K2P2.1 (TREK-1)–activator complexes reveal a cryptic selectivity filter binding site

Abstract

Polymodal thermo- and mechanosensitive two-pore domain potassium (K2P) channels of the TREK1 subfamily generate ‘leak’ currents that regulate neuronal excitability, respond to lipids, temperature and mechanical stretch, and influence pain, temperature perception and anaesthetic responses1,2,3. These dimeric voltage-gated ion channel (VGIC) superfamily members have a unique topology comprising two pore-forming regions per subunit4,5,6. In contrast to other potassium channels, K2P channels use a selectivity filter ‘C-type’ gate7,8,9,10 as the principal gating site. Despite recent advances3,11,12, poor pharmacological profiles of K2P channels limit mechanistic and biological studies. Here we describe a class of small-molecule TREK activators that directly stimulate the C-type gate by acting as molecular wedges that restrict interdomain interface movement behind the selectivity filter. Structures of K2P2.1 (also known as TREK-1) alone and with two selective K2P2.1 (TREK-1) and K2P10.1 (TREK-2) activators—an N-aryl-sulfonamide, ML335, and a thiophene-carboxamide, ML402—define a cryptic binding pocket unlike other ion channel small-molecule binding sites and, together with functional studies, identify a cation–π interaction that controls selectivity. Together, our data reveal a druggable K2P site that stabilizes the C-type gate ‘leak mode’ and provide direct evidence for K2P selectivity filter gating.

Access options

Rent or Buy article

Get time limited or full article access on ReadCube.

from$8.99

All prices are NET prices.

Figure 1: K2P2.1 structures.
Figure 2: K2P2.1–activator interactions.
Figure 3: K2P2.1 C-tail and lipid binding sites.
Figure 4: K2P2.1 activator function.

Accession codes

Primary accessions

Protein Data Bank

References

  1. 1

    Feliciangeli, S., Chatelain, F. C., Bichet, D. & Lesage, F. The family of K2P channels: salient structural and functional properties. J. Physiol. (Lond.) 593, 2587–2603 (2015)

  2. 2

    Devilliers, M. et al. Activation of TREK-1 by morphine results in analgesia without adverse side effects. Nat. Commun. 4, 2941 (2013)

  3. 3

    Vivier, D. et al. Development of the first two-pore domain potassium channel TWIK-related K+ channel 1-selective agonist possessing in vivo anti-nociceptive activity. J. Med. Chem. 60, 1076–1088 (2017)

  4. 4

    Brohawn, S. G., del Mármol, J. & MacKinnon, R. Crystal structure of the human K2P TRAAK, a lipid- and mechano-sensitive K+ ion channel. Science 335, 436–441 (2012)

  5. 5

    Miller, A. N. & Long, S. B. Crystal structure of the human two-pore domain potassium channel K2P1. Science 335, 432–436 (2012)

  6. 6

    Dong, Y. Y. et al. K2P channel gating mechanisms revealed by structures of TREK-2 and a complex with Prozac. Science 347, 1256–1259 (2015)

  7. 7

    Bagriantsev, S. N., Clark, K. A. & Minor, D. L., Jr. Metabolic and thermal stimuli control K2P2.1 (TREK-1) through modular sensory and gating domains. EMBO J. 31, 3297–3308 (2012)

  8. 8

    Bagriantsev, S. N., Peyronnet, R., Clark, K. A., Honoré, E. & Minor, D. L., Jr. Multiple modalities converge on a common gate to control K2P channel function. EMBO J. 30, 3594–3606 (2011)

  9. 9

    Schewe, M. et al. A non-canonical voltage-sensing mechanism controls gating in K2P K+ channels. Cell 164, 937–949 (2016)

  10. 10

    Piechotta, P. L. et al. The pore structure and gating mechanism of K2P channels. EMBO J. 30, 3607–3619 (2011)

  11. 11

    Bagriantsev, S. N. et al. A high-throughput functional screen identifies small molecule regulators of temperature- and mechano-sensitive K2P channels. ACS Chem. Biol. 8, 1841–1851 (2013)

  12. 12

    Su, Z., Brown, E. C., Wang, W. & MacKinnon, R. Novel cell-free high-throughput screening method for pharmacological tools targeting K+ channels. Proc. Natl Acad. Sci. USA 113, 5748–5753 (2016)

