoscale texture (Fig. 1D) on the Fermi surface composed
Systems, Harvard University, Cambridge, MA 02138, USA. of both Ising and Rashba components (18).
*Present address: Institute for Materials Research, Tohoku These spin textures and the resulting physics
University, Sendai, Miyagi 980-8577, Japan.
†Present address: Department of Physics and Astronomy, Center for are suppressed in the bulk limit, where the
Materials Theory, Rutgers University, Piscataway, NJ 08854, USA.
‡Present address: Thermo Fisher Scientific, 5600KA Eindhoven, overall unit cell preserves inversion symmetry.
Netherlands.
§Corresponding author. Email: [email protected] We have synthesized a single-crystal mate-

rial, Ba6Nb11S28, composed of high-quality
H-NbS2 layers and Ba3NbS5 block layers in
which the TMD layers are strongly decoupled
(17). Figure 1E shows a cross section of the
structure imaged by high-angle annular dark-

field scanning transmission electron microscopy

(HAADF-STEM) with the model structure super-

imposed. As determined by electron and powder
x-ray diffraction, the unit cell (space group P3 1c
with a = 10.4 Å, c = 24.5 Å) is composed of two
inversion-related H-NbS2 layers across each
of which mirror symmetry is broken by the

Devarakonda et al., Science 370, 231–236 (2020) 9 October 2020 1 of 6

RESEARCH | REPORT

Fig. 1. 2D superconductivity and Ba6Nb11S28. (A) Survey of superconducting Ising and Rashba coupling. (E) HAADF-STEM image of Ba6Nb11S28 taken along the
materials characterized by anisotropy of the upper critical field Hcc2=Hca2b and ratio ½1 100Š axis (scale bar, 1 nm). A simulation of the model structure is overlaid with one
of the Pippard coherence length to mean free path x0/‘. The boundary between unit cell shaded in green. Ba, Nb, and S atoms are depicted as blue, red, and yellow
the clean and dirty limits is shown as a horizontal line. (B) Crystal structure of H-MX2
projected onto the ab-plane. Lack of inversion symmetry is illustrated by the missing circles, respectively. (F) Resistivity as a function of temperature rxx(T) in Ba6Nb11S28
showing the superconducting transition. Upper inset: Magnified view of the transition in
chalcogen (X) inversion partners (dashed circles). (C) The ab-plane mirror symmetry
rxx(T) and magnetic susceptibility 4pcc measured with zero field cooling (ZFC) and field
in monolayer H-MX2 can be broken by substrates or local fields (∇U). (D) Depiction cooling (FC). rxx(T) is well fit by the Halperin-Nelson model shown in black (see text).
of momentum space spin-orbit texture for monolayer H-MX2 with varying degrees of Lower inset: H-NbS2 layer and mirror symmetry–breaking Ba3NbS5 block layers.

neighboring block layers; the overall unit cell Ba6Nb11S28 exhibits a 3 × 3 in-plane, commen- sity wave transition (17, 27). The upper inset of
surate superstructure caused by the difference Fig. 1F shows a detailed view of the super-
retains inversion symmetry. The H-NbS2 inter- conducting transition, which shows onset near
layer distance d = 8.9 Å is more than three in in-plane lattice constants between the two T = 1.6 K and reaches zero resistance at T = 0.85 K.
times that of 2H-NbS2 (24), leading to a re- layer types (17), which leads to additional mod- At the latter temperature, the magnetic sus-
duction of the interlayer transfer integral t⊥. ification of the electronic structure. ceptibility 4pcc with field along the c axis shows
This amplifies the two-dimensionality of the a Meissner signal reaching a shielding fraction
Figure 1F shows the dependence of elec- of 75% (Fig. 1F, upper inset, dark green ZFC
electronic structure relative to 2H-NbS2 and trical resistivity rxx(T) on temperature T for data points) and a volume fraction of 40%
enables local symmetry breaking–induced Ba6Nb11S28. The system is a metal, eventually (Fig. 1F, upper inset, light green FC data
spin-orbit textures on the H-NbS2 layers showing superconductivity below T = 1 K. points). The reduction in the transition tem-
(25, 26). [H-NbS2 monolayers in Ba6Nb11S28 This can be compared to bulk 2H-NbS2, which perature for the NbS2 layers relative to the
experience local inversion symmetry break- is also metallic and becomes a superconduc- bulk is similar to that observed in organic in-
tor at critical temperature Tc = 5.7 K. Unlike tercalated variants (28) and is consistent with
ing with point group symmetry 32 (D3) (17).] several other related H-MX2 systems, neither
Whereas traditional misfit compounds com- Ba6Nb11S28 nor 2H-NbS2 shows signs of a den-

bine incommensurate layers in a superlattice,

Devarakonda et al., Science 370, 231–236 (2020) 9 October 2020 2 of 6

RESEARCH | REPORT

Fig. 2. Quantum oscillations and electronic structure of Ba6Nb11S28. by 25. (D) DFT calculation of monolayer H-NbS2 Fermi surfaces including
(A) Magnetoresistance as a function of perpendicular field MR ≡ [rxx(m0H⊥)/ spin-orbit coupling (17). (E) Depiction of zone-folding scheme involving the
rxx(0)] – 1 at temperature T = 0.39 K for different field rotation angles q
(geometry defined as shown in the inset). Curves are vertically offset by 150% 3 × 3 superstructure imposed by the Ba3NbS5 block layer where the reduced
of MR for clarity. (B and C) Low-frequency range (B) and full range (C) of Brillouin zone is enclosed by the bold line. (F) Electronic structure of zone-

quantum oscillation amplitude FFT as a function of perpendicular frequency folded monolayer H-NbS2 with experimentally observed Fermi surface
F cos(q). The FFT amplitudes for the higher-frequency pockets are multiplied cross-sectional areas drawn to scale as solid circles. The black box
corresponds to 0.01 Å–2.

a reduction of the electron-phonon coupling tially enhanced relative to rzz/rxx ~ 100 in points of the hexagonal Brillouin zone (Fig. 2D),
2H-NbS2 (17) (fig. S24).
strength inferred from the measured resistiv- and zone folding into the reduced Brillouin
Magnetotransport measurements demon- zone determined by the 3 × 3 superstructure
ity at high T (17).
strate the cleanliness of this material and show imposed by the block layers (Fig. 2E). In par-
The temperature dependence of resis-
further evidence for a 2D electronic structure. ticular, the reduction in the pocket size from
tivity for 0.85 K < T < 1.6 K is well described Figure 2A shows the magnetoresistance MR ≡ monolayer H-NbS2 caused by zone folding,
[rxx(H)/rxx(0)] – 1, measured to 31 T. We approximately one order of magnitude, quan-
by the ðHÀba=lppetffirffiÞi,nw-NheelrseornFxxmaonddelr,NxxrFxaxrðeT Þ¼ observe SdH quantum oscillations that re- titatively captures the size of the observed
rNxx exp the
spond to the component of the magnetic pockets (Fig. 2F) and is further supported by
fluctuation and normal-state resistivity, re- field perpendicular to the ab-plane (the tilt first-principles calculations (17) (fig. S12). An
angle q is measured between the c axis and important aspect of this structure is that the
spectively; t = (T/THN) – 1; and b is a fitting the applied field). The fast Fourier transform large ratio of the spin-orbit coupling to t⊥, evi-
parameter on the order of 1 (Fig. 1F, upper dent from the degree of two-dimensionality,
(FFT) computed after subtracting a monoton- enables local symmetry breaking to affect the
inset, dashed curve) (29). The agreement ically increasing background (17) (fig. S14)
plotted versus inverse field shows this more bulk electronic structure. The zone folding
with the Halperin-Nelson model evidences promotes the Rashba-textured pockets asso-
clearly (Fig. 2, B and C). Here, the oscillation ciated with the G point in monolayer H-NbS2
fluctuations of the superconducting order frequency multiplied by cos(q) has little var- to be larger than the Ising-split pockets at K
iance versus angle, demonstrating the 2D and K´ [supported by comparing the calcu-
parameter above a two-dimensional BKT lated band structure for monolayer H-NbS2
nature of the Fermi surface. This is qualita- with that of the 3 × 3 zone-folded structure (17)
transition. Such behavior is generally rare tively different from 2H-NbS2, for which elec- (figs. S11 and S12)] and has potential implica-
tronic structure calculations indicate warped
in bulk single crystals but has been reported and elliptical Fermi surfaces (31). Owing to the tions for superconducting pairing.
reduced coupling between the TMD layers, the More generally, it is noteworthy that quan-
in La1.875Ba0.125CuO4 and attributed to the observed bands (labeled here as a, b1, b2, g1,
decoupling of superconducting CuO2 planes and g2) can be understood by starting with the tum oscillations have not been reported in 2H-
by stripe order (30). This is a further departure 2D electronic structure of monolayer H-NbS2, NbS2; there, the typical transport mobilities
which consists of bands at the G, K, and K′ reported for bulk single crystals are on the
from the behavior in bulk 2H-NbS2, which
exhibits a sharp superconducting transition

(17) (fig. S23); instead, it closely resembles

those observed in monolayer H-MX2. Further

evidence of increased two-dimensionality is
the resistivity anisotropy rzz/rxx > 103 at low
T in the normal state of Ba6Nb11S28 (where
rzz is the c-axis resistivity), which is substan-

Devarakonda et al., Science 370, 231–236 (2020) 9 October 2020 3 of 6

RESEARCH | REPORT

Fig. 3. 2D superconductivity and Pauli limit breaking in Ba6Nb11S28. superconductivity by Joule heating. For q = 80° and 90°, only low current is used.
(A) Current-voltage characteristics I(V) from T = 0.95 K to T = 0.28 K. The inset shows (C) Angular dependence of upper critical field m0Hc2 measured at T = 0.28 K with fits
the evolution of the power law V º Ia; the horizontal line marks a = 3. (B) Longitudinal to the 2D-Tinkham model, computed using data in the range |q – 90°| < 1.7° (purple
resistivity rxx as a function of field m0H for different values of q. Curves are vertically curve) and |q – 90°| > 1.7° (black curve), respectively. The inset shows a detailed view
offset by 20 mW·cm for clarity (horizontal lines). Vertical ticks separate regions near q = 90° where an enhancement of m0Hc2(q) is observed across the Pauli limit m0HP.
We define m0Hc2 and its error to be when rxx reaches 50 ± 5% of the normal-state value.
measured with low current (7 mA) and higher current (70 mA) to avoid suppression of

order of 1 cm2 V–1 s–1 (27). In Ba6Nb11S28, we see the normal-state value). Figure 3C summa- enhancement at low T and high H quickly dis-
the onset of SdH quantum oscillations in mag- rizes this behavior, with m0Hc2(q) showing a appears as q is moved away from 90°, and by
sharp cusp for in-plane fields. Recent studies q = 86° there is little variation with further
netic fields between 2 and 3 T, indicating quan- of 2D superconductors have shown that a fea- field tilt (Fig. 4A). A distinct feature at all
tum mobilities on the order of 103 cm2 V–1 s–1. ture that distinguishes such systems from an- values of q is the finite ds associated with
isotropic 3D superconductors is the profile of fluctuating superconductivity for low H ex-
Analysis of the quantum oscillations and low- m0Hc2(q) following the 2D Tinkham form tending to T beyond TBKT. To remove this
fluctuation contribution, we plot the differ-
field magnetoresistance indicates an associated ðqÞsin q 2 þ Hc2 ðqÞcos q ¼ 1 ð1Þ ence ds(q = 90°) – ds(q = 84°) in Fig. 4B. The
transport mean free path ‘ = 1.21 mm, which Hc2 Hcc2 expected 2D Ginzburg-Landau behavior is
greatly exceeds the Pippard coherence length
x0 ≈ 0.18ħvF/kBTc = 254 nm (where ħ is the Hca2b shown as a green curve; the transition curve
Planck constant divided by 2p, vF is the Fermi follows this response below TBKT until T/
velocity, and kB is the Boltzmann constant) (17). where Hca2b and Hcc2 are the upper critical fields TBKT ≈ 0.6, below which a considerable en-
This places Ba6Nb11S28 in the clean limit of for field applied in-plane and out-of-plane, hancement is observed. As shown in Fig. 4C,
superconductivity (Fig. 1A).
respectively (33). The response of Ba6Nb11S28 this behavior is confined to low temperature
Turning to properties of the superconduct- can be fit by such a form, contrasting the and to a small angular region dq about the
ab-plane.
ing state, Fig. 3A shows the current-voltage anisotropic 3D character of 2H-NbS2 (17) (fig.
I(V) characteristics of Ba6Nb11S28 across the S25). Furthermore, for Ba6Nb11S28, we observe These observations taken together indicate
superconducting transition. As expected for an enhancement of the scale of m0Hc2(q) for
a BKT transition (29), a linear response at T = angles below 1.7° measured relative to the ab- the appearance of a clean 2D superconduct-
0.95 K and above crosses over to a nonlinear ing state with enhanced stability (larger Hc2)
dependence V º Ia with a ~ 3 at TBKT = 0.82 K, plane. As shown in the inset of Fig. 3C, this when T < 0.5Tc and field H > HP applied very
consistent with THN = 0.85 K. With further close to the layer plane. Various theoretical
examination of the fluctuation conductivity anomalous enhancement coincides with m0Hc2
(q) crossing the Pauli paramagnetic limit m0HP scenarios have been discussed for Pauli break-
and the slope of the power-law exponent close ≈ 1.84TBKT = 1.51 T. We find that two inde-
to TBKT, we find evidence for a vanishingly pendent Tinkham fits trace the data across the ing in 2D superconductors, including spin-
small interlayer coupling in the superconduct- orbit scattering (34), Ising superconductivity
ing state (17, 32). Figure 3B shows the evo- entire angular regime: For |q – 90°| ≥ 1.7° we (19–21), and Fulde-Ferrell-Larkin-Ovchinnikov
lution of rxx(H) as a function of magnetic field have m0Hcc2 = 0.15 T, m0Hca2b = 2.19 T, and for (FFLO) states (1, 2). Given the clean-limit
for different values of q. Whereas for q = 0° |q – 90°| ≤ 1.7° we obtain m0Hcc2 = 0.09 T, superconductivity realized here, spin-orbit
m0Hca2b = 2.55 T.
superconductivity is suppressed with relative- scattering enhancements cannot account for
To further examine the anomaly in m0Hc2,
ly low fields and gives rise to quantum oscil- we measured rxx(H, T) with q systematically the present observations. The dominant local
lations, for larger q the upper critical field tuned away from 90°. Plotted as the excess
m0Hc2 rapidly increases (we define m0Hc2 and conductivityds ≡ 1 À ðrxx=rNxxÞ, the substantial Rashba spin-orbit coupling in the present sys-
its error to be when rxx reaches 50 ± 5% of
tem reduces the importance of the local Ising
coupling (35), and in Ising superconductors no

Devarakonda et al., Science 370, 231–236 (2020) 9 October 2020 4 of 6

RESEARCH | REPORT

Fig. 4. Superconducting phase diagram of Ba6Nb11S28. (A) Excess conductivity of m0Hc2 in a clean 2D system where an enhancement is expected within a critical
relative to the normal state ds(m0H, T) for field angles q near the ab-plane (q = 90°). region |q – 90°| < dq where orbital limiting is quenched and an FFLO state,
(B) Difference between ds(m0H, T) for q = 90° and q = 84°. The temperature axis is characterized by a superconducting gap D of the form exp(iq·r), appears for T/Tc
normalized to TBKT. The green curve represents the 2D Ginzburg-Landau (2D-GL) < 0.55 and H > HP (dark blue solid line). Theoretical studies of 2D FFLO
model of m0Hc2. (C) Angular dependence of m0Hc2 at T/TBKT = 0.3 (orange) and superconductors further predict a cascade of magnetic vortex states that appear
m0Hc2 at T/TBKT = 0.8 (green, magnified by a factor of 3). Inset: Schematic depiction
as a corrugation of m0Hc2(q) within this regime (37) (red dashed line).

