Research Article

Oxidation Products of Biogenic Emissions Contribute to Nucleation of Atmospheric Particles

See allHide authors and affiliations

Science  16 May 2014:
Vol. 344, Issue 6185, pp. 717-721
DOI: 10.1126/science.1243527

Out of the Air

New-particle formation from gaseous precursors in the atmosphere is a complex and poorly understood process with importance in atmospheric chemistry and climate. Laboratory studies have had trouble reproducing the particle formation rates that must occur in the natural world. Riccobono et al. (p. 717) used the CLOUD (Cosmics Leaving Outdoor Droplets) chamber at CERN to recreate a realistic atmospheric environment. Sulfuric acid and oxidized organic vapors in typical natural concentrations caused particle nucleation at similar rates to those observed in the lower atmosphere.


Atmospheric new-particle formation affects climate and is one of the least understood atmospheric aerosol processes. The complexity and variability of the atmosphere has hindered elucidation of the fundamental mechanism of new-particle formation from gaseous precursors. We show, in experiments performed with the CLOUD (Cosmics Leaving Outdoor Droplets) chamber at CERN, that sulfuric acid and oxidized organic vapors at atmospheric concentrations reproduce particle nucleation rates observed in the lower atmosphere. The experiments reveal a nucleation mechanism involving the formation of clusters containing sulfuric acid and oxidized organic molecules from the very first step. Inclusion of this mechanism in a global aerosol model yields a photochemically and biologically driven seasonal cycle of particle concentrations in the continental boundary layer, in good agreement with observations.

Aerosol particles affect climate both directly by scattering and absorbing solar radiation and indirectly by acting as cloud condensation nuclei (CCN) (1). New-particle formation (nucleation) via gas-to-particle conversion is the largest source of atmospheric particles in the atmosphere (2), and it is thought to contribute up to half of the global CCN inventory (3, 4). It is therefore important to provide global models with accurate laboratory measurements of nucleation rates as a function of the concentrations of vapor precursors in order to assess the impact of anthropogenic emissions (5).

Field measurements have shown that nucleation is strongly associated with the ambient vapor concentration of sulfuric acid (H2SO4), which is therefore considered to be a key component of atmospheric nucleation (69). Classical nucleation theories (CNTs) of sulfuric acid, water, and ammonia are often used in global aerosol models (5). However, they grossly underestimate the ambient nucleation rates J observed in the planetary boundary layer (BL) at a given H2SO4 concentration and overestimate the sensitivity of the nucleation rates to H2SO4 concentration when compared with field observations (10, 11). This indicates that additional compounds participate in BL nucleation.

More recently, global aerosol models have shown improved estimation of the particle number concentration in the planetary BL after the CNT nucleation rates were replaced with empirical parameterizations of field measurements using a power-law functional dependence of the form J = k[H2SO4]p (1)where the exponent p ranges between 1 and 2 (7, 9, 11). The empirical prefactor k varies by roughly a factor of 100 at different locations and times, reflecting the variability of the ambient parameters (such as temperature and ionization rates) but also the variability of other unaccounted compounds involved in atmospheric nucleation (7, 9). Oxidized biogenic organic vapors, ubiquitous in the BL, may well be important species involved in new-particle formation (7, 8, 1215), but so far the identity of these additional chemical compounds, or class of compounds, remains unknown.

Observational Background

Identification of the chemical composition of the nucleating clusters in field measurements is difficult for several reasons. Vapor concentrations are extremely low (107 cm−3 or below), and the clusters contain only a few molecules. The presence of abundant “spectator” molecules and clusters in ambient air can obscure the actively nucleating clusters. Empirical correlations, such as the association of nucleation rates with H2SO4, are also strongly affected by the concurrent diurnal variations of other species, which are formed via gas-phase photochemistry.

There is a paucity of studies demonstrating a quantitative overlap between ambient observations and laboratory measurements. Previous comparisons between laboratory experiments and ambient observations have mostly failed to reproduce ambient nucleation rates or to quantify their dependence on H2SO4 concentration in the presence of pure sulfuric acid and water, or even with the addition of ammonia or organic compounds (1622). A notable exception is the study by Chen et al. (23) showing a reasonable agreement of predicted nucleation rates with rates measured in Atlanta and Mexico City when including efficient stabilization of the acids by bases such as amines [amine concentrations exceeded 100 parts per trillion by volume (pptv) in Atlanta but were not measured in Mexico City]. Amines strongly enhance nucleation rates already in the low pptv range (24). Other studies included biogenic organic compounds but did not show direct evidence that organic compounds participate in the nucleation process itself, because they have been unable to distinguish between an effect of organics on nucleation or on the initial growth after nucleation has taken place (25, 26).

