Skip to main content

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

DEAD-box ATPases are global regulators of phase-separated organelles

Abstract

The ability of proteins and nucleic acids to undergo liquid–liquid phase separation has recently emerged as an important molecular principle of how cells rapidly and reversibly compartmentalize their components into membrane-less organelles such as the nucleolus, processing bodies or stress granules1,2. How the assembly and turnover of these organelles are controlled, and how these biological condensates selectively recruit or release components are poorly understood. Here we show that members of the large and highly abundant family of RNA-dependent DEAD-box ATPases (DDXs)3 are regulators of RNA-containing phase-separated organelles in prokaryotes and eukaryotes. Using in vitro reconstitution and in vivo experiments, we demonstrate that DDXs promote phase separation in their ATP-bound form, whereas ATP hydrolysis induces compartment turnover and release of RNA. This mechanism of membrane-less organelle regulation reveals a principle of cellular organization that is conserved from bacteria to humans. Furthermore, we show that DDXs control RNA flux into and out of phase-separated organelles, and thus propose that a cellular network of dynamic, DDX-controlled compartments establishes biochemical reaction centres that provide cells with spatial and temporal control of various RNA-processing steps, which could regulate the composition and fate of ribonucleoprotein particles.

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: The RNA-binding core and the unstructured tails of Dhh1 are required for LLPS and formation of processing bodies.
Fig. 2: Phase separation by DDXs is widespread and evolutionary conserved.
Fig. 3: The catalytic activity of DDXs regulates compartment turnover and RNA accumulation in phase-separated organelles.
Fig. 4: DDX ATPase activity controls RNA partitioning between phase-separated compartments in vivo and in vitro.

Data availability

The datasets generated in this study are available from the corresponding author upon request.

Code availability

Diatrack is available at www.diatrack.org. Additional Matlab codes used in this study are available from www.diatrack.org/DiatrackMaria.zip.

References

  1. 1.

    Shin, Y. & Brangwynne, C. P. Liquid phase condensation in cell physiology and disease. Science 357, eaaf4382 (2017).

    Article  Google Scholar 

  2. 2.

    Banani, S. F., Lee, H. O., Hyman, A. A. & Rosen, M. K. Biomolecular condensates: organizers of cellular biochemistry. Nat. Rev. Mol. Cell Biol. 18, 285–298 (2017).

    CAS  Article  Google Scholar 

  3. 3.

    Linder, P. & Jankowsky, E. From unwinding to clamping—the DEAD box RNA helicase family. Nat. Rev. Mol. Cell Biol. 12, 505–516 (2011).

    CAS  Article  Google Scholar 

  4. 4.

    Mugler, C. F. et al. ATPase activity of the DEAD-box protein Dhh1 controls processing body formation. eLife 5, e18746 (2016).

    Article  Google Scholar 

  5. 5.

    Ozgur, S. et al. The conformational plasticity of eukaryotic RNA-dependent ATPases. FEBS J. 282, 850–863 (2015).

    CAS  Article  Google Scholar 

  6. 6.

    Franzmann, T. & Alberti, S. Prion-like low-complexity sequences: key regulators of protein solubility and phase behavior. J. Biol. Chem. 294, 7128–7136 (2018).

    Article  Google Scholar 

  7. 7.

    Kato, M. et al. Cell-free formation of RNA granules: low complexity sequence domains form dynamic fibers within hydrogels. Cell 149, 753–767 (2012).

    CAS  Article  Google Scholar 

  8. 8.

    Nott, T. J. et al. Phase transition of a disordered nuage protein generates environmentally responsive membraneless organelles. Mol. Cell 57, 936–947 (2015).

    CAS  Article  Google Scholar 

  9. 9.

    Elbaum-Garfinkle, S. et al. The disordered P granule protein LAF-1 drives phase separation into droplets with tunable viscosity and dynamics. Proc. Natl Acad. Sci. USA 112, 7189–7194 (2015).

    ADS  CAS  Article  Google Scholar 

  10. 10.

    Montpetit, B. et al. A conserved mechanism of DEAD-box ATPase activation by nucleoporins and InsP6 in mRNA export. Nature 472, 238–242 (2011).

