Science 13.11.2020

RESEARCH | REPORT

9. E. Lu, D. A. Llano, S. M. Sherman, Hear. Res. 257, 16–23 (2009). 28. L. A. Anderson, J. F. Linden, Hear. Res. 274, 48–60 generously sharing reagents. Funding: Max Planck Society, German
10. L. Petreanu, D. Huber, A. Sobczyk, K. Svoboda, Nat. Neurosci. (2011). Research Foundation (LE 3804/3-1, CRC 1193, and SPP1665 to
J.J.L.), European Research Council (StG 335587 to J.J.L.),
10, 663–668 (2007). 29. Á. Kulik, S. A. Booker, I. Vida, Neuropharmacology 136, 80–91 Human Frontier Science Program (LT000814/2016-L to M.B.P.).
11. J. J. Letzkus, S. B. E. Wolff, A. Lüthi, Neuron 88, 264–276 (2015). (2018). Author contributions: Conceptualization, project administration,
12. G. Major, M. E. Larkum, J. Schiller, Annu. Rev. Neurosci. 36, writing, and funding acquisition: M.B.P. and J.J.L. Investigation
30. P. R. Roelfsema, A. Holtmaat, Nat. Rev. Neurosci. 19, 166–180 and analyses: M.B.P., J.V., T.D., T.S., and D.-L.P. Computational
1–24 (2013). (2018). modeling conceptualization, methodology, and software: L.B.N., F.K.,
13. E. M. Callaway, L. Luo, J. Neurosci. 35, 8979–8985 (2015). and H.S. Visualization: M.B.P. Supervision: J.J.L. Competing interests:
14. T. R. Reardon et al., Neuron 89, 711–724 (2016). 31. D. Castaño-Díez, Freezing Analysis, Github (2018); Authors declare no competing interests. Data and materials
15. D. Cai et al., Cereb. Cortex 29, 2437–2454 (2019). https://doi.org/10.17617/1.8Q. availability: All data are available in the main text or the supplementary
16. C. Herry, J. P. Johansen, Nat. Neurosci. 17, 1644–1654 (2014). materials. Code for analysis is available on GitHub or GitLab (31–35)
17. J. J. Letzkus et al., Nature 480, 331–335 (2011). 32. F. Kretschmer, EyeTracker, MPCDF Gitlab (2020); and code for computational model at Zenodo (36).
18. C. Mezzera, G. López-Bendito, J. Chem. Neuroanat. 75, 32–40 https://doi.org/10.17617/1.8M.
SUPPLEMENTARY MATERIALS
(2016). 33. F. Kretschmer, PylonRecorder, MPCDF Gitlab (2020);
19. R. Batista-Brito, E. Zagha, J. M. Ratliff, M. Vinck, Curr. Opin. https://doi.org/10.17617/1.8N. science.sciencemag.org/content/370/6518/844/suppl/DC1
Materials and Methods
Neurobiol. 52, 172–181 (2018). 34. F. Kretschmer, D. Unzué, AudioGameGUI, MPCDF Gitlab Supplementary Text
20. S. A. Vasquez-Lopez et al., eLife 6, e25141 (2017). (2020); https://doi.org/10.17617/1.8O. Figs. S1 to S14
Table S1
21. S. Oláh et al., Nature 461, 1278–1281 (2009). 35. F. Kretschmer, D. Castaño-Díez, MotionCorrection, MPCDF References (37–49)
22. M. Capogna, J. Physiol. 589, 1875–1883 (2011). Gitlab (2020); https://doi.org/10.17617/1.8P. MDAR Reproducibility Checklist
23. G. Tamás, A. Lörincz, A. Simon, J. Szabadics, Science 299,
36. L. B. Naumann, H. Sprekeler, Model code for frequency View/request a protocol for this paper from Bio-protocol.
1902–1905 (2003). modulation with presynaptic inhibition, Version 1, Zenodo
24. T. Deneux et al., eLife 8, e44006 (2019). (2020); https://doi.org/10.5281/zenodo.3886569. 14 April 2020; accepted 30 September 2020
10.1126/science.abc2399
25. S. M. Sherman, R. W. Guillery, in The Synaptic Organization ACKNOWLEDGMENTS
of the Brain, G. M. Shepherd, Ed. (Oxford Univ. Press, ed. 5,
2004) chap. 8, pp. 311–360. We thank members of the Letzkus and Sprekeler labs,
M. S. Fustinana Gueler, J. Gjorgjieva, S. Romano, and J. Keijser
26. M. M. Halassa, S. Kastner, Nat. Neurosci. 20, 1669–1679 (2017). for discussions; E. Abs, A. Wrana, S. Junek, F. Vollrath, G. Tushev,
27. N. Takahashi, T. G. Oertner, P. Hegemann, M. E. Larkum, and S. Onasch for technical assistance; and L. L. Looger,
J. Akerboom, D. S. Kim, the GENIE Project at Janelia Farm,
Science 354, 1587–1590 (2016). K. Deisseroth, E. S. Boyden, H. Zeng, and K.-K. Conzelmann for

Pardi et al., Science 370, 844–848 (2020) 13 November 2020 5 of 5

RESEARCH

SENSORS tinuum gradient dye pattern, for which we can
obtain continuous spatial resolution of single-
Stretchable distributed fiber-optic sensors mode deformation (fig. S1); and (ii) a discrete
color-block pattern, for which we can measure
Hedan Bai1*, Shuo Li2*, Jose Barreiros3, Yaqi Tu2, Clifford R. Pollock4, Robert F. Shepherd1,2,3† and decouple multiple modes of deformations
with discrete spatial resolution. To demon-
Silica-based distributed fiber-optic sensor (DFOS) systems have been a powerful tool for sensing strain, strate the multifunctionality of stretchable
pressure, vibration, acceleration, temperature, and humidity in inextensible structures. DFOS systems, DFOSs, our work focused on the design with
however, are incompatible with the large strains associated with soft robotics and stretchable electronics. the discrete color-block pattern, which we name
We develop a sensor composed of parallel assemblies of elastomeric lightguides that incorporate continuum the stretchable lightguide for multimodal sen-
or discrete chromatic patterns. By exploiting a combination of frustrated total internal reflection and sing (SLIMS). We compare major features of
absorption, stretchable DFOSs can distinguish and measure the locations, magnitudes, and modes SLIMS, FBG, and intrinsic DFOSs in table S1.
(stretch, bend, or press) of mechanical deformation. We further demonstrate multilocation decoupling and
multimodal deformation decoupling through a stretchable DFOS–integrated wireless glove that can Figure 1, A and B, shows the construction
reconfigure all types of finger joint movements and external presses simultaneously, with only a single of SLIMS with a lightguide consisting of two
sensor in real time. polyurethane elastomeric cores—a dyed core
doped with absorbing dyes at four discrete
L ight has primary properties of intensity, and processed by distinctive mechanoreceptors locations and a clear core with no dyes—
wavelength spectrum, polarization, and (12). Thus, mechanically compliant deforma- separated by silicone cladding. A white light-
propagation direction that can be indi- tion sensors with multiple sensing modalities emitting diode (LED) is coupled to the dyed
vidually detected to reflect disturbances could give intelligent soft systems the capacity core at one end. Two red-green-blue (RGB)
in the propagating medium. On the basis for rich tactile sensations comparable to those sensor chips are placed at the other end, one
of these properties, sensing through optical of mammals. Myriad classes of intrinsically coupled to the dyed core and one coupled to
fibers has been a useful platform for moni- stretchable sensors have been developed on the clear core (see the “sensor design” and
toring mechanical deformation in stiff infra- the basis of resistive (13–15), capacitive (16, 17) “sensor fabrication” sections in the materials
structures (e.g., bridges, roads, buildings, and and optical sensing principles (18–22). These and methods). Our optomechanical sensing
geophysics) (1–3). Distributed fiber-optic sen- highly stretchable sensors have demonstrated approach is based on deformation-induced
sor (DFOS) systems can be categorized into numerous sensing modes for a variety of de- geometric changes in the optical path of light
two approaches: (i) intrinsic, in which a single formations (e.g., stretch, bend, press, and twist). propagation in SLIMS. Chromatic dyes pro-
measurand (e.g., strain or temperature) can The sensors, however, share the same limita- vide wavelength-selective modulation through
be monitored continuously through the length tion: They respond to the different deforma- light absorption variations; the dual-core
of an optical fiber via Raman, Rayleigh, or tion modes with changes of one parameter structure allows frustrated total internal re-
Brillouin scattering; or (ii) quasi, such as fiber (resistance, capacitance, optical intensity, or flection (TIR). By observing the chromati-
Bragg grating (FBG) sensors, in which multi- chromaticity). Therefore, unlike mechanor- city and intensity outputs from the two cores,
ple measurands (e.g., strain, pressure, vibra- eceptors that distinctively react to certain we can determine the location, magnitude,
tion, acceleration, temperature, and humidity) stimuli, these stretchable sensors cannot easily and modes of deformation (Fig. 1C and movies
can be monitored with discrete spatial resolu- differentiate deformation modes. To detect S1 to S3).
tion through sensitizing sections of an optical complex deformation combinations, multiple
fiber for specific measurements (1–4). Although sensors are usually geometrically constrained Because the lightguides have cross sections
commercially available DFOSs have already strategically to ensure that each sensor is de- at the millimeter scale, the wave properties of
shown promise for advanced multifunctional coupled and detects one mode of deformation visible light (with wavelengths <700 nm) be-
sensing in soft materials (5, 6), the inextensible (18, 23). To detect spatial information, a grid- come unimportant. The white LED fitted to
silica optical fibers, bulky laser light source, like distribution of sensors is employed in two the dyed core input has a viewing angle of
and expensive detection equipment required or three dimensions (24–26). As sensing be- 110° so that rays from the same LED also
for resolving the subtle backscattered wave- comes more advanced, the sensor system’s enter the clear core. The doped dyes act as
length shift become barriers to broad adop- architectural complexity increases in terms of color codes for the spatial information. Be-
tion of this technology in research areas such sensor placement and density. Toward this end, cause the dye patterns have a depth equal to
as soft robotics (7), stretchable electronics machine learning has recently been adopted only a fraction (in this case, ¼) of the dyed
(8, 9), and biomedicine (10, 11). in soft sensors that can measure and differen- core height, very few light rays can pass through
tiate between bending and twisting with sim- the dyes in the undeformed configuration (for
To apply the abilities of DFOSs to tactile ple sensor layouts (27). Until now, sensing of a ray diagram, see Fig. 1C, undeformed). Both
sensation in stretchable optoelectronics, we complex soft-body deformations has been ad- cores have white outputs when undeformed.
took inspiration from mammals. Mammals dressed by either complicated system integ-
have soft tissues that allow mechanical stim- ration or computation methods that involve When a dyed region is stretched (ray diag-
uli to propagate to various depths, where dif- model training. ram: see Fig. 1C, stretched), the optical path
ferent deformation modalities are detected length in that region increases, inducing more
We present a multifunctional stretchable op- absorption by the dye and a change in output
1Sibley School of Mechanical and Aerospace Engineering, tomechanical sensor that we call the stretch- light toward the corresponding color. By con-
Cornell University, Ithaca, NY 14853, USA. 2Department able DFOS, inspired by silica-based DFOSs. trast, the clear core does not change color
of Materials Science and Engineering, Cornell University, Stretchable DFOS systems are collateral as- owing to insufficient coupling between the
Ithaca, NY 14853, USA. 3Program of Systems Engineering, semblies of elastomeric lightguides with em- two cores at the stretched dyed region. The
Cornell University, Ithaca, NY 14853, USA. 4School of bedded chromatic patterns. We have designed ray optics simulation (COMSOL Multiphysics,
Electrical and Computer Engineering, Cornell University, two chromatic dye patterns as analogies of the COMSOL Inc.) also confirms this color-changing
Ithaca, NY 14853, USA. two types of conventional DFOSs: (i) a con- response (figs. S4 and S5A). Because stretch-
*These authors contributed equally to this work. ing increases the optical path in both cores,
†Corresponding author. Email: [email protected] according to the Beer-Lambert law, light
outputs for both cores decrease in intensity.

Bai et al., Science 370, 848–852 (2020) 13 November 2020 1 of 5

RESEARCH | REPORT

Fig. 1. SLIMS. (A) Image of a SLIMS tied into a knot. (B) Schematic of the SLIMS showing the discrete close to the light source in the spatial transient
dyed regions, the design of the collateral cores, and its coupling to a light source and color sensors. region of the lightguide (see the “nonlinearity
(C) Optical outputs and ray diagrams of SLIMS when it is undeformed, stretched, bent, and pressed. in sensor response” section in the supplemen-
Scale bars, 1 cm. tary text).

Characterization of SLIMS shows that as SLIMS falls within 2 to 5 dB e−1 (where e is When a dyed region is in bending configu-
any single dyed region is stretched (for de- strain) (see the “sensitivity variation and dye ration (Fig. 1C, bent), most rays in the dyed
tails, see the “sensor characterization” section absorbance” section in the supplementary core can pass through the dye, leading to a
in the materials and methods), no chroma- text) over a range of 150% elongation. We change in the output color. Some colored rays
ticity change in the clear core is observed obtained 0.1-mm resolution (1% strain) with a that fail to satisfy the critical angle escape the
(Fig. 2A), whereas chromaticity change with 0.5-mm prestrain. We found that prestraining dyed core and enter the clear core. These rays,
increasing saturation is observed in the dyed improves the sensor resolution (fig. S6), which however, cannot be coupled to the clear core.
core (Fig. 2D). Both cores have output in- is consistent with the stretch characterization Thus, the clear core still outputs white light
tensities that attenuate linearly in a logarith- results (Fig. 2J). Note the nonlinear intensity (for a ray-optics simulation, see fig. S5B). With
mic scale (Fig. 2, G and J). On the basis of the response when the yellow-green region (YG) each of the four dyed regions bent to a cur-
dyed core’s attenuation, strain sensitivity of is stretched (Fig. 2G): The YG dye is placed too vature up to k = 0.65 cm−1, we measured the
chromaticity and intensity responses of both
cores (Fig. 2, B, E, H, and K). The chromaticity
plots demonstrate that the color response to
bending is similar to the stretching response:
The clear core output remains white, and the
dyed core output changes its color toward the
bent dye region (Fig. 2, B and E). Differentia-
tion between stretching and bending can be
realized by comparing the intensity output of
the clear core, as the intensity is largely un-
affected (Fig. 2H) in bending and attenuates
substantially in stretching. With the dyed core’s
attenuation (Fig. 2K), SLIMS’ bending sensi-
tivity is calculated to be 7 to 24 dB·cm−1. The
bending resolution is measured as 0.5° using a
digital protractor (fig. S7).

Frustrated TIR takes place when SLIMS is
pressed (Fig. 1C, pressed). Rays in the dyed
core escape through the dye and couple to the
clear core, leading to a colored output in the
clear core (Fig. 2C). As the compression force
increases, the clear core has more-saturated
color output. The dyed core output remains
white because there is no increase in optical
path length or number of rays that pass through
the dye (Fig. 2F). On the basis of the intensity
attenuation of the clear core (Fig. 2I), SLIMS
has a sensitivity to pressing of 0.9 to 1.2 dB N−1
in the range of 2 N < F < 5 N (F, force). This
sensitivity can be tuned with the Young’s
moduli of the constructing elastomers. Com-
pression force resolution is found to be 0.14 N.
We observed high repeatability (fig. S8) and
close to linear trend (in logarithmic scale) in
the intensity attenuation for all dyed regions
except YG. This issue is also attributable to the
proximity of the YG region to the light source.

We incorporated SLIMS into a 3D printed
soft glove, where each finger uses only a single
customized SLIMS that allows proprioceptive
sensing of three finger joints’ movements and
exteroceptive sensing of external presses to be
captured simultaneously. The proximal, mid-
dle, and distal joints of the finger are covered
by discrete red, blue, and green dyes, respectively
(Fig. 3A; see the “glove fabrication” section in
the materials and methods).

For proprioception, we first collected the
raw data of RGB intensities from both cores

Bai et al., Science 370, 848–852 (2020) 13 November 2020 2 of 5

RESEARCH | REPORT

puts. More specifically, we can formulate the
algorithm to decouple multilocation bending
as a vector sum model: Any multijoint bend
response [denoted by a random point P (x, y, Y)
within the pyramid] can be related to the three
single-joint bend reference vectors with different
weights, and those weights depict the respec-
tive bending angles at each joint (Fig. 3C)

→→
OP ¼ ðProximal Joint Angle=90°Þ OR

→
þ ðMiddle Joint Angle=90°Þ OB

→
þ ðDistal Joint Angle=90°Þ OG

where point O represents the response of the

undeformed output with white color and max-

imum intensity. Converting the intensity at-

tenuation to a logarithm→ic s→cale (in →units of
decibels) yielded vectors OR, OB, and OG that

correspond to the three linearly fi→tted single-
joint bend reference lines. Vector OP, the rel-

ative response of an arbitrary multijoint bend

configuration, O→BtheRcean,uO→sceaBnv,ecbatneodresx→O→OpGrRe,sws→OeiBtdh,aadsnifda-
vector sum of
f→erent weights.
OG are linearly independent and span the en-

tire 3D space, this model has unique solutions

for the three unknowns: proximal, middle,

and distal joint angles. See the supplementary

text for model validation. With the vector sum

model, we can solve for the three unknown

joint angles, given the measured RGB inten-

sities in the dyed core. Figure 3G and movie S4

show that the glove performs decoupling and

reconstruction of the three finger joints’ move-

ments in real time.

For simultaneous exteroceptive sensing, we

achieved the decoupling of external pressing

from bending by setting thresholds for the

normalized intensity output of the clear core

(Fig. 4A and movie S4). In the case of bending

only (0 to 10 s), the intensity attenuates slight-

Fig. 2. Characterization of SLIMS in different deformation modes. Chromaticity responses of the clear ly; it jumps when the SLIMS is pressed at dif-
core when SLIMS is (A) stretched, (B) bent, or (C) pressed in the dyed regions. Chromaticity responses
of the dyed core when SLIMS is (D) stretched, (E) bent, or (F) pressed in the dyed regions. Intensity responses of ferent joints for multiple times (10 to 30 s). We
the clear core when SLIMS is (G) stretched, (H) bent, or (I) pressed in the dyed regions. Intensity responses of the
dyed core when SLIMS is (J) stretched, (K) bent, or (L) pressed in the dyed regions. Error bars indicate SDs. set this threshold within the range of 0.75 to

1.0. When the normalized intensity exceeds

this range, we could determine the presence

of a press. To simplify the representation of

chromaticity, we extracted the hue value from

as the index finger is bent with different joint and we then repeated this measurement for the RGB intensities of the clear core output
angle combinations (Fig. 3B). When there is all three joints. To decouple intensity and
only one bend from either the proximal or the chromaticity response, we converted the RGB (Fig. 4B). The hue representation enabled us
middle joint (it is difficult to bend only the intensity output into the CIE xyY color space
distal joint), the dyed core immediately changes (Fig. 3D). As each joint bends from 0° to 90° to locate the region that is being pressed. The
color from white to either red or blue. We can with an interval of 10°, the normalized inten-
ascertain both the location and the angle sity output Y (represented by the size of the magnitude of the exerted force during press-
of that bend. When multiple bends coexist, dots) drops and the chromaticity output xy
however, we observed a mixture of colors (represented by the color of the dots) becomes ing also qualitatively scales with this hue value.
from the bent regions. Therefore, we need to more saturated. Each combined response (rep-
develop a mathematical model to decouple the resented by the dots) follows a linear trend, We were able to reconstruct, in real time, the
fusion in both chromaticity and intensity re- producing red, blue, and green single-joint
sponses (Fig. 3C). bend reference lines. The space within the location and magnitude of an external press
pyramid formed by these three reference lines
We first measured the dyed core response thus includes all possible multijoint bend out- when the index finger is bent at multiple loca-
in SLIMS when only a single joint was bent,
tions (Fig. 4C).