abrupt change in Hc2(q) is expected. Instead, expects a sharp enhancement of Hc2(q) as the inset, red dashed line), which may be resolved
the sharp angle–dependent enhancement re- large in-plane magnetic field stabilizes a finite- by higher-resolution measurements. More di-
sembles that of the clean layered organic FFLO momentum pairing state. rectly, relative to clean 2D organic supercon-
candidate b″-(ET)2SF5CH2CF2SO3 [where ET is ductors (Fig. 1A), the robust inorganic nature
bis(ethylenedithio)tetrathiafulvalene] in sim- The close parallel between our observations of Ba6Nb11S28 offers the opportunity to exam-
ilar conditions of H and T arising when the and those expected for a 2D FFLO phase ine the potential real-space modulation of super-
orbital limiting of superconductivity is quenched [along with our ability to fit the phase diagram conductivity—using, for example, scattering
by aligning H close to the layer plane (36). As in Fig. 4B with predictions for such a phase techniques (38). Unlike other clean inorganic
shown schematically in the inset of Fig. 4C, once (17)] highlights it as a promising candidate. systems, Ba6Nb11S28 is derived from a well-
Theoretical studies of 2D FFLO superconduc- studied family of materials, lacks localized
orbital limiting is overcome within a critical tors further predict a cascade of magnetic magnetic moments, and, perhaps most impor-
angular window dq [typically on the order of vortex states with finite momentum for T/Tc tant, has vdW layer bonding that allows for
1° to 2° in organic systems (37)], one naturally < 0.55, H > HP, and |q – 90°| < dq (Fig. 4C,

Devarakonda et al., Science 370, 231–236 (2020) 9 October 2020 5 of 6

RESEARCH | REPORT

exfoliation and integration into superconduct- REFERENCES AND NOTES 39. D. J. Van Harlingen, Rev. Mod. Phys. 67, 515–535 (1995).
40. D. Jena, A. Konar, Phys. Rev. Lett. 98, 136805 (2007).
ing device structures. One exciting conse- 1. L. N. Bulaevskii, Sov. Phys. JETP 38, 634–639 (1973). 41. Y. Nakamura, Y. Yanase, Phys. Rev. B 96, 054501 (2017).
quence could be simplified fabrication of 2. L. N. Bulaevskii, Sov. Phys. JETP 37, 1133–1136 (1973). 42. L. Wang, T. O. Rosdahl, D. Sticlet, Phys. Rev. B 98, 205411
3. B. T. Zhou, N. F. Q. Yuan, H. L. Jiang, K. T. Law, Phys. Rev. B
phase-sensitive junction devices similar to (2018).
those used to study the cuprates and other 93, 180501 (2016). 43. Z. Fei et al., Nat. Phys. 13, 677–682 (2017).
unconventional superconductors (39). 4. C.-X. Liu, Phys. Rev. Lett. 118, 087001 (2017). 44. G. Wang et al., Rev. Mod. Phys. 90, 021001 (2018).
5. A. F. Hebard, A. T. Fiory, Phys. Rev. Lett. 44, 291–294 (1980). 45. A. Devarakonda, H. Inoue, S. Fang, C. Ozsoy-Keskinbora,
We hypothesize that the large enhancement 6. D. B. Haviland, Y. Liu, A. M. Goldman, Phys. Rev. Lett. 62,
of electronic mobility observed for the H-NbS2 T. Suzuki, M. Kriener, L. Fu, E. Kaxiras, D. C. Bell,
layers in Ba6Nb11S28 is attributable to screen- 2180–2183 (1989). J. G. Checkelsky, Data for “Clean 2D superconductivity in a
ing by the highly polarizable block layer akin 7. J. W. P. Hsu, A. Kapitulnik, Phys. Rev. B 45, 4819–4835 (1992). bulk van der Waals superlattice”. Harvard Dataverse (2020);
to that observed in engineered semiconductor 8. N. Reyren et al., Science 317, 1196–1199 (2007). doi:10.7910/DVN/GSYYOA.
heterostructures (40). Additionally, our den- 9. Y. Kozuka et al., Nature 462, 487–490 (2009).
sity functional theory (DFT) calculations sug- 10. C. Park, R. L. Snyder, J. Am. Ceram. Soc. 78, 3171–3194 (1995). ACKNOWLEDGMENTS
11. P. A. Lee, N. Nagaosa, X.-G. Wen, Rev. Mod. Phys. 78, 17–85
gest that the lowest-energy cleavage occurs We thank Y. Tokura, P. A. Lee, M. Nakano, H. Matsuoka,
between the H-MX2 and block layers, implying (2006). N. F. Q. Yuan, V. Mitrovic, and S.-L. Zheng for fruitful discussions,
that mechanically exfoliated Ba6Nb11S28 may 12. A. P. Mackenzie, Y. Maeno, Rev. Mod. Phys. 75, 657–712 (2003). and M. Kamitani and A. Akey for technical support. Funding:
yield naturally encapsulated H-NbS2 mono- 13. J. Singleton, C. Mielke, Contemp. Phys. 43, 63–96 (2002). Supported by the Gordon and Betty Moore Foundation through
layers akin to vdW structures made by stack- 14. M. Smidman, M. B. Salamon, H. Q. Yuan, D. F. Agterberg, Rep. grants GBMF3848 and GBMF9070 to J.G.C. (instrumentation
ing MX2 layers and h-BN (22). [We observe, development); the Office of Naval Research (ONR) under award
using optical and atomic force microscopy, Prog. Phys. 80, 036501 (2017). N00014-17-1-2883 (advanced characterization); the U.S.
15. F. R. Gamble, F. J. Disalvo, R. A. Klemm, T. H. Geballe, Science Department of Energy (DOE) Office of Science, Basic Energy
that standard exfoliation techniques can be Sciences, under award DE-SC0019300 (material development);
used to obtain flakes suitable for device fabri- 168, 568–570 (1970). the STC Center for Integrated Quantum Materials, NSF grant
cation (17).] However, given that bulk Ba6Nb11S28 16. A. Meerschaut, Curr. Opin. Solid State Mater. Sci. 1, 250–259 DMR-1231319 (A.D., S.F., C.O.-K., and E.K.); and the DOE Office
already exhibits two-dimensional physics, we of Basic Energy Sciences under award DE-SC0018945 (L.F.).
(1996). Computations were performed on the Odyssey cluster
propose that insertion of commensurate spacer 17. See supplementary materials. supported by the FAS Division of Science, Research Computing
layers could be an alternative to fabricating 18. D. Xiao, G. B. Liu, W. Feng, X. Xu, W. Yao, Phys. Rev. Lett. 108, Group at Harvard University, and the Extreme Science and
Engineering Discovery Environment (XSEDE), which is
exfoliated nanodevices. There is also scope 196802 (2012). supported by NSF grant ACI-1548562. A portion of this work was
for functionalizing the spacer layer to further 19. J. M. Lu et al., Science 350, 1353–1357 (2015). performed at the National High Magnetic Field Laboratory, which
modulate the H-MX2 layer—for example, by 20. Y. Saito et al., Nat. Phys. 12, 144–149 (2016). is supported by NSF Cooperative Agreement DMR-1157490,
introducing magnetic constituents. The large 21. X. Xi et al., Nat. Phys. 12, 139–143 (2016). the State of Florida, and DOE. Author contributions: A.D.
22. S. Manzeli, D. Ovchinnikov, D. Pasquier, O. V. Yazyev, A. Kis, synthesized and characterized the single crystals. A.D. and
electronic mean free path of Ba6Nb11S28 en- H.I. performed the electrical transport experiments; A.D. and
ables clean-limit superconductivity and can Nat. Rev. Mater. 2, 17033 (2017). M.K. performed the magnetization experiments; C.O.-K. and
23. Y. A. Bychkov, E. I. Rashba, JETP Lett. 39, 78–81 (1984). D.C.B. performed the electronic microscopy experiments; A.D.
potentially realize unconventional phases pre- 24. B. Morosin, Acta Crystallogr. B 30, 551–552 (1974). and S.F. performed theoretical calculations; all authors
dicted in monolayer H-MX2 superconductors 25. X. Zhang, Q. Liu, J. W. Luo, A. J. Freeman, A. Zunger, Nat. Phys. contributed to discussions and writing of the manuscript; and
(3, 4, 41, 42). The physics of other MX2 mate- J.G.C. coordinated the project. Competing interests: The
rials can also benefit from longer electron 10, 387–393 (2014). authors declare no competing interests. Data and materials
26. J. M. Riley et al., Nat. Phys. 10, 835–839 (2014). availability: The data in the manuscript are available from the
mean free paths. In particular, extending the 27. M. Naito, S. Tanaka, J. Phys. Soc. Jpn. 51, 219–227 (1982). Harvard Dataverse (45).
materials family of MX2 natural commensurate 28. Y. Hamaue, R. Aoki, J. Phys. Soc. Jpn. 55, 1327–1335 (1986).
superlattices may, for example, pave the way 29. B. I. Halperin, D. R. Nelson, J. Low Temp. Phys. 36, 599–616 SUPPLEMENTARY MATERIALS
to longer mean free paths, which would enable
(1979). science.sciencemag.org/content/370/6513/231/suppl/DC1
topological edge mode circuitry in WTe2 super- 30. Q. Li, M. Hücker, G. D. Gu, A. M. Tsvelik, J. M. Tranquada, Phys. Materials and Methods
lattices (43) or longer exciton lifetimes in MoS2 Supplementary Text
and related semiconducting TMD materials (44). Rev. Lett. 99, 067001 (2007). Figs. S1 to S26
31. C. Heil, M. Schlipf, F. Giustino, Phys. Rev. B 98, 075120 Tables S1 to S6
References (46–154)
(2018).
32. D. H. Kim, A. M. Goldman, J. H. Kang, R. T. Kampwirth, 28 September 2019; accepted 21 August 2020
10.1126/science.aaz6643
Phys. Rev. B 40, 8834–8839 (1989).
33. M. Tinkham, Introduction to Superconductivity (Dover, ed. 2,

2004).
34. R. A. Klemm, A. Luther, M. R. Beasley, Phys. Rev. B 12, 877–891

(1975).
35. J. Falson et al., Science 367, 1454–1457 (2020).
36. R. Beyer, B. Bergk, S. Yasin, J. A. Schlueter, J. Wosnitza, Phys.

Rev. Lett. 109, 027003 (2012).
37. H. Shimahara, D. Rainer, J. Phys. Soc. Jpn. 66, 3591–3599

(1997).
38. A. Huxley et al., Nature 406, 160–164 (2000).

Devarakonda et al., Science 370, 231–236 (2020) 9 October 2020 6 of 6

RESEARCH

VACCINES centage of IgG-secreting BMPCs producing
influenza-specific antibodies was 0.8%, with
Influenza vaccine–induced human bone marrow an interquartile range (IQR) of 0.52 to 1.3%.
plasma cells decline within a year after vaccination Across the cohort, the percentage of influenza-
specific IgG+ BMPCs correlated significantly
Carl W. Davis1,2, Katherine J. L. Jackson3*, Megan M. McCausland1,2, Jaime Darce4†, Cathy Chang1,2, with influenza-specific IgG titers in the blood
Susanne L. Linderman1,2, Chakravarthy Chennareddy1,2, Rebecca Gerkin2,5, Shantoria J. Brown2,5, (Fig. 1C). These data are consistent with a key
Jens Wrammert1,6, Aneesh K. Mehta2,7, Wan Cheung Cheung4‡, Scott D. Boyd3, role for BMPCs in maintaining humoral immu-
Edmund K. Waller2,5, Rafi Ahmed1,2§ nity to influenza. We also measured influenza-
specific memory B cells (MBCs) in the blood of
A universal vaccine against influenza would ideally generate protective immune responses that are the donors. Influenza-specific cells accounted
not only broadly reactive against multiple influenza strains but also long-lasting. Because long-term for 0.93% (IQR: 0.58 to 1.5%) of IgG MBCs in
serum antibody levels are maintained by bone marrow plasma cells (BMPCs), we investigated the the blood (Fig. 1D). This MBC frequency cor-
production and maintenance of these cells after influenza vaccination. We found increased numbers of related with blood antibody titers (Fig. 1E).
influenza-specific BMPCs 4 weeks after immunization with the seasonal inactivated influenza vaccine, Furthermore, there was a clear trend toward
but numbers returned to near their prevaccination levels after 1 year. This decline was driven by the loss higher numbers of influenza-specific MBCs in
of BMPCs induced by the vaccine, whereas preexisting BMPCs were maintained. Our results suggest those donors with the most influenza-specific
that most BMPCs generated by influenza vaccination in adults are short-lived. Designing strategies to BMPCs (Fig. 1F).
enhance their persistence will be a key challenge for the next generation of influenza vaccines.
We next investigated whether seasonal in-
T he goal of vaccination is to generate long- whom we were able to obtain paired bone fluenza vaccination resulted in the production
lasting protection against infection. Most marrow samples before and 1 month after of new influenza-specific BMPCs. Influenza-
vaccines in clinical use achieve this pro- influenza vaccination, in conjunction with specific BMPCs increased significantly to 1.9%
tection at least in part through the gen- a full schedule of blood samples. In addition, (IQR: 1.2 to 2.9%) of IgG BMPCs by day 28 after
eration of pathogen-specific antibody we obtained a third bone marrow draw at vaccination (Fig. 2A and fig. S1A). The season-
responses. Antibody levels peak in the months ~1 year after vaccination from some of the al influenza vaccine contains hemagglutinin
after vaccination, followed by a decline to a subjects, allowing us to assess the long-term (HA) proteins from multiple viral strains: one
plateau level that may be maintained for dec- maintenance of vaccine responses in the from an H1N1 influenza A strain, one from an
ades with minimal decline (1–3). Animal mod- bone marrow. H3N2 strain, and either one or two HA pro-
els have shown that these plateau antibody teins from influenza B strains. An HA protein
levels are maintained by nondividing, bone BMPC responses were measured in 53 vol- similar or identical to that of the pandemic
marrow–resident long-lived plasma cells (4, 5). unteers who received the inactivated influenza H1N1 strain A/California/7/2009 was included
Studies of antibody synthesis rates have sug- vaccine between 2009 and 2018. Some of these in the vaccine from the 2010–2011 season until
gested that bone marrow plasma cells (BMPCs) 53 donors enrolled in multiple study years (see the 2016–2017 season, allowing us to monitor
are likely to produce most serum immuno- the materials and methods) and were treated responses against a single, consistent antigen
globulin G (IgG) in humans as well (6, 7). as separate enrollments, for a total of 75 vac- in multiple seasons. IgG BMPCs specific for
Consistent with this, total and antigen- cine responses studied. In addition to provid- this HA increased from 0.11% (IQR: 0.07 to
specific serum antibody levels correlate closely ing blood samples on the day of vaccination 0.18%) to 0.36% (IQR: 0.18 to 0.76%) on day
with BMPC numbers in humans (8, 9). Anti- (day 0) and on days 7, 14, 28, and 90 after 28 (Fig. 2A and fig. S1B). We also measured
body titers and protective immunity decline vaccination, the volunteers also underwent influenza-specific IgA BMPCs when plasma
rapidly after seasonal influenza vaccination bone marrow aspiration procedures on days cell yields were high enough to run multiple
(10, 11), suggesting either that the vaccine 0 and 28. In addition, some donors provided antigen-specific ELISpot assays from each
may fail to elicit BMPCs or that these BMPCs blood and bone marrow samples ~1 year after sample. Although we only obtained paired IgA
fail to become long-lived. To investigate these vaccination, either at their reenrollment for data for six donors, there was a trend toward
possibilities, we designed a clinical study of the following year’s study or as their final increasing numbers of IgA BMPCs (Fig. 2A).
healthy adults (20 to 45 years of age) from study time point. This trend was significant in the unpaired
analysis (fig. S1C), which included additional
1Emory Vaccine Center and Department of Microbiology and We first examined influenza-specific immu- donors for whom IgA BMPCs could be mea-
Immunology, Emory University, Atlanta, GA, USA. 2Emory- nity at baseline before vaccination. Using sured at day 0 or day 28 but not both. These
UGA Center of Excellence of Influenza Research and plasma cells enriched from the bone marrow increases were specific to plasma cells rec-
Surveillance (CEIRS), Atlanta GA, USA. 3Department of aspirates, we performed enzyme-linked immuno- ognizing influenza vaccine antigens; no changes
Pathology, Stanford University, Stanford, CA, USA. 4Cell sorbent spot (ELISpot) assays to quantify were observed in the percentage of cells se-
Signaling Technology, Inc., Danvers, MA, USA. 5Department influenza-specific BMPCs (Fig. 1A). Cells se- creting IgG specific for tetanus toxin (Fig. 2A
of Hematology and Oncology, Winship Cancer Institute, creting influenza-specific antibodies (both and fig. S1C). Thus, we found no evidence of
Emory University School of Medicine, Atlanta, GA, USA. IgG and IgA) were readily detectable in study bystander effects of the vaccine on the fre-
6Department of Pediatrics, Division of Infectious Disease, volunteers before vaccination and were quencies of BMPCs specific for unrelated
Emory University School of Medicine, Atlanta, GA, USA. markedly more common than cells producing antigens.
7Division of Infectious Diseases, School of Medicine, Emory antibodies specific for tetanus toxin. Influenza-
University, Atlanta, GA, USA. specific BMPC numbers were also five- to We have previously shown that influenza-
*Present address: Immunology Division, Garvan Institute of 10-fold higher than we have observed for specific antibody-secreting cells (ASCs) appear
Medical Research, Darlinghurst, NSW, Australia. varicella zoster virus (12). The percentage in the blood at ~1 week after vaccination (13).
†Present address: Bluefin BioMedicine, Beverly, MA, USA. of influenza-specific IgG BMPCs on day 0 Because bone marrow aspirate samples will
‡Present address: BioMedicine Design, Medicinal Sciences, Worldwide followed an approximately log-normal dis- unavoidably contain some level of contamina-
Research & Development, Pfizer Inc., Cambridge, MA, USA. tribution (Fig. 1B). The geometric mean per- tion with peripheral blood, we wanted to
§Corresponding author. Email: [email protected] confirm that influenza-specific ASCs were not
in circulation at the time of bone marrow