A critical limitation is that, until recently, nucleation rates were inferred by the appearance rate of particles at roughly 2-nm diameter. The appearance rate depends on the survival probability of nucleated particles growing to the observed size, which in turn depends strongly on their growth rate, making it difficult to disentangle nucleation and growth (27). Thus, there is a critical need for laboratory experiments that mimic the atmosphere under carefully controlled conditions.

Experimental Approach

We studied, under well-controlled laboratory conditions, nucleation in the presence of sulfuric acid and biogenic oxidized organic vapors at ionization rates and vapor concentrations spanning atmospheric values. The CLOUD (Cosmics Leaving Outdoor Droplets) experiment at CERN provides a unique facility to measure the evolution from gas molecules to clusters to particles under atmospheric conditions. The CLOUD chamber is a 26-m3 stainless-steel vessel that enables nucleation experiments to be carried out under extremely stable, reproducible, and essentially contaminant-free conditions (16); a discussion of minute remaining contamination, which does not influence the measured nucleation rates, is given in (28). The gas and particle phases in the chamber are continuously sampled and analyzed by a comprehensive suite of state-of-the-art instruments (28). In these experiments, condensable vapors were formed by gas-phase reactions of hydroxyl radicals (OH), the dominant oxidant in Earth’s atmosphere, with sulfur dioxide (SO2) and pinanediol (PD, C10H18O2). Sulfuric acid was formed from oxidation of SO2, and a broad range of oxidized biogenic compounds was produced from oxidation of PD. PD is a first-generation oxidation product of α-pinene; consequently, its oxidation products (in the following called BioOxOrg) closely represent later-generation oxidation products of biogenic monoterpenes. By using the experimental results, we developed a parameterization describing the dependence of nucleation rates on H2SO4 and BioOxOrg concentrations and implemented it in a global aerosol microphysics model to assess the effect of this process on new-particle formation in the continental BL.

The effects of H2SO4 and BioOxOrg on the nucleation rates were isolated by independently varying the concentration of H2SO4 or BioOxOrg [see (28) for detailed experimental setup and BioOxOrg concentration determination]. The CLOUD facility also benefits from an adjustable pion beam from the CERN Proton Synchrotron (PS) to simulate ionizing cosmic rays and from an electric clearing field of up to 20 kV m–1 when an ion-free environment is needed. For each set of gas conditions, we measured the nucleation rates J under three different ion concentrations: (i) Jn in an ion-free environment (neutral, with the electric field on); (ii) Jgcr in a ground-level ionized environment (gcr, with ions naturally generated by galactic cosmic rays); and (iii) Jπ in a typical high tropospheric environment (pion beam, provided by the CERN PS).

Results and Discussion

Figure 1A shows the measured nucleation rates at 1.7-nm mobility diameter (J1.7). Despite the scatter in the data (caused mainly by the calculation of the BioOxOrg concentration), at BioOxOrg concentrations ≥3 × 106 cm−3 the nucleation rates are similar to those observed in the ambient atmosphere (small squares in Fig. 1A) over the range of H2SO4 concentrations typically observed. At lower BioOxOrg concentrations, the J1.7 values decrease, approaching those observed for nucleation from sulfuric acid either with water alone or with water and ammonia (open markers in Fig. 1, A and B) (16).

Fig. 1 Exploration of nucleation rates.