    ADS  CAS  Article  Google Scholar 

  11. 11.

    Oberer, M., Marintchev, A. & Wagner, G. Structural basis for the enhancement of eIF4A helicase activity by eIF4G. Genes Dev. 19, 2212–2223 (2005).

    CAS  Article  Google Scholar 

  12. 12.

    Hilliker, A., Gao, Z., Jankowsky, E. & Parker, R. The DEAD-box protein Ded1 modulates translation by the formation and resolution of an eIF4F-mRNA complex. Mol. Cell 43, 962–972 (2011).

    CAS  Article  Google Scholar 

  13. 13.

    Valentin-Vega, Y. A. et al. Cancer-associated DDX3X mutations drive stress granule assembly and impair global translation. Sci. Rep. 6, 25996 (2016).

    ADS  CAS  Article  Google Scholar 

  14. 14.

    Xiol, J. et al. RNA clamping by Vasa assembles a piRNA amplifier complex on transposon transcripts. Cell 157, 1698–1711 (2014).

    CAS  Article  Google Scholar 

  15. 15.

    Mathys, H. et al. Structural and biochemical insights to the role of the CCR4-NOT complex and DDX6 ATPase in microRNA repression. Mol. Cell 54, 751–765 (2014).

    CAS  Article  Google Scholar 

  16. 16.

    Gao, Z. et al. Coupling between the DEAD-box RNA helicases Ded1p and eIF4A. eLife 5, 568 (2016).

    Google Scholar 

  17. 17.

    Balagopal, V. & Parker, R. Polysomes, P bodies and stress granules: states and fates of eukaryotic mRNAs. Curr. Opin. Cell Biol. 21, 403–408 (2009).

    CAS  Article  Google Scholar 

  18. 18.

    Teixeira, D., Sheth, U., Valencia-Sanchez, M. A., Brengues, M. & Parker, R. Processing bodies require RNA for assembly and contain nontranslating mRNAs. RNA 11, 371–382 (2005).

    CAS  Article  Google Scholar 

  19. 19.

    Buchan, J. R., Muhlrad, D. & Parker, R. P bodies promote stress granule assembly in Saccharomyces cerevisiae. J. Cell Biol. 183, 441–455 (2008).

    CAS  Article  Google Scholar 

  20. 20.

    Galganski, L., Urbanek, M. O. & Krzyzosiak, W. J. Nuclear speckles: molecular organization, biological function and role in disease. Nucleic Acids Res. 45, 10350–10368 (2017).

    CAS  Article  Google Scholar 

  21. 21.

    Dias, A. P., Dufu, K., Lei, H. & Reed, R. A role for TREX components in the release of spliced mRNA from nuclear speckle domains. Nat. Commun. 1, 97 (2010).

    ADS  Article  Google Scholar 

  22. 22.

    Mor, A. et al. Influenza virus mRNA trafficking through host nuclear speckles. Nat. Microbiol. 1, 16069 (2016).

    CAS  Article  Google Scholar 

  23. 23.

    Nishimura, K., Fukagawa, T., Takisawa, H., Kakimoto, T. & Kanemaki, M. An auxin-based degron system for the rapid depletion of proteins in nonplant cells. Nat. Methods 6, 917–922 (2009).

    CAS  Article  Google Scholar 

  24. 24.

    Kroschwald, S., Maharana, S. & Simon, A. Hexanediol: a chemical probe to investigate the material properties of membrane-less compartments. Matters 3, 1–6 (2017).

    Google Scholar 

  25. 25.

    Nott, T. J., Craggs, T. D. & Baldwin, A. J. Membraneless organelles can melt nucleic acid duplexes and act as biomolecular filters. Nat. Chem. 8, 569–575 (2016).

    CAS  Article  Google Scholar 

  26. 26.

    Langdon, E. M. et al. mRNA structure determines specificity of a polyQ-driven phase separation. Science 157, eaar7432 (2018).

    Google Scholar 

  27. 27.

    Van Treeck, B. et al. RNA self-assembly contributes to stress granule formation and defining the stress granule transcriptome. Proc. Natl Acad. Sci. USA 115, 2734–2739 (2018).

    Article  Google Scholar 

  28. 28.