Davis et al., Science 370, 237–241 (2020) 9 October 2020 1 of 5

RESEARCH | REPORT

aspiration. Influenza-specific IgG ASCs were declined sharply by day 14, and were again would not have been inflated by contamina-
at very low levels or undetectable in the blood very rare or undetectable on day 28 (Fig. 2B). tion with circulating ASCs. The magnitude of
on day 0, with the exception of one donor with Similar results were seen for IgA and IgM the day 7 ASC response correlated with the
a recent history of an influenza-like illness ASCs (fig. S1, D and E). Thus, the percentages increase in influenza-specific BMPCs after
(green symbols). ASC numbers peaked at day 7, of influenza-specific BMPCs on days 0 and 28 vaccination (Fig. 2C). This is consistent with

Fig. 1. Analysis of including repeat study participants; n = 43 unique donors). The means and standard
influenza-specific BMPCs deviations of the log-transformed data are indicated by red lines in (B) and (D). In (C),
before vaccination. (E), and (F), Pearson correlation analysis was performed on the log-transformed
(A) Example ELISpot data BMPC, MBC, and ELISA data. The green dots represent a donor with evidence of recent
using magnetically influenza infection who was excluded from statistical calculations.
enriched BMPCs with
detection of cells secret-
ing total or antigen-
specific IgG or IgA.
(B) Number of BMPCs
producing influenza-
specific IgG before vacci-
nation expressed as a
percentage of the total
number of IgG-secreting
BMPCs. (C) Correlation
between prevaccination
influenza-specific IgG
BMPCs and influenza-
specific IgG levels in the
blood as measured by
Enzyme-linked immuno-
sorbent assay (ELISA).
(D) Influenza-specific IgG
MBCs from the peripheral
blood are shown as a
percentage of total IgG MBCs before vaccination. (E) Correlation between
influenza-specific IgG MBCs and IgG titers in the blood before vaccination.
(F) Correlation between the percentages of influenza-specific IgG BMPCs and
MBCs. The data shown in (B) to (F) are for all study donors for which day
0 influenza-specific BMPC and MBC percentages were measured (n = 59

Fig. 2. Increase in influenza-specific plasma cells in the bone marrow after transformed data. The means and standard deviations of the log-transformed
vaccination. (A) The percentage of IgG- or IgA-secreting BMPCs specific for data are indicated in red. (B) Blood IgG ASC response to the influenza vaccine
the indicated antigens was measured by ELISpot on the day of vaccination (d0) measured by ELISpot. Data are shown for n = 45 vaccine responses from
and 28 days later (d28). Influenza, trivalent or quadrivalent inactivated influenza 35 unique donors. The dotted line represents the lower limit of detection for the
vaccine (n = 46 IgG responses measured from 34 unique donors; n = 6 IgA assay. (C) Pearson correlation between the magnitude of the IgG ASC response
responses from six unique donors); pdm H1 HA, recombinant HA protein from at day 7 and the increase in influenza-specific BMPCs, calculated as the
pandemic H1N1 strain A/California/07/2009 (n = 20 responses from 14 unique percentage of influenza-specific IgG BMPCs at day 28 minus this percentage at
donors); tetanus, tetanus toxin C fragment (n = 6 responses from six unique day 0. Green symbols and lines in (A) to (C) indicate a donor with evidence of
donors). P values shown are for paired, two-tailed t tests using the log- recent influenza infection who was excluded from the statistical analyses.

Davis et al., Science 370, 237–241 (2020) 9 October 2020 2 of 5

RESEARCH | REPORT

models suggesting that BMPCs are derived Fig. 3. Decline of influenza-specific BMPCs at 1 year after vaccination. (A) Percentage of influenza-
from blood ASCs that migrated to the bone specific IgG BMPCs shown over time in volunteers who returned for a third bone marrow aspiration ~1 year
marrow (14). after vaccination. P values shown are for paired, two-tailed t tests of the log-transformed data. (B) Same
data plotted as in (A) except the percentages of influenza-specific BMPCs at day 28 and 1 year are
To determine whether the increase in normalized to the day 0 percentage to visualize the fold change relative to prevaccination levels in each
influenza-specific BMPCs was maintained, we donor. Numbers above the points indicate the geometric means of the fold changes, with the IQRs in
obtained bone marrow samples from a subset parentheses. (C) Influenza-specific blood IgG ELISA titers at day 28, day 90, and 1 year normalized to the
of donors at a late time point (mean 325 days day 0 titer and plotted as in (B). The same n = 18 responses from 15 unique donors are shown in all panels.
after vaccination; range: 220 to 457 days). The
percentage of influenza-specific BMPCs declined types that we identified, only one of them tive polymerase chain reaction (qPCR) probe
significantly between day 28 and the late sample (clonotype 8 from donor 42) contributed and primer sets (fig. S5C). Using this qPCR
(Fig. 3A). In subjects who reenrolled in multiple detectably to H1-binding antibodies found assay, we observed a similar trend where
seasons and were vaccinated a second time, in the serum on day 0 (fig. S5B). The other 10 vaccine-specific antibody clonotypes expanded
influenza-specific BMPCs typically increased H1-specific clonotypes could only be detected and then declined in BMPC RNA samples
after both immunizations, although there was in the serum starting at day 7 or day 14. This (median increase in frequency at 1 year was
substantial heterogeneity among donors (fig. suggests either that there were few long-lived 0.09-fold of the initial increase at day 28;
S2). As shown in Fig. 3B, in the subset of do- plasma cells belonging to these clonotypes IQR: 0- to 0.38-fold). Both the initial increase
nors who provided a long-term follow-up before vaccination or that these preexisting in clonotype frequency among BMPCs and
sample, there was a geometric mean 1.9-fold plasma cells produced antibodies that bound the long-term increase correlated with the
change (IQR: 1.3- to 3.2-fold) in influenza- poorly to the pandemic H1 HA protein. representation of each clonotype among day 7
specific BMPCs between days 0 and 28, sim- ASC sequences (Fig. 4, C and D), which is
ilar to the overall cohort. However, by 1 year, Of the 15 influenza-specific clonotypes that consistent with what we observed by ELISpot
the percentage of influenza-specific BMPCs were expanded in the day 7 ASCs, 14 increased (Fig. 2C) and in keeping with the notion that
had declined to baseline. Analysis of serum significantly among BMPC IgG sequences blood ASCs give rise to BMPCs.
antibody levels revealed similar trends (Fig. between days 0 and 28 after vaccination (P =
3C). The level of decline in BMPCs did not 0.0003 by Wilcoxon signed-rank test; Fig. 4A). Our clonotype-tracking results suggested
appear to correlate with the timing of the Thus, the same antibody families that con- that the contraction in influenza-specific
1-year bone marrow aspiration (fig. S3), tributed to the blood ASC response also con- BMPCs between day 28 and 1 year that we
suggesting that the loss of newly generated tributed to the BMPC compartment, again saw by ELISpot was due to the loss of newly
influenza-specific BMPCs happened within consistent with the idea that blood ASCs generated cells. To determine whether pre-
7 months after vaccination. give rise to BMPCs. Between day 28 and the existing influenza-specific BMPCs were more
late time point (~1 year), the percentage of stable over time, we examined “nonrecruited”
We next examined in detail the clonal com- BMPC IgG sequences belonging to these influenza-specific clonotypes in the two donors
position of the B cell response to the vaccine clonotypes decreased sharply (P = 0.0017). whose H1-specific responses were profiled by
by performing immunoglobulin heavy chain After 1 year, 11 of the 14 clonotypes still made serum proteomics analysis. These clonotypes
repertoire sequencing of BMPCs and ASCs up a higher percentage of BMPC IgG reads were defined as HA-specific antibody families
from four donors. Three out of four donors compared with day 0; however, this difference that were detected in the serum and in IgG
(donors 35, 40, and 42) had a strong day 7 was not statistically significant (P = 0.068). BMPC sequences on day 0, but which repre-
ASC response to the vaccine (fig. S4A), and There was a wide variation in how well the sented <0.1% of day 7 vaccination-induced ASCs
all four donors showed serologic responses initial increase in frequency of each clonotype by antibody repertoire sequencing (tables S4
to each influenza strain in the vaccine (fig. among BMPC sequences was sustained (Fig. and S5). Most of these clonotypes could be
S4B), particularly to the H1N1 component. 4B). The median increase in frequency for detected among both IgG and IgA BMPC se-
To identify influenza-specific antibody fami- these 14 clonotypes after 1 year was only 0.15- quences (see table S5). The nonrecruited clones
lies (or “clonotypes”) in the repertoire sequenc- fold as high as the initial increase seen at showed no significant change in their abun-
ing data, we used two strategies: production day 28 (IQR: 0.01- to 0.29-fold). In other words, dances over 1 year in either donor (Fig. 4, E and
of monoclonal antibodies (mAbs) cloned from for most clones, 70 to 99% of the newly gen- F), indicating that they likely represented a
day 7 ASCs (all four donors) and a proteomics- erated BMPCs were lost over 1 year. As a sep- stable population of long-lived plasma cells
based approach in which antibodies specific to arate strategy to measure clonotype abundances that had been generated by previous influenza
the H1 HA protein were identified from the over time, we designed clone-specific quantita- infection or vaccination.
polyclonal serum (donors 35 and 42) (fig. S5A).
We identified 15 clonotypes that were highly
expanded (>0.5% of day 7 ASC sequences) and
determined to be influenza specific by either
mAb cloning or proteomics analysis (fig. S5B
and table S5). These antibody families already
exhibited extensive somatic hypermutation
in their heavy chain variable region genes at
day 7, indicating that they were likely derived
from preexisting MBCs. All 15 clonotypes
were represented among both IgA- and IgG-
expressing ASCs, although some clonotypes
were enriched for one isotype or the other.
We hypothesize that multiple MBCs from
each clonotype responded to the vaccine,
with varying recruitment of IgA versus IgG
memory cells. Of the 11 H1-specific clono-

Davis et al., Science 370, 237–241 (2020) 9 October 2020 3 of 5

RESEARCH | REPORT

Fig. 4. Longitudinal tracking of influenza–specific antibody lineages (clono- comprised a lower percentage of IgG reads at 1 year compared with day 0.
types) after vaccination. Clonotypes were tracked over time in BMPC and (C) Correlation between the percentage of day 7 ASC IgG sequences belonging to
ASC immunoglobulin sequences. Fifteen clonotypes comprising at least 0.5% of each clonotype and the initial increase in that clonotype among BMPC IgG
day 7 blood ASC IgG sequences were considered to have been “recruited” into transcripts between days 0 and 28. (D) Correlation between the percentage of
the vaccine-specific response. Clonotypes are named by their frequency among day 7 ASC IgG sequences belonging to each clonotype and the long-term
ASC sequences (e.g., “35-2” is the clonotype with second-highest read count in increase of these clonotypes among BMPC sequences between day 0 and 1 year.
IgG sequences from day 7 ASCs from donor 35). (A) Percentage of BMPC IgG (E) Persistence of preexisting, nonrecruited H1 HA-specific clonotypes in donor
sequences belonging to each recruited clonotype shown on day 0, day 28, and 35. Clonotypes were defined as preexisting if they were found in the BMPC
1 year after vaccination. (B) For the 14 recruited clonotypes that increased in the repertoire sequencing data at day 0 and confirmed to be influenza specific by
bone marrow between days 0 and 28, the long-term increase after 1 year is serum proteomics analysis or mAb production. Clonotypes were defined as
shown as a proportion of the initial increase. A clonotype that neither expanded nonrecruited if they accounted for <0.1% of day 7 IgG, IgM, and IgA ASC
nor contracted between day 28 and 1 year would have a value of 1 on this scale, sequences. (F) Persistence of preexisting, nonrecruited H1 HA-specific
whereas a clonotype that declined to its starting frequency would have a value of clonotypes in donor 42. P values shown in (A), (E), and (F) are for the Wilcoxon
0. Values >1 indicate that the clonotype expanded among BMPC sequences signed-rank test. In (C) and (D), Pearson correlation was performed using the
between day 28 and 1 year, and values below 0 indicate clonotypes that nontransformed data, and the best-fit linear regression lines are shown.

