Rotation tracking of genome-processing enzymes using DNA origami rotors

Article metrics

Abstract

Many genome-processing reactions, including transcription, replication and repair, generate DNA rotation. Methods that directly measure DNA rotation, such as rotor bead tracking1,2,3, angular optical trapping4 and magnetic tweezers5, have helped to unravel the action mechanisms of a range of genome-processing enzymes that includes RNA polymerase (RNAP)6, gyrase2, a viral DNA packaging motor7 and DNA recombination enzymes8. Despite the potential of rotation measurements to transform our understanding of genome-processing reactions, measuring DNA rotation remains a difficult task. The time resolution of existing methods is insufficient for tracking the rotation induced by many enzymes under physiological conditions, and the measurement throughput is typically low. Here we introduce origami-rotor-based imaging and tracking (ORBIT), a method that uses fluorescently labelled DNA origami rotors to track DNA rotation at the single-molecule level with a time resolution of milliseconds. We used ORBIT to track the DNA rotations that result from unwinding by the RecBCD complex, a helicase that is involved in DNA repair9, as well as from transcription by RNAP. We characterized a series of events that occur during RecBCD-induced DNA unwinding—including initiation, processive translocation, pausing and backtracking—and revealed an initiation mechanism that involves reversible ATP-independent DNA unwinding and engagement of the RecB motor. During transcription by RNAP, we directly observed rotational steps that correspond to the unwinding of single base pairs. We envisage that ORBIT will enable studies of a wide range of interactions between proteins and DNA.

Access options

Rent or Buy article

Get time limited or full article access on ReadCube.

from$8.99

All prices are NET prices.

Fig. 1: Single-molecule DNA rotation measurement using ORBIT.
Fig. 2: Pausing and backtracking during DNA unwinding induced by RecBCD.
Fig. 3: Initiation of RecBCD helicase activity at dsDNA breaks.
Fig. 4: DNA rotation and rotational steps of a single base pair during transcription by RNAP, detected by ORBIT.

Data availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Code availability

The single-molecule data were analysed using custom Python and Igor Pro code, available at https://github.com/altheimerb/python-sma/.

References

  1. 1.

    Bryant, Z. et al. Structural transitions and elasticity from torque measurements on DNA. Nature 424, 338–341 (2003).

  2. 2.

    Gore, J. et al. Mechanochemical analysis of DNA gyrase using rotor bead tracking. Nature 439, 100–104 (2006).

  3. 3.

    Lebel, P., Basu, A., Oberstrass, F. C., Tretter, E. M. & Bryant, Z. Gold rotor bead tracking for high-speed measurements of DNA twist, torque and extension. Nat. Methods 11, 456–462 (2014).

  4. 4.

    Deufel, C., Forth, S., Simmons, C. R., Dejgosha, S. & Wang, M. D. Nanofabricated quartz cylinders for angular trapping: DNA supercoiling torque detection. Nat. Methods 4, 223–225 (2007).

  5. 5.

    Lipfert, J., van Oene, M. M., Lee, M., Pedaci, F. & Dekker, N. H. Torque spectroscopy for the study of rotary motion in biological systems. Chem. Rev. 115, 1449–1474 (2015).

  6. 6.

    Harada, Y. et al. Direct observation of DNA rotation during transcription by Escherichia coli RNA polymerase. Nature 409, 113–115 (2001).

  7. 7.

    Liu, S. et al. A viral packaging motor varies its DNA rotation and step size to preserve subunit coordination as the capsid fills. Cell 157, 702–713 (2014).

  8. 8.

    Lipfert, J., Wiggin, M., Kerssemakers, J. W. J., Pedaci, F. & Dekker, N. H. Freely orbiting magnetic tweezers to directly monitor changes in the twist of nucleic acids. Nat. Commun. 2, 439 (2011).

  9. 9.

    Dillingham, M. S. & Kowalczykowski, S. C. RecBCD enzyme and the repair of double-stranded DNA breaks. Microbiol. Mol. Biol. Rev. 72, 642–671 (2008).

  10. 10.

    Rothemund, P. W. K. Folding DNA to create nanoscale shapes and patterns. Nature 440, 297–302 (2006).

  11. 11.

    Douglas, S. M. et al. Self-assembly of DNA into nanoscale three-dimensional shapes. Nature 459, 414–418 (2009).