The principles of stretchable DFOSs could

also be applied to detect measurands beyond

those demonstrated in this work and for

different applications. For example, movie S5

shows the dynamic color change that could

be used for slip detection in robotic grippers.

Bai et al., Science 370, 848–852 (2020) 13 November 2020 3 of 5

RESEARCH | REPORT

Fig. 3. SLIMS-integrated soft glove and proprioception of multijoint bending. (A) Image of the SLIMS-integrated soft glove with LED light source and
electronics module. (B) Real-time normalized intensity data from both cores under five different multijoint bending configurations. (C) Vector sum model for
multijoint bend decoupling. (D) Single-joint bend measurements. (E) Two-joint bend measurements. (F) Three-joint bend measurements. (G) Real-time reconstruction
of multijoint bending using the vector sum model that we derived.

The spatial resolution of the discrete dye pat- and color sensor resolution” section in the sup- 3. K. O. Hill, G. Meltz, J. Lit. Technol. 15, 1263–1276 (1997).
tern is limited by the spacing between the plementary text). The cross-sectional dimen- 4. B. Lee, Opt. Fiber Technol. 9, 57–79 (2003).
dyes; distributed sensing could be achieved sions of SLIMS are substantially larger than 5. K. C. Galloway et al., Soft Robot. 6, 671–684 (2019).
through the gradient dye pattern (fig. S1). those of silica optical fibers, owing to the size 6. Y.-L. Park, K. Chau, R. J. Black, M. R. Cutkosky, in Proceedings
We characterized the spatial resolution as of the coupled optoelectronic devices; how-
1 mm for the gradient dye in bending con- ever, the size of SLIMS could be reduced with of the 2007 IEEE International Conference on Robotics and
figuration (fig. S11; see the “distributed sen- a tapered design. The theoretical limit in fiber Automation (IEEE, 2007), pp. 1510–1516.
sing characterization with gradient pattern” diameter is then comparable to that of com- 7. S. Li, H. Bai, R. F. Shepherd, H. Zhao, Angew. Chem. Int. Ed. 58,
section in the supplementary text). Combin- mercial silica multimodal fibers (~50-mm 11182–11204 (2019).
ing the discrete and gradient patterns could diameter; see the “scaling” section in the sup- 8. S. Niu et al., Nat. Electron. 2, 361–368 (2019).
enable multilocation detection with continu- plementary text). SLIMS has an attenuation 9. K. Takei, W. Gao, C. Wang, A. Javey, Proc. IEEE 107, 2155–2167
ous spatial resolution (fig. S12). coefficient of 0.68 dB·cm−1, which limits the (2019).
range of sensing length to tens of centimeters.
Our current SLIMS sensor is limited to de- Multimeter-scale applications can be achieved 10. H. U. Chung et al., Science 363, eaau0780 (2019).
coupling three dyed regions for multibend via a hybrid approach of low-loss optical fi-
applications, owing to the limited output gen- bers for transmission and SLIMS for sensing 11. S.-H. Byun et al., Sci. Adv. 5, eaay0418 (2019).
erated by the RGB color sensor chip (fig. S13A, (see the “scaling” section in the supplemen-
three color bands). If color sensors with better tary text). 12. V. E. Abraira, D. D. Ginty, Neuron 79, 618–639 (2013).
spectrum resolution (e.g., PixelSensor multi- 13. T. Yamada et al., Nat. Nanotechnol. 6, 296–301 (2011).
spectral sensors, PIXELTEQ; fig. S13B, eight REFERENCES AND NOTES
color bands) are used, the number of de- 14. M. D. Dickey, Adv. Mater. 29, 1606425 (2017).
couplable dyed regions can be improved to 1. A. D. Kersey, A. Dandridge, J. Inst. Electron. Radio Eng. 58, S99 15. C. Keplinger et al., Science 341, 984–987 (2013).
the number of bands that the color sensor (1988). 16. A. Frutiger et al., Adv. Mater. 27, 2440–2446 (2015).
can resolve (see the “multi-bend decoupling 17. S. Li, B. N. Peele, C. M. Larson, H. Zhao, R. F. Shepherd,
2. N. J. Lindsey, T. C. Dawe, J. B. Ajo-Franklin, Science 366,
1103–1107 (2019). Adv. Mater. 28, 9770–9775 (2016).
18. H. Zhao, K. O’Brien, S. Li, R. F. Shepherd, Sci. Robot. 1,

eaai7529 (2016).
19. M. Kolle et al., Adv. Mater. 25, 2239–2245 (2013).
20. J. Guo et al., Adv. Mater. 28, 10244–10249 (2016).
21. A. Leber, B. Cholst, J. Sandt, N. Vogel, M. Kolle, Adv. Funct. Mater.

29, 1802629 (2018).
22. R. Ghaffari et al., Adv. Funct. Mater. 30, 1907269 (2019).
23. R. L. Truby et al., Adv. Mater. 30, e1706383 (2018).
24. M. Ramuz, B. C.-K. Tee, J. B.-H. Tok, Z. Bao, Adv. Mater. 24,

3223–3227 (2012).

Bai et al., Science 370, 848–852 (2020) 13 November 2020 4 of 5

RESEARCH | REPORT

Fig. 4. SLIMS-integrated soft-glove decoupling
combined proprioception and exteroception.
(A) Real-time normalized RGB intensities of both
cores under bending and pressing on different joints
from 0 to 30 s. The threshold for press detection is
denoted by the gray box. (B) Real-time hue values
derived from the normalized RGB intensities under the
same deformations from 0 to 30 s. The colored
boxes show hue values that determine the pressed
joints. (C) Real-time reconstruction of combined
proprioception and exteroception.

25. A. Levi, M. Piovanelli, S. Furlan, B. Mazzolai, L. Beccai, Sensors the NSF (grant NNCI-1542081), and at the Cornell Center for Materials conclusions in the paper are present in the main text and the
13, 6578–6604 (2013). Research Shared Facilities, which was supported through the NSF supplementary materials.
MRSEC program (grant DMR-1719875). Author contributions:
26. P. A. Xu et al., Sci. Robot. 4, eaaw6304 (2019). Concept, design, and study direction: H.B. and R.F.S.; device SUPPLEMENTARY MATERIALS
27. I. M. Van Meerbeek, C. M. De Sa, R. F. Shepherd, Sci. Robot. 3, fabrication: H.B., S.L., and J.B.; experimental validation: H.B., S.L., science.sciencemag.org/content/370/6518/848/suppl/DC1
Y.T., and R.F.S.; data analysis: H.B., S.L., and R.F.S.; theoretical Materials and Methods
eaau2489 (2018). modeling: H.B., C.R.P., and R.F.S..; manuscript writing: H.B., S.L., Supplementary Text
and R.F.S.; manuscript editing: H.B., S.L., C.R.P., and R.F.S.; Figs. S1 to S20
ACKNOWLEDGMENTS supervision: C.R.P. and R.F.S.; project administration: R.F.S. Tables S1 and S2
We thank K. Wang for conducting tensile tests and A. Cornwell for Competing interests: The stretchable DFOS systems presented in References (28–31)
helping with circuit soldering. Funding: This work was supported by this work have been filed under an international patent application Movies S1 to S5
the NSF EFRI program (grant EFMA-1830924), the Air Force Office of (no. 62/642,407) for “Waveguide and Sensor Based on Same.” Code S1
Scientific Research (grant FA9550-18-1-0243), Cornell Technology The listed inventors are R.F.S., H.B., S.L., and Y.T. The intellectual
Acceleration and Maturation (CTAM), National Institute of Food and property is being licensed by Organic Robotics Corporation. 12 December 2019; accepted 5 October 2020
Agriculture (grant 2019-67021-29225), and the Office of Naval R.F.S. is the founder of Organic Robotics Corporation. Data 10.1126/science.aba5504
Research (grant N00014-17-1-2837). Part of the study was performed and materials availability: All data needed to evaluate the
at the Cornell NanoScale Facility, a member of the National
Nanotechnology Coordinated Infrastructure, which was supported by

Bai et al., Science 370, 848–852 (2020) 13 November 2020 5 of 5

RESEARCH

TRANSLATION REGULATION expression was stress specific and did not oc-
cur with glutamine or serum starvation (Fig.
A phosphorylation-regulated eIF3d translation 1A). Analysis of Jun mRNA distribution in
switch mediates cellular adaptation polysome fractions revealed that elevation of
to metabolic stress Jun protein synthesis during glucose depri-
vation was due to increased ribosome asso-
Adam M. Lamper*†, Rebecca H. Fleming*, Kayla M. Ladd*, Amy S. Y. Lee‡§ ciation and translation efficiency (Fig. 1, B and
C). By contrast, the eIF4E-dependent control
Shutoff of global protein synthesis is a conserved response to cellular stresses. This general mRNA PSMB6 shifted to monosome and free
phenomenon is accompanied by the induction of distinct gene programs tailored to each stress. RNA fractions under the same glucose depri-
Although the mechanisms driving repression of general protein synthesis are well characterized, vation conditions (Fig. 1, B and C). Thus, the
how cells reprogram the translation machinery for selective gene expression remains poorly ability to engage eIF3d is likely key to enabling
understood. Here, we found that the noncanonical 5′ cap-binding protein eIF3d was activated in the Jun mRNA to escape translation shutoff.
response to metabolic stress in human cells. Activation required reduced CK2-mediated We next investigated whether nutrient starva-
phosphorylation near the eIF3d cap-binding pocket. eIF3d controls a gene program enriched in tion directly changed eIF3d cap-binding activ-
factors important for glucose homeostasis, including members of the mammalian target of ity. An autoinhibitory conformation restricts
rapamycin (mTOR) pathway. eIF3d-directed translation adaptation was essential for cell survival eIF3d and necessitates that the subunit must
during chronic glucose deprivation. Thus, this mechanism of translation reprogramming regulates function as part of the 13-subunit eIF3 com-
the cellular response to metabolic stress. plex for 5′ cap recognition (2). Taking advan-
tage of a natural HIV-1 protease (PR) cleavage
D uring stress, cells undergo substantial initiation factor 4E (eIF4E) (1). How initia- site in eIF3d that occurs immediately before
alterations to protein synthesis. Global tion factor activity is modified during stress the cap-binding domain (fig. S1B) (3), we used
translation is inhibited, whereas the to allow some mRNAs to escape shutoff de- immunoprecipitation and on-bead cleavage to
translation of a subset of mRNAs en- spite carrying a conserved 5′ cap structure isolate eIF3d cap-binding activity separately
coding stress-responsive proteins is con- remains unknown. The eIF3d subunit of the from other eIF3 complex functions (4–6) (fig.
siderably up-regulated. Translation shutoff 13-subunit eukaryotic initiation factor com- S1, C to G). eIF3d cap binding to Jun mRNA
is mediated through inactivation of core plex 3 (eIF3) is a cap-binding protein that increased 10-fold specifically during glucose
translation initiation factors, including the 5′ directs noncanonical translation of the tran- deprivation (Fig. 1D). Thus, a cellular mecha-
cap-binding protein eukaryotic translation scription factor Jun (2), raising the possibility nism can reprogram eIF3d cap-binding activ-
that eIF3d could be harnessed for stress- ity and translation regulation during nutrient
Department of Biology, Brandeis University, Waltham, MA induced translation. depletion.
02453, USA.
*These authors contributed equally to this work. To test this hypothesis, we examined changes Protein phosphorylation is a major mech-
†Present address: Department of Biological Sciences, Columbia to eIF3d-mediated translation of the Jun mRNA anism regulating translation factor function (1).
University, New York, NY 10027, USA. during metabolic stress in human embryonic Using [32P]-orthophosphate labeling, we found
‡Corresponding author. Email: [email protected] kidney (HEK) 293T cells. Conditions of nutrient that 48 hours after glucose starvation, the eIF3
§Present address: Department of Cell Biology, Harvard Medical depletion caused a marked decrease in general complex underwent a phosphorylation switch,
School, Boston, MA 02115, USA, and Department of Cancer protein synthesis (fig. S1A), yet chronic glucose with loss of modification of an ~70-kDa protein
Immunology and Virology, Dana-Farber Cancer Institute, Boston, depletion led to a pronounced increase in Jun (Fig. 1E). Protein complex digestion with HIV-1
MA 02115, USA. protein levels (Fig. 1A). The increase in Jun PR confirmed that the phosphorylated pro-
tein was eIF3d (Fig. 2A). Mass spectrometry
mapped the phosphorylation sites in eIF3d to
positions S528 and S529 (Fig. 2B and fig. S2A).

Fig. 1. eIF3d-specialized translation increases during chronic A – glucose – serum – glutamine D 15
glucose deprivation. (A) Jun and rpS19 protein levels in HEK293T cells
after nutrient deprivation. (B and C) Polysome association of Jun or h: 0 24 36 48 0 24 36 48 0 24 36 48 Jun Cap-Binding 10
PSMB6 mRNA in response to glucose deprivation in HEK293T cells. Jun (Fold to CM)
PSMB6 is a control for eIF4E-dependent translation. mRNA abundance is 5
expressed as a percentage of total transcript recovered from all rpS19
fractions. Results are plotted as the mean ± SD from a representative 0
quantitative reverse transcription polymerase chain reaction (qRT-PCR) BC polysomes CM -glu -s -glut
performed in duplicate. The results of (A) to (C) are representative light heavy
of three biological replicates. (D) eIF3d binding to the 5′ cap of Jun complete media 40 -glu -s -glut
mRNA in HEK293T cells upon nutrient deprivation. eIF3d cap binding Jun (%) 30 0 6 24 48 48 48 kDa
is quantified as levels of Jun transcripts determined by reverse
transcription and quantitative PCR, in HIV-1 PR-treated eIF3d immuno- A254 20 170
precipitation samples compared with total input Jun RNA. Cap binding is 95
normalized to samples prepared from cells grown in complete media 10 E 72
(CM). glu, glucose; s, serum; glut, glutamine. Results are shown as 55
the mean ± SD of three independent experiments. (E) Phosphorylation Frac: 1 2 3 4 5 6 7 8 0 h: 43
of eIF3 subunits after nutrient deprivation. Shown is a phosphor image Frac: 1 2 3 4 5 6 7 8
34
- glucoseA254PSMB6 (%)80 b/c
Frac: 1 2 3 4 5 6 7 8 60 d/l
40
20

0
Frac: 1 2 3 4 5 6 7 8

complete media
-glucose

of SDS–polyacrylamide gel electrophoresis (PAGE) gel resolving eIF3 subunit phosphorylation levels as detected by [32P]-orthophosphate labeling of HEK293T cells

and immunoprecipitation of the eIF3 complex. The results are representative of two independent experiments.

Lamper et al., Science 370, 853–856 (2020) 13 November 2020 1 of 4

RESEARCH | REPORT

Fig. 2. CK2 phosphorylation inhibits eIF3d cap-binding activity. A HIV-1 protease HIV-1 PR B pS528-pS529:
(A) Phosphorylated eIF3 subunit identification by selective cleavage with
HIV-1 PR. HIV-1 PR was incubated with eIF3 complex immunoprecipitated cleavage site - + VYSLPDGTFSSDEDEE
from [32P]-orthophosphate-labeled HEK293T cells. HIV-1 PR cleavage of kDa
eIF3d leads to faster gel migration. (B) Location of phosphorylated eIF3d eIF3d 72 eIF3d
residues S528 and S529 (blue) identified by mass spectrometry. (66 kDa) + HIV1- PR
(C) Validation of eIF3d phosphorylation sites using eIF3d-HA mutants by 55 RNA- cap-binding
[32P]-orthophosphate labeling in HEK293T cells. Coomassie staining is
shown as a loading control. WT, wild-type. (D and E) Fold change in eIF3d 13 kDa 53 kDa binding
binding to the 5′ cap of Jun mRNA by phosphoinhibitory (S528/529N) or
phosphomimetic (S528/529D) mutants in HEK293T cells. Cap-binding C S528/ D 10 E2
activity was measured as in Fig. 1D and normalized to WT eIF3d. Results
are shown as the mean ± SD of three independent experiments. WT S528DS529D 529D
(F) Association of Jun transcripts with ribosomal complexes in HEK293T Jun (%) Jun (%) Jun (%) 8
cells in which endogenous eIF3d was replaced with eIF3d mutants through Jun Cap-Binding 61
coexpression of shRNA-resistant mutants and an eIF3d-targeting shRNA. (Fold to WT)
The results are representative of two independent experiments, and 32P 4 0.5
ribosome association was determined as in Fig. 1C. (G) In vitro 2
phosphorylation of the recombinant 12-subunit eIF3 complex by
incubation with purified CK2 and [g32P] ATP. Autophosphorylated CK2 Coomassie 00
subunits are indicated. (H) CK2 inhibition in HEK293T cells with 10 mM
CX-4945 or CX-5011 decreases eIF3d phosphorylation levels as assessed F 75 WTS528N WTS528D
by [32P]-orthophosphate labeling. The results in (A), (C), (G), and (H) are
representative of three independent experiments. 50 polysomes S529N S529D
25 WT
G CK2: - + -+
0 kDa ba*/c*
75 S528/529D 130 d/l b
50 e/f/g d
25 72 ih/m CK2
55 k
43 CK2

34
26

0 Coomassie 32P
75 H CX- CX-

50 DMSO 4945 5011
S528/529N

25 32P

0 Coomassie
Frac:1 2 3 4 5 6 7 8

Fig. 3. eIF3dTEV Subunit-Seq identifies cap-binding targets enriched in A HA Immunoprecipitation TEV Protease
regulators of cell metabolism. (A) Schematic of eIF3dTEV Subunit-Seq Cleavage
methodology. Position R131 within eIF3d was selected for TEV site insertion A254 TEV Remove non-eIF3d
to minimize perturbation of eIF3d function because this linker region exhibits protease site HA d –cap protected RNA
high phylogenetic variation and is naturally targeted by HIV-1 PR (3)
(fig. S1B). See text for details. (B) Immunoblot of eIF3dTEV Subunit-Seq +TEV + Cap-CLIP
samples. In, Input lysates; IP, anti-HA eIF3d immunoprecipitants; TEV-IP, + XRN-1
IP samples after on-bead incubation with TEV protease. Results are HA d eIF3
representative of three independent experiments. (C) Read mapping to RNA-Seq
respective transcripts from eIF3dTEV Subunit-Seq performed in glucose-
deprived cells. The schematics represent the transcript architecture: Thin B TEV- D Enrichment Score
lines are the 5′ and 3′ UTRs and thick lines are the open reading frames. The
annotated y-axis maximum is set equivalently for within each transcript for In IP IP 012345
input (In) and TEV-IP samples. (D) Gene ontology analysis of eIF3d cap- eIF3d
binding targets from glucose-deprived cells. (E) Phosphor image of SDS- full-length Protein phosphorylation
PAGE gel resolving RNase-protected 32P-cap-labeled 5′ UTRs. Recombinant cleaved Histone methylation
eIF3 complex was cross-linked to target RNAs in vitro and treated with Histone acetylation
RNase. Cap protection by eIF3d leads to the formation of a radiolabeled Actomyosin organization
covalent eIF3d-cap complex. HIV-1 PR cleavage was performed to validate eIF3b Ras signaling
cross-linked subunit identity as eIF3d, as shown through the shifted
migration of the cap-cross-linked complex. (F) In vitro eIF3d cap-cross-linking Integrin signaling
to Raptor and Larp1 5′ UTRs specifically requires a methylated cap Ubiquitination
structure. The competitor methylated cap analog m7GpppG (m7G) blocks rpS19 Cell cycle regulation
eIF3d cross-linking to the 5′ UTR, unlike the unmethylated cap analog
GpppG (G). The results in (E) and (F) are representative of three C Jun Rho signal transduction
independent experiments. Carbon metabolism
In 1720 Glycerolipid metabolism
MAPK/PI3K signaling
TEV-IP FoxO signaling

Cell response to heat
mTOR signaling
rpS19 15000 Insulin signaling Raptor
In
E Jun Larp1
TEV-IP