In recent years, it has become clear that to influenza vaccine are not well maintained. space there and (ii) it must undergo changes in
humans are capable of generating protective We show that influenza-specific titers in the gene expression and metabolism that promote
antibody responses directed against epitopes serum correlate with BMPC numbers before longevity (15–20). Thus, our data suggest that
on the influenza HA and neuraminidase (NA) vaccination, supporting a key role for these most vaccine-induced BMPCs fail at one or
proteins that are broadly conserved across cells in maintaining antibody levels. Vaccina- both of these steps.
diverse viral strains. This has raised hopes that tion does lead to the generation of new BMPCs,
a “universal” influenza vaccine targeting these and some of these persist for at least 1 year. The decline in BMPC and antibody levels
conserved regions could be developed. If such However, most of the newly generated BMPCs observed here was likely not specific to the in-
a vaccine elicited protective antibody responses are lost within 1 year, showing that localiza- fluenza vaccine. One study of children receiv-
that persisted for at least several years, it would tion to the bone marrow is not enough to ing the hepatitis B vaccine showed declines in
eliminate the need for annual immunization, determine longevity. Two steps are required antibody titers of 85 to 90% from peak levels in
which would be a major improvement on for a BMPC to become a long-lived plasma the first year after vaccination (21). Conversely,
current vaccines. The data presented here add cell: (i) The cell must reach the appropriate in a study of adults given a first booster of
to our understanding of why humoral responses survival niche and successfully compete for the tetanus, diphtheria, and acellular pertussis
vaccine (Tdap), IgG antibody levels to the

Davis et al., Science 370, 237–241 (2020) 9 October 2020 4 of 5

RESEARCH | REPORT

vaccine antigens declined only 40 to 66% for boosting long-term numbers of BMPCs 30. T. Arulraj, S. C. Binder, P. A. Robert, M. Meyer-Hermann,
in the first year and another 15 to 25% in Front. Immunol. 10, 2116 (2019).
the following year (22). This is a consider- targeting conserved epitopes.
ably slower rate of decay than we observed, ACKNOWLEDGMENTS
suggesting that there is room for improve- REFERENCES AND NOTES
ment with influenza vaccines. We thank the Emory Flow Cytometry Core for assistance with
1. E. Hammarlund et al., Nat. Med. 9, 1131–1137 (2003). cell sorting, the Emory Integrated Genomics Core for assistance
A few findings from this study may be 2. S. Crotty et al., J. Immunol. 171, 4969–4973 (2003). with MiSeq experiments, and R. Antia and V. Zarnitsyna for
helpful for designing next-generation influ- 3. I. J. Amanna, N. E. Carlson, M. K. Slifka, N. Engl. J. Med. 357, valuable discussions. Funding: NIAID Centers of Excellence
enza vaccines capable of eliciting longer-term for Influenza Research and Surveillance (CEIRS) contract
immunity. First, the blood ASC response af- 1903–1915 (2007). HHSN272201400004C; Influenza Pathogenesis and Immunology
ter vaccination was highly predictive of the 4. R. A. Manz, A. Thiel, A. Radbruch, Nature 388, 133–134 (1997). Research Center (IPIRC, CEIRS) contract HHSN266200700006C;
overall BMPC response, both in the short 5. M. K. Slifka, R. Antia, J. K. Whitmire, R. Ahmed, Immunity 8, NIH grants 1R01AI127877 and 1R01AI130398 (SDB). Author
term and in the long term, and thus may be a contributions: Conceptualization: C.W.D., J.W., E.K.W., R.A.;
useful marker for screening vaccine candi- 363–372 (1998). Data curation: C.W.D., K.J.L.J., J.D., R.G., W.C.C.; Formal analysis:
dates. Because ASC generation after influenza 6. R. McMillan et al., J. Immunol. 109, 1386–1394 (1972). C.W.D., K.J.L.J.; Investigation: C.W.D., K.J.L.J., M.M.M., J.D.,
vaccination can be increased through the use 7. T. Hibi, H. M. Dosch, Eur. J. Immunol. 16, 139–145 (1986). C.Che, C.Cha, S.L.L., W.C.C.; Methodology: C.W.D., K.J.L.J., M.M.M.,
of adjuvants (23–25), it may be useful to assess 8. I. Turesson, Acta Med. Scand. 199, 293–304 (1976). J.D., J.W., W.C.C., S.D.B., E.K.W., R.A.; Project administration:
whether adjuvants are able to improve long- 9. T. Pritz et al., Eur. J. Immunol. 45, 738–746 (2015). E.K.W., R.A.; Funding acquisition: E.K.W., R.A.; Resources: K.J.L.J.,
term BMPC responses in humans. Second, most 10. G. T. Ray et al., Clin. Infect. Dis. 68, 1623–1630 (2019). J.D., R.G., W.C.C., A.K.M., S.D.B., E.K.W.; Software: K.J.L.J.,
of the preexisting H1-specific antibody fam- 11. P. F. Wright et al., Pediatr. Infect. Dis. J. 27, 1004–1008 (2008). S.D.B.; Supervision: W.C.C., S.D.B., E.K.W., R.A.; Visualization:
ilies detected in the serum on day 0 were not 12. C. S. Eberhardt et al., J. Virol. 94, e02127-19 (2020). C.W.D., K.J.L.J.; Writing – original draft: C.W.D., R.A.; Writing –
detected in the day 7 ASC response, and most 13. J. Wrammert et al., Nature 453, 667–671 (2008). Review & Editing: C.W.D., K.J.L.J., S.D.B., E.K.W., R.A. Competing
of the H1-specific clonotypes that were re- 14. M. Odendahl et al., Blood 105, 1614–1621 (2005). interests: The authors declare no competing interests. Data
cruited did not contribute appreciably to serum 15. N. Pengo et al., Nat. Immunol. 14, 298–305 (2013). and materials availability: Antibody high-throughput DNA-
H1-specific titers before vaccination. This may 16. L. D’Souza, D. Bhattacharya, Immunol. Rev. 288, 161–177 (2019). sequencing data have been deposited in the Sequence Read Archive
reflect antibody feedback, in which preexisting 17. S. M. Lightman, A. Utley, K. P. Lee, Front. Immunol. 10, 965 (BioProject accession no. PRJNA649987).
serum antibodies binding to the vaccine block
B cells recognizing the same epitopes from (2019). SUPPLEMENTARY MATERIALS
acquiring antigen (26–30). Designing strat- 18. J. L. Halliley et al., Immunity 43, 132–145 (2015).
egies to overcome this effect will be critical 19. H. E. Mei et al., Blood 125, 1739–1748 (2015). science.sciencemag.org/content/370/6513/237/suppl/DC1
20. A. Radbruch et al., Nat. Rev. Immunol. 6, 741–750 (2006). Materials and Methods
21. V. Gilca et al., Vaccine 27, 6048–6053 (2009). Tables S1 to S5
22. S. van der Lee et al., Front. Immunol. 9, 681 (2018). Figs. S1 to S5
23. R. J. Cox et al., PLOS ONE 10, e0131652 (2015). References (31–42)
24. L. Li, Y. Honda-Okubo, C. Li, D. Sajkov, N. Petrovsky, PLOS ONE
View/request a protocol for this paper from Bio-protocol.
10, e0132003 (2015).
25. A. H. Ellebedy et al., Proc. Natl. Acad. Sci. U.S.A. 117, 16 October 2019; accepted 2 August 2020
Published online 13 August 2020
17957–17964 (2020). 10.1126/science.aaz8432
26. V. I. Zarnitsyna, J. Lavine, A. Ellebedy, R. Ahmed, R. Antia,

PLOS Pathog. 12, e1005692 (2016).
27. S. F. Andrews et al., Immunity 51, 398–410.e5 (2019).
28. D. Hobson, F. A. Baker, R. L. Curry, Lancet 2, 155–156 (1973).
29. Y. Zhang et al., J. Exp. Med. 210, 457–464 (2013).

Davis et al., Science 370, 237–241 (2020) 9 October 2020 5 of 5

RESEARCH

CORONAVIRUS mutant identified as a high-stringency candi-
date gene.
MHC class II transactivator CIITA induces
cell resistance to Ebola virus and All transposon insertions at the second CIS—
SARS-like coronaviruses located on chromosome 16—were upstream
of the gene CIITA and were oriented in the
Anna Bruchez1*†, Ky Sha1*, Joshua Johnson2‡, Li Chen3, Caroline Stefani1, Hannah McConnell1§, sense orientation, consistent with activation
Lea Gaucherand1¶, Rachel Prins1, Kenneth A. Matreyek4†, Adam J. Hume5,6, Elke Mühlberger5,6, of expression (Fig. 1F and fig. S1). CIITA over-
Emmett V. Schmidt7, Gene G. Olinger2,5,6,8, Lynda M. Stuart1,9, Adam Lacy-Hulbert1,10# expression in wild-type U2OS cells increased
cell survival, reduced green fluorescent protein
Recent outbreaks of Ebola virus (EBOV) and severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) (GFP) reporter expression, and completely
have exposed our limited therapeutic options for such diseases and our poor understanding of the cellular inhibited plaque formation, which confirms
mechanisms that block viral infections. Using a transposon-mediated gene-activation screen in human cells, we that CIITA increases resistance to EboGP-VSV
identify that the major histocompatibility complex (MHC) class II transactivator (CIITA) has antiviral activity 100- to 1000-fold (Fig. 2, A to E, and fig. S2).
against EBOV. CIITA induces resistance by activating expression of the p41 isoform of invariant chain CIITA-overexpressing cells were also resistant
CD74, which inhibits viral entry by blocking cathepsin-mediated processing of the Ebola glycoprotein. to EboGP-pseudotyped single cycle viruses
We further show that CD74 p41 can block the endosomal entry pathway of coronaviruses, including SARS- (Fig. 2, F and G), which strongly suggests
CoV-2. These data therefore implicate CIITA and CD74 in host defense against a range of viruses, and they that CIITA inhibits viral entry rather than
identify an additional function of these proteins beyond their canonical roles in antigen presentation. targeting viral transactivators as suggested for
HIV and human T cell leukemia virus (HTLV)
R ecent and ongoing outbreaks of Ebola libraries were treated with Ebola glycoprotein (7, 8). Furthermore, using EboGP virus–like
virus (EBOV) in Africa (1) and the severe (EboGP)–expressing recombinant vesicular particles (EboGP-VLPs) carrying b-lactamase
acute respiratory syndrome coronavirus stomatitis virus (referred to as EboGP-VSV). (9), we found that CIITA did not affect the
2 (SARS-CoV-2) pandemic highlight the Susceptible wild-type U2OS cells died after internalization of EboGP-VLPs into cells (Fig.
need to identify additional treatment 3 to 4 days of treatment, whereas surviving 2H), but it blocked viral fusion, which occurs
strategies for viral infections, including ap- cells could be expanded from mutagenized in the endosome (10) (Fig. 2I). CIITA-expressing
proaches that might complement traditional libraries and exhibited stable resistance to U2OS cells were also highly resistant to in-
antivirals. Of particular interest is the identi- rechallenge with EboGP-VSV (Fig. 1B). These fection by high titers of native EBOV, showing
fication of host-directed therapies that target cells showed no cross-resistance to vesicular reduced reporter gene expression, cell death,
common vulnerabilities and may be efficacious stomatitis virus (VSV) containing the VSV and plaque formation (Fig. 2, J to M). CIITA
against multiple viruses, including those that glycoprotein (VSVg-VSV) (Fig. 1C), which sug- expression did not inhibit replication of an
may emerge in the future. gests that most of the resistance mechanisms EBOV minigenome, which indicates that CIITA
selected in this screen targeted EboGP-mediated does not act on the viral replication complex
We set out to identify host pathways of entry. (fig. S3). Furthermore, CIITA inhibited infec-
cellular resistance to pathogens with pandemic tion mediated by glycoproteins (GPs) from a
potential, using a transposon-mutagenesis– We identified candidate resistance genes by range of EBOV species—including Sudan, Zaire,
forward genetic approach. We used a modified identifying genomic regions with high numbers and Reston—as well as by those from the dis-
PiggyBac (PB) transposon (Fig. 1A), which of transposon insertions [referred to as common tantly related filovirus Marburg virus (Fig. 2G).
stimulates or disrupts the expression of neigh- insertion sites (CISs)] (3). Combining data from Thus, CIITA induces broad antiviral activity
boring genes, thereby allowing an interroga- eight independent screens revealed seven ge- against EBOV and other pathogenic filoviruses
tion of both gene activation and inactivation nomic loci with highly statistically significant through the inhibition of viral GP-mediated
in a single screen (2). Transposon-mutagenized (P < 10−8) CISs that occurred in more than one entry.
screen, representing high-confidence candidate-
1Benaroya Research Institute, Seattle, WA 98101, USA. 2National resistance mutations (Fig. 1D, outer ring). Likely CIITA, also known as NLRA, is a nucleotide-
Institute of Allergy and Infectious Diseases (NIAID) Integrated target genes of transposon insertions were binding oligomerization domain (Nod)–like
Research Facility, Frederick, MD 21702, USA. 3Massachusetts identified on the basis of transposon insertion receptor (NLR) (11), but unlike most other
General Hospital, Boston, MA 02114, USA. 4Department of position and orientation (Fig. 1D and table S1). NLRs, which function as cytosolic sensors,
Genome Sciences, University of Washington, Seattle, WA 98109, We focused on the two genes that were found CIITA is a transcription factor (12). We there-
USA. 5Boston University School of Medicine, Boston, MA 02118, in all eight screens using the most stringent fore hypothesized that its antiviral activity
USA. 6National Emerging Infectious Diseases Laboratories, criteria. occurred through the altered expression of
Boston University, Boston, MA 02118, USA. 7Merck and Co., Inc, host target genes. Supporting this hypothesis,
Kenilworth, NJ 07033, USA. 8MRIGlobal, Gaithersburg, MD The first of these was NPC1, located on mutation of domains required for transcrip-
20878, USA. 9Bill and Melinda Gates Foundation, Seattle, WA chromosome 18. All transposon insertions tional activity completely ablated CIITA anti-
98109, USA. 10Department of Immunology, University of at this site were intragenic in both sense and viral activity (fig. S4). Resistance also required
Washington, Seattle, WA 98109, USA. antisense orientations, and all were predicted NF-Y, a component of the enhanceosome mul-
*These authors contributed equally to this work. to disrupt NPC1 expression (Fig. 1E). This is tiprotein complex, which mediates transcrip-
†Present address: Department of Pathology, Case Western Reserve consistent with the role of NPC1 as the EBOV tional activation by CIITA (13), but resistance
University, Cleveland, OH 44106, USA. receptor (4, 5) and validates our screening was independent of another enhanceosome
‡Present address: AbViro LLC, Bethesda, MD 20814, USA. approach. Notably, U2OS cells are haploid protein, RFX5 (figs. S4 and S5). Antiviral ac-
§Present address: Division of Basic Sciences, Fred Hutchinson at the NPC1 locus (6), and these transposon tivity was therefore mediated by a subset of
Cancer Research Center, Seattle, WA 98109, USA. insertions are therefore predicted to gen- NF-Y–dependent, RFX5-independent CIITA
¶Present address: Graduate Program in Molecular Microbiology, erate NPC1-null cells, which explains why target genes, which includes genes associated
Tufts Graduate School of Biomedical Sciences and Department of NPC1 was the only predicted gene-disruption with antiviral immunity (14). Systematic knock-
Molecular Biology and Microbiology, Tufts University, Boston, MA down of all CIITA target genes identified a
02155, USA. single gene, CD74, required for CIITA-mediated
#Corresponding author. Email: [email protected] resistance (Fig. 3, A and B). This was confirmed

Bruchez et al., Science 370, 241–247 (2020) 9 October 2020 1 of 6

RESEARCH | REPORT

by CRISPR knockout of CD74 expression and Treatment with interferon-g (IFN-g) and lipo- primary immune cells, which can be induced
by exposure to IFN-g and LPS.
function in CIITA-overexpressing cells (Fig. 3C). polysaccharide (LPS) induced expression of
Both CIITA and CD74 are expressed at high CIITA and CD74, and Ciita−/− and Cd74−/− CD74 is the major histocompatibility com-
BMDMs primed with IFN-g and LPS had higher plex class II (MHC-II) invariant chain, and
levels by macrophages and dendritic cells human cells express four main isoforms of
(DCs), which are early targets of EBOV levels of EboGP-VLP fusion than those observed CD74, which differ in the presence of an
(15, 16). To test whether CIITA has antiviral N-terminal endoplasmic reticulum (ER) reten-
activity in immune cells, we used primary in equivalent wild-type cells (Fig. 3, D to G, and tion signal and an internal thyroglobulin do-
bone marrow–derived macrophages (BMDMs) fig. S6). Similar results were seen in Cd74−/− main (Fig. 4A) (17). Only one CD74 isoform,
from Ciita−/− and Cd74−/− mice. Naïve BMDMs bone marrow–derived DCs and in a CD74−/− p41, was able to fully rescue resistance to
did not express high levels of CIITA or CD74, human macrophage-like cell line (differenti- EboGP-VSV infection in CIITA-expressing,
and they showed no difference in viral fusion.
ated THP-1) (figs. S7 and S8). Thus, endoge-

nous CIITA and CD74 have antiviral activity in

Fig. 1. Transposon- A B 100 ** ** C
mediated activation tag-
ging generates mutant Cell Survival 100 Parent U2OS
cells resistant to Ebola. gene activation splice (% total cells) Cell Survival
(A) Modified PB transposon. selection promoter Selected Pool(% total cells)
SV-Puro-pA, puromycin PB donor 50 50
selection cassette; CMV, PB
cytomegalovirus promoter;
SD, splice donor. SV-Puro-pA CMV SD
(B and C) Resistance of
selected cells to EboGP-VSV 0 0
(B) and VSVg-VSV (C). Data 0 24 48 72 96 120 0 24 48 72 96 120
are means ± SD of n = Time (h) Time (h)
3 replicates for one
representative pool. D PDE4B 1 TNFSF18
Student’s t test; **P < 0.01.
(D) Distribution of trans- 1 library
poson insertions. Inner rings
show insertions per 1 Mb for Combined 2 8 libraries
individual libraries (black Y
histograms) and CISs (P <
10−7). Outer ring shows (p--lvoalg[u1e0])
combined insertions for all freq
libraries (black histogram) 15 TGFBRAP1
and lowest P value for CISs
(red bubble plot). Point size X
represents the number of
libraries with the CIS. 19 20 21 22 6
freq, frequency. (E and
F) Cumulative independent Individual
insertions from all eight libraries
libraries mapping to NPC1 4
(E) and CIITA (F). NPC1 3

