Letter | Published:

A global ocean inventory of anthropogenic mercury based on water column measurements

Nature volume 512, pages 6568 (07 August 2014) | Download Citation

Subjects

Abstract

Mercury is a toxic, bioaccumulating trace metal whose emissions to the environment have increased significantly as a result of anthropogenic activities such as mining and fossil fuel combustion1,2. Several recent models have estimated that these emissions have increased the oceanic mercury inventory by 36–1,313 million moles since the 1500s2,3,4,5,6,7,8,9. Such predictions have remained largely untested owing to a lack of appropriate historical data and natural archives. Here we report oceanographic measurements of total dissolved mercury and related parameters from several recent expeditions to the Atlantic, Pacific, Southern and Arctic oceans. We find that deep North Atlantic waters and most intermediate waters are anomalously enriched in mercury relative to the deep waters of the South Atlantic, Southern and Pacific oceans, probably as a result of the incorporation of anthropogenic mercury. We estimate the total amount of anthropogenic mercury present in the global ocean to be 290 ± 80 million moles, with almost two-thirds residing in water shallower than a thousand metres. Our findings suggest that anthropogenic perturbations to the global mercury cycle have led to an approximately 150 per cent increase in the amount of mercury in thermocline waters and have tripled the mercury content of surface waters compared to pre-anthropogenic conditions. This information may aid our understanding of the processes and the depths at which inorganic mercury species are converted into toxic methyl mercury and subsequently bioaccumulated in marine food webs.

Main

Mercury (Hg) is emitted to the atmosphere by natural and human sources primarily as Hg0, which is unusually volatile for a metal1. The elemental form is removed from the atmosphere after oxidation to Hg2+ which is then deposited to land and ocean. Within the ocean, Hg2+ is readily reduced to Hg0, resulting in surface waters being supersaturated in the elemental form with respect to the atmosphere. With an atmospheric lifetime of between a few months and a year as well as the evasion of Hg0 from the ocean to the atmosphere, Hg from any source can be widely dispersed across the globe5. Hg in the ocean is also subject to bioaccumulation and scavenging by organic-rich particles. Such particles eventually sink out of the surface ocean and are respired at deeper depths, transporting carbon, nutrients and metals like Hg in the process. In this way, Hg is very much like carbon dioxide (CO2) in that it is a biologically active gas that exhibits wide dispersal in the atmosphere, vigorous air–sea exchange and vertical transport in the ocean as a result of the particulate “biological carbon pump”10. Like the other group-12 elements (Zn and Cd)11 we might expect Hg distributions in the ocean to mimic macronutrients like phosphate (PO43−) (low in the surface, increasing through the thermocline, higher in the deep Pacific than in the deep Atlantic). As can be seen in some representative vertical profiles of Hg concentrations (Fig. 1), this general trend is indeed observed.

Figure 1: Hg and Premin distributions in the ocean.
Figure 1

a, Representative profiles of total dissolved Hg from the North Atlantic near Bermuda (triangles), from the South Atlantic (diamonds) and the northeast Pacific Ocean (circles; the station labelled ‘SAFe’). b, Global vertical distribution of Premin as interpolated from WOCE data (http://www.ewoce.org). Transect (black lines) is shown to the right. Figure generated using Ocean Data View (http://odv.awi.de/). AIW, Atlantic Intermediate water; NADW, North Atlantic Deep Water; AABW, Antarctic Bottom Water; PDW, Pacific Deep Water.

However, oceanic Hg distributions are a combination of pre-anthropogenic, nutrient-like and transient signals resulting from human activities over the past several centuries. Figure 2 shows the concentrations of Hg and dissolved phosphate released during organic matter remineralization (Premin is the apparent oxygen utilization divided by 170; ref. 12) measured in a variety of water masses from GEOTRACES cruises to the North and South Atlantic Ocean, the Pacific sector of the Southern Ocean, a GEOTRACES Intercalibration cruise to the subtropical northeast Pacific Ocean and non-GEOTRACES cruises to the tropical Pacific Ocean (the ‘Metalloenzyme’ cruise), the North Pacific Ocean (CLIVAR Repeat P16), and the central Arctic Ocean (2011 Polarstern cruise ARK-XXVI/3–TransArc) (refs 13,14,15,16 and K.M.M., C.H.L., G.J.S. & M.A.S., manuscript in preparation, but unpublished data available at http://www.bco-dmo.org or on request). In the water masses other than Northern Hemisphere North Atlantic Deep Water and recently subducted Antarctic Bottom Water (henceforth referred to as ‘unaffected’ deep waters), a striking correlation between Hg and Premin is seen (the reduced major axis regression line in Fig. 2). This correlation offers several important insights: (1) these water masses possess little anthropogenic Hg delivered by the biological pump, otherwise a good correlation and a y-intercept that is almost zero (−0.07 ± 0.03 pmol kg−1 Hg) would not have been observed (see Supplementary Information); (2) the slope of the line is an expression of the Hg/P ratio in sinking organic matter formed in surface waters from before the anthropogenic impact (1.02 ± 0.03 µmol Hg per mole P); (3) the relationship between Premin and Hg allows us to use it as a benchmark against which water masses that do contain anthropogenic Hg can be compared.

Figure 2: The concentration of Hg and Premin in various water masses.
Figure 2

The grey symbols are the data from deep waters (>1,000 m depth) not suspected of containing anthropogenic Hg. The remaining symbols include North Atlantic Deep Water in the North Atlantic Ocean (1), Antarctic Bottom Water sampled between Tasmania and Antarctica (2), thermocline waters from the North Atlantic Ocean (6), South Atlantic Ocean (4), the Southern Ocean between Tasmania and Antarctica (8), the Arctic Ocean (7), the northeast Pacific Ocean (5) and the tropical Pacific Ocean (3).

The impact of anthropogenic Hg emissions in the deep North Atlantic and various thermocline water masses is evident in Fig. 2, with data points that lie above the unaffected deep water regression line, thus showing evidence of anthropogenic Hg contributions; the vertical distance between the data and the line represents the amount of Hg in that water mass contributed from human sources. It is immediately apparent, however, that the degree of Hg perturbation for each water mass is not the same. This can be explored further by dividing the amount of anthropogenic Hg in each water mass by a tracer, preferably a pollutant that has a similar emissions history. This will allow the derived amount of anthropogenic mercury (Hganth) to be cross-checked against expectations as well as greatly simplify the conversion of our measurements to a scaled-up estimate of the total amount of pollutant Hg in the ocean. For this purpose, we have selected the amount of anthropogenic carbon dioxide (CO2, anth) present in each water mass (Table 1). The CO2, anth estimates were derived using the ΔC* method of Gruber and colleagues17 from a variety of data sets18, and then gridded over the whole ocean (the GLODAP database)19. The Hganth/CO2, anth ratios in most of the water masses are not statistically different from either each other or the Hg/CO2 ratio in primary anthropogenic atmospheric emissions (9.6–12.4 Mmol Hg per year; 0.79 ± 0.04 Pmol C per year; Hg/C = 14 ± 2 nmol mol−1)20,21,22. However, shallower water masses appear to have smaller mean Hganth/CO2, anth ratios than either North Atlantic Deep Water (NADW) or Antarctic Bottom Water, which have mean Hganth/CO2, anth ratios exceeding those in most known emissions sources21. The cause of this higher ratio is unclear, but it may be attributed to either high localized rates of atmospheric Hg deposition due to high rates of precipitation (Southern Ocean), enrichment caused by salt rejection during sea-ice formation14, proximity to historically strong regions of Hg emissions in North America and Europe (North Atlantic) or the prevalence of coal burning as a source of CO2 early in the Industrial Revolution. For example, surface waters near Iceland (the site of NADW formation) and Antarctica (Antarctic Bottom Water) are enriched in Hg (about 2 pM)14,23 with respect to average surface waters (0.6 pM; see below), which is consistent with greater mean Hg/CO2 ratios in these deep waters. Some alteration in Hganth/CO2, anth ratios should also be expected from the differential behaviours in the ocean between these two biologically active gases (Hganth will be moved into the thermocline and mode waters by both the biological as well as the solubility pump10, whereas CO2, anth will not be pumped biologically because oceanic primary productivity is not C-limited).

Table 1: Summary of Hg, Premin and CO2, anth data

We used the observed Hganth/CO2, anth ratios in each affected water mass to estimate the inventory of Hganth in the ocean as a whole by multiplying these ratios by the estimated amount of CO2, anth in the ocean (9.8 ± 1.6 Pmol C)19. Given the still small and evolving amount of oceanographic Hg data available, we chose to use one Hganth/CO2, anth ratio for intermediate waters (100–1,000 m; 25 ± 11 nmol mol−1) and another for the deep North Atlantic (66 ± 14 nmol mol−1); we used the GLODAP model estimate for the percentage of CO2, anth in each ocean layer: 15% in surface water, 71% in intermediate waters, and 16% in deep water. This calculation suggests that there are about 170 ± 80 Mmol of anthropogenic Hg between 100 m and 1,000 m depth and about 100 ± 20 Mmol deeper than 1,000 m.

It is not appropriate to use Premin to identify the anthropogenic impact on Hg in waters shallower than 100 m because atmospheric deposition is the primary source of Hg to the surface ocean, not particle remineralization. Alternatively, we estimated Hganth in surface waters by comparing the slope of the regression in Fig. 2 with the Hg/P ratio in contemporary suspended particulate matter. The Hg/P ratio was derived from analysis of Hg and P in mixed-layer particulate matter collected by in situ pumping performed during both the North Atlantic ‘GEOTRACES’ and tropical Pacific ‘Metalloenzyme’ cruises. This ratio is 3.4 ± 1.3 µmol Hg per mole P, indicating a factor of 3.4 ± 1.3 increase in the concentration of Hg in surface ocean particulate matter and presumably in solution as well since industrialization. This degree of secular change of Hg in surface waters is consistent with archives of atmospheric Hg deposition that indicate a 2–5-fold increase worldwide since industrialization24. The data presented here suggest that the total amount of Hg in the top 100 m of the ocean is about 22 Mmol (an average concentration of 0.6 pM). Accordingly, Hganth in this layer is about 16 ± 6 Mmol.

Our overall estimate of 290 ± 80 Mmol (rounded to two significant figures) of Hganth in the ocean is in reasonable agreement with a number of model-based predictions4,7,8,25, but suggests that the highest and lowest model estimates are implausible. On the high end is the prediction of Streets and colleagues2, who estimated an amount of Hganth in the ocean of 1,313 Mmol, which required a major contribution from artisanal and small-scale gold mining at present and in the past. It is important to test this particular inventory2 because it has featured prominently in recent negotiations concerning international efforts to curb emissions of Hg to the environment26. Our measurements and calculations here suggest that either the Streets2 estimate for past Hg anthropogenic releases is too high, or that much of the Hg they predicted to be in the ocean resides elsewhere, such as in soils. Recent work by Jaegle, Zhang and colleagues27 has provided support for this as well by using modelling fits to water column profiles that also suggest that loadings to the ocean are lower than those of Streets and colleagues2. It should be noted that the estimate for total CO2, anth to which we have indexed18 is for the year 1994. Estimates for more recent times and with different methods suggest greater CO2, anth (for example, 12.9 Pmol; ref. 28) which would predict higher values of Hganth as well (380 Mmol). However, this higher estimate is still much less than that of ref. 2.

As noted, we found that about a third of anthropogenic Hg loadings to the ocean are in deep water, particularly NADW. One model with which our results agree quite well is that of Sunderland and Mason7, who used a multi-box model that explicitly included deep water formation in the North Atlantic. In their simulation, 129 Mmol of Hganth are in ocean water shallower than 1,500 m, with another 124 Mmol in deeper waters. Thus, the prevalence of anthropogenic Hg in deep waters of the North Atlantic indicate the importance, as captured by the Sunderland and Mason model7, of deep water formation for sequestration of surface Hg on millennial timescales. This observation also leads to the conclusion, given that Hg emissions from anthropogenic sources are predicted to increase at a rate faster than in the previous few centuries20, that future loadings may somewhat overwhelm the deep water formation sink. We should therefore expect that the rate of increase of Hg in surface waters in the next few decades will be greater than the rate of increase in emissions during the same time period.

The impact of anthropogenic loadings on the oceanic Hg reservoir can be estimated with knowledge of the total amount of Hg in the ocean. Taking the North and South Atlantic concentration profiles each to represent a quarter of the whole ocean and the Pacific profiles to represent the other half, we estimated that the ocean contains 1,390 Mmol of dissolved total Hg, with 22 Mmol in the 0–100 m surface ocean, 292 Mmol in the 100–1,000 m intermediate depths and 1,260 Mmol in waters deeper than 1,000 m (the average concentration in these three layers being 0.6 pM, 0.9 pM and 1.0 pM, respectively13,14,15,16 (G.J.S., C.H.L., M.J.A.R. & C.R.H., manuscript in preparation)). These amounts are smaller than most previous estimates; for example, Sunderland and Mason7 estimated 666 Mmol in water shallower than 1,500 m and 1,095 Mmol in deeper water. Thus, analysis of the new data presented here suggests that the relative impact of human Hg emissions on the ocean is greater than previously thought: waters shallower than 1,000 m appear to have contained 120 Mmol in the pre-industrial past, and exhibit a factor-of-2.6 increase, while the ocean as a whole has experienced a factor-of-1.1 increase.

As our analysis reveals, and as has been noted elsewhere25, the impact of human Hg emissions is not uniform within the ocean. Therefore, the extent to which methyl mercury concentrations in fish have changed since industrialization, and might change in response to further perturbation (perhaps as much as a fivefold increase over pre-industrial levels by 2050)21 can be determined only following studies of the vertical patterns in Hg methylation dynamics as well as basin-scale controls on methylation of anthropogenic Hg.

Methods

Water samples were collected using ultraclean techniques29, including the use of a largely metal-free collection system and pressure filtration to 0.45 µm of water samples directly from the sampling GO-Flo bottles. Aliquots for total ‘dissolved’ Hg were collected in 250-ml, acid-washed, borosilicate glass bottles, oxidized with BrCl and analysed by cold vapour atomic fluorescence spectrometry following SnCl2 reduction and gold-trap pre-concentration30,31,32.

Premin was calculated according to Anderson and Sarmiento12 as the apparent oxygen utilization divided by 170 ± 10, where the apparent oxygen utilization is calculated as [O2]saturated minus [O2]observed, where [O2]saturated is determined from depth, temperature and salinity33.

Particulate Hg and P were determined from subsamples of quartz-fibre or polyethersulphone filters loaded with suspended matter (<51 µm) using McLane pumps. For Hg, the filter aliquots were digested with 2 M HNO3, and the digest treated as dissolved samples34. For P, polyethersulphone filter subsamples were digested in a 3:1 sulphuric acid:hydrogen peroxide solution to oxidize and dissolve the polyethersulphone filter, dried down, and then particles were digested in a 4 M HCl/HNO3/HF mixture35. The digest was analysed for multiple elements including P on a high-resolution inductively coupled plasma mass spectrometer and standardized using multi-element external standards (similar to ref. 36).

Water masses were defined primarily based on depth (as noted in Table 1), in accordance with those suggested by Talley and colleagues37. This definition represents an approximation for more refined definitions made on the basis of temperature, salinity and basin.

References

  1. 1.

    & in Treatise on Geochemistry (eds & ) Vol. 9, Ch. 4, 1–47 (Pergamon, 2003)

  2. 2.

    et al. All-time releases of mercury to the atmosphere from human activities. Environ. Sci. Technol. 45, 10485–10491 (2011)

  3. 3.

    , & The biogeochemical cycling of elemental mercury—anthropogenic influences. Geochim. Cosmochim. Acta 58, 3191–3198 (1994)

  4. 4.

    , , & A non-steady-state compartmental model of global-scale mercury biogeochemistry with interhemispheric atmospheric gradients. Geochim. Cosmochim. Acta 66, 1105–1118 (2002)

  5. 5.

    et al. Global 3-D land-ocean-atmosphere model for mercury: present-day versus preindustrial cycles and anthropogenic enrichment factors for deposition. Glob. Biogeochem. Cycles 22, GB2011 (2008)

  6. 6.

    et al. An improved global model for air-sea exchange of mercury: high concentrations over the North Atlantic. Environ. Sci. Technol. 44, 8574–8580 (2010)

  7. 7.

    & Human impacts on open ocean mercury concentrations. Glob. Biogeochem. Cycles 21, GB4022 (2007)

  8. 8.

    , & Vertical transport of anthropogenic mercury in the ocean. Glob. Biogeochem. Cycles 24, GB4014 (2010)

  9. 9.

    , , & Legacy impacts of all-time anthropogenic emissions on the global mercury cycle. Glob. Biogeochem. Cycles 27, 410–421 (2013)

  10. 10.

    & in The Carbon Cycle and Atmospheric CO2: Natural Variations Archean to Present (eds & ) 99–110 (American Geophysical Union, 1985)

  11. 11.

    , & in Treatise on Geochemistry Vol. 6 The Oceans and Marine Geochemistry (ed. ) 113–143 (Elsevier, 2003)

  12. 12.

    & Redfield ratios of remineralization determined by nutrient data-analysis. Glob. Biogeochem. Cycles 8, 65–80 (1994)

  13. 13.

    , , & Mercury in the North Atlantic Ocean: the U.S. GEOTRACES zonal and meridional sections. Deep Sea Res. II (in the press)

  14. 14.

    et al. Mercury in the Southern Ocean. Geochim. Cosmochim. Acta 75, 4037–4052 (2011)

  15. 15.

    & Vertical methylmercury distribution in the subtropical North Pacific. Mar. Chem. 132–133, 77–82 (2012)

  16. 16.

    , , , & Mercury sources, distribution, and bioavailability in the North Pacific Ocean: insights from data and models. Glob. Biogeochem. Cycles 23, GB2010 (2009)

  17. 17.

    , & An improved method for detecting anthropogenic CO2 in the oceans. Glob. Biogeochem. Cycles 10, 809–837 (1996)

  18. 18.

    et al. The oceanic sink for anthropogenic CO2. Science 305, 367–371 (2004)

  19. 19.

    et al. A global ocean carbon climatology: results from Global Data Analysis Project (GLODAP). Glob. Biogeochem. Cycles 18, GB4031 (2004)

  20. 20.

    , & Projections of global mercury emissions in 2050. Environ. Sci. Technol. 43, 2983–2988 (2009)

  21. 21.

    et al. Global emission of mercury to the atmosphere from anthropogenic sources in 2005 and projections to 2020. Atmos. Environ. 44, 2487–2499 (2010)

  22. 22.

    et al. The global carbon budget 1959–2011. Earth Syst. Sci. Data Discuss. 5, 1107–1157 (2012)

  23. 23.

    , & Mercury in the North Atlantic. Mar. Chem. 61, 37–53 (1998)

  24. 24.

    et al. Modern and historic atmospheric mercury fluxes in both hemispheres: global and regional mercury cycling implications. Glob. Biogeochem. Cycles 16, 1104 (2002)

  25. 25.

    et al. Mercury biogeochemical cycling in the ocean and policy implications. Environ. Res. 119, 101–117 (2012)

  26. 26.

    Global change and mercury cycling: challenges for implementing a global mercury treaty. Environ. Toxicol. Chem. 33, 1202–1210 (2014)

  27. 27.

    , , , & The Past 600 Years: Changing Hg Concentrations in a Global 3D Ocean Tracer Model. In 11th Int. Conf. on Mercury as a Global Pollutant (Edinburgh, 2013); abstr. M4-1000,

  28. 28.

    et al. Global ocean storage of anthropogenic carbon. Biogeosciences 10, 2169–2191 (2013)

  29. 29.

    & . Rapid and noncontaminating sampling system for trace elements in global ocean surveys. Limnol. Oceanogr. Meth. 10, 425–436 (2012)

  30. 30.

    & Picomolar mercury measurements in sea water and other materials using stannous chloride reduction and two-stage gold amalgamation with gas phase detection. Mar. Chem. 20, 227–243 (1987)

  31. 31.

    & Subnanogram determination of mercury by two-stage gold amalgamation applied to atmospheric analysis. Anal. Chem. 51, 1714–1720 (1979)

  32. 32.

    , , & An intercomparison of procedures for the determination of total mercury in seawater and recommendations regarding mercury speciation during GEOTRACES cruises. Limnol. Oceanogr. Meth. 10, 90–100 (2012)

  33. 33.

    Solubility of nitrogen, oxygen and argon in water and seawater. Deep-Sea Res. 17, 721–735 (1970)

  34. 34.

    & Bioaccumulation and trophic transfer of methylmercury in Long Island Sound. Arch. Environ. Contam. Toxicol. 51, 416–424 (2006)

  35. 35.

    & Decomposition and analysis of refractory oceanic suspended materials. Anal. Chem. 48, 886–890 (1976)

  36. 36.

    et al. The flux of bio- and lithogenic material associated with sinking particles in the mesopelagic "twilight zone" of the northwest and North Central Pacific Ocean. Deep-Sea Res. II 55, 1540–1563 (2008)

  37. 37.

    et al. North Pacific Intermediate Water in the Kuroshio Oyashio mixed water region. J. Phys. Oceanogr. 25, 475–501 (1995)

Download references

Acknowledgements

We thank the captains and crews of all cruises, as well as: P. Morton, J. Fitzsimmons, R. Shelley, A. Aguilar-Islas, R. Bundy, P. Morris, S. Owens, K. Wang, S. Rigaud and S. Pike for sample collection during the North Atlantic GEOTRACES cruise; L. Groot, D. Weiss, P. Laan, J. de Jong, R. Middag, L. Pena, A. Hartman, J. M. Godoy, L. Gerringa, M. Boyé and J. Dérot for sample collection during the South Atlantic GEOTRACES cruise; T. Goepfert, E. Bertrand and D. Moran for sampling during the Metalloenzyme cruise; and M. Rutgers van der Loeff and B. Galfond for providing samples from the 2011 Polarstern cruise ARK-XXVI/3–TransArc to the central Arctic Ocean. We are also grateful to D. Cossa and E. Sunderland for providing digital versions of their Southern Ocean and P16 data. We also thank H. Amos, L. Jaegle, B. Jonsson, R. Mason, E. Sunderland and Y. Zhang for discussions and D. Cossa for comments. This work was supported by NSF grant numbers OCE-0825108, OCE-0825157, OCE-0927274, OCE-0928191, OCE-1031271, OCE-1132480 and OCE-1132515. We thank co-Principal Investigators R. Mason and G. Gill. L.-E.H. thanks J. E Sonke for funding Arctic Ocean observations via research grant ERC-2010-StG_20091028 to JES.

Author information

Affiliations

  1. Department of Marine Chemistry and Geochemistry, Woods Hole Oceanographic Institution, Woods Hole, Massachusetts 02543, USA

    • Carl H. Lamborg
    • , Gretchen J. Swarr
    • , Kathleen M. Munson
    • , Daniel C. Ohnemus
    • , Phoebe J. Lam
    •  & Mak A. Saito
  2. Department of Earth and Environmental Sciences, Wright State University, Dayton, Ohio 45435, USA

    • Chad R. Hammerschmidt
    •  & Katlin L. Bowman
  3. Observatoire Midi-Pyrénées, Laboratoire Géosciences Environnement Toulouse, CNRS/IRD/Université Paul-Sabatier, 14, avenue Édouard Belin, 31400 Toulouse, France

    • Lars-Eric Heimbürger
  4. Department of Biological Oceanography, Royal Netherlands Institute for Sea Research, Den Burg, 1790 AB, The Netherlands

    • Micha J. A. Rijkenberg

Authors

  1. Search for Carl H. Lamborg in:

  2. Search for Chad R. Hammerschmidt in:

  3. Search for Katlin L. Bowman in:

  4. Search for Gretchen J. Swarr in:

  5. Search for Kathleen M. Munson in:

  6. Search for Daniel C. Ohnemus in:

  7. Search for Phoebe J. Lam in:

  8. Search for Lars-Eric Heimbürger in:

  9. Search for Micha J. A. Rijkenberg in:

  10. Search for Mak A. Saito in:

Contributions

C.H.L., C.R.H., K.L.B., G.J.S., D.C.O., L.-E.H., M.J.A.R. and M.A.S. participated in the GEOTRACES, Metalloenzyme cruises. C.H.L., C.R.H., K.L.B., G.J.S. and L.-E.H. performed Hg analyses. D.C.O. and P.J.L. designed the particulate sampling experiments and performed P analyses. C.H.L., C.R.H., M.J.A.R., M.A.S. and L.-E.H. designed the Hg-related experiments. C.H.L., C.R.H., K.L.B., G.J.S., K.M.M. and L.-E.H. interpreted the data. All authors contributed to manuscript preparation.

Corresponding author

Correspondence to Carl H. Lamborg.

Supplementary information

PDF files

  1. 1.

    Supplementary Information

    This file contains Supplementary Methods.

About this article

Publication history

Received

Accepted

Published

DOI

https://doi.org/10.1038/nature13563

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.