    Jain, A. & Vale, R. D. RNA phase transitions in repeat expansion disorders. Nature 546, 243–247 (2017).

    ADS  CAS  Article  Google Scholar 

  29. 29.

    Rai, A. K., Chen, J.-X., Selbach, M. & Pelkmans, L. Kinase-controlled phase transition of membraneless organelles in mitosis. Nature 559, 211–216 (2018).

    ADS  CAS  Article  Google Scholar 

  30. 30.

    Ramachandran, V., Shah, K. H. & Herman, P. K. The cAMP-dependent protein kinase signaling pathway is a key regulator of P body foci formation. Mol. Cell 43, 973–981 (2011).

    CAS  Article  Google Scholar 

  31. 31.

    Longtine, M. S. et al. Additional modules for versatile and economical PCR-based gene deletion and modification in Saccharomyces cerevisiae. Yeast 14, 953–961 (1998).

    CAS  Article  Google Scholar 

  32. 32.

    Schindelin, J. et al. Fiji: an open-source platform for biological-image analysis. Nat. Methods 9, 676–682 (2012).

    CAS  Article  Google Scholar 

  33. 33.

    Vallotton, P. et al. Diatrack particle tracking software: review of applications and performance evaluation. Traffic 18, 840–852 (2017).

    CAS  Article  Google Scholar 

Download references

Acknowledgements

We would like to thank G. Stojanovski and M. Martinovic for technical assistance. We are grateful to A. Sachs and J. Liphardt for plasmids, S. Jonas, M. Jinek, P. Kimming, E. Dultz, S. Khawaja, C. Weber and M. Zedan for their critical reading of this manuscript, ScopeM and J. Kusch for help with microscopy, M. Linsenmeier and A. Küffner for discussions of image and protein sequence analysis, and members of the Weis laboratory for discussions and comments. M.H. was supported by a Human Frontier Science Program (HFSP) postdoctoral fellowship (LT000914/2015) and an ETH postdoctoral fellowship (FEL-37-14-2). S.H. and M.H. acknowledge support from an EMBO long-term fellowship (ALTF 290-2014, EMBOCOFUND2012, GA-2012-600394 to S.H.; ALTF 870-2014 to M.H.). This work was supported by the Swiss National Science Foundation (SNF 31003A_179275 and 31003A_159731 to K.W.).

Author information

Affiliations

Authors

Contributions

M.H. and K.W. conceptualized and organized the project and wrote the manuscript. M.H. analysed protein sequences, designed and performed all in vitro experiments and the E. coli imaging. R.S. designed, performed and analysed yeast experiments in Figs. 1c and 4a. M.H., R.S. and S.H. designed, performed and analysed yeast experiments in Fig. 3a. S.H. designed, performed and analysed yeast experiments in Fig. 3f. J.W. and B.M.A.F. designed, performed and analysed experiments on human cell lines. P.V. and M.H. analysed images for the Dhh1 phase diagram and the bacterial protein expression. K.W. provided experimental input, contributed to fluorescence polarization experiments and supervised the study. All authors contributed to editing the manuscript.

Corresponding author

Correspondence to Karsten Weis.

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.

Peer review information Nature thanks Steven McKnight, Rick Russell and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.

Extended data figures and tables

Extended Data Fig. 1 Phase separation behaviour of full-length and tail-less (core) Dhh1 in different pH conditions.

a, Example images for the Dhh1 pH phase diagram. Reactions were assembled in 384-well plates; reaction conditions are described in Supplementary Table 5. Reactions were incubated at room temperature for 20 min and imaged at room temperature on a Nikon wide-field microscope using an automated script (four images per well of one replicate). Dhh1 core denotes Dhh1 (residues 48–425), which lacks the low-complexity tails. Scale bars, 50 μm. b, c, For each well, individual droplets in each image were quantified using Diatrack (see Methods) for their area and mean intensity. The sum of the product of area × mean fluorescence intensity (arbitrary unit: A × I) of all droplets in one image was plotted against the Dhh1 concentration (μM). Data are mean (lines) and s.d. (shaded areas) of the four images recorded per well for one replicate. d, Mean values of the sum of (area × mean fluorescence intensity) of the four images recorded per well were plotted against Dhh1 concentration for all pH values tested. Dhh1 core protein concentrations not tested are marked by a cross.