17 18 14 CLCN3

16 7 PDE4D
6 KIAA0825
CIITA 5SPINK5

15

ABCC4 11 10 LINC100506393
13 8
12 ATG MIR31HG
CNTN1 9 TMC1

LOC100506393 10 kb

E NPC1 2 of 6

ATG

F CIITA

20 kb ATG

Bruchez et al., Science 370, 241–247 (2020) 9 October 2020

RESEARCH | REPORT

CD74-knockout cells (Fig. 4B and fig. S9). CD74 was not limited to U2OS cells, as CD74 in endosomes. Mutant constructs of CD74 re-
p41 conferred resistance independently of p41 similarly inhibited fusion when expressed vealed that only the thyroglobulin domain is
CIITA expression (Fig. 4C), which demonstrates in THP-1 cells (Fig. 4D). The p41 isoform con- essential for antiviral activity, but dissociation
that CD74 p41 expression was sufficient to tains the thyroglobulin domain, lacks the ER from the membrane—either by addition of
induce antiviral activity. This property of retention signal, and normally accumulates a furin cleavage site (labeled furin in Fig. 4E)

A ** ** ** B Cntrl CIITA C 12

100 VSV Ebo VSV Ebo Effective Viral Titer 10
undil (PFU/ml)
Cell Survival (%) 10 -2 1010 **
10 -4
CIITA Viral dilution 10 -6 8 Cntrl
Cntrl 10 -8 CIITA
10-10 10
125
50 CIITA 6 **

10 100

104 75 Cntrl
50 CIITA
0 102
25
0.01 0.1 1 10 0 ND 0

EboGP-VSV MOI 10 VSVg EboGP

VSVg EboGP MLV

VSV

D E 100 F

U2OS Cntrl

Infected cells 10 ** Infectivity
(% total) (% of control)
1 Cntrl
CIITA

0.1

0.01
VSVg LFVGP EboGP

VSV

InfectivityG Cntl CIITA H I VLP fusion activity 2.0 Cntrl
(% of control) (Ratio 460/530) 1.5 CIITA
150 ** ** ** ** ** ** 15 20

100 15
10

10

55
VLP uptake 1.0 * *
(MFIx103 )
50 VLP per cell

0.5

0 00 0.0 no no
VSVg EBOV BDBV TAFV RESTV SUDV MARV env env
Cntrl CIITA Cntrl CIITA 10 1 0.1 10 1 0.1

J uninfected EBOV EboGP-VLP (μl)

Cntrl K ** L ** M

100 100 200
Infected cells **
Cell Survival (%) (% total) 150

Plaques per well
CIITA 50 50 100
50

0 0 0
Cntrl CIITA Cntrl CIITA Cntrl CIITA

Fig. 2. Identification of CIITA as an Ebola restriction factor. (A) Resistance of pseudotyped with VSVg or GP from EBOV, Taï Forest virus (TAFV), Bundibugyo
CIITA-overexpressing and control (Cntrl) U2OS cells EboGP-VSV. MOI, multiplicity of virus (BDBV), Sudan virus (SUDV), Reston virus (RESTV), or Marburg virus
infection. (B and C) Plaque formation assays (B) and effective viral titer (C) for (MARV) (G). (H and I) Internalization (H) and fusion (I) of EboGP-VLPs by control
control and CIITA-overexpressing U2OS cells infected with VSVg-VSV (VSV) and and CIITA-overexpressing U2OS cells. No env, nonenveloped control VLPs.
EboGP-VSV (Ebo). undil, undiluted; PFU, plaque-forming units. (D) Representative (J to M) Infection of control and CIITA-overexpressing U2OS cells by infectious
images of CIITA-transfected (CIITA), control-transfected (Cntrl), and unmanipulated EBOV measured by imaging of GFP reporter (green) and cell nuclei (blue) (J),
U2OS cells (U2OS) infected with mCherry-expressing EboGP-VSV (red) and cell survival (K), infected cells (L), or plaque formation (M). Data are means ±
stained with Hoechst 33342 to resolve cell nuclei (blue). (E to G) Infection of SEM of three independent experiments [(A) to (I)] or experiments with
control and CIITA-expressing U2OS cells by recombinant VSV pseudotyped three independent cell clones [(K) to (M)]. Student’s t test [(A), (C), and (K) to
with EboGP, LFVGP (Lassa virus GP), or VSVg (E); single cycle murine leukemia (M)] or analysis of variance (ANOVA) with Tukey’s multiple comparison test [(E)
virus (MLV) pseudotyped with VSVg and EboGP (F); or single cycle HIV to (I)]; *P < 0.05; **P < 0.01; ND, not detected. Scale bars, 100 mm.

Bruchez et al., Science 370, 241–247 (2020) 9 October 2020 3 of 6

RESEARCH | REPORT

or deletion of the transmembrane sequence that virus-like particles (VLPs) localized proxi- the p41 CTSL binding site (22, 23) by mutation
(No TM in Fig. 4E)—or delivery to the cell mal to CD63 and the ESCRT component Hrs, completely inhibited antiviral activity (Fig. 4J
surface by fusion to a heterologous cyto- which mark MVBs (18, 19) (Fig. 4, G and H). and fig. S10). GP cleavage by endosomal pro-
plasmic and transmembrane sequence from Thus, CIITA and CD74 p41 inhibit fusion by teases facilitates the entry of other viruses,
tetherin (tetherin in Fig. 4E) almost com- arresting viral particles in MVB compartments. including coronaviruses. SARS-CoV and SARS-
pletely removed antiviral activity (Fig. 4E CoV-2 S proteins can be processed by either
and fig. S10). Thus, antiviral activity required EBOV entry requires endosomal cathepsins endosomal cathepsin B and CTSL or alterna-
delivery of the thyroglobulin domain to the (4, 10, 20) (fig. S12), which sequentially pro- tively by cell-surface serine proteases including
endosomal membrane. Electron microscopy cess EboGP (Fig. 4I and fig. S13). The CD74 TMPRSS2 (24, 25). In TMPRSS-expressing
showed that EboGP-VSV virions accumulated thyroglobulin domain inhibits cathepsins (21), cells, such as lung epithelium, inhibition of
in late endosomal multivesicular bodies (MVBs) which suggests that this may be the mecha- both cathepsins and serine proteases is re-
of CIITA- and CD74 p41–expressing cells, with nism for antiviral activity. In support of this, quired to inhibit viral entry, whereas cathep-
some virions within intraluminal vesicles (Fig. CD74 inhibited EboGP processing, similar to sin inhibitors alone block infection in cell
4F and fig. S11). Confocal microscopy confirmed the effects of the cathepsin L (CTSL) inhibitor lines—such as U2OS and Vero cells—that lack
FYDMK (Fig. 4I). Additionally, disruption of

A CD74 B Fold change in Infection**
CIITA (relative to no siRNA control)
210 4
HLA-DOA no siRNA
110 3 GAPDH
HLA-DQB2 NPC1
HLA-DPA1 2 CIITA
HLA-A
HLA-B
HLA-C
HLA-E
HLA-F
HLA-G
HLA-H
HLA-J
HLA-L

HLA-DMA
HLA-DMB
HLA-DOA
HLA-DOB
HLA-DPA1
HLA-DPB1
HLA-DPB2
HLA-DQA1
HLA-DQA2
HLA-DQB1
HLA-DQB2
HLA-DRA
HLA-DRB1
HLA-DRB5
HLA-DRB6

BTN2A2
BTN3A1
BTN3A2
BTN3A3

CD72
CD74
HAPLN4
hCG1651476
LOC554223
HCP5
IFI27
INPP5J
MYBPC2
NCOA2
PARVG
PDCD6IP
PRDM1
PSMB9
TAP1
TAPSAR1
TPP1
TRIM26
TTC28
ZNF672
-log 10 p-value 10 1 **
8 0 **

4 siRNA target
p = 0.05
80
0
60 * *
-2 -1 0 1 2 3 4 5 6 7 8
40
Change in Gene Expression 20
(log 2 fold change) 0

C Cell survival 100 CIITA: - + + + + +
CRISPR: - con CD74 (% starting cells) CRISPR: - - cntrl CD74

CIITA Infected cells
(% total )
CD74 50
HlaDRA
**
actin
0

CIITA: - + + + + +
CRISPR: - - cntrl CD74

D0 E2 F 250 NS G 300 * *

10 ** Ciita expression 10 **Cd74 expression 200 200
(relative to Hprt) (relative to Hprt)
-1 1 150 100
BLAM activity
10 10 (x103 GMFI)100
BLAM activity
-2 0 (x103 GMFI)50

10 10 0 0
-1 NS
-3 wt ko wt ko wt ko
10 CIITA CIITA CD74
10
-2
-4
10
10
-3
IFN-γ/LPS - +
10

IFN-γ/LPS - + - +
CIITA wt wt ko ko

Fig. 3. Transcriptional activity of CIITA and enhanceosome components are verified by immunoblot, and infection and survival were measured after EboGP-VSV
required for resistance. (A) Genes regulated by CIITA in U2OS cells, with challenge. Data are means ± SEM of N = 3 experiments using two independent cell
strongest induced genes identified. Mean of three independent CIITA-expressing clones. (D and E) Ciita and Cd74 expression in wild-type (wt) or Ciita−/− mouse
clones and controls. (B) EboGP-VSV infection of CIITA-expressing cells treated with BMDMs with or without priming by IFN-g and LPS. ko, knockout; NS, not
small interfering RNA (siRNA) against CIITA transcriptional targets. Data are from significant. (F and G) Fusion of EboGP-VLPs in unprimed (F) or primed (G)
two siRNAs per gene, N = 3 independent screens, and bars indicate means with mouse BMDM from Ciita−/− and Cd74−/− mice, measured as geometric mean
95% confidence intervals (CIs). One-way ANOVA with Bonferroni’s multiple fluorescence (GMFI) of cleaved CCF2. BLAM, b-lactamase. Data are means ± SEM
comparisons; *P < 0.05; **P < 0.01. Dotted lines indicate 99% CIs from no siRNA for independent cultures from three mice per group. Student’s t test; *P < 0.05;
control. (C) CD74 CRISPR-targeting in CIITA-overexpressing U2OS cells was **P < 0.01. Similar results were observed in three independent experiments.

Bruchez et al., Science 370, 241–247 (2020) 9 October 2020 4 of 6

RESEARCH | REPORT

A B 100 ** C 100 ** D *

Thyro Cell Survival (%) 100 10 * Cell Survival (%) 100
1 100
10 50 0.1 **
Infected Cells 0.01Infected Cells **
(% total) 0 (% total) Fusion Activity
CLIP 1 * (% of control) 50
0.1
acidic 0.01 50
ER
00

p33 cpn3tr3l cppppn3344tr1533l cppppn3344tr3513l cppppnt3344r3513l cpn3tr3l
p41 p41 p41
p35 p35
p43 p43

E CytoTM Thyro F Uninfected EboGP-VSV

p33 control control control p33
CIITA
p41 ** CIITA p41
**
CT del **
Short **
**
Long
Furin CIITA
Tetherin
No TM

none

0 20 40 60 80

Infected cells (% total)

G EBOV-VLP CD63 DNA H #VLP per cell in CD63 I Therm - + -- ---- -- J Infected cells **
- - - EF - - - - (% total)
Control 25 ** Cat. inhib - - - - - - ++ ++ 80
EBOV-VLP Hrs DNA CIITA cDNA - - - - - - - + ++ 60
20 CD74 Crispr - - - - - - - - p41p33 40
CD74 cDNA - 20
15 EboGP-VSV 0

p41 10 ** Molecular Weight 51
(kDa)
p41 5 NS 39

** 0 28 none
CD74 wt wt ko ko 19 p41
cntl p33 p41 CIITA - + + + PG-RR
PG-DD
EboGP p41 - - - + PG-TT

Control

K 107 105 L Undil Effective viral titer 6
104 (PFU/mL) 10
106 103 -1
105Effective viral titer 102 10SARS-CoV-2 5
(FFU/mL) Viral diltuion 10
-2
10 4
10
-3
** 10 3
** 10
-4
cntl p33 p41 cntl p33 p41 10 2
10
SARS-S WIV1-S -5
10 1 **
10
CD74 cDNA: cntrl p33 p41 0 LOD
104 101 10
CD74 cDNA: cntrl p33 p41
CD74 cDNA: cntrl p33 p41
VSV-G
SARS-CoV-2

Fig. 4. CD74 p41 inhibits cathepsin-mediated cleavage of EboGP. endosomes in U2OS cells expressing CIITA and CD74 as indicated. Each point
(A) Human CD74 isoforms with ER retention signal (ER), CLIP, acidic, and represents a single cell, mean ± SD n ≥ 9. Mann-Whitney U test; **P < 0.01. Similar
p41 thyroglobulin (Thyro) domains. (B and C) EboGP-VSV infection and results were seen in three independent experiments. (I) Immunoblot of EboGP in
survival of Cd74−/− CIITA-expressing (B) or wt (C) U2OS cells expressing EboGP-VSV–infected U2OS cells. EboGP-VSV preparation ± thermolysin (Therm) is
CD74 isoforms. (D) EboGP-VLP fusion in THP-1 macrophage-like cells expressing shown for reference (left). Cells were treated with cathepsin inhibitors (Cat. inhib)
CD74 p33 and p41. (E) EboGP-VSV infection of U2OS cells expressing CD74 E64D (E) or FYDMK (F), or expressed CIITA and CD74. EboGP in virus particles
mutant constructs. Cyto, cytoplasmic domain; TM, transmembrane domain; (arrow), after proteolysis (closed arrowhead), and after partial cleavage (open
Thyro, thyroglobulin domain; CT del, carboxy-terminus deletion; No TM, deletion arrowhead) are indicated. (J) EboGP-VSV infection of U2OS cells expressing p41
of the transmembrane sequence. (F) Transmission electron micrographs of with CTSL binding site mutations. (K) Infection of control, p33-, or p41-expressing
control, CIITA-expressing, and CD74-expressing U2OS cells 3 hours after U2OS cells by HIV-GFP pseudotyped with GPs from VSV, EBOV, SARS-CoV,
infection with EboGP-VSV. Dotted-line regions are enlarged in adjacent panels or WIV1-CoV, measured as focus-forming units per milliliter of virus (FFU/ml).
(as indicated by white arrows). Intraluminal vesicles (black arrowheads) and (L) Infection of control, p33-, or p41-expressing Vero cells by SARS-CoV-2,
internalized EboGP-VSV (black arrows) are marked. Scale bars, 1 mm (left, center showing representative crystal violet-stained monolayers and infection measured
left, and right panels) and 200 nm (center right panels). (G) Confocal microscopy of as plaque-forming units per milliliter of virus (PFU/ml). Except where indicated,
control and p41-expressing U2OS cells showing EBOV-VLP (red), CD63, or Hrs data are means ± SEM of data from ≥3 independent experiments. Student’s t test
(green), and nuclei (white). Scale bars, 10 mm. (H) VLPs associated with CD63 with Benjamini correction; *P < 0.05; **P < 0.01.