HIV-1 PR: - + - + - +

Raptor 27 d full-length
In cleaved

TEV-IP F Raptor

Larp1

Larp1 - m7G G - m7G G
In
200

TEV-IP d

Phosphorylation was reduced with single-site (S528N/S529N) mutations increased eIF3d measured Jun translation efficiency in cells
mutations and completely abolished in an 5′ cap-binding activity in cells (Fig. 2D and engineered to express eIF3d mutants exclu-
eIF3d S528/529D double mutant (Fig. 2C). The S2C). Conversely, phosphomimetic (S528D/ sively through stable knockdown and res-
eIF3d phosphorylation sites are in a location S529D) eIF3d mutations inhibited 5′ cap- cue with short hairpin (shRNA)–resistant
capable of influencing 5′ cap recognition, dependent Jun mRNA interactions (Fig. 2E) plasmids (fig. S2, D and E). Jun mRNA as-
proximal to the cap-binding domain (2) (Fig. and mirrored the reduced ability of eIF3d to sociation was lost in the polysome fractions
2B and fig. S2B). In support of a regulatory interact with Jun when cells were in nutrient- of cells expressing eIF3d S528D/S529D and
role of phosphorylation, phosphoinhibitory replete conditions (Fig. 1, D and E). We next gained in cells expressing eIF3d S528N/S529N

Lamper et al., Science 370, 853–856 (2020) 13 November 2020 2 of 4

RESEARCH | REPORT

(Fig. 2F and fig. S2F). Thus, phosphorylation and C; and table S1). In agreement with a binding competed away by the competitor
under nutrient-replete conditions represses starvation-induced increase in cap binding, ligand m7GpppG (Fig. 3F). Using selective
eIF3d cap binding, and loss of phosphoryl- ~70% of the eIF3d targets exhibited a two- 2′-hydroxyl acylation analyzed by primer
ation upon chronic glucose deprivation relieves fold or higher increase in eIF3d binding upon extension (SHAPE) RNA structure analysis,
this inhibition. glucose deprivation (fig. S4, D to F, and tables we identified a stem loop structure in the
S2 and S3). Gene ontology analysis of the Raptor 5′ UTR that is conserved with the
The eIF3d phosphorylation sites are each eIF3d targets identified an enrichment in cell characterized eIF3–Jun RNA-binding deter-
predicted substrates of the protein kinase metabolism and glucose homeostasis path- minant (4), including a U-rich loop, an upper
CK2, which recognizes the sequence motif ways, including mammalian target of rapa- C-rich internal bulge, and a multihelical junc-
S*/T*XXE/D with an enrichment in down- mycin (mTOR), mitogen-activated protein tion (fig. S5, A and B, and fig. S6). Disruption
stream acidic residues (fig. S3A) (7, 8). CK2 is kinase (MAPK)/phosphatidylinositol 3-kinase of this Raptor 5′ UTR stem loop by deletion
regulated by environmental stimuli (9), and (PI3K), and forkhead box O (FoxO) signal- or mutagenesis was sufficient to block eIF3d
CK2-substrate complex formation is inhibited ing (Fig. 3D). Indeed, eIF3d targets included cap binding (fig. S5C) and translation (fig.
in low-glucose conditions (10), in agreement Regulatory-associated protein of mTOR S5, D to F), suggesting that the presence of
with nutrient-dependent phosphorylation of (Raptor), an essential component of the mTOR conserved secondary structure may allow co-
eIF3d (Fig. 1E). We confirmed that CK2 was complex 1 (mTORC1) (15), and La-related pro- ordinated regulation of the eIF3d-directed
capable of modifying eIF3d in vitro in the tein 1 (Larp1), a protein that blocks translation gene program.
context of the full eIF3 complex (Fig. 2G). of 5′ terminal oligopyrimidine motif-containing
Treatment of cells with the CK2 active-site RNAs during mTOR inhibition (Fig. 3C) (16, 17). We next investigated whether up-regulation
inhibitors CX-4945 or CX-5011 (11) led to a We biochemically validated targets using puri- of the eIF3d-directed gene program through
loss of eIF3d phosphorylation (Fig. 2H). By fied recombinant eIF3 complex and in vitro the phosphorylation switch provides an adapt-
contrast, eIF3d phosphorylation was unaffected transcribed mRNAs (2) and observed that ive response to glucose deprivation. Lack of
by treatment with staurosporine (fig. S3B), a eIF3d bound to and protected the 32P-labeled glucose triggered increased cell death, rounded
broadly acting kinase inhibitor for which CK2 cap structure of Raptor and Larp1 5′ untrans- cell morphology, and poor attachment in cells
is atypically insensitive (12). Additionally, CK2 lated regions (UTRs) from RNase digestion in expressing a phosphomimetic eIF3d mutant
inhibition increased the ability of wild-type, vitro (Fig. 3E and fig. S4G). This interaction (Fig. 4, A and B, and fig. S7A). By contrast,
but not phosphomimetic, eIF3d to bind the 5′ required a 5′ methylated cap structure, with CK2 inhibition enhanced the cell survival
cap (fig. S3C), further demonstrating a role for phenotype in cells expressing wild-type, but
the CK2-dependent phosphorylation in regu-
lating eIF3d. A S528D S528N B – glucose

Activation of eIF3d cap-binding activity WT S529D S529N S528D S528N
during glucose deprivation suggests that an WT S529D S529N
eIF3d-controlled gene program may be critical 7.62 11 8.43
to supporting the cellular response to this
metabolic perturbation. However, because eIF3d CM
is an essential gene and 5′ cap recognition oc-
curs as part of the ~800-kDa 13-subunit eIF3 Cell count 24.5 65.6 33
complex (2, 5), technical restrictions have
prevented a comprehensive identification of D S528N S528D
eIF3d target mRNAs (2, 13, 14). To overcome S529N S529D
this challenge, we developed Subunit-Seq, a – glu
method that allows for the specific release
and isolation of an individual subunit domain Propidium iodide glucose: + – + –
and its associated mRNAs away from a larger Raptor
multisubunit complex. We created a cell line
expressing an engineered eIF3d protein with C S528/529D S528/529N p-mTOR
an internal tobacco etch virus (TEV) protease
cleavage site proximal to the cap-binding do- p-rpS6
main and a C-terminal hemagglutinin (HA)
affinity tag (Fig. 3A and fig. S4A). To perform rpS19
Subunit-Seq, we isolated HA-tagged eIF3 com-
plexes and used TEV protease to induce on- PSMB6 (%) Raptor (%) 100 12% 67% E Nutrient Chronic
bead cleavage and specific isolation of the eIF3d 80 glucose deprivation
cap-binding domain–RNA complexes (Fig. 3, A replete
and B). Contaminating RNAs that did not have 60 CK2
eIF3d-protected 5′ cap structures were removed
with on-bead pyrophosphatase decapping and 40 eIF3
5′ exoribonuclease digestion. This method- d
ology can be readily adapted to separate the 20 d eIF3••••••••••••••••••••••••••••••••••••••••••••••••••••••••
functions of other multisubunit RNA-binding
protein complexes through rational insertion 0
of TEV protease sites.
50 28% 25% 40S
eIF3dTEV Subunit-Seq revealed that eIF3d 40
binds the 5′ cap of 668 transcripts in glucose- 30
deprived HEK293T cells (Fig. 3C; fig. S4, B 20 eIF3d-specialized
10 eIF3d translation
0 cap-binding Cell survival

Frac: 1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8

Fig. 4. Regulation of eIF3d-specialized translation through a phosphorylation switch promotes cell
survival during chronic glucose deprivation. (A) Analysis of apoptotic cell percentage by propidium iodide
staining and flow cytometry in eIF3d phosphorylation mutant cell lines. Results in (A) are representative
of two independent experiments. (B) Representative images of the cell morphology after 48 hours of glucose
deprivation in eIF3d phosphorylation mutant cell lines. Results in (B) are representative of three independent
experiments. (C) Association of Raptor mRNA with ribosomal complexes in eIF3d mutant cell lines during glucose
deprivation. The percentage of total RNA associated with heavy polysomes (i.e., more than three ribosomes) is
indicated. Results are representative of two independent experiments, and ribosome association was determined
as in Fig. 1C. (D) Immunoblot of mTORC1 pathway activation in eIF3d phosphorylation mutant cell lines during
glucose deprivation. Results in (D) are representative of three independent experiments. (E) Model for
phosphorylation-mediated regulation of eIF3d-specialized translation during metabolic stress.

Lamper et al., Science 370, 853–856 (2020) 13 November 2020 3 of 4

RESEARCH | REPORT

not phosphomimetic, eIF3d (fig. S7B). Fur- uncover a parallel between stress-induced 15. D. H. Kim et al., Cell 110, 163–175 (2002).
thermore, dephosphorylated eIF3d was es- regulation of eIF3d and the essential canonical 16. S. Hong et al., eLife 6, e25237 (2017).
sential for the target mRNA Raptor to bypass translation factors eIF2 and eIF4E (1). During 17. R. M. Lahr et al., eLife 6, e24146 (2017).
starvation-induced global shutoff (Fig. 4C glucose deprivation, phosphorylation-regulated 18. G. G. Chiang, R. T. Abraham, J. Biol. Chem. 280, 25485–25490
and fig. S7, C and D), whereas phosphomimetic inhibition of eIF2 and eIF4E, through hypo-
eIF3d-expressing cells exhibited decreased phosphorylation and activation of the 4E-BPs, (2005).
Raptor protein levels (Fig. 4D). Raptor is crit- allows cells to block general protein synthesis. 19. L. Yu et al., Nature 465, 942–946 (2010).
ical for substrate recruitment and mTORC1 The conserved but opposing regulation of eIF4E 20. K. J. Condon, D. M. Sabatini, J. Cell Sci. 132, jcs222570 (2019).
kinase activity (15). Correspondingly, during and eIF3d in response to glucose deprivation 21. B. J. Guan et al., J. Biol. Chem. 289, 12593–12611 (2014).
glucose deprivation, incorrect control of eIF3d reveals that the balance of canonical versus
phosphorylation status led to decreased lev- noncanonical cap-dependent translation is a ACKNOWLEDGMENTS
els of activated mTOR and reduced phospho- key decision point for cellular adaptation to
rylation of rpS6, a downstream target of the metabolic state. We thank S. T. Aoki, K. Chat, P. Garrity, P. J. Kranzusch,
mTOR pathway (15, 18) (Fig. 4D). Although J. K. Nuñez, and members of the Lee laboratory for discussions
mTOR is rapidly inactivated within a few REFERENCES AND NOTES and S. R. Hildebrand, A. D. Yu, W. Xu, and R. Tomaino for
hours of nutrient deprivation, long-term star- experimental advice. Funding: This work was funded by the
vation reactivates it and promotes cell sur- 1. B. Liu, S. B. Qian, Wiley Interdiscip. Rev. RNA 5, 301–315 Charles H. Hood Foundation, the Searle Scholars Program, the
vival (19, 20). A protective response during (2014). Pew Biomedical Scholars Program, and a Sloan Research
chronic endoplasmic reticulum stress is also Fellowship (A.S.Y.L). R.H.F. was supported by NIH T32 GM007122.
dependent on partial rescue of mTOR activity 2. A. S. Lee, P. J. Kranzusch, J. A. Doudna, J. H. Cate, Nature 536, Author contributions: The project was conceived by A.S.Y.L, and
(13, 21). This response is blocked by shRNA- 96–99 (2016). experiments were performed by A.M.L., R.H.F., K.M.L., and A.S.Y.L.
mediated knockdown of eIF3d (13), which The manuscript was written by A.S.Y.L with editing from A.M.L.,
could be due to the inability of Raptor and the 3. S. Jäger et al., Nature 481, 365–370 (2011). R.H.F., and K.M.L., and all authors support the conclusions.
eIF3d-mediated gene program to be trans- 4. A. S. Lee, P. J. Kranzusch, J. H. Cate, Nature 522, 111–114 Competing interests: The authors declare no competing interests.
lated (fig. S7, E and F, and table S1). Data and materials availability: All sequencing data have been
(2015). deposited in the Gene Expression Omnibus under accession
Our results establish a model in which a 5. K. Asano et al., J. Biol. Chem. 272, 27042–27052 (1997). number GSE158249.
phosphorylation-regulated switch in eIF3d 6. Y. Hashem et al., Nature 503, 539–543 (2013).
cap-binding activity enables cellular control 7. F. Meggio, L. A. Pinna, FASEB J. 17, 349–368 (2003). SUPPLEMENTARY MATERIALS
of eIF3d-specialized translation in response 8. N. Blom, T. Sicheritz-Pontén, R. Gupta, S. Gammeltoft,
to metabolic state (Fig. 4E). These findings science.sciencemag.org/content/370/6518/853/suppl/DC1
S. Brunak, Proteomics 4, 1633–1649 (2004). Materials and Methods
9. O. Filhol, C. Cochet, Cell. Mol. Life Sci. 66, 1830–1839 (2009). Figs. S1 to S7
10. S. Welker et al., Int. J. Biochem. Cell Biol. 45, 2786–2795 Tables S1 to S3
References (22–42)
(2013). MDAR Reproducibility Checklist
11. R. Battistutta et al., Biochemistry 50, 8478–8488 (2011).
12. F. Meggio et al., Eur. J. Biochem. 234, 317–322 (1995). View/request a protocol for this paper from Bio-protocol.
13. B. J. Guan et al., Mol. Cell 68, 885–900.e6 (2017).
14. C. de la Parra et al., Nat. Commun. 9, 3068 (2018). 29 January 2020; resubmitted 20 August 2020
Accepted 24 September 2020
10.1126/science.abb0993

Lamper et al., Science 370, 853–856 (2020) 13 November 2020 4 of 4

Corrected 12 November 2020. See full text.

RESEARCH

CORONAVIRUS ACE2 rendered cells susceptible to infec-
tion (Fig. 1A). Although NRP1 did not promote
Neuropilin-1 facilitates SARS-CoV-2 cell entry infection in HEK-293T cells, its coexpres-
and infectivity sion with ACE2 and TMPRSS2 markedly
enhanced infection (Fig. 1, A and B). NRP1 ex-
Ludovico Cantuti-Castelvetri1,2*, Ravi Ojha3*, Liliana D. Pedro1,2*, Minou Djannatian1,2*, Jonas Franz4,5,6*, pression increased infection in Caco-2 cells,
Suvi Kuivanen7*, Franziska van der Meer4, Katri Kallio3, Tug˘berk Kaya1,2,8, Maria Anastasina3,9, which endogenously express ACE2 (12) (Fig. 1C
Teemu Smura7, Lev Levanov7, Leonora Szirovicza7, Allan Tobi10, Hannimari Kallio-Kokko11, and fig. S1D), showing that NRP1 can potentiate
Pamela Österlund12, Merja Joensuu13, Frédéric A. Meunier13, Sarah J. Butcher3,9, infection in the presence of other host factors.
Martin Sebastian Winkler14, Brit Mollenhauer15,16, Ari Helenius17, Ozgun Gokce8, To test the specificity of NRP1-dependent virus
Tambet Teesalu3,19,20, Jussi Hepojoki5,21, Olli Vapalahti7,11,22, Christine Stadelmann4, entry, we developed monoclonal antibodies
Giuseppe Balistreri3,18†, Mikael Simons1,2,23† (mAbs) that were designed to functionally block
the extracellular b1b2 domain of NRP1, which
The causative agent of coronavirus disease 2019 (COVID-19) is the severe acute respiratory syndrome is known to mediate binding to CendR peptides
coronavirus 2 (SARS-CoV-2). For many viruses, tissue tropism is determined by the availability of virus (13). The mAb3 was observed to bind to the re-
receptors and entry cofactors on the surface of host cells. In this study, we found that neuropilin-1 combinant b1b2 domain of wild-type (WT)
(NRP1), known to bind furin-cleaved substrates, significantly potentiates SARS-CoV-2 infectivity, an NRP1 but not to the triple-mutant b1b2 do-
effect blocked by a monoclonal blocking antibody against NRP1. A SARS-CoV-2 mutant with an altered main (S346A, E348A, and T349A in the CendR
furin cleavage site did not depend on NRP1 for infectivity. Pathological analysis of olfactory epithelium binding pocket) (fig. S2A). The potency of the
obtained from human COVID-19 autopsies revealed that SARS-CoV-2 infected NRP1-positive cells mAbs in preventing cellular binding and in-
facing the nasal cavity. Our data provide insight into SARS-CoV-2 cell infectivity and define a potential ternalization of NRP ligands was tested using
target for antiviral intervention. 80-nm silver nanoparticles (AgNP) coated with
the prototypic NRP1-binding CendR peptide
A n outbreak of severe acute respiratory receptor binding sites. Proteolytic cleavage RPARPAROH (9) (fig. S2B). mAb3 efficiently
syndrome coronavirus 2 (SARS-CoV-2) of RRAR^S by furin exposes a conserved C- blocked AgNP-CendR binding (fig. S2C) and
infections has caused a pandemic asso- terminal motif, RXXROH [where R is arginine internalization (fig. S2, D and E), whereas an-
ciated with a severe acute pulmonary and X is any amino acid; R can be substituted other monoclonal antibody, mAb2, had no effect
disease named COVID-19 (coronavirus by lysine (K)], in the S protein. Such C-terminal and was used as a control in further experiments.
disease 2019) (1). A related coronavirus, SARS- sequences that conform to the C-end rule (CendR) Treatment of HEK-293T with mAb3 significantly
CoV, led to a much smaller outbreak in 2003, are known to bind to and activate neuropilin reduced infection by SARS-CoV-2 pseudoviruses
possibly due to infection occurring predomi- (NRP1 and NRP2) receptors at the cell surface in cells expressing ACE2, TMPRSS2, and NRP1
nantly in the lower respiratory system, where- (9). Recent cryo–electron microscopy struc- (Fig. 1D), but not in cells expressing ACE2 and
as SARS-CoV-2 spreads rapidly through active tures of the SARS-CoV-2 S protein demon- TMPRSS2 only (fig. S2F). When SARS-CoV-2
pharyngeal viral shedding (2). Despite these strated that the S1-S2 junction is part of a pseudovirus was preincubated with recom-
differences, uptake of both viruses is mediated solvent-exposed loop and is therefore acces- binant, soluble extracellular b1b2 domain of
by the same cellular receptor: angiotensin- sible for receptor interactions (10, 11). NRP1, the wild type significantly reduced in-
converting enzyme 2 (ACE2) (3–5). One hy- fection compared with the triple mutant (Fig.
pothesis to explain the enhanced spreading To determine whether SARS-CoV-2 can use 1E and fig. S2G).
of SARS-CoV-2 is the presence of a polybasic NRP1 for virus entry and infectivity, we gen-
furin-type cleavage site, RRAR^S, at the S1-S2 erated lentiviral particles pseudotyped with Next, we explored the role of NRP1 using
junction in the SARS-CoV-2 spike (S) protein the SARS-CoV-2 S protein. Pseudoviruses are SARS-CoV-2 isolated from COVID-19 patients
that is absent in SARS-CoV (6). Similar sequences well suited for virus entry assays, as they allow from the Helsinki University Hospital. We used
are found in the S proteins of many other path- viral entry to be distinguished from other vi- WT SARS-CoV-2 and a cleavage-impaired SARS-
ogenic human viruses, including Ebola, HIV-1, rus life-cycle steps. Human embryonic kidney CoV-2 mutant that was isolated from Vero-E6
and highly virulent strains of avian influenza 293 T (HEK-293T) cells, which have almost no cells, which rapidly accumulate mutations at
(6, 7). The presence of the polybasic cleavage detectable ACE2 and NRP1 transcripts (fig. S1), the furin cleavage site of the S protein dur-
site in SARS-CoV-2 results in enhanced path- were transfected with plasmids that encode ing passaging (Fig. 2, A and B) (14). First, we
ogenicity by priming the fusion activity (8) and the two established host factors (4), human confirmed that furin cleaved the WT, but not
could potentially create additional cell surface ACE2 and the transmembrane protease serine the mutant, SARS-CoV-2 S protein by analyz-
2 (TMPRSS2), or NRP1. When expressed alone, ing S protein processing in Chinese hamster
ovary cells with functional (parental) or deficient
(FD11) furin enzyme (fig. S3) (15). Next, we

1Institute of Neuronal Cell Biology, Technical University Munich, Munich, Germany. 2German Center for Neurodegenerative Diseases (DZNE), Munich, Germany. 3Faculty of Biological and
Environmental Sciences, Molecular and Integrative Biosciences Research Program, University of Helsinki, Helsinki, Finland. 4Department of Neuropathology, University Medical Center Göttingen,
Göttingen, Germany. 5Campus Institute for Dynamics of Biological Networks, University of Göttingen, Göttingen, Germany. 6Max Planck Institute for Experimental Medicine, Göttingen, Germany.
7Department of Virology, Medicum, University of Helsinki, Helsinki, Finland. 8Institute for Stroke and Dementia Research (ISD), University Hospital, LMU Munich, Munich, Germany. 9Helsinki
Institute of Life Sciences–Institute of Biotechnology, University of Helsinki, Helsinki, Finland. 10Laboratory of Cancer Biology, Institute of Biomedicine and Translational Medicine, University of
Tartu, Tartu, Estonia. 11Department of Virology, University of Helsinki and Helsinki University Hospital, Helsinki, Finland. 12Department of Health Security, Finnish Institute for Health and Welfare
(THL), Helsinki, Finland. 13Clem Jones Centre for Ageing Dementia Research, Queensland Brain Institute, The University of Queensland, Brisbane, Queensland, Australia. 14Department of
Anesthesiology and Intensive Care Medicine, University Medical Center Göttingen, Göttingen, Germany. 15Department of Neurology, University Medical Center Göttingen, Göttingen, Germany.
16Paracelsus-Elena-Klinik Kassel, Kassel, Germany. 17Institute of Biochemistry, ETH Zürich, Zürich, Switzerland. 18The Queensland Brain Institute, The University of Queensland, Brisbane,
Queensland, Australia. 19Cancer Research Center, Sanford Burnham Prebys Medical Discovery Institute, La Jolla, CA, USA. 20Center for Nanomedicine and Department of Molecular, Cellular, and
Developmental Biology, University of California, Santa Barbara, Santa Barbara, CA, USA. 21Institute of Veterinary Pathology, Vetsuisse Faculty, University of Zürich, Zürich, Switzerland.
22Department of Veterinary Biosciences, University of Helsinki, Helsinki, Finland. 23Munich Cluster of Systems Neurology (SyNergy), Munich, Germany.