  12. 12.

    Nomidis, S. K., Kriegel, F., Vanderlinden, W., Lipfert, J. & Carlon, E. Twist-bend coupling and the torsional response of double-stranded DNA. Phys. Rev. Lett. 118, 217801 (2017).

  13. 13.

    Roman, L. J. & Kowalczykowski, S. C. Characterization of the helicase activity of the Escherichia coli RecBCD enzyme using a novel helicase assay. Biochemistry 28, 2863–2873 (1989).

  14. 14.

    Bianco, P. R. et al. Processive translocation and DNA unwinding by individual RecBCD enzyme molecules. Nature 409, 374–378 (2001).

  15. 15.

    Dohoney, K. M. & Gelles, J. Chi-sequence recognition and DNA translocation by single RecBCD helicase/nuclease molecules. Nature 409, 370–374 (2001).

  16. 16.

    Perkins, T. T., Li, H.-W., Dalal, R. V., Gelles, J. & Block, S. M. Forward and reverse motion of single RecBCD molecules on DNA. Biophys. J. 86, 1640–1648 (2004).

  17. 17.

    Liu, B., Baskin, R. J. & Kowalczykowski, S. C. DNA unwinding heterogeneity by RecBCD results from static molecules able to equilibrate. Nature 500, 482–485 (2013).

  18. 18.

    Saikrishnan, K., Griffiths, S. P., Cook, N., Court, R. & Wigley, D. B. DNA binding to RecD: role of the 1B domain in SF1B helicase activity. EMBO J. 27, 2222–2229 (2008).

  19. 19.

    Farah, J. A. & Smith, G. R. The RecBCD enzyme initiation complex for DNA unwinding: enzyme positioning and DNA opening. J. Mol. Biol. 272, 699–715 (1997).

  20. 20.

    von Hippel, P. H., Johnson, N. P. & Marcus, A. H. Fifty years of DNA “breathing”: reflections on old and new approaches. Biopolymers 99, 923–954 (2013).

  21. 21.

    Wu, C. G. & Lohman, T. M. Influence of DNA end structure on the mechanism of initiation of DNA unwinding by the Escherichia coli RecBCD and RecBC helicases. J. Mol. Biol. 382, 312–326 (2008).

  22. 22.

    Carter, A. R. et al. Sequence-dependent nanometer-scale conformational dynamics of individual RecBCD–DNA complexes. Nucleic Acids Res. 44, 5849–5860 (2016).

  23. 23.

    Dillingham, M. S., Webb, M. R. & Kowalczykowski, S. C. Bipolar DNA translocation contributes to highly processive DNA unwinding by RecBCD enzyme. J. Biol. Chem. 280, 37069–37077 (2005).

  24. 24.

    Forde, N. R., Izhaky, D., Woodcock, G. R., Wuite, G. J. L. & Bustamante, C. Using mechanical force to probe the mechanism of pausing and arrest during continuous elongation by Escherichia coli RNA polymerase. Proc. Natl Acad. Sci. USA 99, 11682–11687 (2002).

  25. 25.

    Adelman, K. et al. Single molecule analysis of RNA polymerase elongation reveals uniform kinetic behavior. Proc. Natl Acad. Sci. USA 99, 13538–13543 (2002).

  26. 26.

    Neuman, K. C., Abbondanzieri, E. A., Landick, R., Gelles, J. & Block, S. M. Ubiquitous transcriptional pausing is independent of RNA polymerase backtracking. Cell 115, 437–447 (2003).

  27. 27.

    Abbondanzieri, E. A., Greenleaf, W. J., Shaevitz, J. W., Landick, R. & Block, S. M. Direct observation of base-pair stepping by RNA polymerase. Nature 438, 460–465 (2005).

  28. 28.

    Righini, M. et al. Full molecular trajectories of RNA polymerase at single base-pair resolution. Proc. Natl Acad. Sci. USA 115, 1286–1291 (2018).

  29. 29.

    Kabsch, W., Sander, C. & Trifonov, E. N. The ten helical twist angles of B-DNA. Nucleic Acids Res. 10, 1097–1104 (1982).

  30. 30.

    Kopperger, E. et al. A self-assembled nanoscale robotic arm controlled by electric fields. Science 359, 296–301 (2018).

Download references

Acknowledgements

This work was supported in part by the National Institutes of Health. B.D.A. was supported by a National Institutes of Health Training Grant for the Graduate Program in Biophysics at Harvard University and a National Science Foundation Graduate Research Fellowship. M.D. was supported by a Howard Hughes Medical Institute International Student Research Fellowship. X.Z. is a Howard Hughes Medical Institute Investigator.

Author information

P.K., B.D.A. and X.Z. conceived of the project and designed experiments. P.K. and B.D.A. performed ORBIT experiments and analysed the data. M.D., P.K. and P.Y. designed the origami structures. B.D.A., M.D. and P.K. prepared the origami structures. P.K. and M.D. performed atomic force microscopy and transmission electron microscopy imaging. P.K., B.D.A. and X.Z. wrote the manuscript with input from M.D. and P.Y.

Correspondence to Xiaowei Zhuang.

Ethics declarations

Competing interests

The authors declare no competing interests.

Additional information

Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Extended data figures and tables

Extended Data Fig. 1 Origami rotor and anchor designs.

a, Routeing diagram of the origami rotor consisting of two 160-nm arms (Supplementary Table 1). The intact arm (a six-helix bundle) passes through a break in the orthogonal arm (two half-length six-helix bundles). Additional helices stabilize the junction. Two of these helices contain staple strands (black) that are extended from the centre of rotor (extension not shown; see c). Six staples within 14 nm of the end of the intact arm (light green) are labelled with Cy3 at their 3′ ends. b, Three-dimensional rendering of the rotor design. c, Magnified view showing the two staple strands (red) extending from the centre of the rotor, forming a 14-bp dsDNA segment and a 12-nt ssDNA overhang for ligation. d, The overhang is ligated to a longer piece of dsDNA, which serves as the substrate for the DNA-interacting enzyme. e, An atomic force microscopy image with large field of view of the origami rotors. Representative of more than ten independent biological replicates. Scale bar, 1 μm. f, Routeing diagram of the origami anchor consisting of three 20-nm wings, each made of a short 6-helix bundle motif (Supplementary Table 2). Several staple strands were extended with binding sites for biotin-labelled secondary oligomers for surface attachment. From the centre of the structure, three strands (black) were used to make an adaptor to allow ligation to additional DNA. Following the final strand crossover, the adaptor consists of 26 bp of dsDNA followed by a 12-nt ssDNA overhang. g, Three-dimensional renderings of the origami anchor. h, Origami structure used for characterizing the Brownian dynamics. The origami rotor, anchor and a dsDNA segment (as needed) were ligated together. The origami anchor is attached to the microscope surface using multiple biotin tags. i, Representative images of the origami anchors from one transmission electron microscopy experiment.

Extended Data Fig. 2 Characterization of the angular and radial-position uncertainty of the origami rotor.

a, Power spectrum showing the Brownian fluctuation in the angular position of the rotor attached to the anchor by 52 bp of dsDNA. Representative of 3 independent biological replicates. Red line shows the modified Lorentzian fit, as described in Supplementary Discussion, Supplementary Equation (1). This fit yields the torsional stiffness (κ) and hydrodynamic drag (γ). b, Dependence of the inverse of κ on the length of dsDNA segment between the rotor and origami. The slope of the linear fit yields the torsional stiffness per unit length of the dsDNA in the absence of an applied force (Supplementary Discussion, Supplementary Equation (3)), C = 200 ± 10 pN nm2 rad−1, which is consistent with previous measurements under zero force12. The inverse of the y-offset from this fit, κother = 30 ± 8 pN nm rad−1 (Supplementary Discussion, Supplementary Equation (3)), represents the torsional stiffness of the remainder of the structure, which is the equivalent of about 20 bp of dsDNA. c, Dependence of the relaxation time, τ = γ/κ, on the length of the dsDNA segment between the rotor and anchor calculated using κ and γ derived from the power spectrum fit (Supplementary Discussion, Supplementary Equation (1)). Data in b, c are mean ± s.e.m. (n = 203, 195 and 133 (from left to right) rotor–anchor complexes from 3 independent biological replicates). d, Dependence of γ of the origami rotor on the viscosity of the buffer. The origami rotor was connected by a 92-bp dsDNA segment to the anchor. The different viscosities were achieved using 0%, 10% and 25% glycerol. Data are mean ± s.e.m. (n = 195, 210 and 150 (from left to right) rotor–anchor complexes from 3 independent biological replicates). e, Standard deviation of the angular positions of the rotor as a function of integration time. Black line shows the s.d. measured from a single rotor connected to the anchor with a 52-bp dsDNA segment tracked at 3 kHz after down-sampling to the indicated integration time. Representative of 3 independent biological replicates. Red and blue curves show predicted precision with and without (respectively) taking into account the contribution of localization uncertainty (Supplementary Discussion, Supplementary Equation (2)). κ = 7.8 pN nm rad−1, γ = 5.0 fN nm s and localization uncertainty per frame \({\sigma }_{L}^{2}\) = 0.038 rad2. κ and γ were derived from the measurements of multiple (n = 203) rotors with 52-bp dsDNA segment connecting the rotor to the anchor. \({\sigma }_{L}^{2}\) was estimated using the measurement uncertainty in radial position, and converted to an angular value using the radius of the circular trajectory. The crossing points of the top and bottom dashed lines with the s.d. versus integration time curve give the integration times required for detection of single-base-pair rotation (34.6°) with a signal-to-noise ratio of 1 and 3, respectively. f, Radial localization uncertainty (s.d.) during processive unwinding by RecBCD as a function of the average fluorescence signal intensity (mean) from individual rotors. Representative of 3 independent biological replicates. We apply a localization uncertainty threshold of 16 nm (0.1 pixel), shown in red, to select only trajectories with relatively high localization precision. All measurements were performed with 0% glycerol, except for d.

Extended Data Fig. 3 Unwinding rate distributions.

af, Histograms of the average unwinding rate of individual molecules at various concentrations of ATP concentrations (solution pH 8): 15 μM ATP (a), 25 μM ATP (b), 50 μM ATP (c), 75 μM ATP (d), 150 μM ATP (e) and 300 μM ATP (f).

Extended Data Fig. 4 RecBCD unwinding rate and pausing characteristics at solution pH 6.

a, Average unwinding rate as a function of ATP concentrations fit to a Michaelis–Menten dependence with vmax = 290 ± 10 bp s−1, KM = 130 ± 10 μM. Data are mean ± s.e.m. (n = 47, 94, 80, 37 and 110 (from left to right) trajectories from at least 3 independent biological replicates for each condition). b, ATP concentration dependence of the pause frequency. Pause frequency was determined as the average number of pauses per second for each single-molecule trajectory. Data are mean ± s.e.m. (n = 47, 94, 80, 37 and 110 (from left to right) trajectories, from at least 3 independent biological replicates for each condition). c, Median duration of pauses during forward unwinding at various concentrations of ATP. The error bars are the s.d. of the median derived from resampling (n = 76, 62, 27 and 18 (from left to right) events, from at least 3 independent biological replicates for each condition). d, Mean backtracking distance at various concentrations of ATP. Data are mean ± s.e.m. (n = 20, 22, 16 and 10 (from left to right) events, from at least 3 independent biological replicates for each condition). e, Median recovery pause duration after a backtracking event at various concentrations of ATP. The error bars are the s.d. of the median derived from resampling. The P values for the differences between the 25-μM-ATP data point and the 50-, 75- and 300-μM-ATP data points are 0.05, 0.004 and 0.09, respectively, derived from two-sided Kolmogorov–Smirnov test of the distributions of the pause durations (n = 20, 22, 16 and 10 events for the 25, 50, 75, and 300 μM data points, respectively). Note that not all trajectories contain a backtracking event. Data were acquired at 10% glycerol, solution pH 6, 500 Hz. f, Schematic of a kinetic model of DNA unwinding induced by RecBCD. During DNA unwinding, pausing occurs frequently and some pauses lead to enzyme backtracking; the enzyme can exit the backtracking state and resume DNA unwinding through a recovery pause intermediate, which is distinct from the initial pause state.