Extended Data Fig. 2 Phase separation behaviour of full-length and tail-less (core) Dhh1 in different salt concentrations.

a, Example images for the Dhh1 salt phase diagram at pH 6.4 and 7.0. Reactions were assembled in 384 well plates; reaction conditions are described in Supplementary Table 5. Reactions were incubated at room temperature for 20 min and imaged at room temperature on a Nikon wide-field microscope using an automated script (nine images per well of one replicate). Scale bars, 50 μm. b, c, For each well, individual droplets in each image were quantified using Diatrack (see Methods) for their area and mean intensity. The sum of the product of area × mean fluorescence intensity (A × I) of all droplets in one image was plotted against the Dhh1 concentration. Data are mean (lines) and s.d. (shaded areas) of the nine images recorded per well for one replicate. d, Mean values of the (area × mean fluorescence intensity) sum of the nine images recorded per well are plotted against the Dhh1 concentration for all conditions tested. Dhh1 core protein concentrations not tested are marked by a cross.

Extended Data Fig. 3 Phase separation behaviour of full-length and tail-less (core) Dhh1 in different ATP concentrations.

a, Example images for the Dhh1 ATP phase diagram at pH 6.4 and 7.0. Reactions were assembled in 384-well plates; reaction conditions are described in Supplementary Table 5. Reactions were incubated at room temperature for 20 min and imaged at room temperature on a Nikon wide-field microscope using an automated script (nine images per well of one replicate). Scale bars, 50 μm. be, For each well, individual droplets in each image were quantified using Diatrack (see Methods) for their area and mean intensity. The sum of the product of area × mean fluorescence intensity (A × I) of all droplets in one image was plotted against the Dhh1 concentration. Data are mean (lines) and s.d. (shaded areas) of the nine images recorded per well for one replicate. pH 6.4 (b, d) and pH 7.0 (c, e). f, Mean values of the (area ×mean fluorescence intensity) sum of the nine images recorded per well are plotted against the Dhh1 concentration for all conditions tested. Dhh1 core protein concentrations not tested are marked by a cross.

Extended Data Fig. 4 Phase separation behaviour of full-length and tail-less (core) Dhh1 in different polyU concentrations.

a, Example images for the Dhh1 polyU phase diagram at pH 6.4 and 7.0. Reactions were assembled in 384-well plates; reaction conditions are described in Supplementary Table 5. Reactions were incubated at room temperature for 20 min and imaged at room temperature on a Nikon wide-field microscope using an automated script (nine images per well of one replicate). Scale bars, 50 μm. b, c, For each well, individual droplets in each image were quantified using Diatrack (see Methods) for their area and mean intensity. The sum of the product of area × mean fluorescence intensity (A × I) of all droplets in one image was plotted against the Dhh1 concentration. Data are mean (lines) and s.d. (shaded areas) of the nine images recorded per well for one replicate. d, Mean values of the (area × mean fluorescence intensity) sum of the nine images recorded per well are plotted against the Dhh1 concentration for all conditions tested. Dhh1 core protein concentrations not tested are marked by a cross.

Extended Data Fig. 5 Dhh1, Ded1 and orthologues have unstructured low-complexity sequences in their N- and C-terminal tails.

a, High-magnification images of the samples represented in Fig. 1c. Scale bars, 5 μm. Dcp2, the catalytic subunit of the Dcp1–Dcp2 mRNA decapping complex, is used as a marker of processing bodies. Images are representative of more than four independent experiments. b, Clustal Omega sequence alignment of yeast DDXs (Dhh1 and Ded1) and their human counterparts (DDX6X and DDX3X, respectively). The RecA core is displayed in green; asterisks indicate sequence identity, dots represent sequence similarity. c, Schematic representation of low-complexity sequence distribution in the unstructured tails of Arabidopsis thaliana DDXs. Clustal Omega sequence alignment of yeast Dhh1 with its three A. thaliana orthologues (AT4G00660, AT3G61240 and AT2G45810).