Bruchez et al., Science 370, 241–247 (2020) 9 October 2020 5 of 6

RESEARCH | REPORT

TMPRSS2 (25). p41 inhibited the entry of 4. M. Côté et al., Nature 477, 344–348 (2011). at NIAID Integrated Research Facility was funded by contract
viruses pseudotyped with S proteins from 5. J. E. Carette et al., Nature 477, 340–343 (2011). no. HHSN272200700016I to Battelle Memorial Institute (BMI).
SARS-CoV and a related bat virus, WIV1-CoV, 6. J. Barretina et al., Nature 483, 603–607 (2012). J.J. performed this work as an employee of BMI. SARS-CoV-2
into U2OS cells, which demonstrates that p41 7. H. Okamoto, K. Asamitsu, H. Nishimura, N. Kamatani, work was performed in the BSL3 at Case Western Reserve
inhibited S protein processing (Fig. 4K). To University (CWRU), which is supported by the CWRU and
determine whether p41 exhibited antiviral T. Okamoto, Biochem. Biophys. Res. Commun. 279, 494–499 University Hospitals Center for AIDS research grant P30AI36219.
activity against authentic SARS coronavirus, (2000). Author contributions: A.B. performed most of the experiments.
we challenged p41-expressing Vero E6 cells 8. G. Tosi et al., Proc. Natl. Acad. Sci. U.S.A. 103, 12861–12866 Screen and data analysis tools were developed by K.S. BSL4
with SARS-CoV-2. CD74 p41 expression com- (2006). experiments were performed by J.J. and G.G.O., minigenome
pletely inhibited plaque formation, which dem- 9. C. J. Shoemaker et al., PLOS ONE 8, e56265 (2013). experiments were performed by A.J.H. and E.M., and K.A.M.
onstrates that this antiviral activity extended 10. J. S. Spence, T. B. Krause, E. Mittler, R. K. Jangra, K. Chandran, designed all CD74 and CIITA mutations. H.M., R.P., and
beyond filoviruses (Fig. 4L). mBio 7, e01857-15 (2016). L.G. provided technical assistance. C.S. assisted with data
11. J. A. Harton, M. W. Linhoff, J. Zhang, J. P. Y. Ting, J. Immunol. analysis and visualization. G.G.O. and E.M. provided assistance
Here, we identify the antiviral activity of 169, 4088–4093 (2002). with experimental planning and data interpretation. L.C., E.V.S.,
CIITA and CD74. We show that CIITA in- 12. R. S. Accolla et al., J. Exp. Med. 164, 369–374 (1986). L.M.S., and A.L.-H. conceived the study. The manuscript was
duces resistance by up-regulation of the p41 13. X. S. Zhu et al., Mol. Cell. Biol. 20, 6051–6061 (2000). written by A.L.-H. and L.M.S. with assistance from A.B. and K.S.
isoform of CD74, which blocks cathepsin- 14. D. Wong et al., Genome Biol. 15, 494 (2014). Competing interests: E.V.S. is presently an employee of Merck
mediated cleavage of viral GPs, thereby pre- 15. T. W. Geisbert et al., Am. J. Pathol. 163, 2347–2370 and Co., Inc., Kenilworth, NJ, and holds stock in Merck and Co. This
venting viral fusion. This antiviral activity (2003). work was conducted before E.V.S.’s affiliation with Merck. The
protects against a wide range of cathepsin- 16. M. Bray, T. W. Geisbert, Int. J. Biochem. Cell Biol. 37, authors declare no other competing interests. Data and materials
dependent viruses, including filoviruses and 1560–1566 (2005). availability: Full analysis of screen results is presented in the
coronaviruses; functions in macrophages and 17. M. Strubin, C. Berte, B. Mach, EMBO J. 5, 3483–3488 supplementary materials. DNA and RNA sequencing data are
DCs that are early targets of infection (15, 16); (1986). deposited at Gene Expression Omnibus (under accession
and is activated by IFN-g. We demonstrate 18. T. Kobayashi et al., Mol. Biol. Cell 11, 1829–1843 (2000). nos. GSE156598 and GSE155204, respectively). The PB transposon
that CIITA and CD74 mediate the endosomal 19. K. G. Bache, A. Brech, A. Mehlum, H. Stenmark, J. Cell Biol. was obtained under a material transfer agreement with the
sequestration of certain viruses as a mecha- 162, 435–442 (2003). Wellcome Trust Sanger Institute. All other data are available in
nism of cellular host defense. We speculate 20. E. H. Miller et al., EMBO J. 31, 1947–1960 (2012). the manuscript or the supplementary materials. This work is
that this activity is evolutionarily ancient and 21. M. Mihelič, A. Dobersek, G. Guncar, D. Turk, J. Biol. Chem. 283, licensed under a Creative Commons Attribution 4.0 International
precedes their better-known role in antigen 14453–14460 (2008). (CC BY 4.0) license, which permits unrestricted use, distribution,
processing. We anticipate that the applica- 22. G. Gunčar, G. Pungerčič, I. Klemenčič, V. Turk, D. Turk, EMBO and reproduction in any medium, provided the original work is
tion of this transposon screening approach to J. 18, 793–803 (1999). properly cited. To view a copy of this license, visit https://
other models of infection will reveal addi- 23. A. M. Lennon-Duménil et al., EMBO J. 20, 4055–4064 creativecommons.org/licenses/by/4.0/. This license does not
tional mechanisms that have eluded conven- (2001). apply to figures/photos/artwork or other content included in the
tional screening strategies. article that is credited to a third party; obtain authorization from
24. G. Simmons et al., Proc. Natl. Acad. Sci. U.S.A. 102, the rights holder before using such material.
REFERENCES AND NOTES 11876–11881 (2005).
1. S. K. Gire et al., Science 345, 1369–1372 (2014). SUPPLEMENTARY MATERIALS
2. L. Chen et al., BMC Cancer 13, 93 (2013). 25. M. Hoffmann et al., Cell 181, 271–280.e8 (2020).
3. T. L. Bergemann et al., Nucleic Acids Res. 40, 3822–3833 (2012). science.sciencemag.org/content/370/6513/241/suppl/DC1
ACKNOWLEDGMENTS Materials and Methods
Figs. S1 to S14
We thank M. Mason, M. Rosasco, S. Presnell, and the Tables S1 to S5
Bioinformatics Department at Benaroya Research Institute (BRI) References (26–43)
for support in data analysis and V. Gersuk and the BRI genomics
core for sequencing. We thank B. Schneider and S. MacFarlane View/request a protocol for this paper from Bio-protocol.
from the Electron Microscopy Resource at Fred Hutch for help with
transmission electron microscopy experiments and L. Eisenlohr 21 February 2020; accepted 20 August 2020
and M. O’Mara at Children’s Hospital of Philadelphia for providing Published online 27 August 2020
Cd74-knockout mouse bone marrow. Funding: This work was 10.1126/science.abb3753
supported by National Institutes of Health grants R33AI102266,
U01AI070330, and R33AI119341 (to A.L.-H. and L.M.S.);
U19AI125378-04S1 (to A.L.-H.); and R21AI135912 (to E.M.). Work

Bruchez et al., Science 370, 241–247 (2020) 9 October 2020 6 of 6

RESEARCH

NEUROSCIENCE rat running backward rather than turning
around and running forward in the opposite
Alternating sequences of future and past behavior direction. The direction-momentum relation-
encoded within hippocampal theta oscillations ship in reverse theta sequences thus bears a
similarity to reverse replay observed within
Mengni Wang1, David J. Foster2, Brad E. Pfeiffer1* rest-based sharp-wave ripples (1, 2).

Neural networks display the ability to transform forward-ordered activity patterns into reverse-ordered, Given that the reverse window was confined
retrospective sequences. The mechanisms underlying this transformation remain unknown. We discovered to phases of the theta oscillation associated
that, during active navigation, rat hippocampal CA1 place cell ensembles are inherently organized to produce with minimal firing across the hippocampal
independent forward- and reverse-ordered sequences within individual theta oscillations. This finding may network, we investigated whether the back-
provide a circuit-level basis for retrospective evaluation and storage during ongoing behavior. Theta phase ward portion of the theta sequence was en-
procession arose in a minority of place cells, many of which displayed two preferred firing phases in theta coded by a different population of neurons
oscillations and preferentially participated in reverse replay during subsequent rest. These findings reveal than the forward component—a population
an unexpected aspect of theta-based hippocampal encoding and provide a biological mechanism to that might selectively fire while the remain-
support the expression of reverse-ordered sequences. der is relatively silent. Although most neurons
(which we termed “unimodal cells”) displayed
E xperience necessarily occurs in a sequen- both the linear track and open field tasks typi- a canonical (6) unimodal relationship between
tial manner, and hippocampal function is cally produced a virtual spatial path progress- firing rate and theta phase, a subset of neurons
critical for representation and storage of ing from the rat’s current location ahead of (“bimodal cells”) displayed a bimodal relation-
sequential information. Adaptive behav- the animal during periods of active movement ship with theta phase (Fig. 2, A and B, and
ior requires the ability to analyze experi- (Fig. 1A and figs. S1 and S2). However, theta fig. S9), consistent with the firing patterns
ence, both prospectively and retrospectively. It oscillations rarely progressed uniformly in a reported for deep CA1 pyramidal neurons (14).
remains unclear how forward-ordered neural single direction. They consisted of two distinct The activity of bimodal cells across the theta
activity can facilitate storage or expression of components, one that traveled ahead of the oscillation was separated into two windows: a
reverse-ordered sequences, which are observed rat and a second that moved backward in the major peak between phases 200° and 430°/70°,
in ripple-based reverse replay (1, 2) and may reverse direction of the animal’s actual move- comparable to the forward window, and a mi-
underlie human episodic memory retrieval (3). ment (Fig. 1, A and B, and figs. S1 and S2). The nor peak between 80° and 190°, consistent with
Accurate, precise, and stable spatial represen- reverse component consistently occurred near the reverse window (Fig. 2B). Spatial represen-
tation across the hippocampal network of place the peak of theta oscillation (Fig. 1B), a time win- tation during the reverse window, but not the
cells is insufficient to support goal-directed spatial dow associated with minimal hippocampal forward window, was more strongly influenced
navigation in the absence of theta oscillations population activity (6). To ensure that the ob- by bimodal cell activity as opposed to unimodal
(4, 5). This indicates that the precise, population- servation of a backward theta sequence was activity (fig. S10). Bimodal cells were more
level patterns of activity organized by theta not a trivial result of low spike counts biasing likely to fire in ripples than unimodal cells
oscillations, termed theta sequences (6–9), are the decoding algorithm, we initially restricted (Fig. 2C) and further displayed an increased
critical for hippocampal-dependent, memory- our analysis to the third of theta oscillations in likelihood to participate in reverse replay as
guided behavior (10). We examined ensemble each session with the highest firing rates within opposed to forward replay (Fig. 2D), indicating
place cell activity recorded from bilateral hip- that window. On the basis of troughs in pos- a neuronal link between reverse theta sequen-
pocampal area CA1 in rats engaged in reward- terior probability distributions (Fig. 1B), we ces and reverse replay. Bimodal cells had sim-
seeking exploration of both linear tracks and defined two windows in the theta oscillation: ilar cluster quality measurements to those of
open arenas (11, 12). Large numbers of simul- a forward window from phases 250° to 420°/ unimodal cells (figs. S11 and S12) and were
taneously monitored place cells (per-session 60° and a reverse window from 80° to 230°, as- thus unlikely to be a trivial consequence of
mean ± SEM = 144.3 ± 14.0 putative excitatory sociated with statistically significant forward and poor single-unit isolation. Similar to deep CA1
units) allowed a memoryless, uniform-prior backward virtual trajectories, respectively (Fig. 1C pyramidal neurons (15), bimodal cells were
Bayesian decoding algorithm to accurately ex- and figs. S3 and S4). Virtual representations of more likely to shift their preferred phase of
tract the spatial information encoded by the forward or reverse trajectories were observed firing within theta during REM (rapid eye
recorded ensemble in each session (table S1). within single theta oscillations (fig. S5) and movement) sleep (fig. S13). Furthermore, like
We investigated the temporally compressed were directly and independently correlated with deep CA1 neurons (16), bimodal cells displayed
spatial trajectories expressed within theta os- population activity in the forward or reverse increased spatial information (fig. S12).
cillations during active exploration (speed windows, respectively (fig. S6). The reverse com-
≥10 cm/s), examining a total of 115,845 theta ponent of the theta sequence was not a trivial We next sought to identify the cellular mech-
oscillations (7240.3 ± 1126.7 per session). consequence of our analysis methods or reset- anism that produces the reverse component of
ting of the forward trajectory (figs. S7 and S8). theta sequences. The predominant model ac-
We quantified sequential spatial representa- counting for the well-studied forward compo-
tion within theta sequences by aligning each Taking advantage of the fact that place cells nent of theta sequences is theta phase precession
decoded time frame to the rat’s current move- display direction-selective firing patterns on (6, 7), the iterative firing of a place cell at pro-
ment direction every 5 ms. Theta sequences in linear tracks (13), we observed that both the gressively earlier phases of the theta oscillation
forward and reverse components were pre- as the rat passes through the neuron’s firing
1Neuroscience Graduate Program, Department of dominantly encoded by neurons representing field (17). We reasoned that the backward
Neuroscience, Peter O’Donnell Jr. Brain Institute, University the rat’s current movement direction (Fig. 1, D trajectory encoded within theta sequences may
of Texas Southwestern Medical Center, Dallas, TX 75390, and E, and fig. S2). Thus, the reverse compo- be explained by the opposite process: phase
USA. 2Department of Psychology and Helen Wills nent of a theta sequence bears a direction- procession within individual place cells during
Neuroscience Institute, University of California, Berkeley, CA momentum relationship consistent with the the reverse window. When we examined the
94720, USA. relationship of theta firing phase to the rat’s
*Corresponding author. Email: [email protected] location within each cell’s place field, we iden-
tified clear phase precession during the major

Wang et al., Science 370, 247–250 (2020) 9 October 2020 1 of 4

RESEARCH | REPORT

A LFP D Linear Track, UP Runs

60Relative 0 0.040 0 0.25 60 UP Fields DOWN Fields
Loc. (cm)
0 Relative 0.048
Loc. (cm)
-60 0
0 Time (s)
0.46 0 0.16 0 0.13 -60 0.054 0
B All Open Field
0.036 All Linear Track
60
Linear Track, DOWN Runs
Relative
Loc. (cm) UP Fields DOWN Fields

0 60
Relative
0 Loc. (cm) 0
-60
Max 3.6 3.8
Prob
(× 10-2) -60 0
0
1.8 2.2 360 720 0 360 720
Theta Theta Phase (°)

0 360 720 0 360 720 E Forward Window

Theta Phase (°) Reverse Window
Reverse Forward
Window Window UP Runs DOWN Runs UP Runs DOWN Runs

C Open Field Linear Track Weighted Correlation 0.5 0.4 0.1 0.1

Forward Window Forward Window

0.12 p < 0.002 0.09 p < 0.002 0
0
Shuffle Proportion
0 0 0.2 0 0 0.25
-0.1 Reverse Window -0.1 Reverse Window

0.12 p < 0.002 0.09 p < 0.002

0 0 0
-0.1 0 -0.4 -0.35
-0.15 0 0.1 -0.25 0 0.1
Decoded With UP Fields Decoded With DOWN Fields
Location-Phase Weighted Correlation

Actual Data Cell ID Shuffle Phase Shuffle

Fig. 1. Forward and reverse components of theta sequences. (A) Examples windows. (Bottom) Idealized theta oscillation. (C) Actual data and distribution
of theta sequences displaying both reverse and forward components. (Top) of weighted correlation values for 500 theta phase or cell ID shuffles per
Raw (gray) and theta-filtered (black) local field potential (LFP) trace. Loc., theta oscillation in forward and reverse windows. The maximum Monte-Carlo
location. (Bottom) Decoded spatial representation probability relative to the rat’s P value for either shuffle displayed is noted. (D) As in the top portion of (B),
current location (blue line) and movement direction (positive values). Dashed but for only linear track sessions separated by runs across the track in the UP
lines mark the 70° theta phase. (B) (Top) Across-session probability histogram (top) or DOWN (bottom) direction, decoded with place fields calculated during up
of theta phase (10° bins) with maximum posterior probability at each position (left) or down (right) runs. (E) Per-session box-and-whisker plot (box, quartile;
relative to rat position (2-cm bins) for the third of theta oscillations with highest whiskers, extreme range), mean (red solid line), and median (red dashed line)
firing near the peak of theta phase. Dashed blue lines indicate best fit (least- of weighted correlations per session for forward and reverse windows during
squares method). (Middle) Smoothed (Gaussian, s = 10°) maximum value up or down runs decoded with up or down fields. Dashed lines represent
of each theta bin above. Vertical dashed lines indicate troughs in maximum individual session averages. *P < 0.05, **P < 0.01, ***P < 0.001; Wilcoxon
probability (Max Prob) density (70° and 240°) used to define forward and reverse rank-sum test; n = 10 sessions.

A Unimodal Cells B Unimodal Bimodal C D
80 100
80 0.5 0.2 0.09
Firing Rate 0.03
Spike Index
Count
Ripples/Run
000 Spike Ratio
Unimodal 1041
Bimodal 557
Replay/Run
Unimodal 569
Bimodal 275
Replay/Run
For. Replay 275
Rev. Replay 275
Bimodal Cells 0.1
0 360 720
100 120 120 Theta Phase (°)

Spike 0 00
Count
Minor Major
0 0 0 Peak Peak

0 360 720 0 360 720 0 360 720
Theta Phase (°) Theta Phase (°) Theta Phase (°)

Fig. 2. Unimodal and bimodal cells. (A) Raw (black) and smoothed (red) spikes emitted during runs (velocity ≥10 cm/s) for unimodal or bimodal
histograms of action potential count per theta phase for example unimodal cells across all sessions. (D) (Left) Mean ± SEM ratio of spikes in significant
(top) and bimodal (bottom) cells. (B) Mean ± SEM firing rate index versus theta replay events to spikes during runs for unimodal or bimodal cells across all
phase for all unimodal (red; n = 1041) and bimodal (blue; n = 557) cells across linear track sessions. (Right) Mean ± SEM ratio of spikes in forward (solid) or
all open field and linear track sessions. Vertical lines mark troughs of mean reverse (checkered) replay to spikes during runs for bimodal cells across all
firing for bimodal cells used to separate theta oscillations into major and minor linear track sessions. *P < 0.05, ***P < 0.001; Student’s t test; cell number (n)
peak windows. (C) Mean ± SEM ratio of spikes emitted across all ripples to is listed on the bar graphs.

Wang et al., Science 370, 247–250 (2020) 9 October 2020 2 of 4

RESEARCH | REPORT

A Unimodal Cells Bimodal Cells B All Unimodal Cells All Bimodal Cells

100 180 60 120 3.6x10 -4 3.0x10 -3 3.2x10 -4 2.6x10 -3

Spike 720
Count
0 Theta Phase (°)
0 0 0 0 720 360
720 0 720 0
720 720 0

Theta Phase (°)

Theta Phase ( °) 0 1.7x10 -3 3.2x10 -4 2.2x10 -3

3.6x10 -4

190

0 0.004 80
720

-1 0 1 -1 0 1

Normalized Position in Place Field

0 0
-1 0
1 -1 0 1 -1 0 1 -1 0 1 C Unimodal Cells Bimodal Cells

Normalized Position in Place Field 0.25 Major Peak Window 0.18 Major Peak Window
p < 0.001 p < 0.001

Fig. 3. Phase precession and procession. (A) Example unimodal and bimodal Shuffle Proportion 00
cells displaying phase precession and/or phase procession. (Top) Spike count per
theta phase. (Bottom) Spike plot and smoothed (Gaussian, s = 2 bins) probability -0.15 0 0.05 -0.15 0 0.05
heatmap of theta phase versus normalized location (−1 is entering the field, 1 is
leaving). (B) (Top) Smoothed firing probability per theta phase bin (10°) and 0.18 Minor Peak Window Minor Peak Window
normalized position in place field bin (0.1), averaged across all cells for open field p < 0.001 0.14 p < 0.001
and linear track sessions. Dotted lines mark major and minor peak boundaries.
(Bottom) As in top panel, but showing only the minor peak window with a rescaled 0 0 0.1
colormap. (C) Actual data and distribution of weighted correlation values for -0.05 0 0.1 -0.05 0
1000 shuffles of position or theta phase in major peak and minor peak windows.
The maximum Monte-Carlo P value for either shuffle displayed is noted. Phase-Position Weighted Correlation

Actual Data Position Shuffle Phase Shuffle

peak of activity for both unimodal and bi- itively correlated with the power of the hip- forward, prospective sequence, whereas anti-
modal cells (Fig. 3, A and B, and fig. S14). pocampal theta oscillation (Fig. 4 and fig. S16). phase EC3 input selectively facilitates phase
However, during the minor peak of activity Activity at the minor peak, but not the major procession and drives the reverse, retrospec-
associated with the reverse window, many peak, was correlated with the power of the tive sequence. Consistent with this interpreta-
bimodal cells displayed significant phase pro- beta oscillation (Fig. 4 and fig. S16). In ad- tion, theta oscillations with significant reverse
cession (Fig. 3C and figs. S14 and S15). The dition, beta power was increased in theta os- sequences have increased power in the fast
average phase-versus-location plots for bimodal cillations expressing reverse theta sequences and medium gamma bands (fig. S17), which is
cells contained two distinct clusters rather than (fig. S17). Although some prior studies suggest associated with EC input to CA1 (14, 25), but
the single linear or curved relationship ob- that the hippocampus may receive both theta- not the slow gamma band, which is associated
served for unimodal neurons (Fig. 3B), as in- frequency and beta-frequency inputs (19–21), with CA3 input (25).
dicated in earlier work (6, 18). Although their several others indicate that two independent
total firing rates were lower in the minor peak theta-frequency inputs that are roughly anti- Although several models have been pro-
window, unimodal cells also displayed phase phase to one another drive activity in area CA1 posed to explain the observation of reverse
procession within this period (Fig. 3, B and C, (14, 22–24), which can produce beta-frequency replay (1, 26, 27), it remains unclear how
and figs. S14 and S15). Thus, phase procession oscillatory activity. Our data are consistent reverse-ordered sequences can arise from
during the reverse window is not a character- with the second model (fig. S18) and support forward-ordered activity. Given that activity
istic aspect of bimodal cells but is likely a com- the hypothesis that activity in the forward and within the reverse window is linked to parti-
mon feature of hippocampal activity during reverse windows is driven by two independent, cipation in reverse replay events (Fig. 2D),
this time frame. antiphase theta-frequency inputs. our data suggest that synaptic inputs arriv-
ing during this window, likely originating
Finally, we sought to identify whether the CA1 pyramidal neurons receive two primary, from EC3, facilitate synaptic changes across
forward and reverse theta sequences were anatomically segregated excitatory inputs: the hippocampal network that underlie the
driven by a common source or were instead the Shaffer collaterals originating in hippo- expression of reverse replay. CA1 synapses re-
regulated by independent inputs. Because for- campal area CA3 and the perforant path from ceiving input from area CA3 have synaptic and
ward and reverse sequences were respectively layer 3 of the entorhinal cortex (EC3). The plastic properties that are distinct from those
correlated with population activity during the timing of these respective inputs is also seg- of synapses receiving input from EC3 (28, 29).
forward and reverse windows (fig. S6), we rea- regated, with EC3 input arriving near the peak The relative timing of EC3 input also affects
soned that if a common input accounted for of CA1 theta oscillation (during the reverse win- plasticity at Shaffer collateral synapses (30, 31).
both the forward and reverse components, pop- dow) and CA3 input arriving nearer the trough The increased likelihood of observing reverse
ulation activity in both windows would be of CA1 theta oscillation (during the forward replay immediately after a behavioral trajec-
strongly correlated. If forward and reverse window) (14, 22). Thus, the relative timing and tory (2) and the lower level of reverse replay
theta sequences were instead driven by two independent expression that we observe in the reported during sleep (32) suggest that plas-
distinct inputs, activity in these two windows forward and reverse components of theta se- ticity generated in the reverse window may be
should be independent. Neural activity within quences indicate that CA3 input selectively stronger but more temporary than plasticity
the major peak, but not the minor peak, pos- facilitates phase precession and drives the generated in the forward window. The power

Wang et al., Science 370, 247–250 (2020) 9 October 2020 3 of 4

RESEARCH | REPORT

Fig. 4. Independent A Major Peak r = 0.084; p = 0.32 14. A. Fernández-Ruiz et al., Neuron 93, 1213–1226.e5 (2017).
activity in forward and 0.3 15. K. Mizuseki, K. Diba, E. Pastalkova, G. Buzsáki, Nat. Neurosci.
reverse components r = 0.78; p < 10-30
of theta sequences. For 14, 1174–1181 (2011).
all open field and linear 0.35 16. N. B. Danielson et al., Neuron 91, 652–665 (2016).
track sessions, firing rate 17. J. O’Keefe, M. L. Recce, Hippocampus 3, 317–330
index (FRI) in the major FRI FRI0.05 0.1
peak window (A) or minor (1993).
peak window (B) as a -1.5 2.5 -2 4 18. Y. Yamaguchi, Y. Aota, B. L. McNaughton, P. Lipa, J.
function of theta (6 to Theta Power (Z-score) Beta Power (Z-score)
12 Hz) power (left) or Neurophysiol. 87, 2629–2642 (2002).
beta (15 to 20 Hz) power B Minor Peak r = 0.38; p < 10-5 19. C. S. Lansink et al., J. Neurosci. 36, 10598–10610
(right), separated into 0.18
10 divisions per session. r = -0.047; p = 0.57 (2016).
Colored lines indicate 20. J. D. Berke, V. Hetrick, J. Breck, R. W. Greene, Hippocampus
data and best fit for each 0.18
session; black lines indicate 18, 519–529 (2008).
best fit across all sessions. 00 21. L. M. Rangel, A. A. Chiba, L. K. Quinn, Front. Syst. Neurosci. 9,
Statistics are based on -1.5 2.5 -2 4
repeated measures of 96 (2015).
correlation; n = 16 sessions. Theta Power (Z-score) Beta Power (Z-score) 22. E. W. Schomburg et al., Neuron 84, 470–485 (2014).
23. M. Valero, L. M. de la Prida, Curr. Opin. Neurobiol. 52, 107–114
of a beta-frequency oscillation, which we show spond to the forward window (9). Although
to be correlated with the reverse theta sequence these findings align with our observations, a (2018).
(Fig. 4 and figs. S16 and S17), is increased notable contrast is that here we have spe- 24. A. Navas-Olive et al., Nat. Commun. 11, 2217 (2020).
during early exploration of novel environments 25. L. L. Colgin et al., Nature 462, 353–357 (2009).
(20) and may represent enhanced EC3 input, cifically identified two distinct and inde- 26. R. A. Koene, M. E. Hasselmo, Neural Netw. 21, 276–288
which facilitates the rapid expression of reverse pendent information streams within each
replay (1). (2008).
theta cycle: one that represents a possible 27. T. Haga, T. Fukai, eLife 7, e34171 (2018).
Although most studies of theta sequences future outcome and one that represents prior 28. E. Arrigoni, R. W. Greene, Br. J. Pharmacol. 142, 317–322
have reported exclusively forward-ordered tra-
jectories (6, 7), a small number have suggested behavior in the reverse order. Our findings (2004).
the presence of retrospective information en- therefore facilitate understanding and inte- 29. A. Aksoy-Aksel, D. Manahan-Vaughan, Front. Synaptic Neurosci.
coded within some hippocampal theta oscil-
lations (8, 32–34). Most prior studies lacked gration of prior work into a broad, cohesive 5, 5 (2013).
the cell yield necessary to accurately quantify model of hippocampal circuit function, de- 30. H. Takahashi, J. C. Magee, Neuron 62, 102–111
spatial sequences on a fine time scale. They
were thus unable to observe the consecutive, monstrating alternating prospective and (2009).
phase-locked forward and reverse sequences retrospective evaluation of behavior within 31. J. Basu et al., Science 351, aaa5694 (2016).
we report here, particularly during the reverse 32. A. M. Wikenheiser, A. D. Redish, Hippocampus 23, 22–29
window when population activity is low. Bi- individual theta oscillations.
modal firing rates and theta phase–versus– (2013).
location relationships of individual neurons REFERENCES AND NOTES 33. K. W. Bieri, K. N. Bobbitt, L. L. Colgin, Neuron 82, 670–681
have been previously reported (6, 14, 18). How-
ever, these earlier studies did not predict se- 1. D. J. Foster, M. A. Wilson, Nature 440, 680–683 (2006). (2014).
quential representation of prior experience 2. K. Diba, G. Buzsáki, Nat. Neurosci. 10, 1241–1242 (2007). 34. A. Cei, G. Girardeau, C. Drieu, K. E. Kanbi, M. Zugaro,
within individual theta oscillations. Finally, 3. G. E. Wimmer, Y. Liu, N. Vehar, T. E. J. Behrens, R. J. Dolan,
a recent study reported that the forward com- Nat. Neurosci. 17, 719–724 (2014).
ponent of successive theta sequences alter- Nat. Neurosci. 23, 1025–1033 (2020). 35. M. Wang, D. Foster, B. Pfeiffer, Data from “Alternating
nates between possible future paths or heading 4. D. Robbe, G. Buzsáki, J. Neurosci. 29, 12597–12605 (2009).
directions during phases of theta that corre- 5. K. A. Bolding, J. Ferbinteanu, S. E. Fox, R. U. Muller, sequences of future and past behavior encoded within
hippocampal theta oscillations,” Zenodo (2020); doi:10.5281/
Hippocampus 30, 175–191 (2020). zenodo.3972156.
6. W. E. Skaggs, B. L. McNaughton, M. A. Wilson, C. A. Barnes,
ACKNOWLEDGMENTS
Hippocampus 6, 149–172 (1996).
7. G. Dragoi, G. Buzsáki, Neuron 50, 145–157 (2006). Funding: This work was supported by NIH R01-MH085823,
8. A. S. Gupta, M. A. van der Meer, D. S. Touretzky, A. D. Redish, NIH R01-NS104829, the Alfred P. Sloan Foundation, and the
Southwestern Medical Foundation. Author contributions:
Nat. Neurosci. 15, 1032–1039 (2012). B.E.P. and D.J.F. conceived the original experiments. B.E.P.
9. K. Kay et al., Cell 180, 552–567.e25 (2020). performed the original experiments and collected the original
10. C. Drieu, R. Todorova, M. Zugaro, Science 362, 675–679 (2018). data. M.W. and B.E.P. conceived, designed, and performed
11. B. E. Pfeiffer, D. J. Foster, Nature 497, 74–79 (2013). the analyses. M.W. and B.E.P. wrote the manuscript.
12. B. E. Pfeiffer, D. J. Foster, Science 349, 180–183 (2015). Competing interests: None declared. Data and materials
13. B. L. McNaughton, C. A. Barnes, J. O’Keefe, Exp. Brain Res. 52, availability: Data analysis code (MATLAB) is available on
GitHub (github.com/Brad-E-Pfeiffer/ThetaForwardReverseCode)
41–49 (1983). and Zenodo (35). All (other) data needed to evaluate the
conclusions in the paper are present in the paper or the
supplementary materials.