*These authors contributed equally to this work.

†Corresponding author. Email: [email protected] (G.B.); [email protected] (M.S.)

Cantuti-Castelvetri et al., Science 370, 856–860 (2020) 13 November 2020 1 of 5

Corrected 12 November 2020. See full text.

RESEARCH | REPORT

A control ACE2 NRP1 TMPRSS2 1.5 **** B GFP merge 6*
**** **** **
SARS-2 relative infection NRP1+ACE2+TSS2 ACE2+TSS2 relative infection
1.0 4

2
0.5

VSV-G 0.0 0
ctrl ACE2 NRP1 TSS2
ACE2 - + +
TSS2 - + +
NRP1 - - +

C SARS-2 3* D GFP merge 20 E GFP merge 1.2 **

control relative infection 2 control Ab infected cells [%] 15 mut b1b2 relative infection 1.0

* 0.8

10 0.6

NRP1 1 0.4 mAb3 wt b1b2
5
0.2

0 NRP1 0 0.0
ctrl ctrl mAb3 mut wt

b1b2 b1b2

Fig. 1. NRP1 facilitates the cellular entry of SARS-CoV-2 pseudotyped particles. vector. Data are normalized to the respective infectivity of SARS-2 and VSV-G
(A) Representative images and quantification of SARS-CoV-2 S protein (SARS-2) pseudotype in control cells. Two-way ANOVA was carried out with Sidak’s
(blue bars) and vesicular stomatitis virus glycoprotein (VSV-G) pseudotype (gray correction for multiple comparisons. (D and E) HEK-293T cells transiently
bars) infectivity in HEK-293T cells transiently expressing control (ctrl) vector, expressing NRP1, ACE2, and TMPRSS2 were inoculated with SARS-2 pseudotype
ACE2, NRP1, or TMPRSS2 (TSS2). Data are normalized to the respective in the presence of mAb3 antibody against NRP1 [(D), mAb3] or control mAb2
infectivity of SARS-2 and VSV-G pseudotype in ACE2-expressing cells. Two-way [(D), ctrl Ab] and in the presence of soluble NRP1 wild-type b1b2 domain
analysis of variance (ANOVA) was carried out with Tukey’s correction for multiple [(E), wt b1b2] or NRP1 mutant b1b2 domain [(E), mut b1b2]. Data in (E) are
comparisons. (B) HEK-293T cells transiently expressing ACE2 and TMPRSS2 normalized to untreated cells expressing NRP1, ACE2, and TMPRSS2. Two-tailed
or NRP1, ACE2, and TMPRSS2 were inoculated with SARS-2 pseudotype. Data are unpaired Student’s t test was performed. All data are represented as means ± SDs
normalized to SARS-2 infectivity in cells expressing ACE2 and TMPRSS2. One- from three independent experiments [(A) to (C)] or three biological replicates
way ANOVA was performed with Tukey’s correction for multiple comparisons. [(D) and (E)]. *P < 0.05, **P < 0.01, ****P < 0.0001. All images show GFP-positive,
(C) SARS-2 pseudotype infectivity in Caco-2 cells expressing NRP1 or control infected cells (magenta) and Hoechst stain (cyan). Scale bars, 100 mm.

validated that exogenous ACE2 expression ren- with TQTNSPRRAROH peptide or different NRP1 expression correlated with the detection
dered HEK-293T cells susceptible to infection control peptides, including one with a ter-
with SARS-CoV-2 (Fig. 2, C and D). NRP1 ex- minal amide group (TQTNSPRRARNH2), which of virus RNA in single-cell transcriptomes. For
pression alone caused lower levels of infection, reduces NRP1 binding (9) (Fig. 3A). We found these analyses, we used published single-cell
which were detectable only with increasing that AgNP-TQTNSPRRAROH, but not control
virus titer (Fig. 2, C and D). We then compared AgNPs, were efficiently taken up by HEK-293T RNA sequencing (scRNA-seq) datasets of cul-
the ability of WT and mutant SARS-CoV-2 to cells expressing NRP1 (Fig. 3, B and C). Next, we
infect HEK-293T that stably express ACE2; determined whether AgNP-TQTNSPRRAROH tured experimentally infected human bron-
ACE2 and TMPRSS2; or ACE2, TMPRSS2, and particles were also internalized into cells in vivo. chial epithelial cells and cells isolated from
NRP1. Infection of these cell lines by the WT, We chose to study nanoparticle entry in the
but not the mutant, virus increased in the pres- mouse olfactory epithelium, owing to the known bronchoalveolar lavage fluid (BALF) of severely
ence of NRP1, providing further evidence that expression of NRP1 in the olfactory system (16), affected COVID-19 patients (17). Among the
NRP1 requires a furin-cleaved substrate for its including olfactory neuronal cells of the epithe- proposed entry and amplification factors, NRP1,
effects (Fig. 2, E and F). We studied the effect of lium (fig. S4). AgNPs-TQTNSPRRAROH and con- FURIN, and TMPRSS11A were enriched in SARS-
the NRP1-blocking antibody, mAb3, on infec- trol AgNP-TQTNSPRRARNH2 were administered CoV-2–infected cells compared with noninfected
tion of Caco-2 cells by WT and mutant SARS- into the nose of anesthetized adult mice. Six cells (fig. S6). We also detected increased ex-
CoV-2 and found that preincubation with hours after administration, we observed a sig-
NRP1-blocking antibody reduced WT virus nificantly larger uptake of AgNP-TQTNSPRRAROH pression of these proteins after infection (fig.
infection by ~40%, whereas the control mAb2 than of AgNP-TQTNSPRRARNH2 into the ol- S6). In addition, RNA expression of NRP1 and
had no effect (Fig. 2, G and H). NRP1-blocking factory epithelium (Fig. 3, D and E) and, un- its homolog NRP2 was elevated in SARS-CoV-
antibody had no effect on the infection by the expectedly, into neurons and blood vessels of 2–positive cells compared with adjacent cells
mutated virus (Fig. 2, G and H). the cortex (Fig. 3, F and G). Similar results in the BALF of severely affected COVID-19 pa-
were obtained for AgNPs coated with the pro- tients (fig. S7).
Cleavage of SARS-CoV-2 S protein at the totypic NRP1-binding CendR peptide RPAR-
S1-S2 site generates the C-terminal end se- PAROH (fig. S5). Because the availability of virus receptors
quence TQTNSPRRAROH. To determine wheth- and entry cofactors on the surface of host cells
er this specific sequence can function as a Having obtained evidence for a role of NRP1
substrate for NRP1, we used AgNPs coated in cell entry of SARS-CoV-2, we examined whether determines infectivity, we compared the ex-
pression patterns of ACE2 and NRP1 in pub-
lished scRNA-seq datasets of human lung tissue
(18) and human olfactory epithelium (19).
Whereas ACE2 was detected at very low lev-
els, both NRP1 and NRP2 were abundantly ex-
pressed in almost all pulmonary and olfactory

Cantuti-Castelvetri et al., Science 370, 856–860 (2020) 13 November 2020 2 of 5

Corrected 12 November 2020. See full text.

RESEARCH | REPORT

Fig. 2. A blocking antibody against A B cwmtoutnStrASolARSR-S2-2§ §§

the b1b2 domain of NRP1 reduces

infection by wild-type SARS-CoV-2 Sequence Furin cleavage site %

(SARS-2-wt) but not a mutant with a NC_045512.2_Wuhan-Hu-1WT ASYQTQTNSPRRAR^SVA 100

deletion at the furin-cleavage site FIN-UH25M-Nasopharynx ASYQTQTNSPRRAR^SVA 100 180 S
(SARS-2-mut). (A) Sequence analysis of Finland/1/2020-Nasopharynx ASYQTQTNSPRRAR^SVA ND 130 S1
viruses isolated at different passages FIN-UH25M-Caco-2 P3 § ASYQTQTNSPRRAR^SVA 100 100 tubulin
(P) from different cell types. The first Finland/1/2020-VeroE6 P1 ASY - - - - - SPRRAR^SVA 53 *
sequence is the reference from the ****

Wuhan isolate (NC_045512.2). The Finland/1/2020-VeroE6 P3 ASY - - - - - SPRRAR^SVA 78 70 5.000

sequence abundance in each virus pop- Finland/1/2020-VeroE6 P7 §§ ASY - - - - - SPRRAR^SVA 100 55 **
ulation is indicated as a percentage.
wt mut
ND, not determined. §, SARS-2-wt; §§, C MOI 2.5 MOI 5 D SARS-2
SARS-2-mut. A, Ala; S, Ser; Y, Tyr; Q,
Gln; T, Thr; N, Asn; P, Pro; R, Arg; V, Val. 1.2 **** mAb3

(B) A deletion adjacent to the furin- NRP1 1.0

cleavage site abrogates the enzymatic relative infection 0.8 ****
cleavage of the S protein. Immunoblot
analysis was carried out on cell lysates ****
from Vero-E6 cells infected for 16 hours
0.6

****

with two viral populations (§ and §§). 0.4

Numbers indicate protein size (in kilo- ACE2 0.2
daltons). (C and D) Representative

images and quantification of SARS-2-wt 0.0

infectivity in HEK-293T cells stably mock 0.313 0.625 1.250 2.500
expressing ACE2 (blue bars) or NRP1
(orange bars) compared with nontrans- SARS-CoV-2 titer [MOI]

fected cells (gray bars). Different virus E ACE2 TSS2+ACE2 NRP1+TSS2+ACE2 F **
***
titers were used. Data are normalized to 3 ***

the infectivity in ACE2-expressing cells wt SARS-2§ relative infection

at multiplicity of infection (MOI) = 5. *
Two-way ANOVA was done with Tukey’s 2

correction for multiple comparisons.

(E and F) Representative images (E) and 1
quantification (F) of HEK-293T cells
mutant SARS-2§§
stably expressing the indicated combi-

nations of ACE2, TMPRSS2 (TSS2), and 0 wt mut wt mut
NRP1 after inoculation with SARS-2-wt wt mut
(wt; blue bars) or SARS-2-mut (mut; ACE2 ACE2 ACE2
gray bars). Data are normalized to the TSS2 TSS2
respective infectivity in ACE2-expressing NRP1

cells. Two-way ANOVA was performed G PBS control Ab mAb3 H
with Tukey’s correction for multiple
comparisons. (G and H) Caco-2 cell 1.5 **

infection in the presence of control wt SARS-2§ relative infection

mAb2 (ctrl. Ab) or mAb3 blocking anti- 1.0

bodies against NRP1 after SARS-2-wt

(wt; blue bars) or SARS-2-mut (mut; 0.5
gray bars) inoculation. Data are normal-
mutant SARS-2 §§
ized to the respective vehicle control

(phosphate-buffered saline) sample. 0.0

Two-way ANOVA was performed with wt mut wt mut wt mut
Tukey’s correction for multiple compar- control
isons. Data are means ± SDs from three SARS-2 SARS-2
independent experiments. *P < 0.05, PBS ctrl. Ab

**P < 0.01, ***P < 0.001, ****P < 0.0001. Magenta, SARS-2-wt– and SARS-2-mut–infected cells; Hoechst stain, cyan. Scale bars, 50 mm.

cells, with the highest levels of expression in of the human respiratory and olfactory epi- transcription factor 2 (OLIG2), which is mostly
endothelial cells (figs. S8 and S9). We con- thelium (fig. S10A). ACE2 was hardly de- expressed by olfactory neuronal progenitors
firmed these results by examining NRP1 im- tectable in these tissues (fig. S10B). Within (fig. S10C).
munoreactivity in human autopsy tissue and the olfactory epithelium, NRP1 was also ob-
detected NRP1 in the epithelial surface layer served in cells positive for oligodendrocyte In light of the widely reported disturbance
of olfaction in a large fraction of COVID-19

Cantuti-Castelvetri et al., Science 370, 856–860 (2020) 13 November 2020 3 of 5

Corrected 12 November 2020. See full text.

RESEARCH | REPORT

A B CoV2-pre C AgNP-positive cells [%] *** ***

SARS-2 S protein CoV2-Ala 100

S1 S2 80
...TQTNSPRRAR SVA...

60

nanoparticles SARS-2-Ala biot-X-TQTNSPRRAA-OH CoV2-post CoV2-post amide 40
SARS-2-pre biot-X-TQTNSPRRAR^SVA-OH 20
SARS-2-post biot-X-TQTNSPRRAR-OH
SARS-2-post amide biot-X-TQTNSPRRAR-NH2 0

biotin -Ala -pre -post -post
-amide

SARS-2

D AgNP NeuN + AQP4 merge E *** F AgNP NeuN + AQP4 merge G
2.5 0.8

AgNP-positive area [%] AgNP-positive area [%] *

control 2.0 control
0.6
SARS2 peptide SARS2 peptide
1.5
0.4

1.0

0.2
0.5

0.0 0.0
-post -post -post -post
-amide -amide
SARS-2 SARS-2

Fig. 3. NRP mediates entry of nanoparticles coated with SARS-CoV-2 cells (magenta) and Hoechst stain (cyan). One-way ANOVA was carried out with
(SARS-2) S–derived CendR peptides into cultured cells, olfactory epithe- Tukey’s correction for multiple comparisons. (D to G) Representative images and
lium, and the central nervous system of mice. (A) Peptide sequences used quantification of the main olfactory epithelium [(D) and (E), respectively] and
for AgNP coating. The peptides mimic SARS-2 S protein after furin cleavage cortex [(F) and (G), respectively] 6 hours after intranasal administration of
(post) and as controls; S protein before cleavage (pre), in which the terminal AgNPs coated with SARS2-post and SARS2-post amide peptides. n = 4 replicates
amino acid is replaced by alanine (Ala); or with an amide terminus (post for (C); n = 5 (E) and n = 4 (G) mice per condition. Data are means ± SDs.
amide). X, any amino acid. The arrow indicates the cleavage site. (B and C) Two-tailed unpaired Student’s t test; *P < 0.05, ***P < 0.001. Magenta,
Representative images and quantification of the internalization of peptide-coated AgNPs; cyan, Hoechst stain; green, NeuN (neuronal marker); yellow, AQP4.
AgNPs in HEK-293T cells expressing NRP1. Merged images show AgNP-positive Scale bars, 100 mm (B), 20 mm [(D) and (F)].

Fig. 4. SARS-CoV-2 patients (20, 21) and the enrichment of NRPs
infects the olfactory in the olfactory epithelium, we analyzed a series
epithelium. (A) Costain- of autopsies from six COVID-19 patients and
ing of S protein (brown) eight noninfected control patients to determine
and NRP1 (purple) in the whether SARS-CoV-2 could infect NRP1-positive
apical olfactory epithe- cells (Fig. 4 and table S1). Using antibodies
lium (OE) in a nonin- against the S protein, we detected infection of
fected control (left) and the olfactory epithelium in five of six COVID-19
in the apical OE (middle) patients. The infected olfactory epithelial cells
and adjacent mucosa showed high expression of NRP1 (Fig. 4, A
(right) in a COVID-19 and B). Additional costaining indicated in-
patient. LP, lamina fection of cells positive for OLIG2 (Fig. 4B
propria; HB, horizontal and fig. S11).
basal cells. (B) Costain-
ing of NRP1 (magenta) or There is limited knowledge about the virus–
OLIG-2 (magenta) with host interactions that determine cellular entry
S protein (yellow) in of SARS-CoV-2. Viruses display considerable
OE cells. Nuclei are redundancy and flexibility because they can
shown in blue. Scale exploit weak multivalent interactions to en-
bars, 10 mm. hance affinity. To date, studies of SARS-CoV-2
entry have focused almost entirely on ACE2,
which is expressed at very low protein levels
in respiratory and olfactory epithelial cells
(22). This raises the possibility that cofactors
are required to facilitate virus–host cell in-
teractions in cells with low ACE2 expression.
NRP1 could represent such an ACE2 poten-
tiating factor by promoting the interaction
of the virus with ACE2. The reason a number
of viruses (23–26) use NRPs as entry factors
may be their high expression on epithelia facing
the external environment and their function in

Cantuti-Castelvetri et al., Science 370, 856–860 (2020) 13 November 2020 4 of 5

Corrected 12 November 2020. See full text.