Extended Data Fig. 6 Three of the five E. coli DEAD-box ATPases contain low-complexity sequences, undergo phase separation in vitro and form foci in vivo.

a, Clustal Omega sequence alignment of the five E. coli DEAD-box ATPases. The RecA core is displayed in green; asterisks indicate sequence identity, dots represent sequence similarity. b, In vitro phase separation of E.coli SrmB–mCherry (6 µM) and DbpA–mCherry (6 µM) in the presence of ATP and RNA. Scale bars, 25 μm. Images are representative of three independent experiments. c, Individual imaging channels of the composite images presented in Fig. 2e. Images are representative of more than three independent experiments. Scale bars, 2 μM. d, High-magnification view of expression of E. coli DDX–mCherry samples. Scale bar, 15 µm. Images are representative of three independent experiments. e, Quantification of foci in E. coli samples for four images per construct. Cells were segmented and for each individual cell, the mean fluorescence intensity and number of foci was quantified. Cells with 0 or 1 foci were grouped for technical reasons (see Methods). Cells were binned based on the mean fluorescence intensity, which represents their expression level, and the three highest bins were excluded from further analysis because they contain cells in which fluorescence intensity has reached saturation. The percentage of cells containing 0/1, 2, 3 or more than 3 foci are plotted for each bin. There is no correlation between expression levels and focus formation in the various strains.

Extended Data Fig. 7 The catalytically deficient mutant Ded1DQAD forms constitutive stress granules, without being compromised for RNA or ATP binding.

a, Ded1DQAD–mCherry, but not wild-type Ded1–mCherry, forms stress granules (arrows) in unstressed cells. Images are representative of more than three independent experiments. Scale bars, 5 µm. b, Fluorescence polarization analysis to measure binding of MANT-ATP to either wild-type Ded1 or Ded1DQAD. c, Fluorescence polarization analysis to measure binding of a fluorescein-UTP-labelled, 100-bp-long RNA to either wild-type Ded1 or Ded1DQAD. In b and c, data are mean and s.d. from n = 3 technical replicates. Nonlinear fit (on site-binding curve) calculated using Prism (GraphPad).

Extended Data Fig. 8 The ATPase activity of Ded1 controls disassembly of stress granules.

High-magnification view of the experiment in Fig. 1b, including more time points and dimethylsulfoxide (DMSO)-treated control samples. Scale bars, 5 μm. Images are representative of three biological replicates.

Extended Data Fig. 9 DDX ATPase activity controls the turnover of nuclear compartments in human and yeast.

a, b, Depletion of the DDX ATPase UAP56 leads to an increase in nuclear speckle size. This is consistent with the model that UAP56, which does not contain LCDs and is not an essential ‘building block’ for nuclear speckles, is required for RNA turnover in speckles, and its absence would thus lead to an increased residence time of RNA in the compartment and a subsequent increase in the size of pre-existing compartments. a, A549 cells were transfected with control siRNA or UAP56 siRNA that targets the 3′ untranslated region. After 48 h, cells were infected with influenza virus (strain WSN) at a multiplicity of infection of 10 for 6 h. Cells were subjected to single-molecule RNA FISH to label viral M mRNA, and nuclear speckles were detected by immunostaining with an anti-SON antibody. ‘Viral M RNA’ is an influenza virus transcript that has been described to traffic through nuclear speckles22 and is used as a model to represent poly-adenylated, spliced cellular transcripts. Insets denote enlargements of the white squares. Scale bar, 10 μm. Images are representative of three independent experiments. b, The percentage of viral M mRNA at nuclear speckles was plotted against the nuclear speckle volume (423 nuclear speckles for each condition—control and UAP56 siRNA. c, Stably transfected A549 cells expressing wild-type or UAP56 mutant (E197A) were treated with UAP56 siRNA, and cells were subjected to immunostaining with an anti-SON antibody. Scale bars, 10 μm. Images are representative of three independent experiments. d, e, Selection and characterization of stably transfected A549 cells expressing wild-type or mutant (E197A) UAP56. For gel source data, see Supplementary Fig. 1. Images are representative of three independent experiments. d, Several cell clones stably expressing wild-type or mutant UAP56 were tested by western blot analysis using an anti-Flag antibody. Clone 2 of wild-type UAP56, and clone 3 of mutant UAP56 were selected for further studies. e, Immunofluorescence with an anti-Flag antibody shows similar expression levels of exogenous UAP56 or UAP56(E197A) in the selected stable cell lines. Scale bar, 10 μm. f, Merged images with nuclear envelope marker protein staining for experiments in Fig. 3f. Images are representative images of five (Sub2-degron) or six (Sub2) biological replicates. g, Cells were treated as in Fig. 4g. At time point t = 60 min after induction of reporter, induction cells were treated with either water or 5% 1.6 hexanediol for 20 min. Images are representative of three biological replicates. h, Quantification of the percentage of cells displaying either distinct nuclear RNA foci (transcription foci, TF; up to two to account for mitotic cells) or diffuse nuclear RNA signal in Sub2-depleted cells. Images are representative of five biological replicates with n > 380 cells per replicate. ***P = 0.0009, **** P < 0.0001, two-tailed unpaired t-test. Data are mean and s.e.m. Dots represent the mean of individual replicates.