  13. 13

    Brohawn, S. G., Campbell, E. B. & MacKinnon, R. Domain-swapped chain connectivity and gated membrane access in a Fab-mediated crystal of the human TRAAK K+ channel. Proc. Natl Acad. Sci. USA 110, 2129–2134 (2013)

  14. 14

    Brohawn, S. G., Campbell, E. B. & MacKinnon, R. Physical mechanism for gating and mechanosensitivity of the human TRAAK K+ channel. Nature 516, 126–130 (2014)

  15. 15

    Lolicato, M., Riegelhaupt, P. M., Arrigoni, C., Clark, K. A. & Minor, D. L., Jr Transmembrane helix straightening and buckling underlies activation of mechanosensitive and thermosensitive K2P channels. Neuron 84, 1198–1212 (2014)

  16. 16

    Cohen, A., Ben-Abu, Y., Hen, S. & Zilberberg, N. A novel mechanism for human K2P2.1 channel gating. Facilitation of C-type gating by protonation of extracellular histidine residues. J. Biol. Chem. 283, 19448–19455 (2008)

  17. 17

    Sandoz, G., Douguet, D., Chatelain, F., Lazdunski, M. & Lesage, F. Extracellular acidification exerts opposite actions on TREK1 and TREK2 potassium channels via a single conserved histidine residue. Proc. Natl Acad. Sci. USA 106, 14628–14633 (2009)

  18. 18

    Bagnéris, C. et al. Prokaryotic NavMs channel as a structural and functional model for eukaryotic sodium channel antagonism. Proc. Natl Acad. Sci. USA 111, 8428–8433 (2014)

  19. 19

    Tang, L. et al. Structural basis for inhibition of a voltage-gated Ca2+ channel by Ca2+) antagonist drugs. Nature 537, 117–121 (2016)

  20. 20

    Kearney, P. C. et al. Molecular recognition in aqueous-media – new binding-studies provide further insights into the cation-pi interaction and related phenomena. J. Am. Chem. Soc. 115, 9907–9919 (1993)

  21. 21

    Hibbs, R. E. & Gouaux, E. Principles of activation and permeation in an anion-selective Cys-loop receptor. Nature 474, 54–60 (2011)

  22. 22

    Yelshanskaya, M. V. et al. Structural bases of noncompetitive inhibition of AMPA-subtype ionotropic glutamate receptors by antiepileptic drugs. Neuron 91, 1305–1315 (2016)

  23. 23

    McClenaghan, C. et al. Polymodal activation of the TREK-2 K2P channel produces structurally distinct open states. J. Gen. Physiol. 147, 497–505 (2016)

  24. 24

    Chemin, J. et al. A phospholipid sensor controls mechanogating of the K+ channel TREK-1. EMBO J. 24, 44–53 (2005)

  25. 25

    Chemin, J. et al. Up- and down-regulation of the mechano-gated K2P channel TREK-1 by PIP 2 and other membrane phospholipids. Pflugers Arch. 455, 97–103 (2007)

  26. 26

    Murbartián, J., Lei, Q., Sando, J. J. & Bayliss, D. A. Sequential phosphorylation mediates receptor- and kinase-induced inhibition of TREK-1 background potassium channels. J. Biol. Chem. 280, 30175–30184 (2005)

  27. 27

    Patel, A. J. et al. A mammalian two pore domain mechano-gated S-like K+ channel. EMBO J. 17, 4283–4290 (1998)

  28. 28

    Maingret, F. et al. TREK-1 is a heat-activated background K+ channel. EMBO J. 19, 2483–2491 (2000)

  29. 29

    Honoré, E., Maingret, F., Lazdunski, M. & Patel, A. J. An intracellular proton sensor commands lipid- and mechano-gating of the K+ channel TREK-1. EMBO J. 21, 2968–2976 (2002)

  30. 30

    Fink, M. et al. Cloning, functional expression and brain localization of a novel unconventional outward rectifier K+ channel. EMBO J. 15, 6854–6862 (1996)

  31. 31

    Veale, E. L. et al. Influence of the N terminus on the biophysical properties and pharmacology of TREK1 potassium channels. Mol. Pharmacol. 85, 671–681 (2014)

  32. 32

    Tertyshnikova, S. et al. BL-1249 [(5,6,7,8-tetrahydro-naphthalen-1-yl)-[2-(1H-tetrazol-5-yl)-phenyl]-amine]: a putative potassium channel opener with bladder-relaxant properties. J. Pharmacol. Exp. Ther. 313, 250–259 (2005)

  33. 33

    Hardy, J. A. & Wells, J. A. Searching for new allosteric sites in enzymes. Curr. Opin. Struct. Biol. 14, 706–715 (2004)

  34. 34

    Kawate, T. & Gouaux, E. Fluorescence-detection size-exclusion chromatography for precrystallization screening of integral membrane proteins. Structure 14, 673–681 (2006)

  35. 35

    Drew, D. et al. GFP-based optimization scheme for the overexpression and purification of eukaryotic membrane proteins in Saccharomyces cerevisiae. Nat. Protocols 3, 784–798 (2008)

  36. 36

    Newstead, S., Kim, H., von Heijne, G., Iwata, S. & Drew, D. High-throughput fluorescent-based optimization of eukaryotic membrane protein overexpression and purification in Saccharomyces cerevisiae. Proc. Natl Acad. Sci. USA 104, 13936–13941 (2007)

  37. 37

    Kirchhofer, A. et al. Modulation of protein properties in living cells using nanobodies. Nat. Struct. Mol. Biol. 17, 133–138 (2010)

  38. 38

    Shaya, D. et al. Voltage-gated sodium channel (NaV) protein dissection creates a set of functional pore-only proteins. Proc. Natl Acad. Sci. USA 108, 12313–12318 (2011)

  39. 39

    Kabsch, W . Xds. Acta Crystallogr. D 66, 125–132 (2010)

  40. 40

    Evans, P. R. & Murshudov, G. N. How good are my data and what is the resolution? Acta Crystallogr. D 69, 1204–1214 (2013)

  41. 41

    Diederichs, K. & Karplus, P. A. Better models by discarding data? Acta Crystallogr. D 69, 1215–1222 (2013)

  42. 42

    Karplus, P. A. & Diederichs, K. Linking crystallographic model and data quality. Science 336, 1030–1033 (2012)

  43. 43

    Emsley, P. & Cowtan, K. Coot: model-building tools for molecular graphics. Acta Crystallogr. D 60, 2126–2132 (2004)

  44. 44

    Collaborative Computational Project, Number 4. The CCP4 suite: programs for protein crystallography. Acta Crystallogr. D 50, 760–763 (1994)

  45. 45

    Adams, P. D . et al. PHENIX: a comprehensive Python-based system for macromolecular structure solution. Acta Crystallogr. D 66, 213–221 (2010)

  46. 46

    Chen, V. B. et al. MolProbity: all-atom structure validation for macromolecular crystallography. Acta Crystallogr. D 66, 12–21 (2010)

  47. 47

    Zhou, Y., Morais-Cabral, J. H., Kaufman, A. & MacKinnon, R. Chemistry of ion coordination and hydration revealed by a K+ channel-Fab complex at 2.0 Å resolution. Nature 414, 43–48 (2001)

  48. 48

    Guizouarn, H., Gabillat, N., Motais, R. & Borgese, F. Multiple transport functions of a red blood cell anion exchanger, tAE1: its role in cell volume regulation. J. Physiol. (Lond.) 535, 497–506 (2001)

  49. 49

    Rapp, C., Goldberger, E., Tishbi, N. & Kirshenbaum, R. Cation-π interactions of methylated ammonium ions: a quantum mechanical study. Proteins 82, 1494–1502 (2014)

  50. 50

    Crowley, P. B. & Golovin, A. Cation-pi interactions in protein-protein interfaces. Proteins 59, 231–239 (2005)

  51. 51

    Payandeh, J. & Minor, D. L., Jr. Bacterial voltage-gated sodium channels (BacNa(V)s) from the soil, sea, and salt lakes enlighten molecular mechanisms of electrical signaling and pharmacology in the brain and heart. J. Mol. Biol. 427, 3–30 (2015)

Download references

Acknowledgements

We thank K.Brejc, S. Capponi, M.Grabe and L. Jan for comments, and A. Renslo for comments and synthesis advice. This work was supported by grants R01-MH093603 to D.L.M., and to M.L. and C.A. from the American Heart Association.

Author information

M.L., C.A., C.B. and D.L.M. conceived the study and designed the experiments. T.M. and Y.S. conceived and ran thallium flux assays. M.L. and C.A. performed experiments. M.L. and K.A.C. expressed and purified proteins. M.L. performed crystallization and structure determination. M.L. and C.A. performed electrophysiological experiments and analysed the data. C.B. designed synthetic routes, synthesized, and purified the compounds. D.L.M analysed data and provided guidance and support. M.L., C.A. and D.L.M. wrote the paper.

Correspondence to Daniel L. Minor Jr.

Ethics declarations

Competing interests

T.M and Y.S. are employees of Ono Pharmaceutical Co. Ltd. M.L., C.A., K.A.C., C.B. and D.L.M. declare no competing financial interests.

Additional information

Reviewer Information Nature thanks E. Honoré and the other anonymous reviewer(s) for their contribution to the peer review of this work.

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 Figure 1 K2P2.1cryst function and structure.

a, Exemplar recording from K2P2.1cryst expressed in Xenopus oocytes. Current was elicited from a −80 mV holding potential followed by a 500 ms ramp from −150 mV to +50 mV. b, K2P2.1cryst potassium selectivity recorded in Xenopus oocytes in K+/N-methyl-d-glucamine solutions (98.0 mM total) at pHo = 7.4. Data represent mean ± s.e.m. (n = 4). Dashed grey line represents Nernst equation Erev = RT/F × log([K+]o/[K+]i), where R and F have their usual thermodynamic meanings, z is equal to 1, and T = 23 °C, assuming [K+]i = 108.6 mM (ref. 48). c, Exemplar 2Fo − Fc electron density (1.0σ) for the C-tail region of K2P2.1cryst. Select residues and channel elements are indicated. d, Extracellular view of K2P2.1cryst showing environment of His126 and Ile148 (purple). Select residues are labelled. The extracellular proton sensor His12616,17 is supported by a highly conserved residue, Trp127, and contacts a gain-of-function mutant site, Ile1488, that interacts with the selectivity filter residue Asn147. This network of physical interactions indicates how changes at His12616,17 or Ile1488 could affect the C-type gate. e, f, Exemplar L2/L3 lipid electron density for K2P2.1–ML335. 2Fo − Fc (e; blue, 1.0σ) and Fo − Fc (f; pink, 3.0σ). Chains are coloured smudge and light orange. Channel elements and select residues are labelled. g, Crystal lattice packing for K2P2.1cryst showing that the C-tail makes lattice interactions stabilized by a cadmium ion coordinated between His313 of adjacent symmetry mates. Asymmetric unit is coloured smudge (chain A) and orange (chain B). Symmetry related channels are shown in slate (chain A) and cyan (chain B). Insets show the anomalous difference map (5.0σ) and locations of Cd2+ ions and their ligands. Colours defined according to PyMol, see https://pymolwiki.org/index.php/Color_Values for reference.

Extended Data Figure 2 K2P2.1cryst modulator binding pocket densities and K2P2.1cryst functional properties.

ae, Exemplar electron densities for the modulator binding pockets. ac, 2Fo − Fc densities (blue) for K2P2.1–ML335 (a; 1.5σ), K2P2.1–ML402 (b; 1.0σ) and K2P 2.1 (c; 1.0σ). Offset angle for the cation–π interactions for Lys271–ML335 and Lys271–ML402 is shown and adopts an oblique geometry common to cation–π interactions49,50. d, e, Fo − Fc densities (pink, 3.0σ) for K2P2.1–ML335 (d) and K2P2.1–ML402 (e). Final models are shown in all panels and select residues are shown and labelled. fh, Exemplar current traces for K2P2.1cryst (black) with 40 μM ML335 (orange) (f), K2P2.1cryst (black) with 80 μM ML402 (green) (g) and K2P2.1(G137I) (black) with 80 μM ML335 (blue) (h). i, Dose–response curves for K2P2.1–ML335 (black), EC50 = 14.3 ± 2.7 μM (n ≥ 5); K2P2.1cryst–ML335, EC50 = 10.5 ± 2.7 μM (n ≥ 3) (yellow orange) K2P2.1cryst–ML402, EC50 = 14.9 ± 1.6 μM (n ≥ 3); and K2P2.1(G137I)–ML335 (blue).

Extended Data Figure 3 Comparison of K2P modulator and VGIC antagonist sites.

a, Superposition of the K2P2.1–ML335 complex (smudge and orange) with the BacNaV ‘pore-only’ NaVMs structure18 (magenta and pink). Bromine site (Br) from labelled sodium channel antagonists is shown as a firebrick sphere. b, Superposition of the pore domains of the K2P2.1–ML335 complex (smudge and orange) with the pore domain of the BacNaV CaVAb (5KMD) bound to the inhibitor amlodipine (AMLOD)19, a site normally occupied by lipid19,51. Select residues of the K2P modulator pocket are shown as sticks and are labelled. CaVAb subunits are coloured cyan, marine, slate and dark blue. ML335 (yellow) and amlodipine (cyan) are shown in space filling representation.

Extended Data Figure 4 K2P modulator pocket structure and conservation.

a, b, Details of ML335 (a) and ML402 (b) interactions with K2P2.1. c, d, Representative K2P channel sequence comparisons for the M4 face (c) and P1 face (d). Purple bar and orange shading on sequence identifiers denotes the thermo- and mechanosensitive K2P2.1 subfamily. Protein secondary structure is marked above the sequences. Selectivity filter region is in red. Residues involved in direct interactions with ML335 and ML402 are orange and marked with an orange asterisk. Conserved positions are highlighted. K2P2.1 is the mouse protein used for this study. K2P2.1H (TREK-1) is the human homologue. All other K2P sequences are human origin. Sequences and identifiers are as follows: K2P2.1, NP_034737.2; human K2P2.1, NP_001017424.1; K2P10.1, NP_612190.1; K2P4.1, NP_001304019.1; K2P3.1 (TASK-1), NP_002237.1; K2P9.1 (TASK-3), NP_001269463.1; K2P5.1 (TASK-2), NP_003731.1; K2P1.1 (TWIK-1), NP_002236.103812.2; K2P6.1 (TWIK-2), NP_004823.1; K2P7.1 (KCNK7), AAI03812.2; K2P16.1 (TALK-1), NP_001128577.1; K2P17.1 (TALK-2), NP_113648.2; K2P12.1 (THIK-2), NP_071338.1; K2P12.1 (THIK-1), NP_071337.2; K2P15.1 (TASK-5), NP_071753.2; and K2P18.1 (TRESK), NP_862823.1. The dot in the K2P16.1 (TALK-1) sequence in c denotes the following, non-conserved sequence that was removed to avoid a long alignment gap: NFITPSGLLPSQEPFQTPHGKPESQQIP.

Extended Data Figure 5 K2P structure comparisons.

K2P modulator pocket views coloured by B-factor for K2P2.1 (a), K2P2.1–ML335 (b) and K2P2.1–ML402 (c). Bars show B-factor scale.

Extended Data Figure 6 K2P structure comparisons.

a, Backbone atom superposition of K2P2.1 (smudge, up), K2P2.1–ML335 (yellow, up), K2P2.1–ML402 (cyan, up), K2P10.1 (4BW5) (pink, up)6, K2P10.1 (4XDJ) (magenta, down)6, K2P10.1–norfluoxetine (4XDK) (purple, down)6, K2P4.1 (4I9W) (limon, up)13, K2P4.1(G124I) (4RUE) (marine, down)15, and K2P4.1(W262S) (4RUF) (lime green, down)15. ‘Up’ or ‘down’ denotes M4 conformation. Selectivity filter ions for K2P2.1 (smudge), K2P2.1–ML335 (yellow), and K2P2.1–ML402 (cyan) are shown as spheres. ML335 and ML402 are shown as sticks. Select channel elements are labelled. b, Superposition showing K2P2.1 chain A (smudge) and chain B (light orange). Sites of gain-of-function mutations, G137I (orange)7, and Trp2758 are indicated. c, K2P2.1–ML335 chain A (pink) and chain B (deep salmon), d, K2P2.1–ML402 chain A (blue) and chain B (teal). e, K2P2.1 chain A (smudge) and chain B (orange), K2P10.1 (4BW5) (pink)6, K2P10.1 (4XDJ) (magenta)6, K2P10.1–norfluoxetine (4XDK) (purple)6. f, K2P2.1–ML335 (pink), K2P4.1 (4I9W) (limon)13, K2P4.1(G124I) (4RUE) (blue)15, K2P4.1(W262S) (4RUF) (lime green)15. G124I from K2P4.1(G124I)15 is shown as sticks. In bf, Phe134, His126, Lys271, Trp275 and their equivalents in K2P10.1, K2P4.1 and K2P4.1(G124I), are shown as sticks. In c and d, ML335 and ML402 are shown as sticks.

Extended Data Figure 7 K2P activator responses.

a, b, Exemplar current traces for K2P10.1 (black) with 20 μM ML335 (yellow orange) (a) and K2P10.1 (black) with 20 μM ML402 (cyan) (b). c, Dose–response curves for K2P10.1 with ML335 (EC50 = 5.2 ± 0.5 μM (n > 3)) (yellow orange) and ML402 (EC50 = 5.9 ± 1.6 μM (n ≥ 4) (cyan). dg, Exemplar current traces for K2P2.1 K271Q (black) and with 20 μM ML335 (purple) (d); K2P4.1(Q258K) (black) and with 50 μM ML335 (orange) (e); K2P2.1(K271Q) (black) and with 50 μM ML402 (purple) (f); K2P4.1(Q258K) (black) and with 50 μM ML402 (orange) (g). Currents were evoked from Xenopus oocytes expressing the indicated channels from a −80 mV holding potential followed by a 500 ms ramp from −150 mV to +50 mV. Compound structures are shown. h, i, Exemplar current traces for HEK293 cell inside-out patches expressing K2P2.1 (h) and K2P2.1(K271Q) (i) to stimulation by 10 μM arachidonic acid (AA) (green). j, Current potentiation measured in HEK cells at 0 mV in response to 10 μM arachidonic acid for K2P2.1 (n = 5) and K2P2.1(K271Q) (n = 4). k, Current potentiation measured in Xenopus oocytes at 0 mV for K2P4.1 (white), K2P2.1 (black), K2P4.1(Q258K) (cyan) and K2P2.1(K271Q) (grey) in response to 10 μM BL-1249, 30 μM ML67-33, and 20 μM ML335. For all experiments (n ≥ 4). Data are mean ± s.e.m.

Extended Data Figure 8 K2P channel patch clamp recordings.

a, Dose response for K2P2.1 to ML335 (black circles) and ML402 (open circles) measured in HEK293 cells by whole-cell patch clamp. EC50 values are 5.2 ± 0.8 μM and 5.9 ± 1.6 μM for ML335 and ML402, respectively (n ≥ 3). b, c, Representative current traces and voltage–current relationship from inside-out patches on HEK293 cells expressing K2P4.1 (b) and K2P4.1(G124I) (c) with a 350-ms voltage-step protocol from −100 mV to +100 mV in 150 mM /150 mM . d, Rectification coefficients (I+100 mV/I−100 mV) calculated from n ≥ 3 current recordings obtained from the same conditions in b and c.

Extended Data Table 1 Data collection and refinement statistics
Extended Data Table 2 K2P2.1 B-factor and structure comparisons

Supplementary information

Supplementary Information

This file contains Supplementary Methods, additional references and Supplementary Data. (PDF 1785 kb)

41586_2017_BFnature22988_MOESM2_ESM.mp4

K2P2.1(TREK 1) subunit chains are colored orange and smudge. ML335 is shown in yellow sticks. Residues His126, Phe134, Asn147, Lys271, and Trp275 from the orange subunit are shown as sticks. (MP4 29247 kb)

Morph between the K2P2.1(TREK 1) and K2P2.1(TREK 1):ML335 structures

K2P2.1(TREK 1) subunit chains are colored orange and smudge. ML335 is shown in yellow sticks. Residues His126, Phe134, Asn147, Lys271, and Trp275 from the orange subunit are shown as sticks. (MP4 29247 kb)

PowerPoint slides

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Lolicato, M., Arrigoni, C., Mori, T. et al. K2P2.1 (TREK-1)–activator complexes reveal a cryptic selectivity filter binding site. Nature 547, 364–368 (2017). https://doi.org/10.1038/nature22988

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.