(A) Nucleation rates J1.7 as a function of H2SO4 concentration for the whole range of BioOxOrg concentrations experimentally explored (color-coded with log[BioOxOrg]), and nucleation rates observed in the BL (small green squares) (79). Contaminants NH3 and dimethylamine are <2 and <1 pptv, respectively. Bold diamonds correspond to those runs for which the APi-TOF mass defect plots are shown in Fig. 2 and figs. S2 to S4. (B) Nucleation rates J1.7 as a function of H2SO4 concentration for a limited range of BioOxOrg concentration (1.0 ± 0.7 × 108 cm−3, ~4 pptv). The marked area represents the region of BL observations. (C) Nucleation rates as a function of BioOxOrg concentration for a limited range of H2SO4 concentrations (1.9 ± 0.7 × 106 cm−3, ~0.1 pptv). (D) Percentage of ion-induced nucleation as a function of the total nucleation, calculated from the differences of the corresponding experiments under identical conditions. In (B), (C), and (D), blue, green, and red circles correspond to Jn, Jgcr, and Jπ, respectively. The bars represent 1σ total errors, but the overall systematic scale uncertainty on H2SO4 (about factor 2) and BioOxOrg concentrations is not shown. In (A) and (B), the binary (H2SO4 and H2O, open squares) and ternary (H2SO4, H2O, and NH3, open triangles) GCR nucleation rates from CLOUD are also shown (16). The black lines correspond to unconstrained least-squares fits for GCR conditions. The gray band in (D) is shown to guide the eye.

Figure 1, B and C, shows J1.7 as a function of H2SO4 and BioOxOrg concentrations, with each other held nearly constant. For quasi-constant concentrations of BioOxOrg (~4 pptv, well within the atmospheric range) (14), we find the same power-law dependency on H2SO4 concentration (Eq. 1) as observed in field campaigns (marked area in Fig. 1B). The unconstrained least-squares fit (solid line) yields an exponent p with the corresponding 90% confidence interval of 2.17 ± 0.14 for the GCR conditions (green circles). Figure 1C shows that oxidized organic compounds contribute to J1.7 at atmospheric H2SO4 mixing ratios (0.1 pptv). The unconstrained least-squares fit of J = k[BioOxOrg]q (solid line) yields an exponent q with the corresponding 90% confidence interval of 0.80 ± 0.23 for the GCR conditions (green circles).

Last, Fig. 1D reveals the role of ions at atmospheric H2SO4 and BioOxOrg concentrations. It shows the percentage of the nucleation rate attributable to ions as a function of the total nucleation rate under GCR and pion-beam conditions (shown in green and red, respectively). At lower-tropospheric GCR conditions, the ion-induced fraction is about 60% at low nucleation rates (J1.7 ≤ 0.01 cm−3 s−1), falling to below 10% at high nucleation rates (>10 cm−3 s−1). For upper-tropospheric ionization (pion-beam conditions), the ion-induced fraction is about 10% higher than for GCR conditions. This decreasing ion contribution with increasing total nucleation rate (i.e., at higher H2SO4 and BioOxOrg concentrations) is consistent with an emerging picture that charge, bases, and oxidized organics all compete in stabilizing small H2SO4 clusters (16, 24). This picture suggests that ambient new-particle formation may involve any, or several, of these stabilizing agents, along with H2SO4, depending on meteorological conditions and trace gas levels.

The dependence of J on H2SO4 and BioOxOrg concentration can be summarized in terms of a multicomponent power lawJ = km[H2SO4]p[BioOxOrg]q (2)where km is the multicomponent prefactor. Nucleation occurs through a sequence of collisions between the growing clusters and vapor molecules where only a fraction of the BioOxOrg compounds are able to form stabilized clusters. We label the clusters with (n, m), where n and m correspond to the number of molecules of H2SO4 and BioOxOrg, respectively, present in the cluster. In the absence of coagulation or wall losses, the survival probability of a cluster (n, m) will depend on the competition between its evaporation rate (which depends on the cluster composition) and the collision rate of the cluster with a condensable vapor molecule (which depends on the vapor concentration). A critical cluster is often defined to have a size at which it is equally likely to grow or to evaporate. According to the first nucleation theorem for multicomponent systems (29), the critical cluster composition is directly linked to the exponents p and q of Eq. 2.

However, the critical cluster does not apply to systems where highly stabilized clusters are formed from the beginning: the slopes (exponents p and q of Eq. 2) are then defined by other, more important loss processes (i.e., the chamber walls in the CLOUD experiment or the preexisting aerosol in the ambient atmosphere) (30). This also means that different conditions would yield different slopes. At very low condensing molecule concentrations, the time to reach 1.7 nm is increased, and the wall loss or coagulation effects are accordingly enhanced (30), resulting in steeper slopes. At high concentrations, saturation effects similar to those found for amine concentrations (24) may be expected, resulting in shallower slopes or even complete saturation. Nevertheless, because our concentrations of H2SO4 and BioOxOrg are within the atmospheric range and our condensation sink (~3 × 10−3 s−1) is similar to typical pristine BL values (8), we conclude that our slopes are representative of atmospheric nucleation. However, this hypothesis will need to be tested in future experiments.

Confirmation that BioOxOrg compounds participate in nucleation from the very first step is provided by APi-TOF (atmospheric pressure interface time-of-flight) mass spectrometer data (28). Figure 2A shows an APi-TOF mass-defect plot where the deviation from the nominal (integer) mass (i.e., the “mass defect”) of a compound is plotted versus the exact mass (31, 32). This plot reveals the molecular composition of the nucleating negatively charged clusters (up to ~1400 Th) and shows the presence of heteromers composed of various numbers of H2SO4 and BioOxOrg molecules (n, m), in addition to sulfuric acid monomers, dimers, and trimers (red circles). Especially prominent are (1, 1) (H2SO4, BioOxOrg) (yellow circles, first band), (2, 1) (orange circles, first band), (1, 2) and (2, 2) (yellow and orange circles, respectively, second band). These heteromers dominate the negative ion signals during the nucleation events. Pure sulfuric acid clusters above the trimer are absent. Several other mass-defect plots for different conditions are shown in (28).

Fig. 2 Experimental growth in cluster sizes.

(A) Mass-defect plot of negative clusters measured by the APi-TOF and (B) the time series of the normalized concentration of growing clusters grouped in bands and the time series of the particle concentration measured with the diethylene glycol condensation particle counter (DEG CPC) with D50 = 2 nm and the CPC with D50 = 3 nm. The experiment was started by turning the electric clearing field off at 06:02 and the ultraviolet lights on at 06:11, so most of the observed growth (especially of the larger bands) is caused by formation of BioOxOrg, rather than by diffusional charging of already existing BioOxOrg. This is different from the experiments shown in Fig. 1, B and C, where the sequence neutral→GCR→pion beam was always followed. The clusters are grouped in bands, from 1 to 4, according to the number m of BioOxOrg molecules present in the clusters of band m. In the mass-defect plot, the sulfuric acid monomers, dimers, and trimers, (H2SO4)0-2HSO4, are shown in red. m oxidized organic molecules bound to HSO4 (1, m) are shown in yellow. m oxidized organic molecules bound to sulfuric acid dimers (2, m) and trimers (3, m) are shown in lighter and darker orange, respectively. In (A), the circle diameters are proportional to (count rates)½. The unidentified ions are shown in gray and nitrate-BioOxOrg clusters in dark violet. Water molecules evaporate rapidly in the APi-TOF and are not detected, as described in (31, 32). The experimental conditions for this specific run were J = 4.38 cm−3 s−1, [H2SO4] = 2.41 × 106 cm−3, [BioOxOrg] = 3.81 × 108 cm−3, temperature = 278 K, relative humidity = 39%.

The APi-TOF measurements confirm that the negatively charged clusters grow with time. Figure 2B shows the time series of the normalized cluster concentration of the four bands seen in Fig. 2A. The number m of the band corresponds to the number of BioOxOrg molecules present in the clusters of that band. The clusters additionally contain mainly one or two sulfuric acid molecules (one of which accommodates the negative charge by deprotonation to HSO4). Each new band appears with a delay of 3 to 10 min relative to the previous band, indicating that the clusters grow by the consecutive addition of BioOxOrg molecules. However, the preferential pathway for growing clusters will depend on the collision and evaporation rates, which in turn depend on the relative vapor concentrations, the oxidation state of the BioOxOrg molecules, and the charge on the cluster. Figure 2B also shows the appearance of 2- and 3-nm particles, providing evidence that the nucleated clusters indeed grow into larger particles sequentially detected by particle counters with higher cut-offs. The APi-TOF measured only negatively charged clusters, and the pathways for neutral clusters could be different. However, Schobesberger et al. (32) showed similar growth rates for charged and neutral clusters involving BioOxOrg molecules, which explains the good agreement with the particle counters in Fig. 2B.

The large enhancement of the nucleation rate with BioOxOrg is due to the formation of clusters with much lower evaporation rates than those of binary sulfuric acid clusters. Quantum chemical calculations (28) indicate that the evaporation rate of a neutral sulfuric acid dimer is much higher than that of a highly oxidized BioOxOrg molecule [with an oxygen-to-carbon ratio (O:C) of up to 1 (28)] clustered in a heterodimer with one sulfuric acid molecule (e.g., by a factor of 104 for a 3-methyl-1,2,3-butane-tricarboxylic acid molecule). Zhang et al. (2009) (18) investigated heterodimers with cis-pinonic acid; however, because of the lower O:C of 0.3, these clusters are substantially less stabilized (28), and therefore much higher concentrations were needed to observe nucleation.

Notwithstanding the caveats on the slopes of the nucleation rates versus H2SO4 and BioOxOrg concentrations, which may vary with experimental conditions, we have parameterized these dependencies to investigate the resulting particle production in a global aerosol model. Setting p = 2 and q = 1 in Eq. 2, we obtain km = 3.27 × 10−21 cm6 s−1, with a 90% confidence interval of 1.73 × 10−21 to 6.15 × 10−21 cm6 s−1. A nucleation rate in the form J = km[H2SO4]2[BioOxOrg] can be compared to rates in the form J = k′[H2SO4]2 that have been fitted empirically to ambient measurements (7, 9), which yield an effective prefactor k′ that lies in the range of 0.01 to 10 × 10−12 cm3 s−1. For example, with α-pinene mixing ratios of 200 pptv [e.g., (33)], a condensation sink rate of monomers of 10−3 s−1, and [OH] = 106 cm−3, we estimate k′ = 1.3 × 10−12 cm3 s−1, which is well within measured values (28). Although both parameterizations are consistent with ambient observations (7, 9), our new parameterization contains an explicit dependence on organics. The parameterized CLOUD data can thus be used to quantitatively test the hypothesis that organic-sulfuric acid nucleation plays an important role in the planetary BL, with relatively few assumptions, because the parameterization has its basis in laboratory measurements made under realistic, albeit limited, atmospheric conditions.

We included this parameterization in a three-dimensional global aerosol model, GLOMAP (34, 35). The concentration of BioOxOrg was calculated assuming that it derives from a two-stage oxidation of α-pinene to first-generation products and then to BioOxOrg, with the α-pinene emitted from the terrestrial biosphere (36). The new parameterization resulted in a 57% (90% confidence interval range from 46 to 70%) increase in the global annual mean particle (>3 nm in diameter) number concentration in the BL compared with a control simulation that included only binary homogenous nucleation, BHN [from a global annual mean value of 450 cm−3 for BHN to 706 cm−3 (range from 654 to 765 cm−3) for the new parameterization]. The increase in particle concentrations occurs mostly over land areas, where α-pinene is emitted; the mean fractional increase in particle number concentration is 97% (range from 78 to 119%) over land and 49% (range from 40 to 60%) over the ocean. In contrast, when the nucleation rate in the BL is defined as J = kACT[H2SO4] the enhancement is 79% over land and 137% over oceans, which is likely to be an unrealistic marine/land contrast because it neglects the modulation caused by lower organic compound concentrations over remote marine regions.

Figure 3 shows the seasonal cycle in simulated and observed particle number concentrations across 19 locations in the Northern Hemisphere continental BL (11); see (28) for further details. Model simulations based on the assumption that only H2SO4 controls nucleation in the boundary layer [J = kACT[H2SO4], with rate coefficients determined from ambient data as in Fig. 1A (7, 9)] tend to predict peak particle concentrations in early spring and autumn, with summer nucleation being suppressed by the higher condensation sink. Our parameterization (dark green line in Fig. 3) improves both the simulated magnitude and the seasonal variation of particle concentration (table S3), including the summer peak caused by the strong seasonal variation in biogenic emissions and OH concentration (OH concentration appears to the power of three in the experimentally determined rate because it accounts for the oxidation both of organics to BioOxOrg and of SO2 to H2SO4). This difference in seasonality between J = kACT[H2SO4] and our mechanism is especially apparent at sites located in or near forests (table S3) but less apparent at polluted sites where biogenic emissions play a smaller role. At the forest sites, the mean correlation coefficient between the modeled and observed monthly mean concentrations is 0.7 using our mechanism but only 0.35 using J = kACT[H2SO4]. Our mechanism can therefore explain about 50% of the temporal variability in particle concentrations, but activation nucleation can explain only 12%. The Metzger et al. (25) mechanism (J = kMET[H2SO4][NucOrg]) also improves the seasonal cycle, but it assumes that the anthropogenic organic compound used in those experiments is representative of α-pinene in the model, which may not be the case.

Fig. 3 Simulated and observed monthly mean total particle number concentration (N) across 19 Northern Hemisphere continental BL locations.

Multiannual observations, represented by the black line, are derived from measurements with minimum cut-off diameters ranging from 3 to 14 nm (11). Simulations using three different nucleation mechanisms are shown as colored lines: blue represents BHN plus activation BL nucleation (BHN + k[H2SO4]); red represents BHN plus Metzger et al. (25) in the BL; dark green represents BHN plus our mechanism (Eq. 2) in the BL [see (28) for a definition of these experiments]. Light green shading around the dark green line for our mechanism represents simulated particle concentrations at the edge of the 90% confidence interval for the value of km. Pearson correlation coefficient (R) given for each site (colored according to nucleation mechanism) at the top of each individual plot.

There are many other uncertain model processes that could control the seasonality of particle concentrations. However, in a model study sampling from the 28-dimensional uncertainty space of model processes and emissions (37, 38), the summer dip in particle concentrations was shown to be a robust feature of the model when nucleation is driven only by H2SO4. This problem seems to be overcome here by including a biogenically controlled nucleation mechanism. Figure 3 shows a remaining model-observation bias in particle concentrations in winter, which may reflect other uncertain processes or emissions (37) or may indicate that other nucleation mechanisms are operating that are controlled by nonbiogenic compounds. However, at present, there are no measurements that enable the competing effects of compounds like ammonia, amines, and oxidized organics to be accounted for in a model.


Recent experimental results and quantum chemical calculations have shown that amines and ions can also effectively stabilize the sulfuric acid clusters, reducing evaporation rates and enhancing the nucleation rates at low H2SO4 concentrations (24, 39, 40). Thus, the dominant nucleation pathway may ultimately depend on the local atmospheric concentration of H2SO4, ions, and amines and on the concentration and functionalization of BioOxOrg, all of which vary considerably over time and space. At present, these compounds cannot be combined in a single mechanism because of the unknown way in which they compete with each other to stabilize clusters. Challenging laboratory and field measurements, as well as accurate modeling, of all these variables are required to predict the dominant nucleation pathway at different locations and eventually to understand the global effect of new-particle formation on climate. However, our model simulations show that the nucleation of sulfuric-acid and oxidized biogenic organic compounds (BioOxOrg) explains some features of the observed seasonal cycle of new particles in the continental BL that cannot be explained by sulfuric acid alone. Given the experimental evidence and modeling results presented here, it appears that highly oxidized biogenic organic vapors and sulfuric acid together play a major role in new-particle formation in the BL.

Supplementary Materials

Materials and Methods

Figs. S1 to S8

Tables S1 to S4

References (4171)

Data S1 to S4

References and Notes

  1. Materials and methods are available as supplementary materials on Science Online.
  2. Acknowledgments: We thank CERN for supporting CLOUD with important technical and financial resources and providing a particle beam from the CERN PS; J.-L. Agostini, P. Carrie, L.-P. De Menezes, F. Josa, I. Krasin, R. Kristic, O. S. Maksumov, S.V. Mizin, R. Sitals, A. Wasem, and M. Wilhelmsson for their important contributions to the experiment; and C. Kuang and V.-M. Kerminen for providing the observational data sets used in Fig. 1A and the authors that contributed to them. This research was funded by the European Commission Seventh Framework Programme (Marie Curie Initial Training Network “CLOUD-ITN,” grant no. 215072), the European Research Council Advanced Grant “ATMNUCLE” (no. 227463), the Academy of Finland via the Centre of Excellence program (project no. 1118615) and grant no. 1133872, the German Federal Ministry of Education and Research (project no. 01LK0902A), the Swiss National Science Foundation (project no. 206621_125025 and 206620_130527), the Austrian Science Fund (project no. P19546 and L593), the Portuguese Foundation for Science and Technology (project no. CERN/FP/116387/2010), the Russian Foundation for Basic Research (grant N08-02-91006-CERN), the Davidow Foundation, the Royal Society Wolfson Award, and the NSF (grants AGS1136479 and CHE1012293). The data reported in this paper are available online as .txt files in the supplementary materials).

Stay Connected to Science

Navigate This Article