RESEARCH | REPORT

enabling cell, vascular, and tissue penetration 26. M. Raaben et al., Cell Host Microbe 22, 688–696.e5 (2017). and provided tools. L.C.C., R.O., L.D.P., M.D., J.F., S.K.,
(9, 13). T.K., C.S., T.S., M.J., F.A.M., S.J.B., J.H., and O.V. analyzed
ACKNOWLEDGMENTS the data or supervised data acquisition. L.C.C., R.O.,
REFERENCES AND NOTES L.D.P., M.D., J.F., S.K., T.K., and O.G. visualized the data.
We thank R. Müller, K. Schulz, and U. Scheidt for expert T.K. and O.G. performed the scRNA-seq data analysis.
1. N. Zhu et al., N. Engl. J. Med. 382, 727–733 (2020). technical assistance; S. Osborne for proofreading the M.S.W., B.M., C.S., and H.K.K. provided human samples.
2. R. Wölfel et al., Nature 581, 465–469 (2020). manuscript, and the DNA Dream Lab facility and K. Kogan for G.B. and M.S. wrote the manuscript. G.B. and M.S.
3. P. Zhou et al., Nature 579, 270–273 (2020). design and cloning of plasmids. Funding: The work in Munich supervised the project. Competing interests: T.T., O.V.,
4. M. Hoffmann et al., Cell 181, 271–280.e8 (2020). and Göttingen was supported by grants from the German and G.B. have a pending patent on the monoclonal
5. N. J. Matheson, P. J. Lehner, Science 369, 510–511 (2020). Research Foundation (SPP2191, TRR128-2, TRR274-1, antibody 3 (mAb3) against the NRP1 b1 domain for
6. B. Coutard et al., Antiviral Res. 176, 104742 (2020). SyNergy Excellence Cluster, EXC2145, Projekt ID390857198, SARS-CoV-2 inhibition. Data and materials availability: All
7. L. V. Tse, A. M. Hamilton, T. Friling, G. R. Whittaker, J. Virol. 88, EXC 2067/1- 390729940, and STA 1389/5-1), the ERC data needed to evaluate the conclusions in the paper are
(Consolidator Grant to M.S.), and the Dr. Miriam and present in the paper or the supplementary materials.
1673–1683 (2014). Sheldon G. Adelson Medical Research Foundation. The This work is licensed under a Creative Commons Attribution
8. M. Hoffmann, H. Kleine-Weber, S. Pöhlmann, Mol. Cell 78, work at the University of Helsinki was supported by the 4.0 International (CC BY 4.0) license, which permits
University of Helsinki and by donations from Finnish unrestricted use, distribution, and reproduction in any medium,
779–784.e5 (2020). colleagues to whom we are very grateful. The Academy of provided the original work is properly cited. To view a
9. T. Teesalu, K. N. Sugahara, V. R. Kotamraju, E. Ruoslahti, Finland supported G.B. (318434), O.V. (336490), S.J.B. copy of this license, visit https://creativecommons.org/
(315950 and 336471), and J.H. (1308613 and 1314119). licenses/by/4.0/. This license does not apply to figures/
Proc. Natl. Acad. Sci. U.S.A. 106, 16157–16162 (2009). O.V. was supported by Jane and Aatos Erkko Foundation, photos/artwork or other content included in the article
10. D. Wrapp et al., Science 367, 1260–1263 (2020). EU Horizon 2020 program VEO (874735), and Helsinki that is credited to a third party; obtain authorization from the
11. A. C. Walls et al., Cell 181, 281–292.e6 (2020). University Hospital Funds (TYH2018322). S.J.B. was rights holder before using such material.
12. K. Liao, D. Sikkema, C. Wang, T. N. Lee, J. Immunol. Methods supported by the Swedish Research Foundation and M.A.
by the Marie Sklodowska-Curie Actions (799929). SUPPLEMENTARY MATERIALS
389, 52–60 (2013). M.J. is supported by The Australian Research Council’s
13. A. Plein, A. Fantin, C. Ruhrberg, Microcirculation 21, 315–323 Discovery Early Career Researcher Award (DE190100565). science.sciencemag.org/content/370/6518/856/suppl/DC1
F.A.M. is supported by an Australian National Health and Materials and Methods
(2014). Medical Research Council Senior Research Fellowship Figs. S1 to S11
14. A. D. Davidson et al., Genome Med. 12, 68 (2020). (GNT1155794). T.T. and A.T. are supported by the European Table S1
15. N. G. Ravindra et al., bioRxiv 2020.05.06.081695 [Preprint]. Regional Development Fund (project 2014-2020.4.01.15-0012), References (27–37)
Wellcome Trust International Fellowship WT095077MA, MDAR Reproducibility Checklist
13 July 2020. European Research Council grant GLIOGUIDE, and the
16. A. Kawakami, T. Kitsukawa, S. Takagi, H. Fujisawa, J. Neurobiol. Estonian Research Council (grants PRG230 and EAG79 to T.T.). View/request a protocol for this paper from Bio-protocol.
Author contributions: G.B., M.S., and A.H. conceived the
29, 1–17 (1996). project. L.C.C., R.O., L.D.P., M.D., J.F., S.K., F.v.d.M., K.K., 12 June 2020; accepted 12 October 2020
17. M. Liao et al., Nat. Med. 26, 842–844 (2020). M.A., and L.S. designed and carried out experiments. A.T., Published online 20 October 2020
18. X. Han et al., Nature 581, 303–309 (2020). T.T., L.L., O.V., J.H., O.G., H.K.K., P.O., and M.J., developed 10.1126/science.abd2985
19. M. A. Durante et al., Nat. Neurosci. 23, 323–326 (2020).
20. L. Mao et al., JAMA Neurol. 77, 683–690 (2020).
21. K. W. Cooper et al., Neuron 107, 219–233 (2020).
22. F. Hikmet et al., Mol. Syst. Biol. 16, e9610 (2020).
23. D. Ghez et al., J. Virol. 80, 6844–6854 (2006).
24. N. Martinez-Martin et al., Cell 174, 1158–1171.e19 (2018).
25. H. B. Wang et al., Nat. Commun. 6, 6240 (2015).

Cantuti-Castelvetri et al., Science 370, 856–860 (2020) 13 November 2020 5 of 5

RESEARCH

CORONAVIRUS noma cell line endogenously expressing ACE2
and widely used in COVID-19 studies, the
Neuropilin-1 is a host factor for SARS-CoV-2 infection suppression of NRP1 expression by short
hairpin RNA (shRNA) greatly reduced SARS-
James L. Daly1*, Boris Simonetti1*†, Katja Klein2*, Kai-En Chen3‡, Maia Kavanagh Williamson2‡, CoV-2 infection at both 7 and 16 hpi, respec-
Carlos Antón-Plágaro1‡, Deborah K. Shoemark4, Lorena Simón-Gracia5, Michael Bauer6, tively, whereas that of vesicular stomatitis
Reka Hollandi7, Urs F. Greber6, Peter Horvath7,8, Richard B. Sessions1, Ari Helenius9, virus (VSV) pseudotyped with VSV-G was
Julian A. Hiscox10,11, Tambet Teesalu5, David A. Matthews2, Andrew D. Davidson2, Brett M. Collins3, unaffected (Fig. 1D and figs. S1H and S2A). To
Peter J. Cullen1†, Yohei Yamauchi2,12† determine if NRP1 was required for early
virus infection, we established a sequential
Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), the causative agent of coronavirus staining procedure using antibodies against
disease 2019 (COVID-19), uses the viral spike (S) protein for host cell attachment and entry. The host SARS-CoV-2 S and N proteins to distinguish
protease furin cleaves the full-length precursor S glycoprotein into two associated polypeptides: S1 and extracellular and intracellular viral particles
S2. Cleavage of S generates a polybasic Arg-Arg-Ala-Arg carboxyl-terminal sequence on S1, which (fig. S2B). Although NRP1 depletion did not
conforms to a C-end rule (CendR) motif that binds to cell surface neuropilin-1 (NRP1) and NRP2 affect SARS-CoV-2 binding to the Caco-2 cell
receptors. We used x-ray crystallography and biochemical approaches to show that the S1 CendR motif surface (Fig. 1E), virus uptake was halved in
directly bound NRP1. Blocking this interaction by RNA interference or selective inhibitors reduced SARS- NRP1-depleted cells compared to control cells
CoV-2 entry and infectivity in cell culture. NRP1 thus serves as a host factor for SARS-CoV-2 infection after 30 min of internalization (Fig. 1F). Thus,
and may potentially provide a therapeutic target for COVID-19. NRP1 enhances SARS-CoV-2 entry and infection.

S evere acute respiratory syndrome corona- motif, termed the “C-end rule” (CendR) (fig. We also observed that SARS-CoV-2–infected
virus 2 (SARS-CoV-2) is the coronavirus S1B) (8). CendR peptides bind to neuropilin-1 HeLawt+ACE2 cells displayed a multinucleated
(NRP1) and NRP2, transmembrane receptors syncytia cell pattern, as reported by others (Fig.
responsible for the current coronavirus that regulate pleiotropic biological processes, 1C) (5). Using an image analysis algorithm and
including axon guidance, angiogenesis, and supervised machine learning (fig. S2, C to F)
disease 2019 (COVID-19) pandemic (1, 2). vascular permeability (8–10). To explore the (11), we quantified syncytia of infected HeLawt+
A marked difference between the spike possibility that the SARS-CoV-2 S1 protein ACE2 and HeLaNRP1KO+ACE2 cells. At 16 hpi,
may associate with neuropilins, we generated the majority of HeLawt+ACE2 cells formed syn-
(S) protein of SARS-CoV-2 and SARS-CoV is the a green fluorescent protein (GFP)–tagged S1 cytia, whereas in HeLaNRP1KO+ACE2 cells, this
construct (GFP-S1) (fig. S1C). When expressed phenotype was reduced (fig. S2G). When in-
presence, in the former, of a polybasic sequence in human embryonic kidney 293T (HEK293T) fected with a SARS-CoV-2 isolate lacking the
cells engineered to express the SARS-CoV-2 furin cleavage site (SARS-CoV-2 DS1/S2) (fig.
motif, Arg-Arg-Ala-Arg (RRAR), at the S1/S2 receptor angiotensin-converting enzyme 2 S1A), the differences in infection and syncy-
(ACE2), GFP-S1 immunoprecipitated endog- tia formation were less pronounced (fig. S2,
boundary. It provides a cleavage site for a host enous NRP1 and ACE2 (Fig. 1A). We tran- H and I). However, a significant decrease in
siently coexpressed NRP1-mCherry and either infection of HeLaNRP1KO+ACE2 was still ob-
proprotein convertase, furin (3–5) (fig. S1A). GFP-S1 or GFP-S1 DRRAR (a deletion of the served at 16 hpi, indicating that NRP1 may
The resulting two proteins, S1 and S2, remain terminal 682RRAR685 residues) in HEK293T additionally influence infection through a
cells. NRP1 immunoprecipitated the S1 pro- CendR-independent mechanism (fig. S2H).
noncovalently associated, with the serine pro- tein, and deletion of the CendR motif re-
duced this association (Fig. 1B). Comparable The extracellular regions of NRP1 and NRP2
tease TMPRSS2 further priming S2 (6). Furin- binding was also observed with mCherry- are composed of two CUB domains (a1 and
mediated processing increases infectivity and NRP2, a receptor with high homology to a2), two coagulation factor domains (b1 and
NRP1 (fig. S1, D and E). In both cases, residual b2), and a MAM domain (9). Of these, the b1
affects the tropism of SARS-CoV-2, whereas binding was observed with the DRRAR mutant, domain contains the specific binding site for
indicating an additional CendR-independent CendR peptides (fig. S3A) (12). Accordingly,
furin inhibition diminishes SARS-CoV-2 entry, association between neuropilins and the S1 the mCherry-b1 domain of NRP1 immunopre-
protein. cipitated GFP-S1, and a shortened GFP-S1
and deletion of the polybasic site in the S pro- construct spanning residues 493 to 685 (figs.
To probe the functional relevance of this S1C and S3B). Isothermal titration calorime-
tein reduces syncytia formation in cell culture interaction, we generated HeLa wild-type and try (ITC) established that the b1 domain of
NRP1 knockout (KO) cell lines stably ex- NRP1 directly bound a synthetic S1 CendR
(3–5, 7). pressing ACE2, designated as HeLawt+ACE2 peptide (679NSPRRAR685) with an affinity of
The C terminus of the S1 protein generated and HeLaNRP1KO+ACE2, respectively (the level 20.3 mM at pH 7.5, which was enhanced to
of ACE2 expression was comparable between 13.0 mM at pH 5.5 (Fig. 2A). Binding was not
by furin cleavage has an amino acid sequence these lines) (fig. S1F). Using a clinical isolate observed to an S1 CendR peptide in which the
(682RRAR685) that conforms to a [R/K]XX[R/K] SARS-CoV-2 (SARS-CoV-2/human/Liverpool/ C-terminal arginine was mutated to alanine
REMRQ001/2020), we performed viral infec- (679NSPRRAA685) (Fig. 2A). We cocrystallized
1School of Biochemistry, Faculty of Life Sciences, Biomedical tion assays and fixed the cells at 6 and 16 hours the NRP1 b1 domain in complex with the S1
Sciences Building, University of Bristol, Bristol BS8 1TD, UK. postinfection (hpi). SARS-CoV-2 infection was CendR peptide (Fig. 2B). The resolved 2.35-Å
2School of Cellular and Molecular Medicine, Faculty of Life reduced in HeLaNRP1KO+ACE2 relative to structure revealed four molecules of b1 with
Sciences, Biomedical Sciences Building, University of Bristol, HeLawt+ACE2 (Fig. 1C). HeLa cells lacking electron density of the S1 CendR peptide
Bristol BS8 1TD, UK. 3Institute for Molecular Bioscience, the ACE2 expression were not infected (fig. S1G). clearly visible in the asymmetric unit (fig.
University of Queensland, St. Lucia, QLD 4072, Australia. In Caco-2 cells, a human colon adenocarci- S3C). S1 CendR peptide binding displayed
4School of Biochemistry and BrisSynBio Centre, Faculty of strong similarity to the previously solved
Life Sciences, Biomedical Sciences Building, University of structure of NRP1 b1 domain in complex with
Bristol, Bristol BS8 1TD, UK. 5Laboratory of Cancer Biology,
Institute of Biomedicine and Translational Medicine,
University of Tartu, Tartu, Estonia. 6Department of Molecular
Life Sciences, University of Zurich, Winterthurerstrasse 190,
8057 Zürich, Switzerland. 7Synthetic and Systems Biology
Unit, Biological Research Centre (BRC), Szeged, Hungary.
8Institute for Molecular Medicine Finland, University of
Helsinki, Helsinki, Finland. 9Institute of Biochemistry, ETH
Zurich, Zurich, Switzerland. 10Institute of Infection, Veterinary
and Ecological Sciences, University of Liverpool, Liverpool, UK.
11Singapore Immunology Network, Agency for Science,
Technology, and Research, 138648, Singapore. 12Division of
Biological Science, Graduate School of Science, Nagoya
University, Furo-cho, Chikusa-ku, Nagoya, 464-8601, Japan.
*These authors contributed equally to this work.
†Corresponding author. Email: [email protected] (B.S.); pete.
[email protected] (P.J.C.); [email protected] (Y.Y.)
‡These authors contributed equally to this work.

Daly et al., Science 370, 861–865 (2020) 13 November 2020 1 of 5

RESEARCH | REPORT

Fig. 1. NRP1 Interacts with S1 and enhances SARS-CoV-2 infection. tailed unpaired t test; P = 0.0005 and 0.00032. Scale bar, 500 mm. (E) Caco-2 shSCR
(A) HEK293T cells transduced to express ACE2 were transfected to express or shNRP1 cells were inoculated with a multiplicity of infection (MOI) = 50 of SARS-
GFP or GFP-tagged S1 and lysed after 24 hours. The lysates were subjected to CoV-2 and incubated in the cold for 60 min, and fixed. A two-step antibody staining
GFP-nanotrap, and the immune isolates were blotted for ACE2 and NRP1 (N = 3 procedure was performed with antibodies against S and N to distinguish external
independent experiments). (B) HEK293T cells were cotransfected to express GFP- (green) and total (red) virus particles, and the binding of particles per cell was quantified
tagged S1 or GFP-S1 DRRAR and mCherry or mCherry-tagged NRP1 and subjected to for >3300 particles per condition (N = 3 independent experiments). Two-tailed
GFP-nanotrap (N = 5 independent experiments). Two-tailed unpaired t test; P = unpaired t test; P = 0.6859. (F) Caco-2 shSCR or shNRP1 cells were bound with
0.0002. (C) HeLawt+ACE2 and HeLaNRP1 KO+ACE2 cells were infected with SARS-CoV-2. SARS-CoV-2 as in (E), followed by incubation at 37°C for 30 min. The cells were
Cells were fixed at 6 or 16 hpi and stained for N protein (magenta) and Hoechst (cyan), fixed and stained as in (E). Viral uptake was quantified for >4200 particles
and virus infectivity was quantified (N = 3 independent experiments). Two-tailed per condition (N = 3 independent experiments). Two-tailed unpaired t test; P =
unpaired t test; P = 0.00002 and 0.00088. Scale bar, 200 mm. (D) Caco-2 cells 0.00079. Scale bars [(E) and (F)], 10 mm and 200 nm (magnified panels). The
expressing shRNA against NRP1 or a nontargeting control (SCR) were infected with square regions were enlarged. The bars, error bars, and circles and triangles
SARS-CoV-2 and fixed at 7 or 16 hpi. The cells were stained for N protein (magenta) and represent the mean, SEM (B) and SD [(C) to (F)], and individual data points,
Hoechst (cyan), and infectivity was quantified (N = 3 independent experiments). Two- respectively. ***P < 0.001, ****P < 0.0001. ns, not signficant.

its endogenous ligand VEGF-A164 (Fig. 2B and GFP-S1493-685 immunoprecipitation by mCherry- and ACE2 expression levels were comparable
fig. S3D) (12). The key residues responsible for b1, confirming the critical role of the C-terminal and both constructs retained similar cell surface
contacting the C-terminal R685 of the CendR arginine (Fig. 2C). Mutagenesis of the T316 localization (fig. S3, G and H). SARS-CoV-2
peptide —Y297, W301, T316, D320, S346, T349 residue within the mCherry-b1 domain of infection was significantly enhanced in cells
and Y353—are almost identical between the NRP1 to arginine also reduced association expressing NRP1 wt-GFP compared to GFP
two structures (Fig. 2B and fig. S3D). The R682 with GFP-S1493-685, consistent with its inhib- control, whereas it was not enhanced in cells
itory impact on VEGF-A164 binding (12) (Fig. expressing the T316R mutant (Fig. 2E). Thus,
and R685 side chains together engage NRP1 2D). Accordingly, incubation of mCherry-b1 the SARS-CoV-2 S1 CendR and NRP1 interac-
via stacked cation-p interactions with NRP1 with VSV particles pseudotyped with trimeric tion promotes infection.
side chains of Y297 and Y353. By projecting S resulted in immunoprecipitation of processed
these findings onto the structure of the NRP1 forms of S1, which was dependent on the T316 To establish the functional relevance of the
residue (fig. S3F). Next, we transiently expressed S1 CendR-NRP1 interaction, we screened mono-
ectodomain, the b1 CendR binding pocket ap- either GFP, full-length NRP1 wt-GFP, or full clonal antibodies (mAb#1, mAb#2, mAb#3)
pears to be freely accessible to the S1 CendR length NRP1-GFP harboring the T316R muta- raised against the NRP1 b1b2 ectodomain.
peptide (fig. S3E) (13). tion in HeLaNRP1KO+ACE2 cells. GFP expression All three bound to the NRP1 b1b2 domain,
displayed staining by immunofluorescence
Site-directed mutagenesis of the S1 R685

residue to aspartic acid drastically reduced

Daly et al., Science 370, 861–865 (2020) 13 November 2020 2 of 5

RESEARCH | REPORT

Fig. 2. Molecular basis for CendR binding of SARS-CoV-2 S1 with NRP1. to mCherry-nanotrap (N = 5 independent experiments). Two-tailed unpaired
(A) Binding of NRP1 b1 with native (green line) and mutant (orange line) form t test; P < 0.0001. (D). HEK293T cells were cotransfected with combinations of
of S1 CendR peptide (corresponding to residues 679 to 685) by ITC at two GFP-tagged S1493-685 and mCherry, mCherry-NRP1 b1 or mCherry-NRP1 b1
different pH conditions (N = 3 independent experiments). All ITC graphs T316R mutant, and subjected to mCherry-nanotrap (N = 5 independent
represents the integrated and normalized data fit with 1-to-1 ratio binding. experiments). Two-tailed unpaired t test; P < 0.0001. (E) HeLaNRP1KO + ACE2
(B) (Left) NRP1 b1–S1 CendR peptide complex superposed with NRP1 b1– cells transfected with GFP, NRP1 wt-GFP, or NRP1 T316R-GFP constructs were
VEGF-A fusion complex (PDB ID: 4DEQ). Bound peptides are shown in stick infected 24 hours later with SARS-CoV-2. At 16 hpi, the cells were fixed and
representation. RMSD, root mean square deviation. (Right) Enlarged view stained for SARS-CoV-2-N, and viral infection was quantified in the GFP-positive
highlighting the binding of S1 CendR peptide b1. Key binding residues on subpopulation of cells (N = 3 independent experiments). The percentage of
b1 are shown in stick representation. Abbreviations for the amino acid residues are infection was normalized to that of GFP-transfected cells. Two-tailed unpaired
as follows: A, Ala; D, Asp; E, Glu; N, Asn; P, Pro; R, Arg; S, Ser; T, Thr; W, Trp; and t test; P = 0.002. The bars, error bars, and circles represent the mean, SEM
Y, Tyr. (C). HEK293T cells were cotransfected with combinations of GFP-tagged [(C) and (D)] and SD (E), and individual data points, respectively. **P < 0.01,
S1493-685 and S1493-685 R685D, and mCherry or mCherry-NRP1 b1, and subjected ****P < 0.0001. ns, not signficant.