Extended Data Fig. 5 RecBCD unwinding measured using ensemble stopped-flow assay.

a, Design of the DNA substrate for the stopped-flow experiments21. The Cy3 on strand A is initially quenched by the Cy5 on strand B. RecBCD activity causes the dissociation of strand A, which results in an increase in fluorescence. The hairpin on the left side (strand C) ensures that activity can only begin from the right side. b, Dual-mixing stopped-flow experiment design. The RecBCD and DNA were mixed together for 200 ms in the delay loop before mixing with ATP and heparin, which prevents additional RecBCD–DNA binding after single turnover. c, Stopped-flow fluorescence measurements on blunt-end substrates with strand A having either 26 bp (black) or 52 bp (green), at 50 μM ATP and solution pH 8, 10% glycerol. Representative of at least 3 independent biological replicates. The ratio of the difference in strand A lengths for the two samples to the difference in measured half-rise times of the two samples allows the unwinding rate to be estimated as about 100 bp s−1, assuming that the initiation kinetics are not different for these two substrates (because they have the same geometry and sequence at the double-stranded break). This unwinding rate is comparable to the 85 bp s−1 value determined by our single-molecule ORBIT measurements at the same concentrations of ATP.

Extended Data Fig. 6 Distributions of durations of RecBCD initiation phase on various types of substrates at 50 μM ATP, solution pH 8.

Reproduction of Fig. 3e, but with individual data points of the durations of initiation phase overlaid as dot plots. Bar graph statistics are described in Fig. 3e. Note that the distributions of the durations of the initiation phase are expected to be long-tailed exponential-like distributions, owing to the stochastic nature of kinetic transitions.

Extended Data Fig. 7 RecBCD initiation kinetics at 300 μM ATP, solution pH 8.

a, Example trajectories of initiation for the blunt-end DNA substrate and DNA substrates with 6-nt 3′ or 10-nt 5′ overhangs. Representative of at least 3 independent biological replicates. The green and magenta colour-coding is as described in Fig. 3a. b, Mean duration of initiation phase (total time from substrate binding until processive unwinding) for the blunt-end substrates and substrates with 6-nt 3′ or 10-nt 5′ overhangs. The green and magenta portions indicate the mean cumulative dwell times in the green and magenta states, respectively. Error bars are s.e.m. (n = 80, 44 and 85 (from left to right) trajectories from at least 3 independent biological replicates). Individual data points of the durations of initiation phase are overlaid as dot plots. Data were acquired at 300 μM ATP, solution pH 8, 10% glycerol, 500 Hz.

Extended Data Fig. 8 RecBCD unwinding kinetics measured using ensemble stopped-flow assay.

a, Design of the DNA substrate for the stopped-flow experiments, as described in Extended Data Fig. 5a, b, except that either a 3′ or a 5′ ssDNA overhang was added as indicated (dashed lines), to prepare the respective substrates with overhangs. b, Stopped-flow fluorescence measurements for blunt-end substrates and substrates with 6-nt 3′ or 10-nt 5′ overhangs, showing slower kinetics for the substrate with the 5′ overhang. c, Stopped-flow fluorescence measurements for substrates with 10-nt 5′ overhang or 10-nt 5′ overhang with G-C pairs in the initial 5 base pairs converted to A-T. d, Comparison of predicted ensemble time course (red line) based on the initiation and unwinding rates derived from single-molecule data (collected at a solution pH 8) to measured stopped-flow time courses (coloured symbols) on the blunt-end substrate at several pH values. To simulate the ensemble time course without fit parameters, we modelled initiation as a single exponential process using our measured initiation rates from the single-molecule data, and unwinding as a series of 1-bp unwinding steps with each molecule in the simulation having an unwinding rate drawn from our experimentally measured distribution of unwinding rates from the single-molecule data. As expected, owing to the fact that the silica coverslip surface is charged (leading to local pH shift of about 2 units; see Supplementary Discussion), the predicted curve from the initiation and unwinding rates measured by single-molecule experiments at solution pH 8 (surface pH 6) matches the stopped-flow data measured at pH 6. e, Comparison of stopped-flow data (red symbols) for blunt-end substrates at pH 6 to the time courses predicted from both initiation and unwinding rates (red) or from unwinding rates alone (black line), derived from single-molecule data obtained at a solution pH 8 (surface pH 6). Thus, the inclusion of the initiation phase in the simulation is required to match the experimental results. f, g, Comparison of stopped-flow data at pH 6 (coloured symbols) shown in b, c to predicted ensemble time courses (lines with matched colour) using initiation and unwinding rates derived from the single-molecule data at a solution pH 8 (surface pH 6), as described for d. All stopped-flow experiments were conducted at 50 μM ATP, 10% glycerol, pH 6 (except for d, for which the pH is indicated in the legend) and are representative of at least 3 individual experiments for each condition. Plots with the same colours are duplicates of the same dataset placed in different panels for the purposes of comparison.

Extended Data Fig. 9 RecBCD initiation kinetics at different pH values.

a, Single-molecule trajectories showing RecBCD initiation on blunt-end substrates at 50 μM ATP, solution pH 6, 10% glycerol, recorded at 500 Hz. Arbitrary vertical offsets are applied to different trajectories for display purposes. The region in the dashed box is magnified below, in which the wound and unwound angle states are marked in magenta and green, respectively. b, Single-molecule substrates showing RecBCD initiation on substrates with 6-nt 3′ or 10-nt 5′ overhangs, under the same conditions. The region demarcated by the dashed box is magnified in the inset. Data in a, b are representative of at least 3 independent biological replicates. c, d, Mean duration of initiation phase (total time from substrate binding until processive unwinding) for the blunt-end substrate and substrates with 3′ or 5′ overhangs at solution pH 6 (c) and solution pH 7 (d), 50 μM ATP, 10% glycerol. The green and magenta portions indicate the cumulative dwell times in the green and magenta states, respectively. Error bars are s.e.m. (n = 40, 27 and 44 (from left to right) trajectories in c; n = 149, 154 and 108 (from left to right) trajectories in d, from at least 3 independent biological replicates for each condition). Individual data points of the durations of initiation phase are overlaid as dot plots.

Extended Data Fig. 10 Additional analysis of RNAP base-pair stepping.

Full probability distribution of forward step sizes (>5°) from the hidden Markov model analysis of the single-molecule trajectories of RNAP-induced DNA rotation. Here, all detected step probabilities in individual single-molecule trajectories are used to construct the histogram (instead of using the most-probable step size of each trajectory, as shown in the inset of Fig. 4d).

Supplementary information

Supplementary Information

A PDF of the Supplementary Discussion, containing additional discussions of technical points, Supplementary Methods containing all method details related to this paper, Supplementary References, Supplementary Tables 3 and 4, and Captions of all Supplementary Tables.

Reporting Summary

Supplementary Table 1

A .xlsx file with Supplementary Table 1 | DNA oligomers used for the origami rotor. This table contains a list of the DNA oligomer sequences used to fold the DNA origami rotor.

Supplementary Table 2

A .xlsx file with Supplementary Table 2 | DNA oligomers used for the origami base. This table contains a list of the DNA oligomer sequences used to fold the origami base.

Supplementary Video 1

A single-molecule ORBIT trajectory of RecBCD-induced processive DNA unwinding. A .mov file of Supplementary Video 1 | A single-molecule ORBIT trajectory of RecBCD-induced processive DNA unwinding. The origami-dsDNA complex diffuses from solution and binds to RecBCD. After a relatively brief initiation period, the origami rotates processively due to the unwinding of the dsDNA substrate by RecBCD at 50 µM ATP. The localizations of the dyes at the tip of the origami rotor are shown. Trajectory is representative of at least three independent experiments.

Supplementary Video 2

A single-molecule ORBIT trajectory of RecBCD-induced processive DNA unwinding with a long initiation phase. A .mov file with Supplementary Video 2 | A single-molecule ORBIT trajectory of RecBCD-induced processive DNA unwinding with a long initiation phase. The origami-double stranded DNA complex diffuses from solution and binds to RecBCD. After an initiation phase showing several reversible transitions between two angular positions ~170° apart due to reversible, ATP-independent unwinding transitions of the first 5 bp, the origami begins rotating processively due to the unwinding of the DNA substrate by RecBCD at 50 µM ATP. The localizations of the dyes at the tip of the origami rotor are shown. Trajectory is representative of at least three independent experiments.

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Kosuri, P., Altheimer, B.D., Dai, M. et al. Rotation tracking of genome-processing enzymes using DNA origami rotors. Nature 572, 136–140 (2019) doi:10.1038/s41586-019-1397-7

Download citation

Further reading

Comments

By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.