SUPPLEMENTARY MATERIALS

science.sciencemag.org/content/370/6513/247/suppl/DC1
Materials and Methods
Figs. S1 to S18
Table S1
References (36–38)
MDAR Reproducibility Checklist

View/request a protocol for this paper from Bio-protocol.

21 February 2020; accepted 21 August 2020
10.1126/science.abb4151

Wang et al., Science 370, 247–250 (2020) 9 October 2020 4 of 4

LIFE SCIENCE TECHNOLOGIES Produced by the Science/AAAS Custom Publishing Office

new products: genetics

Adeno-Associated Virus Titer Kit collection, nucleic acid preservation, and inactivation of pathogens).
Gyros Protein Technologies The kit also features a buffer system that facilitates complete viral-
introduces the Gyrolab AAVX Titer particle lysis for efficient nucleic acid isolation. Small (>50 nucleotide)
Kit for adeno-associated virus (AAV) and large (>200 kilobase) DNA and RNA are bound to magnetic
vector titer determination in AAV beads, washed, and eluted. The isolated high-quality nucleic acids
vector manufacturing. The kit for are ready for all downstream applications, such as next-generation
physical titer determination, one of sequencing, hybridization-based PCR detection, and quantitative
the critical quality attributes of the reverse-transcription PCR detection.
U.S. Food and Drug Administration Zymo Research
process validation guidelines, has For info: 888-882-9682
a broad applicability across numerous AAV serotypes and adds to www.zymoresearch.com/products/quick-dna-rna-viral-magbead
our wide range of ready-to-use kits that are already employed by
scientists and bioengineers in the rapidly growing cell and gene Automated Gene Synthesis
therapy market, where AAV vectors are commonly used. The kit Codex DNA introduces the BioXp 3250 system, the next-generation,
supports scientists in the development and production of cell and automated synthetic biology workstation for rapidly building DNA
gene therapies to maximize performance and productivity, and clones, variant libraries, gene fragments, and genomes.
reduces the time needed to produce patient treatments. With Unlike traditional approaches that can take weeks or months to
compressed development timelines for these therapies, the new kit generate gene fragments, clones, or libraries, BioXp 3250 achieves
generates results 4 times faster and requires 10 times less sample these workflows in a single, overnight run. With increased gene-
compared to traditional ELISA methods, accelerating AAV vector fragment lengths of up to 7,000 base pairs and an extensive selection
workflows and delivering high-quality data from small sample of off-the-shelf vectors, BioXp 3250 offers scientists greater flexibility
volumes. Featuring an analytical range of >3 logs, Gyrolab AAVX to generate full-length DNA constructs, clones, and libraries with
Titer Kit also minimizes repeat testing compared to ELISA. unprecedented speed. In addition, the benchtop system allows
Gyros Protein Technologies laboratories to maintain complete control of intellectual property
For info: +46-(0)-18-56-63-00 relating to their proprietary vectors. As a result, this single platform
www.gyrosproteintechnologies.com provides a secure method to exponentially accelerate discovery of
candidate drugs for infectious disease, cancer, and answering gene-
RNA Extraction for COVID-19 Diagnostic Testing function questions.
Porvair Sciences announces a special version of its popular Codex DNA
Homogenizer Spin Column optimized for RNA extraction from For info: 858-228-4115
buccal swabs. A large number of current COVID-19 diagnostic tests codexdna.com/products/bioxp-system
are reliant on detection of viral RNA from buccal swabs by PCR. The
column has been adapted to routinely improve yield and quality of KAPA Target Enrichment
viral RNA from nasal and throat swab samples. It provides a fast Roche Sequencing announces its KAPA Target Enrichment
procedure for downstream RNA extraction, achieving great results Portfolio, a new collection of powerful, hybridization-based target
in just minutes. The novel dual-frit extraction design reduces lysate enrichment tools for clinical, translational, and basic research.
viscosity and captures insoluble debris by centrifugation, so the The portfolio includes (1) HyperDesign, a new, easy-to-use online
homogenized lysate sample is ready for RNA extraction. Using tool for the design of custom probe panels that leverages Roche’s
the Homogenizer increases RNA yield as compared to alternative renowned design expertise; (2) KAPA Target Enrichment Probes,
shredding columns or if no homogenization is undertaken. The high-performing probes that provide greater sequencing coverage,
column ensures faster homogenization than is possible with lower duplication rates, and improved coverage uniformity; (3)
traditional methods using syringe and needles, with reduced waste KAPA HyperChoice, fully customizable probe panels for human
and minimal loss of sample. samples; (4) KAPA HyperExplore, fully customizable probe panels
Porvair Sciences for nonhuman samples; (5) KAPA HyperExome, a compact whole-
For info: 800-552-3696 exome panel with broad coverage of human genomic and medical
www.chromatrap.com/homogeniser-spin-column research databases (CCDS, RefSeq, Ensembl, GENCODE, ClinVar); and
(6) HyperCap Workflow v3.0: a streamlined, automatable solution
Viral DNA/RNA Purification for integrated library preparation and target enrichment using
Quick-DNA/RNA Viral MagBead kit is designed for high-throughput KAPA HyperPrep or KAPA HyperPlus Kits, now updated for greater
purification of viral DNA and/or RNA from plasma, serum, urine, cell efficiency with KAPA Target Enrichment Probes.
culture media, blood, saliva, cellular suspensions, biopsies, and swab Roche Sequencing
and fecal samples stored in our DNA/RNA Shield medium (for sample For info: 855-527-2246
www.roche.com

Electronically submit your new product description or product literature information! Go to www.sciencemag.org/about/new-products-section for more information.

Newly offered instrumentation, apparatus, and laboratory materials of interest to researchers in all disciplines in academic, industrial, and governmental organizations are featured in this
space. Emphasis is given to purpose, chief characteristics, and availability of products and materials. Endorsement by Science or AAAS of any products or materials mentioned is not
implied. Additional information may be obtained from the manufacturer or supplier.

254 sciencemag.org/custom-publishing SCIENCE

Faculty Position in Interdis- online @sciencecareers.org

Aj int nitiat ve bet een the Max Planck S c ety and ciplinary Cancer Research

MAX PLANCK SCHOOLS _ lead;ng German Unwers;t;e, together ;th at the Ecole polytechnique fédérale
Fraunh fer·Gesells hafFG Helmh ltz Ass c ationG and de Lausanne (EPFL)
University C llege London
The School of Life Sciences of EPFL (Ecole polytechnique fédérale de
2021 Schaefer Research Scholars Program Awards Lausanne) invites applications for a tenure track assistant professor
position in the field of Interdisciplinary Cancer Research.
The Vagelos College of Physicians and Surgeons (VP&S) is pleased
to announce the 2021 Schaefer Research Scholars Program Awards. The This search is part of major initiatives in the Lake Geneva region to
Awards, made possible through a bequest from Dr. Ludwig Schaefer, promote cancer research, which is increasingly driven by integrating
are made annually to four research scientists who have distinguished research from different fields ranging from basic sciences (life sci-
themselves in human physiology, as broadly defined, and whose current ences, physics, chemistry and engineering) to clinical research and
work is of outstanding merit. The proposed research must have the treatment.
potential to illuminate the field. Two awards are made to research
scientists residing or working in North or South America, and two awards The successful candidate will: join the faculty of the Swiss Institute for
are made to research scientists residing or working outside North or South Experimental Cancer Research (ISREC) http://isrec.epfl.ch; develop an
America. Each award consists of a $50,000 cash prize and up to $200,000 independent internationally prominent research program in the broad
in direct research support. domain of interdisciplinary cancer research and its potential thera-
Applications must include a cover sheet; research proposal (one peutic applications; participate in both undergraduate and graduate
page); a research budget (not to exceed $200,000 in total direct costs) teaching; and supervise PhD students and postdoctoral fellows. Candi-
delineated by cost category (salary, fringe, supplies, etc.) for one year dates may work in any cancer-related area, including cancer genetics,
(7/1/2021–6/30/2022); a curriculum vitae (not to exceed 10 pages); and functional genomics and genome instability, epigenetic regulation of
a page summarizing applicant’s research support. Internal candidates cancer genotypes, cancer metabolism, computational cancer genet-
must obtain a nomination letter from their Department Chair. External ics, bioengineering of cell-based therapies and oncolytic viruses, and
candidates must present letters from the Dean or equivalent in their home synthetic biology, systems biology, chemical biology of cancer. Exper-
institution as well as from the Columbia University Irving Medical Center tise in data science is also encouraged.
collaborator, if applicable.
Nomination deadline: November 19, 2020, at 5:00 p.m. (EST) The successful candidate will be a part of EPFL’s cancer research insti-
To apply visit https://www.ps.columbia.edu/schaefer and submit all tute (ISREC), and is expected to perform and coordinate highly interac-
required materials. tive biomedical research, reaching out and taking advantage of EPFL’s
Awardees will be notified in February 2021. interdisciplinary campus (Schools of Basic Sciences, Engineering, and
Information and Communication Technologies) and its involvement
in the multi-institutional Swiss Cancer Center Leman, which brings
together EPFL, the Universities of Lausanne and Geneva, and clinical
and translational research components of the University Hospitals of
Lausanne and Geneva.

Applications should include a cover letter, a curriculum vitae, a list of
publications (annotated to indicate the candidate’s contributions) a
synopsis of major accomplishments, and a concise statement of future
research agenda and teaching interests, along with the contacts of 3-5
referees who can provide letters of recommendations. Applications
should be uploaded as PDF files to the recruitment web site:

https://facultyrecruiting.epfl.ch/position/23691279

Formal evaluation of candidates will begin on January 8, 2021, and
continue until the position is filled.

Enquiries may be sent to:
Prof. Freddy Radtke
Search Committee Chair
E-mail: [email protected]

For additional information on EPFL, the ISREC institute and the school
of life sciences, please consult: www.epfl.ch, sv.epfl.ch

EPFL is an equal opportunity employer and family friendly university.
It is committed to increasing the diversity of its faculty. It strongly
encourages women to apply.

online @sciencecareers.org ESE Tenured or Tenure-track myIDP:
Faculty Openings, 2020 - 2021 A career plan customized
for you, by you.
The School of Engineering and Applied Science at the University of
Pennsylvania is growing its faculty by 33% over a fve-year period. As For your career in science, there’s only one
part of this initiative, the Department of Electrical and Systems
Engineering is engaged in an aggressive, multi-year hiring effort Features in myIDP include:
for multiple tenure-track positions at all levels. Candidates must
hold a Ph.D. in Electrical Engineering, Computer Engineering,  Exercises to help you examine your skills,
Systems Engineering, or related area. The department seeks
individuals with exceptional promise for, or proven record of, interests, and values.
research achievement, who will take a position of international
leadership in defning their feld of study and who will excel in  A list of 20 scientifc career paths with a
undergraduate and graduate education. Leadership in cross-
disciplinary and multi-disciplinary collaborations is of particular prediction of which ones best ft your skills
interest. We are interested in candidates in all areas that enhance and interests.
our research strengths in:
 A tool for setting strategic goals for the
• Nanodevices and nanosystems (nanoelectronics, MEMS/
NEMS, power electronics, nanophotonics, nanomagnetics, coming year, with optional reminders to
quantum devices, integrated devicesand systems at nanoscale); keep you on track.
https://apptrkr.com/2007344
 Articles and resources to guide you through
• Circuits and computer engineering (analog, RF, mm-wave,
digital circuits, emerging circuit design, computer engineering, the process.
IoT, beyond 5G, andcyber-physical systems);
https://apptrkr.com/2007395  Options to save materials online and print

• Information and decision systems (control, optimization, them for further review and discussion.
robotics, data science, machine learning, communications,
networking, information theory, signal processing).  Ability to select which portion of your IDP
https://apptrkr.com/2008072
you wish to share with advisors, mentors,
Prospective candidates in all areas are strongly encouraged to or others.
address large-scale societal problems in energy, transportation,
health, agriculture, food and water, economic and fnancial  A certifcate of completion for users that
networks, social networks, critical infrastructure, and national
security. We are especially interested in candidates whose fnish myIDP.
interests are aligned with the school's strategic plan,
https://www.seas.upenn.edu/about/penn-engineering-2020/ Visit the website and start
planning today!
Diversity candidates are strongly encouraged to apply. Interested myIDP.sciencecareers.org
persons should submit an online application by following the links
above and include curriculum vitae, research, teaching, and In partnership with:
diversity statements, and at least three references. Review of
applications will begin on January 4, 2021.

The University of Pennsylvania values diversity and seeks talented
students, faculty and staff from diverse backgrounds. The University
of Pennsylvania is an equal opportunity and affrmative action
employer. Candidates are considered for employment without
regard to race, color, sex, sexual orientation, gender identity,
religion, creed, national or ethnic origin, citizenship status, age,
disability, veteran status or any other legally protected class.

online @sciencecareers.org

FACULTY POSITION TENURE-TRACK POSITION
BACTERIAL PATHOGENESIS.
The Department of Physiology invites outstanding scientists with Ph.D.,
The Department of Microbiology at UT Southwestern Medical Center is M.D., or equivalent degrees to apply for tenure-track faculty positions at
seeking a new faculty member in bacterial pathogenesis at the Assistant the level of Assistant Professor.
Professor (tenure track) level. Appointment rank will be commensurate
with academic accomplishments and experience. The appointee will be Candidates who bring innovative approaches to the study of any under-
expected to develop a front-rank, competitive, independent research explored/unexplored questions broadly related to physiology are
program on a medically relevant bacterial pathogen(s) and/or on concepts encouraged to apply. The scientific excellence of the candidates is more
relevant to the human microbiome. An important academic responsibility important than the specific area of research. These positions are part of the
also will be the instruction and mentoring of graduate students. An continuing growth of the Department at one of the country’s leading
attractive start-up package, including a competitive salary and generous academic medical centers. They will be supported by significant laboratory
laboratory space in a modern building, is available to conduct space, competitive salaries, state-of-the-art core facilities and exceptional
research within a highly dynamic environment of a leading medical start-up packages. The University of Texas Southwestern Medical Center is
microbiology department (https://www.utsouthwestern.edu/education/ the scientific home to six Nobel Prize laureates and many members of the
medical-school/departments/microbiology). National Academy of Sciences and Institute of Medicine. UT Southwestern
conducts more than 3,500 research projects annually totaling more than
Candidates will be considered for our $2M Endowed Scholars (start- $417 million. Additional information about the Department of Physiology
up) Program (http://www.utsouthwestern.edu/education/programs/ can be found at http://www.utsouthwestern.edu/education/medical-
nondegree-programs/other-programs/endowed-scholars/index.html). school/departments/physiology/index.html.

Candidates should have a Ph.D. and/or M.D. degree with at least 3-4 Information regarding careers can be found at:
years of postdoctoral experience and an exceptional publication record. https://jobs.utsouthwestern.edu/.
Please send a cover letter, C.V., contact information for three letters of
recommendation, and a brief summary of future research to: Bacterial Applicants should submit a CV, a brief statement of current and proposed
[email protected]. research, and a summary of your two most significant publications
describing the importance of the work (100-150 words each). Please
UT Southwestern Medical Center is an Affirmative Action/Equal arrange to have three letters of recommendation sent on his/her behalf. All
Opportunity Employer. Women, minorities, veterans and individuals items should be submitted to: http://academicjobsonline.org/ajo/jobs/
16617. Completed applications will be reviewed starting November 1,
with disabilities are encouraged to apply. 2020. You may email questions to [email protected].

UT Southwestern Medical Center is an Equal Opportunity/Affirmative
Action Employer. Women, minorities, veterans and individuals
with disabilities are encouraged to apply.