in NRP1-expressing PPC-1 (human primary 3C). As a comparison, Caco-2 and Calu-3 cells at 7 and 16 hpi (Fig. 3F). Thus, the SARS-CoV-2
prostate cancer) cells but not in M21 (human were incubated with soluble ACE2, which in- interaction with NRP1 can be targeted to re-
melanoma) cells that do not express NRP1 hibited SARS-CoV-2 infection in both cases duce viral infectivity in relevant human cell
(fig. S4A) (8), and stained the extracellular (fig. S4C). lines (fig. S5).
domain of NRP1-GFP expressed in cells (fig.
S4B). Of these antibodies, mAb#3, and to a Next, we turned to the small molecule Cell entry of SARS-CoV-2 depends on prim-
lesser extent mAb#1, bound to the CendR- EG00229, a selective NRP1 antagonist that ing by host cell proteases (5, 6, 15). Our data
binding pocket with high specificity, as de- binds the b1 CendR binding pocket and in- indicate that a component of SARS-CoV-2 S
fined by reduced ability to bind to a b1b2 hibits VEGF-A binding (Fig. 3D) (14). ITC protein binding to cell surface neuropilins oc-
mutant that targets residues (S346, E348, established that EG00229 bound to the NRP1 curs via the S1 CendR motif generated by the
T349) at the opening of the binding pocket b1 domain with a dissociation constant (Kd) furin cleavage of S1/S2. Though not affecting
(Fig. 3A) (12). Incubation of Caco-2 cells with of 5.1 and 11.0 mM at pH 7.5 and 5.5, respec- cell surface attachment, this interaction promotes
mAbs#1 and 3 reduced SARS-CoV-2 infection tively (Fig. 3E). EG00229 inhibited the direct entry and infection by SARS-CoV-2 in physiolog-
compared to a control mAb targeting avian binding between b1 and the S1 CendR pep- ically relevant cell lines widely used in the study
influenza A virus (H11N3) hemagglutinin (Fig. tide, and the immunoprecipitation of GFP- of COVID-19. The molecular basis for the effect is
3B). Consistent with this, mAb#3 inhibited S1493-685 by mCherry-b1 (Fig. 3E and fig. S4D). unclear, but neuropilins are known to mediate
binding of GFP-S1493-685 and mCherry-b1 (Fig. Finally, incubation of Caco-2 cells with EG00229 the internalization of CendR ligands through an
reduced the efficiency of SARS-CoV-2 infection endocytic process resembling macropinocytosis,

Daly et al., Science 370, 861–865 (2020) 13 November 2020 3 of 5

RESEARCH | REPORT B C

A

DE F

Fig. 3. Selective inhibition of the S1-NRP1 interaction reduces SARS-CoV-2 b1–S1 CendR peptide complex superimposed with NRP1 b1–EG00229 inhibitor
infection. (A) Enzyme-linked immunosorbent assay of anti-NRP1 monoclonal complex (PDB ID:3I97). Key binding residues on b1, bound peptides, and EG00229
antibodies (mAb#1, mAb#2, mAb#3) at 3 mg/ml using plates coated with are shown in stick representation. (E) ITC analysis of EG00229 binding to b1 domain
NRP1 b1b2 wild type, b1b2 mutant (S346A, E348A, T349A), or bovine serum of NRP1 at two different pH conditions. Preincubation with EG00229 blocks S1
albumin (BSA), used as a control (N = 3 independent experiments). Binding is CendR peptide binding (orange line), and the CendR peptide can reduce binding of
represented as arbitrary units of absorbance at 655 nm. Two-tailed unpaired EG00229 (green line) (N = 3 independent experiments). All ITC graphs represent
t test; P = 0.0207, 0.2430, 0.0007. (B) Cells were first treated with anti-H11N3 the integrated and normalized data fit with 1-to-1 ratio binding. (F). Cells were
(100 mg/ml) (Ctrl) mAb, mAb#1, mAb#2, or mAb#3 for 1 hour before infection first treated with 100 mM EG00229 or dimethyl sulfoxide before infection with
with SARS-CoV-2. Cells were fixed at 16 hpi and stained for N protein (magenta) SARS-CoV-2. Cells were fixed at 7 and 16 hpi and stained for N protein (magenta)
and Hoechst (cyan) (N = 3 independent experiments). Two-tailed unpaired t test; and Hoechst (cyan) (N = 3 independent experiments). The square regions were
P = 0.015, 0.36, 0.0003. Scale bar, 500 mm. (C) HEK293T cells were cotransfected enlarged. Scale bars, 500 mm and 100 mm (magnified panels). Two-tailed unpaired
with combinations of mCherry or mCherry-b1 and GFP-tagged S1493-685 and t test; P = 0.0059 and 0.0013. The bars, error bars, and circles and triangles
subjected to mCherry-nanotrap with or without coincubation with mAb#3 represent the mean, SEM (C) and SD [(A), (B), and (F)], and individual data
(N = 3 independent experiments). Two-tailed unpaired t test; P = 0.0143. (D) NRP1 points, respectively. *P < 0.05, **P < 0.01, ***P < 0.001.

(8, 16, 17). Notably, gene expression analysis has 2. E. Dong, H. Du, L. Gardner, Lancet Infect. Dis. 20, 533–534 (2020). 18. M. Ackermann et al., N. Engl. J. Med. 383, 120–128 (2020).
revealed an up-regulation of NRP1 and NRP2 in 3. D. Wrapp et al., Science 367, 1260–1263 (2020). 19. S.-Y. Lau et al., Emerg. Microbes Infect. 9, 837–842 (2020).
lung tissue from COVID-19 patients (18). A SARS- 4. A. C. Walls et al., Cell 181, 281–292.e6 (2020).
CoV-2 virus with a natural deletion of the S1/S2 5. M. Hoffmann, H. Kleine-Weber, S. Pöhlmann, Mol. Cell 78, ACKNOWLEDGMENTS
furin cleavage site demonstrated attenuated We thank the Bristol Synthetic Biology Centre and the Advanced
pathogenicity in hamster models (19). NRP1 779–784.e5 (2020). Computing Research Centre for provision of HPC (Bluegem), and
binding to the CendR peptide in S1 is thus 6. M. Hoffmann et al., Cell 181, 271–280.e8 (2020). the University of Bristol Wolfson Bioimaging Facility. We thank the
likely to play a role in the increased infectivity 7. J. Shang et al., Proc. Natl. Acad. Sci. U.S.A. 117, 11727–11734 (2020). University of Queensland Remote Operation Crystallisation and
of SARS-CoV-2 compared with SARS-CoV. 8. T. Teesalu, K. N. Sugahara, V. R. Kotamraju, E. Ruoslahti, Proc. X-ray facility (UQ-ROCX) and the staff for their support with the
The ability to target this specific interaction crystallization experiments, and the staff of the Australian
may provide a route for COVID-19 therapies. Natl. Acad. Sci. U.S.A. 106, 16157–16162 (2009). Synchrotron for assistance with x-ray diffraction data collection.
Funding: J.L.D. was supported by a Wellcome Trust studentship
REFERENCES AND NOTES 9. H. F. Guo, C. W. Vander Kooi, J. Biol. Chem. 290, 29120–29126 from the Dynamic Molecular Cell Biology Ph.D. program (203959/
1. WHO Coronavirus disease, 2019 (COVID-19) Weekly (2015). Z/16/Z), C.A.P. was supported by Beca Fundación Ramón Areces
Estudios Postdoctorales en el Extranjero, and M.K.W. was
Epidemiological Update – 31 August 2020. https://www.who. 10. A. Plein, A. Fantin, C. Ruhrberg, Microcirculation 21, 315–323 (2014). supported by an MRC grant (MR/R020566/1) awarded to A.D.D.
int/docs/default-source/coronaviruse/situation-reports/ 11. R. Hollandi et al., Cell Syst. 10, 453–458.e6 (2020). This project has received funding from the MRC (MR/P018807/1),
20200831-weekly-epi-update-3.pdf?sfvrsn=d7032a2a_4 12. M. W. Parker, P. Xu, X. Li, C. W. Vander Kooi, J. Biol. Chem. Wellcome Trust (104568/Z/14/2), Lister Institute of Preventive
Medicine, and Elizabeth Blackwell Institute for Health Research
287, 11082–11089 (2012). Rapid Response Call (COVID-19) awarded to P.J.C, the European
13. B. J. Janssen et al., Nat. Struct. Mol. Biol. 19, 1293–1299 (2012). Research Council under the European Union’s Horizon 2020
14. A. Jarvis et al., J. Med. Chem. 53, 2215–2226 (2010).
15. J. K. Millet, G. R. Whittaker, Virology 517, 3–8 (2018).
16. M. Simons, E. Gordon, L. Claesson-Welsh, Nat. Rev. Mol. Cell

Biol. 17, 611–625 (2016).
17. H. B. Pang et al., Nat. Commun. 5, 4904 (2014).

Daly et al., Science 370, 861–865 (2020) 13 November 2020 4 of 5

RESEARCH | REPORT

research and innovation program (No 856581 - CHUbVi), and A.D.D. isolated SARS-CoV-2 strains used for the work. K.C., C.A.P., which permits unrestricted use, distribution, and reproduction in
from MRC-AMED (MR/T028769/1) awarded to Y.Y., the Swiss M.B., L.S.G., U.F.G., K.K., R.B.S., D.K.S., J.A.H., and T.T. did any medium, provided the original work is properly cited. To
National Science Foundation and Kanton Zurich awarded to experimental work and/or provided essential reagents. R.H. and view a copy of this license, visit https://creativecommons.org/
U.F.G. B.M.C. is supported by an Australian National Health P.H. performed image analysis. B.S., A.D.D., B.M.C., P.J.C., licenses/by/4.0/. This license does not apply to figures/
and Medical Research Council (NHMRC) Senior Research and Y.Y. supervised the research. J.L.D., B.S., A.D.D., P.J.C., and photos/artwork or other content included in the article that is
Fellowship (APP1136021) and Project Grant (APP1156493), and Y.Y. wrote the manuscript and made the figures. All authors credited to a third party; obtain authorization from the rights
the United States Food and Drug Administration grant no. read and approved the final manuscript. Competing interests: holder before using such material.
HHSF223201510104C “Ebola Virus Disease: correlates of T.T. is an inventor of patents on CendR peptides and a shareholder
protection, determinants of outcome and clinical management” of Cend Therapeutics Inc., a company that holds a license for SUPPLEMENTARY MATERIALS
amended to incorporate urgent COVID-19 studies awarded the CendR peptides and is developing the peptides for cancer science.sciencemag.org/content/370/6518/861/suppl/DC1
to J.A.H., A.D.D., and D.A.M. R.H. and P.H. acknowledge support therapy. J.A.H. is a member of the Department of Health, Materials and Methods
from the LENDULET-BIOMAG Grant (2018-342), from H2020- New and Emerging Respiratory Virus Threats Advisory Group Figs. S1 to S5
discovAIR (874656), and from Chan Zuckerberg Initiative, Seed (NERVTAG) and the Department of Health, Testing Advisory Tables S1 to S3
Networks for the HCA-DVP. T.T. was supported by the European Group. U.F.G. is a consultant to F. Hoffmann–La Roche Ltd, References (20–33)
Regional Development Fund (Project no. 2014-2020.4.01.15-0012), Switzerland. All other authors declare no competing interests. MDAR Reproducibility Checklist
by European Research Council grant GLIOGUIDE and Estonian Data and materials availability: Coordinates and structure
Research Council (grants PRG230 and EAG79, to T.T.). Author factors for the NRP1 b1-S1 CendR peptide complex have been View/request a protocol for this paper from Bio-protocol.
contributions: J.L.D., B.S., A.H., P.J.C., and Y.Y. conceived the deposited at the Protein Data Bank (PDB) with accession code
study. J.L.D., B.S., K.K, and Y.Y. performed most of the 7JJC. All other data are available in the manuscript or the 14 June 2020; accepted 12 October 2020
experiments. K.K., M.K.W., D.A.M., and A.D.D. performed all work supplementary materials. This work is licensed under a Creative Published online 20 October 2020
with infectious SARS-CoV-2 supervised by A.D.D. M.K.W. and Commons Attribution 4.0 International (CC BY 4.0) license, 10.1126/science.abd3072

Daly et al., Science 370, 861–865 (2020) 13 November 2020 5 of 5

RESEARCH

PREBIOTIC CHEMISTRY action of GCA could yield serine under pre-

Prebiotic synthesis of cysteine peptides that catalyze biotic conditions, the Strecker reaction of BMA
peptide ligation in neutral water would not play a role in prebiotic cysteine

Callum S. Foden*, Saidul Islam*, Christian Fernández-García, Leonardo Maugeri, synthesis. To overcome this inherent sta-
Tom D. Sheppard, Matthew W. Powner† bility problem, we considered an alternative

Peptide biosynthesis is performed by ribosomes and several other classes of enzymes, but a simple biomimetic pathway for cysteine synthesis.
chemical synthesis may have created the first peptides at the origins of life. a-Aminonitriles—prebiotic We suspected that clues to the prebiotic
a–amino acid precursors—are generally produced by Strecker reactions. However, cysteine’s
aminothiol is incompatible with nitriles. Consequently, cysteine nitrile is not stable, and cysteine has synthesis of cysteine might remain embedded
been proposed to be a product of evolution, not prebiotic chemistry. We now report a high-yielding, within the extant biological pathway and that
prebiotic synthesis of cysteine peptides. Our biomimetic pathway converts serine to cysteine by
nitrile-activated dehydroalanine synthesis. We also demonstrate that N-acylcysteines catalyze peptide GCA, rather than BMA, was the key prebiotic
ligation, directly coupling kinetically stable—but energy-rich—a-amidonitriles to proteinogenic amines. precursor of cysteine.
This rare example of selective and efficient organocatalysis in water implicates cysteine as both
catalyst and precursor in prebiotic peptide synthesis. The principal mechanism by which reduced
inorganic sulfur is incorporated into bioor-
P eptides and proteins are essential to all functions in catalysis, redox sensing, and elec-
life on Earth, and their biosynthesis is tron transfer, as well as being an essential lig- ganic compounds is through cysteine biosyn-
achieved by a highly evolved system of and in ancient iron-sulfur proteins (11, 13, 14). thesis (17, 18). In plants and various archaea
enzyme catalysts (1, 2). Although the It therefore seems almost inconceivable that and bacterial species, cysteine biosynthesis be-
origin of protein synthesis in biology re- cysteinyl thiols were not present during the
mains a mystery, life’s exploitation of peptides development of nascent biological processes gins with the enzymatic conversion of serine
may have predated the evolution of the com- on early Earth, and yet this is not the prevail- to O-acetylserine (SerAc) [or O-phosphoserine
plex enzymes that are now required to co- ing view (15–18). Numerous, unsuccessful at- (Sep)] followed by a pyridoxal-5′-phosphate
ordinate biosynthesis. Simple chemical processes tempts to synthesize and isolate cysteines under (PLP)–dependent acetic acid b-elimination,
could have furnished prebiotic peptides and prebiotically plausible conditions (7, 19–21) have followed by sulfide b-addition before disasso-
peptide catalysts (3). To elucidate such chem- led to a widely held belief that cysteine is a ciation of the cysteine-enzyme complex (Fig. 1A).
ical mechanisms, it is essential that we reflect biological invention (17, 18), as well as a late Throughout this sequence, the a-amine re-
on the biochemical strategies of extant biology addition to the genetic code (15–18). Here, we mains covalently bound to PLP to promote
to inform the systematic evaluation of prebiotic report a high-yielding prebiotic synthesis of
chemistry. For example, we recently demon- cysteines and demonstrate that these cysteine elimination and to prevent the release and
strated that a-peptide synthesis in water could peptides catalyze nonenzymatic CPL in neu- rapid decomposition of highly unstable de-
be achieved by H2S-mediated stoichiometric tral water. Our results support the hypothesis hydroalanine (Dha) (26). The nitrile equivalent
ligation (4). Our synthesis exploited a biomi- that cysteine was available at the origins of life of Dha, Dha nitrile (Dha-CN), has been syn-
metic N- to C-terminal chain-growth mech- as a secondary product of serine nitrile (Ser-CN)
anism. This overcame long-standing problems synthesis and that cysteines would have been thesized by Eschenmoser and co-workers
that had prohibited the coupling of a-amino- a cornerstone of early catalytic activity. but was found to be extremely unstable and
nitriles (5–7) and avoided the irrevocable side-
chain modifications caused by electrophilic The Strecker synthesis of aminonitriles is failed to react with H2S to produce Cys-CN,
agents required to activate amino acids (8, 9). widely believed to play an important role in instead undergoing rapid degradation, even
Further reflection on the deep-seated role of the prebiotic origins of amino acids, and we under anhydrous conditions (27). However,
thiols in nonribosomal peptide synthesis and had previously identified glycolaldehyde (GCA) we have recently shown that N-acylation of
core metabolism (10, 11) has now led us to hy- and b-mercaptoacetaldehyde (BMA) as key a-aminonitriles is a crucial element in initiat-
pothesize that cysteine may have originated nodes in the chemical network required for ing peptide synthesis in water, and it prevents
as a secondary product of sulfide-mediated abiogenesis of RNA and peptides (3, 22, 23),
peptide synthesis. We suspected that cysteine and specifically BMA as a Strecker precursor hydantoin-, diketopiperazine-, and imidazole-
could be used to deliver catalytic peptide liga- of cysteine (22). However, although Strecker induced peptide degradation (4–6). Sim-
tion (CPL) in water and a rare example of reactions are generally highly efficient (23, 24), ilarly, N-acylation would stabilize Dha-CN.
prebiotic synthesis directly generating cata- BMA forms intractable and insoluble mixtures N-acylation would simultaneously prevent the
lytic function. in Strecker reactions (fig. S1) (7). This is in highly favorable, but unwanted, enamine-to-imine
stark contrast to GCA, which undergoes the tautomerization that precludes sulfur addi-
Cysteine is the primary organic source of Strecker reaction in excellent yield to afford tion to Dha (27) and prevent the degradation
sulfide in biology and the feedstock for es- a stable aminonitrile product, Ser-CN (fig. S2) of Cys-CN that is brought about by its free
sential cofactors such as glutathione and co- (23, 24). The observed disparity between con- a-amine (7). Consequently, we recognized that
enzyme A (CoA) (12). It is also an important geners GCA and BMA is likely due to the rapid serine diacylation presents a simple biomi-
residue within enzyme active sites, with vital reaction of b-aminothiols with nitriles in water
(25), and because cysteine nitrile (Cys-CN) metic strategy for prebiotic Dha synthesis,
Department of Chemistry, University College London, 20 is both a b-aminothiol and an a-aminonitrile, and we identified N,O-diacetyl-serine nitrile
Gordon Street, London WC1H 0AJ, UK. it is inherently unstable (7). These observations (Ac-SerAc- CN) as a key intermediate for pre-
*These authors contributed equally to this work. suggested to us that whereas the Strecker re-
†Corresponding author. Email: [email protected] biotic cysteine synthesis (Fig. 1B). In addition
to the electron-withdrawing effects of the
a-nitrile, N,O-diacetylation would further
enhance the acidity of the a-proton of Ac-SerAc-
CN as well as activate the serine hydroxyl moiety
as a leaving group. We envisaged that these

combined effects would promote Ac-Dha-CN
synthesis at neutral pH, without recourse to the

highly alkaline (pH > 13) conditions typically
required for Dha formation in water (28)
that would also promote peptide degrada-
tion (29).

We have a long-standing interest in (pre-
biotic) acylation (30, 31), but it was not clear if
the hydroxyl moiety of Ser-CN could be

Foden et al., Science 370, 865–869 (2020) 13 November 2020 1 of 5

RESEARCH | REPORT

selectively acetylated in water. Therefore, (Fig. 1B and fig. S36). We next investigated tral pH (4). To this end, we incubated Ac-Dha-

we were pleased to observe chemoselective the prebiotic acetylating agent, AcSH, as a CN, AcSH, and Gly-CN in phosphate buffer at

N,O-acetylation of Ser-CN with thioacetic more water-soluble sulfide source at pH 7. pH 7 and room temperature. We observed in

acid (AcSH) and ferricyanide (4, 30) to Incubation of Ac-Dha-CN with AcSH led to situ thioester formation and near-quantitative
produce Ac-SerAc-CN in up to 91% yield within quantitative thioester formation after 12 hours acetyl transfer from Ac-CysAc-CN to Gly-CN,

1 hour at room temperature (supplementary in phosphate buffer (Fig. 1B and fig. S38). producing Ac-Gly-CN (81%) and Ac-Gly-SNH2
materials). Acetylation of Ser-CN with N- The cysteine residue was then rapidly liberated (13%) after 3 days (fig. S67).

acetylimidazole (NAI) (30, 31) was equally by ammonolysis yielding Ac-Cys-CN (77%; Prebiotic syntheses, much like biosyntheses,
effective for Ac-SerAc-CN synthesis in neu- Fig. 1B and fig. S44) or thiolysis to yield Ac- are necessarily multistep chemical pathways

tral water. We initially observed O-acetylser- Cys-SNH2 in 95% yield (Fig. 1B and fig. and, therefore, like all multistep processes, are
ine nitrile (SerAc-CN) as the major product S39). Prolonged incubation of Ac-CysAc-CN susceptible to diminished overall yields unless

(<10 min) with NAI (fig. S8) and attribute the with H2S gradually yielded cysteine thioacid the individual reactions proceed with high
nucleophilicity of the b-hydroxyl of Ser-CN to efficiency (3). In biosynthesis, enzyme catal-
the pronounced electron-withdrawing effect of Ac-Cys-SH and amide Ac-Cys-NH2 (2:1) (figs. ysis achieves exquisite selectivity and high
S47 and S48). High-yielding Ser-CN→CysAc-

the a-nitrile and the low basicity (pKaH) of CN conversion was also observed for N- yields, but modern enzymes are a product of
this amino alcohol (supplementary materials). acetylvalinylserine nitrile (Ac-Val-Ser-CN) billions of years of evolution. In the absence of

We also observed the formation of a stable (supplementary materials), demonstrating the enzymes, prebiotic chemistry had to initially

Dha, Ac-Dha-CN, at near-neutral pH for the efficacy of cysteine synthesis within a sterically exploit unevolved and directly accessible al-

first time during these acetylation reactions. encumbered peptide substrate. ternatives, such as organocatalysts (34). Small-

Optimal formation of Ac-Dha-CN from Ac- Acetyl-CoA is biosynthesized from cysteine molecule catalysts could have functioned as
SerAc-CN occurred at pH 8, and after 4 days at and is a universally conserved acetylating agent rudimentary “enzymes” at the origins of life, and

room temperature, an 85% conversion to Ac- for protein, carbohydrate, and lipid metabo- although they have been highly sought after since

Dha-CN was observed (Fig. 1B and fig. S30). lism. The relative simplicity of thioester Ac- the watershed rediscovery of proline catalysis,
CysAc-CN suggests it may have been exploited progress has been greatly hampered by the gen-
This elimination is all the more notable be- erally poor activity of organocatalysts in water
cause N,O-diacetylserinamide (Ac-SerAc-NH2) as an acetyl-CoA analog and an activated source (35). However, the importance of nitriles at the
underwent near-exclusive hydrolysis at pH 8 origins of life (3), as well as their low background
(figs. S34 and S35) (32). The switch in reac- of acetate in (proto)metabolism (10), so we veri-
fied the proficiency of Ac-CysAc-CN as a source

tivity between a-nitrile (acetate elimination) of activated acetate. We suspected that acetyl reactivity, suggested they may be ideal substrates
and a-amide (acetate hydrolysis) demonstrates transfer from Ac-CysAc-CN to a-aminonitriles for a highly selective and high-yielding CPL in

the benefits of a-nitrile activation for acetate (AA-CN) would be promoted by their low water and warranted further investigation.

elimination and indicates that Ac-Dha-CN is pKaH (e.g., Gly-CN pKaH = 5.4), which renders Peptide fragment ligations, such as native

predisposed to form in near-neutral water AA-CN neutral and highly nucleophilic at neu- chemical ligation, are important reactions in

(28, 29).

We next investigated Ac-Dha-

CN synthesis via phosphate elim-

ination. Acetylation of Sep-CN (33)

in water with AcSH and ferricya-

nide at pH 7 yielded Ac-Sep-CN

(80%) after 1 hour. Phosphoserines

typically require alkaline pH and
Ba2+ to promote Dha formation

(29), but the formation of Ac-Dha-

CN (8%) was nonetheless observed

after heating Ac-Sep-CN at 60°C

for 3 days at pH 7. The sluggish

rate of phosphate elimination was
enhanced by Mg2+, yielding Ac-

Dha-CN (24%) after 1 day at 60°C

(fig. S33). This demonstrates that

prebiotically plausible Dha for-

mation can be achieved by serine

acetylation or phosphorylation.

Ac-Dha-CN was found to be

highly stable, and we did not ob-

serve the addition of acetate, phos-

phate, or hydroxide (even at pH 11)

to this Dha, setting the stage for

selective addition of inorganic sul-

fur to synthesize cysteine.

Ac-Dha-CN underwent near-

quantitative conversion to Ac-Cys-

SNH2 upon incubation with H2S, Fig. 1. Prebiotic cysteine synthesis. Biomimetic conversion of serinyl nitriles to cysteinyl nitriles. (A) Biosynthesis: PLP-
yielding the first prebiotic synthe- dependent enzymatic cysteine synthesis pathway. (B) Prebiotic synthesis: Nitrile-activated synthesis of cysteines in water

sis of a stable cysteinyl residue starting from Ser-CN, the stable Strecker product of GCA. Enz, enzyme; Ac, acetyl.

Foden et al., Science 370, 865–869 (2020) 13 November 2020 2 of 5

RESEARCH | REPORT

chemical biology and synthetic chemistry, fa- 7 days (fig. S86). Very little cysteine-catalyzed challenging CPL with N-acetylglutamine di-
cilitating rapid synthesis of longer peptides nitrile (Ac-Glx-CN). Finally, we only observed
from smaller subunits (36–38). Fragment liga- hydrolysis was observed in the absence of an
tions have been proposed to play an important amine nucleophile; incubating Ac-Gly-CN with coupling of proteinogenic alanine in competi-
role in prebiotic chemistry (39), but the plau- tion with a,a-disubstituted (nonproteinogenic
sibility of these are diminished by the need Ac-Cys-OH (30 mol %, pH 7, 60°C) resulted in amino acid) a–aminoisobutyric acid (Aib).
for synthetically prepared C-terminal thioesters only 6% hydrolysis after 24 hours (fig. S87).
(40). We recently demonstrated high-yielding Thiol-catalyzed coupling of a-amidonitriles
prebiotic fragment ligations (4) using stoichio- These results underscore the outstanding ki- with a–amino acids is highly general; all pro-
metric H2S and an activating agent (such as netic stability of a-amidonitriles and the gen- teinogenic a–amino acids coupled with Ac-Gly-
ferricyanide or cyanoacetylene) (Fig. 2A, i). erality of thiol-catalysis, with only a severely CN to give peptidyl amidines Ac-GlyN-AA-OH
Ideally, prebiotic fragment ligations through hindered tertiary thiol failing to catalyze CPL
direct coupling of nitrile and amine fragments in good yields at pH 7 and 60°C (Table 1).
would sidestep thioamide and thioacid inter- (fig. S85).
mediates (4). Therefore, we became intrigued CPL is specific and selective for proteino- However, dipeptides derived from serine, thre-
by the potential of thiols, such as N-acylcysteines,
to act as organocatalysts for CPL. Reversible genic a-peptide synthesis (Fig. 3). For example, onine, and asparagine underwent pronounced
thiol addition to an a-amidonitrile could render the reaction of Ac-Gly-CN and Ac-b-Ala-CN (1:1;
peptide synthesis catalytic and redox-neutral 200 mM) with glycine (200 mM) and Ac-Cys-OH amidine hydrolysis to the corresponding pep-
via a highly reactive, but transient, thioimidate (30 mol %) results in exclusive a-amidonitrile tides (Ac-GlyN-AA-OH→Ac-Gly-AA-OH). In
intermediate (Fig. 2A, ii). We anticipated that coupling to furnish Ac-GlyN-Gly-OH (65%) with the reactions of serine and threonine, we ob-
these thioimidates would be substantially no detectable b-amidonitrile coupling of Ac-
more reactive than the thioesters convention- b-Ala-CN. We also observed only a-ligation upon served oxazoline intermediates, suggesting
ally exploited in peptide ligations (36, 37) but
would also be inherently protected from hy- that intramolecular catalysis by the amino
drolysis by reversible thiol-to-nitrile addition
and elimination back to the stable nitrile acid side chain was responsible for rapid
substrate.
Fig. 2. Thiol-catalyzed peptide synthesis in neutral water. (A) (i) Previous work: Stoichiometric
Eschenmoser and co-workers had previ-
ously reported a cysteine-catalyzed ammo- ferricyanide-activated peptide thioacid ligation, which requires temporally separated thiolysis and activation
nolysis of an a-amidonitrile in methanol (41)
to produce a C-terminal primary amidine. In steps (4). (ii) This work: Thiol-catalyzed peptide ligation via a transient thioimidate intermediate that is
water, however, the addition of cysteine to
a-amidonitriles near-quantitatively yields chemoselectively intercepted by an amine nucleophile in water to form a peptidyl amidine directly from
thiazolines (supplementary materials), fol-
lowing the mechanism implicated in Cys-CN a stable nitrile without any activating agents. R, aminoacyl side chain; R1, aminoacyl side chain. (B) (i)
self-degradation (7). Thiazoline formation is, Peptidyl amidines persist if a-amidonitriles are coupled with an a–amino acid (AA), except serine, threonine,
however, completely suppressed by cysteine and cysteine (see Fig. 2B, ii). Intramolecular amide-assisted hydrolysis of the peptidyl amidine yields
N-acylation (42), suggesting that our biomi- the native peptide bond (e.g., asparagine or R2 = H or peptide). (ii) Intramolecular cyclization of serinyl
metic cysteine synthesis—which necessarily (R3 = H; X = O), threoninyl (R3 = CH3; X = O), or cysteinyl (R3 = H; X = S) residues promote stereoretentive
yields N-acyl-cysteines—is predisposed to fur- hydrolysis of peptidyl amidines to native peptides.
nish catalytically active cysteines. Therefore,
we next tested Ac-Cys-OH as a catalyst for
peptide ligation. We incubated Ac-Gly-CN and
glycine with Ac-Cys-OH and observed CPL in
water for the first time. This reaction yielded
peptidyl amidine Ac-GlyN-Gly-OH (60%) after
24 hours at 60°C and pH 7 (fig. S85). No acti-
vating agents were required to induce ligation.
Moreover, peptide ligation no longer requires
a C-terminal thioacid (4) or thioester and an
N-terminal cysteine residue as the nucleophilic
ligation partner (36, 37, 40). A broad spectrum
of cysteine derivatives were excellent catalysts
for peptidyl amidine synthesis in water, includ-
ing Ac-Cys-NH2 (69%) and N-acylcysteine pep-
tides (57 to 71%), as well as CoA (65%) and
simple thiols such as coenzyme M (79%) (fig.
S85). Catalysis is essential to promoting this
ligation, and the catalytic potency of simple
thiols is highlighted by the lack of background
reactivity. In control experiments (with no thiol
catalyst), no coupling was observed, and Ac-
GlyN-Gly-OH (<1%) could only be detected after

Foden et al., Science 370, 865–869 (2020) 13 November 2020 3 of 5

RESEARCH | REPORT

Fig. 3. Chemo- and regioselective organocatalytic proteinogenic a-peptide ligation. (A to C) CPL amidine hydrolysis (Fig. 2B). Having also
observed that the peptidyl amidine derived
selectively yields native a-peptidyl bonds. Selective ligation of (A) Ac-Gly-CN (200 mM) with glycine (200 mM) from asparagine hydrolyzed, we envisaged
affords Ac-GlyN-Gly-OH in a stoichiometric competition with Ac-b-Ala-CN (200 mM), (B) Ac-Gly-CN (200 mM) that amino amides (and, therefore, peptides)
with alanine (200 mM) affords Ac-GlyN-Ala-OH in a stoichiometric competition with Aib (200 mM), and would behave similarly. We next examined
(C) Ac-Glx-CN (200 mM) with glycine (200 mM) affords Ac-GlyN-Gly-OH. whether amides intramolecularly catalyze
amidine hydrolysis by coupling proteinogenic
Table 1 . Organocatalytic ligation of N-acetylglycine nitrile with a–amino acids and a–amino a–amino amides (AA-NH2). All AA-NH2 cou-
amides. Yields for Ac-Cys-OH (30 mol %)–catalyzed formation of peptidyl amidines (Ac-GlyN-AA1-X) plings resulted in selective dipeptide (Ac-Gly-
and peptides (Ac-Gly-AA1-X) by coupling of Ac-Gly-CN (200 mM) with AA1-X (1 equiv., pH 7, 60°C, AA-NH2) synthesis, irrespective of their side
chain (Table 1). This observed hydrolysis is
24 hours). See tables S6 and S7 for improved yields obtained by varying reaction conditions and important because, although peptidyl ami-
dines undergo racemization as expected (41),
further examples with other a-amidonitriles Ac-AA-CN. “–” indicates not observed. N-terminal serine, threonine, and cysteine
undergo stereoretentive coupling (Fig. 2B).
Ac-GlyN- Ac-Gly- Ac-GlyN- Ac-Gly- The origins of biological homochirality re-
AA1-OH (%) AA1-OH (%) AA1-NH2 (%) AA1-NH2 (%) main a formidable challenge, requiring an as
yet unknown symmetry-breaking event, and
until that event, prebiotic syntheses must
produce racemic mixtures (43). The observed
racemization and epimerization of peptidyl
amidines may have limited impact in a race-
mic (prebiotic) environment; however, the
stereoretentive couplings of peptides with nu-
cleophilic side chains and the intramolecular
amide-catalyzed hydrolysis of peptidyl ami-
dines offer a route to investigate dynamic
kinetic resolution of peptide stereochemistry
that may underpin peptide chiral resolution
in oligomers, rather than a–amino acid (or
a-aminonitrile) monomers (44).

Having demonstrated the tolerance of all
proteinogenic aminoacyl residues at the li-
gation junction in thiol-catalyzed dipeptide
synthesis, we explored the feasibility of or-
ganocatalytic peptide fragment ligations. We
recently demonstrated the prebiotic synthe-
sis of N-acylpeptide nitriles (Ac-AAn-CN) in
water by sulfide-mediated iterative ligation of
a-aminonitriles (Fig. 2A, i) (4), and therefore

AA1-OH AA1-NH2

Gly 60 2 Gly 21 52 Table 2. Organocatalytic peptidyl-nitrile
..................................................................................................................................................................................................................... to peptide fragment ligations. Yields for
MPA (160 mM)–catalyzed ligation of Ac-Gly3-
-Ala 43 – -Ala 3 63D...L................................................................................................................D.................................................................................................. CN (100 mM) and peptides AAn-OH (100 mM,
24 hours, 60°C, pH 7.0).
Arg 37 – Arg 14 56
.....................................................................................................................................................................................................................

Asn 9 45 Asn – 72
.....................................................................................................................................................................................................................

Asp 58 – Asp 6 58
.....................................................................................................................................................................................................................

Gln 56 – Gln – 43
.....................................................................................................................................................................................................................

Glu 58 – Glu – 64
.....................................................................................................................................................................................................................

His 73 – His – 67 Ac-Gly3-AAn-OH
..................................................................................................................................................................................................................... AAn-OH (%)

Ile 55 – Ile 12 47
.....................................................................................................................................................................................................................

Leu 53 – -Leu 5 54..................................................................................................................D................................................................................................... Met-Gly 80
– Lys 25† 52‡
Lys 70* .............................................................................................

..................................................................................................................................................................................................................... Ala-Ala-Ala 90

-Met 72 – Met 5 62.D..L.................................................................................................................................................................................................................. .............................................................................................

Phe 21 – Phe 8 52 Ala-Gly-Ala 84
.............................................................................................

..................................................................................................................................................................................................................... Gly-Ala-Gly 87

Pro 58 – Pro – 21 .............................................................................................

..................................................................................................................................................................................................................... Gly-Gly-Gly 89

Ser – 61 Ser – 68 .............................................................................................

..................................................................................................................................................................................................................... Gly-Gly-His 89

Thr – 51 Thr – 69 .............................................................................................

..................................................................................................................................................................................................................... Leu-Leu-Leu 76*

Trp 32 5 Trp 4 45 .............................................................................................

..................................................................................................................................................................................................................... Met-Ala-Ser 77

Tyr 20 – Tyr 3 62 .............................................................................................
77†
..................................................................................................................................................................................................................... Phe-Gly-Gly

Val 42 6 Val 7 50 .............................................................................................

.....................................................................................................................................................................................................................

*N2-(Ac-GlyN)-Lys-OH (43%), N6-(Ac-GlyN)-Lys-OH (24%), N2,N6-bis(Ac-GlyN)-Lys-OH (3%). †Only the N6-amidine, *Forty-eight hours. †Ac-Gly2-GlyN-Phe-Gly2-OH (12%)

N6-(Ac-GlyN)-Lys-NH2, was observed. ‡Only the N2-peptide, N2-(Ac-Gly)-Lys-NH2, was observed. was also observed. Total ligation yield was 89%.

Foden et al., Science 370, 865–869 (2020) 13 November 2020 4 of 5

RESEARCH | REPORT

we chose to investigate Ac-Gly3-CN as a pre- tide ligation in water. The inherent catalytic 27. G. Ksander et al., Helv. Chim. Acta 70, 1115–1172 (1987).
biotically plausible peptide nitrile in fragment 28. I. Photaki, J. Am. Chem. Soc. 85, 1123–1126 (1963).
ligations. Peptide nitrile Ac-Gly3-CN (100 mM) activity of simple cysteinyl peptides makes 29. M. F. Byford, Biochem. J. 280, 261–265 (1991).
readily coupled with various peptides with 30. F. R. Bowler et al., Nat. Chem. 5, 383–389 (2013).
3-mercaptopropionic acid (MPA) as a cata- them an excellent candidate for catalysts for 31. C. Fernandez-Garcia, M. W. Powner, Synlett 28, 78–83
lyst at neutral pH to give excellent yields of
peptides, even with stoichiometric (1:1) cou- protometabolic reactions in an abiotic en- (2017).
pling partners (Table 2). The simplicity and 32. B. M. Anderson, E. H. Cordes, W. P. Jencks, J. Biol. Chem. 236,
generality of CPL underscore the privileged vironment. Finally, our data support a sce-
nature of thiol-catalyzed peptide ligations 455–463 (1961).
in water. nario in which nitriles served as an early energy 33. A. J. Coggins, M. W. Powner, Nat. Chem. 9, 310–317 (2017).
34. C. F. Barbas III, Angew. Chem. Int. Ed. 47, 42–47 (2008).
We have combined systems chemistry (3, 45) currency on the primordial Earth, perhaps act- 35. M. P. van der Helm, B. Klemm, R. Eelkema, Nat. Rev. Chem. 3,
with analysis of the strategy of modern bio- ing as a forerunner to adenosine 5′-triphosphate
synthesis to resolve two long-standing co- and thioesters that drive reactions in extant 491–508 (2019).
nundrums at the origins of life: the chemical biology (10, 11). 36. P. E. Dawson, T. W. Muir, I. Clark-Lewis, S. B. Kent, Science
origins of cysteine and catalytic peptide syn-
thesis. We note that extant cysteine biosynthe- REFERENCES AND NOTES 266, 776–779 (1994).
sis (17, 18) bears a strong resemblance to the 37. S. S. Kulkarni, J. Sayers, B. Premdjee, R. J. Payne, Nat. Rev. Chem.
prebiotic synthesis we have outlined. However, 1. P. Nissen, J. Hansen, N. Ban, P. B. Moore, T. A. Steitz, Science
the nonenzymatic reactions described here are 289, 920–930 (2000). 2, 0122 (2018).
contingent on an a-nitrile, rather than com- 38. V. R. Pattabiraman, J. W. Bode, Nature 480, 471–479 (2011).
plex enzymes and cofactors. The a-amidonitrile 2. M. A. Fischbach, C. T. Walsh, Chem. Rev. 106, 3468–3496 39. J. Söding, A. N. Lupas, BioEssays 25, 837–846 (2003).
moiety is not only prebiotically plausible but (2006). 40. D. H. Lee, J. R. Granja, J. A. Martinez, K. Severin, M. R. Ghadiri,
also ideally poised to activate Dha formation at
near-neutral pH and provides the in-built en- 3. S. Islam, M. W. Powner, Chem 2, 470–501 (2017). Nature 382, 525–528 (1996).
ergy required to drive peptide synthesis. The 4. P. Canavelli, S. Islam, M. W. Powner, Nature 571, 546–549 41. H. Moser et al., Helv. Chim. Acta 69, 1224–1262 (1986).
a-amidonitrile has a balance of kinetic stab- 42. U. E. W. Lange, B. Schäfer, D. Baucke, E. Buschmann, H. Mack,
ility and thermodynamic reactivity to deliver (2019).
aqueous ligation without requiring any elec- 5. M. S. Chadha, L. Replogle, J. Flores, C. Ponnamperuma, Tetrahedron Lett. 40, 7067–7070 (1999).
trophilic activation, and the highly selective 43. J. E. Hein, D. G. Blackmond, Acc. Chem. Res. 45, 2045–2054
and reversible reaction that thiols undergo with Bioorg. Chem. 1, 269–274 (1971).
a-amidonitriles makes thiol-catalyzed nitrile 6. I. Shalayel, S. Coulibaly, K. D. Ly, A. Milet, Y. Vallée, Life 8, (2012).
ligation a powerful and selective system for 44. D. Beaufils, G. Danger, L. Boiteau, J.-C. Rossi, R. Pascal,
peptide ligation in water. The amidine inter- 47 (2018).
mediates observed during CPL suggest that 7. I. Shalayel et al., Eur. J. Org. Chem. 2020, 3019–3023 Chem. Commun. 50, 3100–3102 (2014).
ligation occurs by intermolecular addition of 45. J. W. Szostak, Nature 459, 171–172 (2009).
the nucleophilic coupling partner to a catalyst- (2020).
bound thioimidate (Fig. 2A, ii). Our ligation 8. L. Leman, L. Orgel, M. R. Ghadiri, Science 306, 283–286 ACKNOWLEDGMENTS
[in contrast to thioester acylations (36, 37)]
tolerates all amino acid side-chain residues (2004). We thank K. Karu, M. Puchnarewicz (UCL Mass Spectrometry
without protection to deliver high yields in 9. H. Griesser, M. Bechthold, P. Tremmel, E. Kervio, C. Richert, Facility), E. Samuel (Mass Spectrometry, UCL School of Pharmacy),
neutral water and is not limited to ligation of A. E. Aliev (NMR spectroscopy), P. Canavelli, B. Thoma, and
N-terminal cysteine residues. Angew. Chem. Int. Ed. 56, 1224–1228 (2017). J. Fairchild for preliminary work and experimental assistance.
10. C. De Duve, Blueprint for a Cell: The Nature and Origin of Life Funding: We thank the Engineering and Physical Sciences
It is notable that a single amino acid resi- Research Council (EPSRC) (EP/K004980/1, EP/M507970/1, and
due, cysteine, provides robust catalysis for pep- (N. Patterson, Burlington, NC, 1991). EP/P020410/1), the Simons Foundation (318881FY19 and
11. J. E. Goldford, H. Hartman, T. F. Smith, D. Segrè, Cell 168, 493895), and the Volkswagen Foundation (94743) for financial
support. Author contributions: M.W.P. conceived the research.
1126–1134.e9 (2017). C.S.F., S.I., C.F.-G., T.D.S., and M.W.P. designed and analyzed the
12. L. B. Poole, Free Radic. Biol. Med. 80, 148–157 (2015). experiments. C.S.F., S.I., C.F.-G., and L.M. conducted the
13. R. V. Eck, M. O. Dayhoff, Science 152, 363–366 (1966). experiments. S.I. assembled and wrote the supplementary
14. C. Bonfio et al., Nat. Chem. 9, 1229–1234 (2017). materials with contributions from C.S.F. and C.F.-G. C.S.F. and S.I.
15. J. T.-F. Wong, Proc. Natl. Acad. Sci. U.S.A. 72, 1909–1912 contributed equally to this work and are listed alphabetically. S.I.
and M.W.P. wrote the manuscript with assistance from T.D.S.
(1975). Competing interests: The authors declare no competing
16. E. N. Trifonov, Gene 261, 139–151 (2000). financial interests. Data and materials availability: All data are
17. A. Sauerwald et al., Science 307, 1969–1972 (2005). available in the main text and the supplementary materials.
18. K. Fujishima et al., Sci. Rep. 8, 1776 (2018).
19. B. N. Khare, C. Sagan, Nature 232, 577–579 (1971). SUPPLEMENTARY MATERIALS
20. A. L. Weber, S. L. Miller, J. Mol. Evol. 17, 273–284 (1981).
21. E. T. Parker et al., Proc. Natl. Acad. Sci. U.S.A. 108, 5526–5531 science.sciencemag.org/content/370/6518/865/suppl/DC1
Materials and Methods
(2011). Figs. S1 to S209
22. M. W. Powner, S.-L. Zheng, J. W. Szostak, J. Am. Chem. Soc. Tables S1 to S14
References (46–51)
134, 13889–13895 (2012).
23. S. Islam, D.-K. Bučar, M. W. Powner, Nat. Chem. 9, 584–589 29 June 2020; accepted 1 October 2020
10.1126/science.abd5680
(2017).
24. B. H. Patel, C. Percivalle, D. J. Ritson, C. D. Duffy,

J. D. Sutherland, Nat. Chem. 7, 301–307 (2015).
25. Z. Chen et al., Angew. Chem. Int. Ed. 59, 3272–3279

(2020).
26. A. J. Borchert, D. C. Ernst, D. M. Downs, Trends Biochem. Sci.

44, 849–860 (2019).

Foden et al., Science 370, 865–869 (2020) 13 November 2020 5 of 5

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

new products

Cell-Freezing Media of the measured flow rates. The current flow rate is also displayed on
AMS Biotechnology announces two the flowmeter’s integral high-resolution organic LED display, allowing
chemically defined, cryopreservation easy control of current flow value. Extraordinary high resolution and
excipient solutions: STEM-CELLBANKER and wide dynamic range make the TESTA flowmeter the perfect flow-
HSC-BANKER. Available in both dimethyl monitoring tool for the most demanding HPLC and GPC/SEC systems.
sulfoxide (DMSO)-containing and DMSO- TESTA Analytical Solutions
free formulations, STEM-CELLBANKER is For info: +49-(0)-30-864-24076
a chemically defined freezing media optimized for stem-cell and www.testa-analytical.com
induced pluripotent stem-cell storage as well as fragile primary cell
storage. Published data supports its ability to cryopreserve orga- Application-Optimized Machine Vision Lenses
noids and tissues to allow the recovery of viable cells. Free from Resolve Optics offers a service to design, develop, and supply
animal-derived components, this popular medium contains only application-optimized machine vision lenses for any industrial
chemically defined U.S. Pharmacopeia (USP)-, European Pharma- and robotic image-processing application. High-quality construc-
copeia (EP)-, and Japanese Pharmacopeia (JP)-grade ingredients. tion, coupled with precision engineering, results in outstanding
Manufactured to be completely free of serum and animal-derived machine vision lenses that deliver sharp, high-resolution, optically
components, HSC-BANKER contains only EP- or USP-grade ingredi- precise images. Our lenses provide wide fields of view with little
ents, making it highly suitable for storage of hematopoietic stem or no distortion, optical designs and coatings that are balanced to
cells developed for cell-therapy applications. Recently the master give best performance at a desired wavelength or waveband, and
files of HSC-BANKER were accepted by the Center for Biologics Evalu- compact lens designs where the target application is space-limited.
ation and Research within the U.S. Food and Drug Administration. Resolve Optics
HSC-BANKER is supplied ready-to-use and requires no special de- For info: +44-(0)-1494-777100
vices, such as a controlled-rate freezer, to achieve consistently high www.resolveoptics.com/machine-vision
viabilities following resuscitation from cryopreservation, even over
extended long-term storage. HSC-BANKER significantly increases cell In Vitro Pyrogen Testing
viability while maintaining cell pluripotency, normal karyotype, and Lonza Bioscience announces the release of the PyroCell Monocyte
proliferation ability after freeze-thaw. It is evaluated for endotox- Activation Test (MAT) System, a sustainable, reliable solution for
ins, pH, osmolarity, and mycoplasma contaminants to ensure GMP in vitro pyrogen testing. The new offering ensures the safety of
equivalent quality. parenteral pharmaceuticals during development, manufacture, and
AMS Biotechnology product release. It provides sensitive pyrogen detection without the
For info: 617-945-5033 use of experimental animals, thereby supporting sustainability objec-
www.amsbio.com/cellbanker-cell-freezing-media tives while helping to deliver safe products to the market. Through a
collaboration with Sanquin Reagents B.V., the PyroCell MAT System
2D Field-Flow Fractionation consists of pooled, cryopreserved peripheral blood mononuclear
The EAF2000 series from Postnova Analytics is a powerful and field- cells (PBMCs) specifically developed for use with the MAT. These
proven 2D field-flow fractionation (FFF) system, which can be applied as PyroCell Kit PBMCs eliminate the need to qualify blood donors and
either traditional asymmetric flow FFF or alternatively as a hybrid FFF undertake cell isolation for each test run. With the PyroCell MAT Sys-
system utilizing both separation forces in the same channel simultane- tem, the cells can be available whenever the need arises.
ously. This allows particle size and molar mass separations induced Lonza
by the crossflow field as well as charge separations by electrophoretic For info: +41-(0)-61-316-81-11
mobility. As particle and molecule charge play a primary role in many www.lonza.com/mat
applications, such as protein aggregation, polymer flocculation, particle
agglomeration, and pharmaceutical formulations, the EAF2000 enables CCD Cameras
significantly better understanding of these phenomena and will help to The Atik VS series of scientific cooled charge-coupled device (CCD)
establish more efficient product development and QC processes. cameras is flexible enough to meet the needs of both scientific OEMs
Postnova Analytics and microscopy users. It provides a versatile solution for any low-light
For info: 801-521-2004 application, which simplifies the designer’s task of integration and
www.postnova.com/product/systems/eaf2000-electrical-af4.html has an optional five-position integrated filter wheel allowing filters
to be loaded without opening the camera. The VS series delivers the
Chromatography Flowmeter highest-quality images under low-light conditions, thanks to 16-bit
TESTA Analytical Solutions has developed a new flowmeter that enables digitization that offers 65,536 gray levels, plus high-performance ther-
continuous measurement of flow rate without interference in chroma- moelectric cooling to –35°C below ambient. At 6 MP/s, the VS series
tography systems. Flow rate is one of the most important parameters offers the fastest digitization of any Atik camera. This range supports
in any liquid chromatography system; it determines retention time or a wide variety of Sony CCD sensors, including ICX285, ICX274, ICX674,
volume and has a major influence on reproducibility. Compatible with ICX655, ICX694, and ICX814. The VS255 and VS825 cameras are
all high-performance liquid chromatography (HPLC) and gel permeation specifically for use with quantitative PCR instruments for accurate,
chromatography/size-exclusion chromatography (GPC/SEC) solvents, reliable COVID-19 testing.
the TESTA flowmeter is conveniently sized and powers itself from a USB Atik Cameras
connection. A PC-based app allows continuous recording and storage For info: 888-733-3034
www.atik-cameras.com/product/atik-vs-range

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.

870 13 NOVEMBER 2020 • VOL 370 ISSUE 6518 sciencemag.org/custom-publishing SCIENCE

POSTDOCTORAL RESEARCH FELLOWS POSTDOCTORAL RESEARCH ASSOCIATES online @sciencecareers.org

Air Force Science & Technology Fellowship Program NRC Research Associateship Programs

The National Academies of Sciences, Engineering, and Medicine The National Academies of Sciences, Engineering, and Medicine
administers postdoctoral and senior fellowship awards at the U.S. administers postdoctoral and senior research awards at participating
Air Force Research Laboratory (AFRL), the U.S. Air Force Institute of federal laboratories and affiliated institutions at locations throughout the
Technology (AFIT), and the U.S. Air Force Academy (USAFA) under the U.S and abroad.
Air Force Science & Technology Fellowship Program (AF STFP).
We are seeking highly qualified candidates who hold, or anticipate
We are seeking highly qualified candidates who are U.S. citizens and earning, a doctorate in a variety of fields of science or engineering.
hold, or anticipate earning, a doctorate in a variety of fields of science Degrees from foreign universities should be equivalent in training and
or engineering. research experience to a doctoral degree from a U.S. institution.
Citizenship eligibility varies among the sponsoring laboratories.

Application deadline dates (four annual review cycles): Application deadline dates (four annual review cycles):

• February 1 • May 1 • February 1 • May 1

• August 1 • November 1 • August 1 • November 1

Awardees have the opportunity to: Awardees have the opportunity to:
• Conduct independent research in an area compatible with the interests of • Conduct independent research in an area compatible with the interests of

the Air Force laboratories the sponsoring laboratory
• Devote full-time effort to research and publication • Devote full-time effort to research and publication
• Access the excellent and often unique Air Force research facilities
• Collaborate with leading scientists and engineers • Access the excellent and often unique facilities of the federal research
enterprise

• Collaborate with leading scientists and engineers at the sponsoring
laboratories

Awardee benefits: Awardee benefits include:
• Base stipend starting at $76,542; may be higher based on experience • Stipends ranging from $45,000 to $86,000; may be higher based on
• Health insurance (including dental and vision), relocation benefits, and
experience
professional travel allowance • Health insurance (including dental and vision), relocation benefits, and a

For detailed program information, to search for AFRL, AFIT, and USAFA professional travel allowance
Research Opportunities, and to contact prospective Research Adviser(s),
visit www.nas.edu/afstfp. For detailed program information, to search Research Opportunities, and to
contact prospective Research Adviser(s) visit www.nas.edu/rap.

WhoÕs the top
employer for 2020?

Science Careers’ annual survey reveals
the top companies in biotech & pharma
voted on by Science readers.

Read the article and employer profles
at sciencecareers.org/topemployers

online @sciencecareers.org

Confused about your next
career move?

Download Free Career
Advice Booklets!
ScienceCareers.org/booklets

YOU’D
SMILE TOO
IF YOU JUST
ADVANCED
YOUR
CAREER.

Find your next job at ScienceCareers.org
There’s scientific proof that when you’re happy
with what you do, you’re better at what you do. Access
career opportunities, see who’s hiring and take
advantage of our proprietary career-search tools.
Get tailored job alerts, post your resume and manage
your applications all in one place: sciencecareers.org

WORKING LIFE

By Sasha McDowell

Difficult conversations

I t was 11 o’clock on a Friday morning—time to meet with my supervisor. Normally, I would have
hurried over to my computer, coffee in hand, casually adjusting my bed head in order to appear at
least somewhat professional. But this May morning was different. I tried to pull myself together,
but I hadn’t slept much, and nothing could relieve my puffy eyes. A few days earlier, George Floyd
had been killed by the police. His homicide made me even more aware that I am a Black student
at a predominantly white institution. To my knowledge, I was the only Black graduate student in
my department at that time. Would I be able to talk about the killing with my white male supervisor?
Did I want to talk about it with him? Or would we ignore the elephant in the room?

Prior to the COVID-19 outbreak, semesters when I wasn’t teaching.

our meetings had typically been It was refreshingly open discourse.

limited to research. When the Despite this new casual cama-

time came, I would anxiously put raderie, the prospect of discussing

together whatever data I had and George Floyd’s killing was daunt-

head over to his office. Between ing. We had spoken about many

grading assignments for the class global issues, but nothing as sen-

in which I was a teaching assistant sitive as this one. I calmed myself

(TA), taking a course of my own, with the thought that it probably

and learning difficult techniques wouldn’t come up.

in the lab, I didn’t have much But on that morning, my super-

research progress to share, and visor started our meeting a little

I worried he thought I was un- differently from usual. Typically, he

productive. Yet I hesitated to tell asked, “How’s it going?” to which

him I felt overwhelmed for fear I’d routinely reply, “It’s going OK.”

that I would seem incompetent. This time, though, he asked, “How

After our lab closed in response are you doing?”

to the pandemic, that dynamic “It was as though an invisible At first, it was awkward tiptoeing
shifted dramatically. Without new barrier had been broken.” around race, but his genuine inter-
data to discuss, our meetings veered est in my well-being made me feel

toward casual conversations about as though I was in a safe space to

topics such as the latest news on the pandemic, how my discuss Black issues and Black representation in science.

supervisor was juggling teaching his young daughter from We acknowledged my minority status. I volunteered that I

home while working, and whether my furniture from Ikea was lucky to not experience racism at our institution, but

had finally arrived. It was as though an invisible barrier that I was interested in promoting equity, diversity, and

had been broken. He was no longer just the person over- inclusion (EDI) practices at our university. After this con-

seeing my research; he was another human being trying versation, my supervisor seemed to make it his mission to

to navigate the difficult times the COVID-19 pandemic had connect me with others with similar goals. As a result, I

imposed on us all. am now part of a departmental trainee group to discuss

When we did discuss research, being separated by a such issues and a member of an EDI committee—which my

screen somehow gave me the confidence to bring up chal- supervisor decided to join as well.

lenging topics and ask questions that felt naïve. I was work- Although the pandemic has restricted my research, it has

ing on my Ph.D. proposal and wasn’t yet adept at literature unexpectedly strengthened my relationship with my super-

searches—something a supervisor might assume graduate visor. It has enabled me to become more involved in issues

students know how to do. Before the pandemic, I likely of social justice and, hopefully, to make a contribution to ILLUSTRATION: ROBERT NEUBECKER

wouldn’t have brought up my fears about overlooking cru- science and academia that goes beyond the lab bench. Just

cial papers. But now I felt comfortable sharing my concerns as important, I think the impact has been mutual. j

and asking for advice. We also discussed how I could reserve

time for research while also teaching, whether by finding a Sasha McDowell is a graduate student at the University of British

less time-intensive course to TA or making the most of the Columbia,Vancouver. Send your career story to [email protected].

874 13 NOVEMBER 2020 • VOL 370 ISSUE 6518 sciencemag.org SCIENCE

Published by AAAS