Extended Data Fig. 10 DDX ATPase activity regulates transfer of RNA molecules between phase-separated compartments in vivo and in vitro.

a, b, In vivo, stress granule assembly after treatment with 0.5% sodium azide was monitored by Ded1–eGFP in cells expressing untagged wild-type Dhh1 or Dhh1DQAD as the sole copy, and in a dhh1Δ background. Quantification of stress granules per cell was performed using Diatrack. Images are representative of four biological replicates, with at least 855 (WT), 755 (dhh1Δ) or 106 (Dhh1DQAD) cells per replicate. Data are mean and s.d. ***P = 0.0003 (dhh1Δ), ***P = 0.0001 (Dhh1DQAD), two-tailed unpaired t-test. Dots represent mean of individual replicates. Scale bars, 5 µm. c, d, RNA transfer between Dhh1 and Ded1 droplets. c, Forward reaction: Dhh1–mCherry droplets were assembled with Cy5-labelled RNA and added to Ded1–GFP droplets. After the addition of Not1MIF4G, Dhh1 droplets dissolve and the Cy5-RNA accumulates in the Ded1 droplets. d, Inverse reaction: Ded1–mCherry droplets were assembled with Cy5-labelled RNA and added to Dhh1–GFP droplets. After addition of recombinant eIF4GC-terminus, Ded1 droplets dissolve, but the Cy5-RNA does not accumulate in the Dhh1 droplets. In contrast to the reaction shown in Fig. 4, no stabilizing agents (such as BSA or PEG) were added in c and d to make the results in the forward and inverse reaction comparable. The fluorescence intensity scaling was adjusted for the first image (before Not1 or eIF4G addition) to account for the sample dilution after the addition of Not1 or eIF4G. However, scaling of the Cy5 channel in the first image, and in the subsequent frames (20–180 s), is identical for the forward and inverse reactions to enable a direct visual comparison. e, Quantification of the reactions presented in c and d. For each experiment, the mean Cy5-RNA intensity accumulating in six Dhh1–GFP droplets is plotted over time after the addition of Not1 or eIF4G. For background correction, six identically sized areas outside of Dhh1–GFP droplets were quantified and subtracted from the intensity measured inside the Dhh1–GFP droplets. These experiments were repeated at least three times, with comparable results. Data are mean (line) and s.d. (shaded area) of six large droplets per movie and forward and inverse reaction. At t = 180 s, 16.7 ± 2.7% of the Cy5-RNA is enriched in Ded1–GFP droplets that occupy 5–7% of the surface area (n = 3 movies). f, Line scan of the Cy5 channel (raw data), at time point t = 180 s through Ded1 droplets shown in c. In the ‘forward’ reaction, Ded1 droplets enrich Cy5-RNA 2–3-fold over background.

Supplementary information

Supplementary Information

This file contains Supplementary Tables 1-5 and the gel source data.

Reporting Summary

Supplementary Data

Sequence analysis of DDX proteins from human, yeast and E. coli.

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Hondele, M., Sachdev, R., Heinrich, S. et al. DEAD-box ATPases are global regulators of phase-separated organelles. Nature 573, 144–148 (2019). https://doi.org/10.1038/s41586-019-1502-y

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.

Search

Quick links

Nature Briefing

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing