Volume 216, Issue 4 p. 1090-1103
Full paper
Free Access

Terrestrial biosphere models underestimate photosynthetic capacity and CO2 assimilation in the Arctic

Alistair Rogers

Corresponding Author

Alistair Rogers

Environmental and Climate Sciences Department, Brookhaven National Laboratory, Upton, NY, 11973-5000 USA

Author for correspondence:

Alistair Rogers

Tel: +1 631 344 2948

Email: [email protected]

Search for more papers by this author
Shawn P. Serbin

Shawn P. Serbin

Environmental and Climate Sciences Department, Brookhaven National Laboratory, Upton, NY, 11973-5000 USA

Search for more papers by this author
Kim S. Ely

Kim S. Ely

Environmental and Climate Sciences Department, Brookhaven National Laboratory, Upton, NY, 11973-5000 USA

Search for more papers by this author
Victoria L. Sloan

Victoria L. Sloan

Environmental Sciences Division, Oak Ridge National Laboratory, Oak Ridge, TN, 37831-6301 USA

Search for more papers by this author
Stan D. Wullschleger

Stan D. Wullschleger

Environmental Sciences Division, Oak Ridge National Laboratory, Oak Ridge, TN, 37831-6301 USA

Search for more papers by this author
First published: 06 September 2017
Citations: 58


  • Terrestrial biosphere models (TBMs) are highly sensitive to model representation of photosynthesis, in particular the parameters maximum carboxylation rate and maximum electron transport rate at 25°C (Vc,max.25 and Jmax.25, respectively). Many TBMs do not include representation of Arctic plants, and those that do rely on understanding and parameterization from temperate species.
  • We measured photosynthetic CO2 response curves and leaf nitrogen (N) content in species representing the dominant vascular plant functional types found on the coastal tundra near Barrow, Alaska.
  • The activation energies associated with the temperature response functions of Vc,max and Jmax were 17% lower than commonly used values. When scaled to 25°C, Vc,max.25 and Jmax.25 were two- to five-fold higher than the values used to parameterize current TBMs. This high photosynthetic capacity was attributable to a high leaf N content and the high fraction of N invested in Rubisco. Leaf-level modeling demonstrated that current parameterization of TBMs resulted in a two-fold underestimation of the capacity for leaf-level CO2 assimilation in Arctic vegetation.
  • This study highlights the poor representation of Arctic photosynthesis in TBMs, and provides the critical data necessary to improve our ability to project the response of the Arctic to global environmental change.


Carbon (C) uptake and loss from high-latitude ecosystems are highly sensitive to climate change, and these processes are poorly represented in Earth system models. The Arctic has experienced the greatest regional warming (Kaufman et al., 2009) and is projected to warm twice as much as the rest of the planet by the end of the century (IPCC, 2013). The resulting widely observed permafrost thaw and degradation are projected to continue as the region warms, leading to the release of large amounts of stored C into the atmosphere (Jorgenson et al., 2006; Schuur et al., 2009, 2015; Koven et al., 2015). At the same time, rising carbon dioxide concentration ([CO2]), increasing temperature and increased nitrogen (N) availability may stimulate CO2 uptake and lead to a continued and enhanced ‘greening’ of the Arctic landscape, creating critical uncertainty over the future of the Arctic C cycle (Sturm et al., 2001; Tape et al., 2006; Frost & Epstein, 2014; Salmon et al., 2016). Accurate projection of C fluxes at high latitudes will require improved model representation of these processes in terrestrial biosphere models (TBMs). A recent analysis has demonstrated that, in current TBMs, photosynthesis remains a dominant source of C cycle uncertainty in the Arctic (Fisher et al., 2014).

Leaf-level photosynthesis is represented in many TBMs by the Farquhar, von Caemmerer and Berry (FvCB) model (Farquhar et al., 1980; von Caemmerer, 2000). Two key parameters required by the FvCB model are the maximum carboxylation rate (Vc,max) by the photosynthetic enzyme Rubisco (EC number and the maximum electron transport rate (Jmax) associated with regeneration of the CO2 acceptor molecule ribulose-1,5-bisphosphate (RuBP). Currently, most TBMs use plant functional types (PFTs) to represent the vegetation present in different biomes, and these PFTs are parameterized with traits that are used to model the CO2 uptake of the vegetation in a given biome. There are a range of alternative approaches that seek to replace PFTs with a ‘trait-based’ approach in future TBMs (Wullschleger et al., 2014). However, these approaches, such as trait–environment linkages (e.g. van Bodegom et al., 2014), trait filtering (e.g. Fisher et al., 2012) and optimality approaches (e.g. Xu et al., 2012), still use an FvCB approach to represent photosynthesis, and still need data to inform parameterization, model development and to evaluate prognostic traits. In current TBMs, Vc,max at 25°C (Vc,max.25) is typically a PFT-specific model input, whereas Jmax at 25°C (Jmax.25) is commonly calculated from the PFT-specific Vc,max.25 using a TBM-specific ratio between Jmax.25 and Vc,max.25 (JVratio.25). The JVratio has also been used in the opposite direction to estimate Vc,max from remotely sensed chlorophyll content (Alton, 2017). Based on previous analysis (Wullschleger, 1993), JVratio.25 has long been assumed to be constant for all PFTs, but has been shown to decrease with increasing growth temperature (Kattge & Knorr, 2007). Critically, JVratio.25 can have a significant impact on photosynthetic CO2 responsiveness, with a higher JVratio.25 enabling a more responsive Rubisco-limited increase in photosynthesis as [CO2] rises (Rogers et al., 2017c). In TBMs, photosynthesis is particularly sensitive to the temperature response functions (TRFs) used to scale both Vc,max and Jmax from the reference temperature, usually 25°C, to growth temperature. However, these TRFs are typically assumed to be identical for all PFTs (Rogers et al., 2017c). Mounting evidence has shown that this is not a valid assumption and there is an ongoing effort in the community to capture PFT-specific TRFs for use in TBMs (Medlyn et al., 2002; Varhammar et al., 2015; Galmes et al., 2016). This is particularly important for the Arctic, because widely used TRFs have been derived from measurements made on temperate species and do not include measurements made below 10°C, a typical growth temperature for Arctic species (Bernacchi et al., 2001, 2003).

Sensitivity analysis, model simulations and efforts to identify model parameter uncertainty have repeatedly shown that model projections of gross and net primary productivity are particularly sensitive to Vc,max.25, and the parameters used to estimate it (Friend, 2010; Bonan et al., 2011; Lebauer et al., 2013; Sargsyan et al., 2014). Moreover, previous studies have also shown that model uncertainty in plant productivity can be significantly reduced through the incorporation of PFT-specific measured values of Vc,max (Dietze et al., 2014). However, a recent examination of the derivation of Vc,max in 10 TBMs revealed that most models do not have an explicit Arctic PFT and the four models that do include an Arctic PFT rely on limited or inappropriate datasets to estimate Arctic Vc,max.25 (Rogers, 2014).

There is a rich history of ecological research in the Arctic, including many studies that have measured photosynthesis (Wookey et al., 1995; Chapin & Shaver, 1996; Muraoka et al., 2002, 2008; Starr & Oberbauer, 2003; Starr et al., 2004; Reich et al., 2009; Albert et al., 2011; Boesgaard et al., 2012; Fletcher et al., 2012; Leffler & Welker, 2013; Patankar et al., 2013; Heskel et al., 2014; Souther et al., 2014; Saarinen et al., 2016). However, at the time we began this study, there were no published data on Vc,max or Jmax measured in Arctic vegetation that could be used to inform the representation of Arctic photosynthesis in TBMs and, to the best of our knowledge, only one other study emerged after we began our work (van de Weg et al. 2013). In short, the data needed to understand and parameterize a key process that has been demonstrated to be driving marked uncertainty in TBM projections of the C cycle in a critical biome are essentially missing.

We hypothesized that current TBM representation and parameterization of photosynthesis, which is based largely on knowledge gained in temperate systems, will be markedly different for Arctic vegetation. The goal of this study was to increase our understanding of photosynthesis in the Arctic and to provide new data and insights that could be used to reduce uncertainty in TBM projections of photosynthesis in the Arctic, either directly through improved parameterization of the Arctic PFT, or indirectly through the evaluation of emergent model states that result from alternative ‘trait-based’ modeling approaches (prognostic photosynthetic parameters). In this study, we provide the first Arctic dataset on the critical photosynthetic traits Vc,max.25 and Jmax.25, their TRFs and associated biochemical and structural traits; and compare these data with the parameterization currently used in TBMs, including the use of plant traits to derive photosynthetic parameters.

Materials and Methods

Plant material

This study was conducted on the coastal tundra at the Barrow Environmental Observatory (BEO), near Barrow, AK (71.3°N, 156.5°W; note that, on 1 December 2016, Barrow was officially renamed Utqia⋗vik following the original Inupiat name), USA. The BEO landscape is characterized by small thaw ponds and low- and high-centered polygons with a low vascular plant species diversity that is dominated by Carex aquatilis (Brown et al., 1980). Mean annual air temperature is −12°C (annual range, 31°C) and mean annual precipitation is 106 mm, with the majority falling as rain during the short summer. Soils are generally classified as Gelisols, underlain by permafrost which extends to depths of 300 m or greater, with an active layer thickness of 20–70 cm (Brown et al., 1980; Bockheim et al., 1999; Shiklomanov et al., 2010).

The measurement of leaf traits and gas exchange was conducted over an area of c. 1 km2 centered at 71.28°N, 156.65°W. This area was characterized by zones of disturbance and significant permafrost degradation, standing water, dry, high-centered polygons as well as relatively undisturbed low-centered polygonal ground that collectively provided diverse microtopography and drainage, and therefore highly suitable habitats for large stands of the different species of interest. Our goal was to measure key traits in the dominant vascular plants in this landscape, but also in those plants that represented key Arctic PFTs (Chapin et al., 1996). Our choice of species was also constrained by practical limitations of the gas exchange instrumentation: for example, the ability to clamp on a leaf with sufficient leaf area to provide an acceptable signal-to-noise ratio that enabled us to measure a CO2 response curve. We studied seven species covering four Arctic PFTs: grasses, Arctagrostis latifolia (R.Br.) Griseb, Arctophila fulva (Trin.) Andersson, Dupontia fisheri R.Br.; sedges, Carex aquatilis Wahlenb., Eriophorum angustifolium Honck.; forbs, Petasites frigidus (L.) Fr.; and deciduous shrubs, Salix pulchra Cham. As a result of the clonal nature of these species, it was not possible to determine whether individual ramets were genetically distinct. Therefore, stands separated by geomorphological features, i.e. different polygonal units or thaw ponds, were chosen to increase the likelihood that they were not members of the same clonal colony (Shaver et al., 1979). Measurements were taken between 10 July and 10 August (2012–2016), a period characterized by cool temperatures and continuous daylight. The bulk of our measurements were focused on the period of the peak biomass (mid to late July), when the first mature leaves in these species are available for gas exchange, but before the onset of leaf senescence. As a result of the scarcity of individuals of some species (S. pulchra), additional data collection associated an effort to link leaf spectral signatures with physiology (Serbin et al., 2012) and the challenge of taking good measurements (D. fisheri), our replication within a species varied (A. latifolia,= 13; A. fulva,= 26; C. aquatilis,= 36; D. fisheri,= 8; E. angustifolium,= 43; P. frigidus,= 44; S. pulchra,= 9).

Gas exchange and derived parameters

Gas exchange measurements were made in the field using two to five LI-6400XT gas exchange systems (Li-Cor, Lincoln, NE, USA) that were zeroed at the field site with a common nitrogen standard (99.9998% N2, CO2 < 0.5 ppm, H2O < 0.5 ppm; Alphagaz 2, Air Liquide American Specialty Gases LLC, Anchorage, AK, USA). We measured the response of photosynthesis (A) to intercellular CO2 concentration (Ci), commonly called A–Ci curves. Leaf chamber temperature was maintained at ambient temperature using the Peltier-based temperature control of the gas exchange system by setting the chamber block temperature to match the air temperature measured using the leaf thermocouple with the chamber open. The water vapor pressure of air entering the chamber was not controlled and therefore matched ambient conditions. However, on cold, high-humidity days, the block temperature was set to 0.5–1.5°C above ambient air temperature to increase the differential between the dew point temperature and the chamber block temperature in order to eliminate the chance of condensation inside the instrument. Even with this slight increase in temperature, the leaf water vapor pressure deficit (VPDleaf) was always below 1.0 kPa and, typically, c. 0.3 kPa. All measurements were made on fully expanded leaves. When the leaf did not completely fill the leaf chamber of the instrument, which was the case for all the graminoid species, the leaf material protruding from the chamber was marked at the edge of the gasket to identify the section that was enclosed in the chamber. The leaf was then removed from the chamber and plant, and the enclosed section was measured indoors with a ruler and hand lens to allow the determination of the leaf width to the nearest 0.25 mm. The leaf width was then used to calculate the leaf area enclosed by the leaf chamber, and gas exchange data were recomputed using the measured leaf area.

During the period of measurement, the Arctic experiences 24 h of daylight. As a result, transient decreases in chloroplast inorganic phosphate concentration and photosystem II efficiency, which can occur shortly after initial illumination, and which may alter the response of A to Ci, are not a concern (Ainsworth et al., 2003). In addition, attempts to remove plant material from the field and to conduct measurements indoors, where marked temperature manipulation would be possible, were unsuccessful. Therefore, A–Ci curves were all measured in situ. Preliminary light response curves, in which we carefully controlled for CO2 concentration and leaf temperature, indicated that, despite expectations, these Arctic species did not photosaturate below 1500 μmol m−2 s−1. Therefore, we used an irradiance of 2000 μmol m−2 s−1 for our light-saturated measurements. The differential between sample and reference infrared gas analyzers (IRGAs) was maximized using the 2 × 3-cm2 leaf chamber to increase potential leaf area, and by lowering the flow rate (typically to 350 μmol s−1) until the CO2 differential between sample and reference chambers was c. 10 μmol CO2 mol−1 at a CO2 reference chamber set point of 400 μmol mol−1 and saturating irradiance. This ensured a good signal-to-noise ratio at low [CO2]. After clamping on a leaf, the chamber was leak tested by vigorously blowing through a tube directed at the margins of the gasket. When leaks were identified (fluctuations in [CO2] in the sample cell > 1 μmol mol−1 over 15 s), the leaf was repositioned or leaks were sealed with a silicone compound (Molykote 111; Dow Corning, MI, USA). Following the established procedure (Long & Bernacchi, 2003; Bernacchi et al., 2006), each leaf was first allowed to achieve steady-state CO2 and water vapor exchange. The minimum amount of time allowed for stabilization was 20 min, but, typically, we waited more than 30 min. The reference [CO2] was then reduced stepwise to 50 μmol mol−1, returned to 400 μmol mol−1 and then increased stepwise to 1800 μmol mol−1. Each individual curve consisted of 13 separate CO2 set points and included multiple set points at a [CO2] value of 400 μmol mol−1 to allow for potential recovery from low [CO2]. Response curves were measured rapidly to avoid acclimation to a given set point. Following an adjustment to a new [CO2] value, data were logged as soon as the [CO2] in the reference cell was stable (SD < 0.75 μmol mol−1 over 20 s). Following equilibration, each curve took c. 30 min to complete.

In addition to making measurements at ambient growth temperature in all species, we also measured A–Ci response curves at multiple temperatures on the same leaf in P. frigidus and E. angustifolium in order to expand the temperature range of our dataset. This was achieved using the cooling and heating feature of the gas exchange system. In some cases, we focused on the initial slope of this response to speed up measurements and, as a result, collected more data on the temperature response of Vc,max than Jmax. In each case, following the completion of a full or partial CO2 response curve, the leaf temperature was increased by c. 5°C and allowed to stabilize for a minimum of 20 min before a new response curve was measured. At our high-temperature limit (25°C), VPDleaf did not rise above 1.75 kPa. Analysis of the data from a nearby flux tower (Torn et al., 2014 and unpublished data) showed that mean daily air temperatures on the days preceding our measurements were similar (3.7 ± 2.5 SD (5-d average), 4.3 ± 2.0 SD (10-d average), 4.3 ± 1.7 SD (30-d average)), giving us confidence that, when combining our full dataset for the analysis of temperature response functions, there would be no marked impact of potential thermal acclimation to the air temperature of the days preceding our measurements.

The photosynthetic parameters Vc,max and Jmax were estimated based on the equations originally described by Farquhar et al. (1980), where A is the minimum of the RuBP-saturated CO2 assimilation rate (Ac) and the RuBP-limited CO2 assimilation rate (Aj, Eqn 1).

urn:x-wiley:0028646X:media:nph14740:nph14740-math-0001(Eqn 1)

Triose phosphate utilization (TPU) limitation can also be a third limitation on A, but usually occurs at a higher than physiologically relevant CO2 concentration (Sharkey, 1985). We saw no evidence of TPU limitation of A in our ACi curves, despite the low measurement temperatures. One possible explanation for the absence of TPU limitation is the large root : shoot ratio (Iversen et al., 2015), and hence sink capacity, in Arctic species which would limit the potential feedback inhibition of photosynthesis by minimizing the buildup of triose phosphate in the leaves. Therefore, we did not include TPU limitation in our model formulation for the estimation of Vc,max and Jmax. We did not account for the influence of mesophyll conductance on the estimates of Vc,max and Jmax, and thus our reported ‘apparent’ values are based on intercellular as opposed to chloroplastic [CO2].

Apparent Vc,max and apparent Jmax were determined based on Ci, as described previously, using the commonly employed method of separate fitting of A to Ci for Rubisco and RuBP regeneration-limited A (Farquhar et al., 1980; Bernacchi et al., 2013). Ac was modeled based on Eqn 2, where Ci and Oi are the intercellular CO2 and O2 concentrations (Oi = 210 mmol mol−1), respectively, Γ* is the CO2 compensation point in the absence of non-photorespiratory mitochondrial respiration in the light, and Kc and Ko are the Michaelis–Menten coefficients of Rubisco activity for CO2 and O2, respectively. We used the NADPH-limited version of the equation used to describe the electron transport rate in the determination of the RuBP-limited CO2 assimilation rate (Aj, Eqn 3), and employed the values and temperature sensitivities of Kc, Ko and Γ* from Bernacchi et al. (2001).

urn:x-wiley:0028646X:media:nph14740:nph14740-math-0002(Eqn 2)
urn:x-wiley:0028646X:media:nph14740:nph14740-math-0003(Eqn 3)

We parameterized Eqns 2 and 3 of the FvCB model, using a custom program (https://github.com/TESTgroup-BNL/R-GasExchange/releases/tag/0.8) developed within the R statistical environment (R Development Core Team, 2013), to calculate the optimum apparent Vc,max, apparent Jmax and leaf respiration for each A–Ci curve. Our model fitting utilized the derivative evolution (DE) algorithm (Price et al., 2006), implemented in the R package ‘DEoptim’ (Ardia, 2009), to minimize the difference between the modeled and observed photosynthetic rate to derive the optimum apparent Vc,max and apparent Jmax for each A–Ci curve. The DE algorithm is a stochastic, population-based optimizer, which seeks the global minimum of the objective function without the need for arbitrary initial parameter values which can result in errors in optimization (Dubois et al., 2007). Our parameter optimization strategy was implemented to avoid the co-limited region of the A–Ci response curve, which, for these Arctic species, was at a high Ci: Vc,max was estimated from the lower portion of the A–Ci curve (Ci < 400 Pa) and Jmax was estimated from the upper portion of the curve (Ci > 650 Pa).

The temperature responses of Vc,max and Jmax were determined by fitting measured values against the mean Tleaf for each individual A–Ci response curve, as described previously (Medlyn et al., 2002), which allowed us to estimate the activation energy (Ea) associated with Vc,max and Jmax. We then used Eqn 4 (shown for Vc,max) to scale the measured Vc,max at Tleaf (Vc,max.T) to Vc,max at 25°C (Vc,max.25). Tleaf was expressed in K and R is the universal gas constant (8.314 J mol−1 K−1). Again, we utilized the ‘DEoptim’ parameter optimization algorithm in R to fit the temperature response model (Eqn. 4).
urn:x-wiley:0028646X:media:nph14740:nph14740-math-0004(Eqn 4)

Biochemical analysis

The leaf area enclosed by the leaf chamber of the IRGA was removed, placed in a paper envelope and dried to a constant mass (70°C, Lindberg Blue M, Thermo Scientific, Waltham MA, USA) before shipping to Brookhaven National Laboratory. Leaf samples were weighed to obtain the leaf mass area (LMA, g m−2) and then ground to a fine powder using a ball mill (2000 Geno Grinder; Spex Sample Prep, Cridersville, OH, USA). A 1.50–2.50-mg aliquot was weighed in 0.1-ml tin foil vials (AX26DR; Mettler Toledo, Columbus, OH, USA) and used to determine the C to N ratio (CNratio) and elemental N content employing a CHNS/O elemental analyzer operated in CHN mode, according to the manufacturer's instructions (2400 Series II CHNS/0 Analyzer; Perkin Elmer, Waltham, MA, USA).

The fraction of leaf N invested in Rubisco

The fraction of leaf N invested in Rubisco (FLNR) can be calculated as described previously (Thornton & Zimmermann, 2007) following Eqn 5, where leaf N content on an area basis (Na, g m−2) and Vc,max.25 are PFT-specific inputs, and the specific activity of Rubisco (αR25) and the mass ratio of total Rubisco molecular mass to N in Rubisco (FNR) are considered to be global constants.
urn:x-wiley:0028646X:media:nph14740:nph14740-math-0005(Eqn 5)

We derived FLNR from the mean Vc,max.25 (Fig. 2a), mean Na (Fig. 4a) and the values for FNR (6.22 g Rubisco g−1 N in Rubisco) and αR25 (47.3 μmol CO2 g−1 Rubisco s−1) provided by Rogers (2014).

Models considered

We looked at the model parameterization for all the TBMs represented in the fourth and fifth phases of the Coupled Climate Carbon Cycle Model Intercomparison Project (Friedlingstein et al., 2006, 2014) and those identified in a recent review of global scale models (Smith & Dukes, 2013). Only four models included explicit parameterization of an Arctic PFT and were considered in detail here (Table 1). In order to compare the parameterization of these four models with the data presented in this study and, importantly, to avoid mixing and matching temperature response functions and kinetic constants (Rogers et al., 2017c), we scaled our measured photosynthetic parameters to 25°C using the temperature response functions and Q10 values or activation energies for Vc,max and Jmax used by the models (Table 1). For the Atmosphere–Vegetation Interaction Model (AVIM), it was necessary to refit the A–Ci response curves using the kinetic constants and TRFs employed by Collatz et al. (1991), and then scale Vc,max.T to 25°C using the Q10 temperature response function for AVIM presented in Table 1. To account for the temperature acclimation formulations in the Biosphere Energy Transfer Hydrology Scheme (BETHY) and Community Land Model (CLM) used in the calculation of JVratio.25 and ΔS (Table 1), we employed data from a nearby flux tower to provide mean daily air temperature for the proceeding 10 (CLM) and 30 (BETHY) days. For CLM, this 10-d mean temperature was below the restricted temperature range implemented in the model, and therefore the 11°C default value was used in our calculations (Oleson et al., 2013).

Table 1. Parameters and equations used by the four terrestrial biosphere models (TBMs) that include an Arctic plant functional type (PFT); the Atmosphere–Vegetation Interaction Model (AVIM), the Biosphere Energy Transfer Hydrology Scheme (BETHY), the Community Land Model v4.5 (CLM4.5) and Hybrid version 6.5 (Hybrid6.5)
AVIM BETHY CLM4.5 Hybrid6.5
Form of FvCB model


Collatz et al. (1991)

urn:x-wiley:0028646X:media:nph14740:nph14740-math-0007 urn:x-wiley:0028646X:media:nph14740:nph14740-math-0008


Kull & Kruijt (1998)

Vc,max.25 (μmol m−2 s−1) 55 20 78 37
Temperature acclimation of JVratio.25 (source) NA


In CLM, acclimation to low temperature is restricted where (Tg − Tf) ≥ 11°K

Kattge & Knorr (2007)

JV ratio.25.Tg NA 2.44 2.21 NA
Jmax.25 (μmol m−2 s−1) NA 49 172 209*
Source of kinetic constants Collatz et al. (1991) Bernacchi et al. (2001, 2003), Kattge & Knorr (2007) Bernacchi et al. (2001, 2003), Kattge & Knorr (2007) Bernacchi et al. (2001), Friend (1995)
Temperature response functions urn:x-wiley:0028646X:media:nph14740:nph14740-math-0011

urn:x-wiley:0028646X:media:nph14740:nph14740-math-0012ΔS = 668.39 − 1.07(Tg − Tf)

In CLM, acclimation to low temperature is restricted where (Tg − Tf) ≥ 11

Kattge & Knorr (2007)



Temperature response parameters for Vc,max Q10 = 2.4

Ea = 71.510 kJ mol−1

Ed = 200 kJ mol−1

ΔS = 650 J K−1 mol−1

Ea = 72 kJ mol−1

Ed = 200 kJ mol−1

Ea = 65.33 kJ mol−1

c = 26.35

Temperature response parameters for Jmax NA NA NA

αT = 3.486 × 1013 mol mol−1 s−1

Ea = 79.5 kJ mol−1

Ed = 199 kJ mol−1

ΔS = 650 J K−1 mol−1

CNratio NA NA 25 NA
Na (g m−2) NA NA 1.3 1.6
Apparent FLNR NA NA 0.1365 0.078
Key model reference Lu & Ji (2006) Ziehn et al. (2011) Oleson et al. (2013) Friend & Kiang (2005)
  • A, CO2 assimilation rate; Ac, RuBP saturated CO2 assimilation; Aj, RuBP limited CO2 assimilation rate; Ap, triose phosphate export limited rate of CO2 assimilation; Tg, the preceding 10 d (CLM, 4.25°C ) and 30 d (BETHY, 4.28°C) mean air temperature (°K) calculated in this study from data from a nearby flux tower (Torn et al., 2014); Tf, freezing point of water (°K); JVratio.25.Tg, JVratio.25 acclimated to growth temperature; Ea, activation energy; Ed, deactivation energy; R, universal gas constant; ΔS, entropy factor; jmax, light saturated potential electron transport (mol electrons mol−1 chlorophyll s−1) rate; αT, a constant linking electron to chlorophyll content. *Jmax was calculated from jmax as described previously (Friend, 1995). Apparent FLNR for Hybrid6.5 was calculated from Na and Vc,max.25 based on Rogers (2014). NA, not applicable.

Statistical analysis

Significant (< 0.05) variation among species was identified using a one-way analysis of variance (ANOVA). Significant differences between an individual species (or the seven species mean) and an individual TBM input were identified using a one-sample t-test. Significant differences between the measured photosynthetic parameters and the model inputs from the four TBMs, and differences between the leaf-level photosynthesis modeled with the measured parameterization and TBM parameterization, were identified using a two-tailed Student's t-test.


Activation energy values associated with the temperature response functions of Vc,max and Jmax

Fitting an Arrhenius temperature response function (Eqn 4, Fig. 1) to our full datasets of Vc,max.T and Jmax.T showed that the derived Ea values associated with the temperature response functions of Vc,max and Jmax were c. 17% lower than the values reported by Bernacchi et al. (2001, 2003). If the Ea values derived by Bernacchi et al. (2001) were used to scale model inputs of Vc,max.25 and Jmax.25 values to a typical Arctic growth temperature (5°C), Vc,max and Jmax would be underestimated by c. 25%. Similarly scaling our measured values to 25°C using our Ea values would result in a lower Vc,max.25 and Jmax.25 than if those same values were scaled to 25°C using the TRFs of Bernacchi et al. (2001, 2003). Therefore, we used the Ea values determined in this study (Fig. 1) to scale our data to a common reference temperature of 25°C. Values of Ea generated from small species-specific datasets, or datasets with a limited temperature range, have the potential to introduce additional sources of uncertainty. Therefore, we adopted the community-level Ea values presented in Fig. 1 rather than use species-specific Ea values.

Details are in the caption following the image
Apparent maximum carboxylation rate (Vc,max, a) and apparent maximum electron transport rate (Jmax, b) measured on individual ramets at ambient growth temperature (circles, seven species) and at multiple leaf temperatures on the same ramet (upward pointing triangles, two species) in Arctagrostis latifolia (pink), Dupontia fisheri (green), Arctophila fulva (blue), Carex aquatilis (cyan), Eriophorum angustifolium (yellow), Petasites frigidus (red) and Salix pulchra (white) growing on the Barrow Environmental Observatory, Barrow, Alaska. An Arrhenius temperature response (black line) was fitted to the data in order to calculate an activation energy (Ea) for both Vc,max and Jmax.

Photosynthetic traits and key model inputs

Estimates of the key TBM inputs, apparent Vc,max.25 and apparent Jmax.25 (Fig. 2a,b), were determined from A–Ci response curves made at growth temperature and scaled to 25°C using the temperature response function shown in Eqn 4 and the Ea values for Vc,max and Jmax determined in this study (Fig. 1). These values were used to calculate the apparent JVratio at 25°C (JVratio.25, Fig. 2c). Apparent mean Vc,max.25 (Fig. 2a) showed significant (F6,173 = 19.2, < 0.001) variation between species, ranging from 69 μmol m−2 s−1 in D. fisheri to 113 μmol m−2 s−1 in A. fulva. Apparent JVratio.25 did not differ significantly between species (F6,142 = 1.9, = 0.08, Fig. 2c), and therefore species variation in apparent Jmax.25 mirrored apparent Vc,max.25, ranging from 183 μmol m−2 s−1 in D. fisheri to 297 μmol m−2 s−1 in A. fulva (F6,142 = 14.4, < 0.001, Fig. 2b). In order to compare the shape of the A–Ci curves between species that were measured at a range of temperatures, we modeled the response of A to Ci at two leaf temperatures (5 and 15°C) using Eqns 13 and the mean species-specific values for Vc,max.25 and Jmax.25 presented here (Fig. 2a,b). In all seven species at both 5°C (Fig. 3a) and 15°C (Fig. 3b), photosynthesis was Rubisco limited (RuBP saturated) at current [CO2]. At 5°C, photosynthesis was Rubisco limited throughout the A–Ci curve. Even at 15°C, photosynthesis did not become RuBP limited in any species until the atmospheric CO2 concentration reached c. 785 μmol mol−1 (a Ci of c. 550 μmol mol−1, Fig. 3b).

Details are in the caption following the image
Apparent maximum carboxylation rate (Vc,max.25, a) and apparent maximum electron transport rate (Jmax.25, b) scaled to 25°C using the Ea values determined from the temperature response of Vc,max and Jmax presented in Fig. 1, and the ratio of Jmax.25 to Vc,max.25 (JVratio.25, c). Data were measured in seven species located on the Barrow Environmental Observatory, Barrow Alaska. All data presented in this figure were derived from measurements made at ambient temperature. Box plots show the interquartile range (box), median (solid line) and mean (broken line). The whiskers show the lowest and highest datum still within 1.5 × interquartile range of the lower and upper quartiles (= 8–44 plants).
Details are in the caption following the image
Synthetic ACi curves modeled at 5°C (a) and 15°C (b). The response of photosynthesis (A) to rising intercellular CO2 concentration (Ci) was modeled based on the data presented in Figs 1 and 2 and Eqns 13 for the seven species considered in this study: Arctagrostis latifolia (pink), Dupontia fisheri (green), Arctophila fulva (blue), Carex aquatilis (cyan), Eriophorum angustifolium (dark yellow), Petasites frigidus (red) and Salix pulchra (black).

Leaf N content (Na, Fig. 4a), calculated from the leaf mass area (LMA, Fig. 4b) and mass-based N content (Nm, data not shown), also varied significantly between species (F6,151 = 37.6, < 0.001), but did not show the same pattern as Vc,max.25 or Jmax.25. The LMA and CNratio (Fig. 4c) both varied significantly between species (LMA, F6,162 = 40.4, < 0.001; CNratio, F6,153 = 64, < 0.001). Our paired gas exchange and Na dataset allowed us to calculate FLNR (Fig. 5), which varied significantly between species (F6,156 = 80, < 0.001) and was notably high in A. fulva, where 34% of leaf N was invested in Rubisco.

Details are in the caption following the image
Leaf nitrogen content (Na, g m−2, a), leaf mass area (LMA, g m−2, b) and the ratio of leaf carbon to nitrogen content (CNratio, c). Data were collected in parallel with gas exchange in seven species located on the Barrow Environmental Observatory, Barrow Alaska. The broken lines in (a) indicate the mean value of Na used by the Hybrid (short dashes) and Community Land Model (CLM) (long dashes) to parameterize the Arctic plant functional type (PFT). The broken line in (c) indicates the CNratio used by CLM to parameterize the Arctic PFT. Box plots show the interquartile range (box), median (solid line) and mean (broken line). The whiskers show the lowest and highest datum still within 1.5 × interquartile range of the lower and upper quartiles (= 8–44 plants).
Details are in the caption following the image
The apparent fraction of leaf nitrogen invested in Rubisco (FLNR) calculated using Eqn 5 from the data provided in Figs 2 and 4 and Rogers (2014). The broken lines show the value of FLNR used in the Community Land Model (CLM) to parameterize the Arctic plant functional type (PFT, long dashes) and the FLNR calculated for Hybrid based on Na and Vc,max.25 in Table 1 (short dashes). Box plots show the interquartile range (box), median (solid line) and mean (broken line). The whiskers show the lowest and highest datum still within 1.5 × interquartile range of the lower and upper quartiles (= 8–44 plants).

Vc,maxleaf N relationship

We found no clear relationship between apparent Vc,max.25 and Na (Fig. 6). We doubled the size of our Vc,maxNa dataset by leveraging estimates of apparent Vc,max.25 derived from an additional dataset of light-saturated A (Asat) using the one-point method for the estimation of Vc,max.25 (De Kauwe et al., 2016) and coupled measurements of Nm and LMA. When we included the derived apparent Vc,max.25 and Na data in our analysis, we still found that Na could explain no more than 2% of the variation in Vc,max.25 (Fig. 6).

Details are in the caption following the image
Apparent maximum carboxylation rate scaled to 25°C (Vc,max.25) plotted against the area-based leaf nitrogen content (Na). Estimates of apparent Vc,max.25 were scaled from gas exchange measurements made at growth temperature. These estimates were derived from ACi curves (circles) made between 2012 and 2015, and from steady-state, light-saturated photosynthesis using the one-point method made in 2016 (downward pointing triangles). Measurements were made on individual ramets of Arctagrostis latifolia (pink), Dupontia fisheri (green), Arctophila fulva (blue), Carex aquatilis (cyan), Eriophorum angustifolium (yellow), Petasites frigidus (red) and Salix pulchra (white) growing on the Barrow Environmental Observatory, Barrow, Alaska. No significant correlation was found.

Comparison with TBM parameterization

To enable fair comparison of the gas exchange data presented here with the model parameterization in the four TBMs considered, we scaled our data from growth temperature to 25°C using model-specific temperature response functions and parameterization (Table 1). All four models markedly and significantly (one-sample t-tests, < 0.05) underestimated Vc,max.25 in all seven species (Fig. 7), ranging from CLM (Fig. 7c, long dashes), where the model parameterization was 42% lower than the seven species mean, to BETHY (Fig. 7b), where the model parameterization was five times lower than the seven species mean. The model comparison with measured Jmax.25 (Fig. 8) shows that, for BETHY (Fig. 8a) and CLM (Fig. 8b), Jmax.25 was markedly and significantly higher than the model parameterization: 55% higher in CLM (Fig. 8b) and over five-fold higher in BETHY (Fig. 8a; Table 1). In TBMs, Jmax.25 is typically calculated from the model estimate of Vc,max.25 and JVratio.25. The species mean JVratio.25 (2.53 ± 0.04 SE, Fig. 2c) is 4% and 13% higher than JVratio.25 used by BETHY (2.44) and CLM (2.21), respectively (one-sample t-tests, < 0.05, Fig. 2c, Table 1). The use of our Arctic Ea values resulted in a small (c. 10%), but significant (t(6) = 8.1, < 0.001), increase in the observed JVratio.25 when compared with the JVratio.25 that would have been obtained using the Ea values provided by Bernacchi et al. (2001, 2003). The Hybrid model uses a different approach to estimate Jmax.25 based on Na (Table 1); as a result, the pattern of derived Jmax.25 for Hybrid follows that of Na (Fig. 4a). The Hybrid Jmax.25 estimate was significantly higher than A. latifolia, D. fisheri and A. fulva, but significantly lower than C. aquatilis, E. angustifolium, P. frigidus and S. pulchra (one-sample t-tests, < 0.05).

Details are in the caption following the image
Apparent maximum carboxylation rate measured at growth temperature and scaled to 25°C (apparent Vc,max.25) using the temperature response functions and parameterization associated with the four terrestrial biosphere models (TBMs) considered in this study (Table 1). The TBM model inputs are shown with a broken line: Atmosphere–Vegetation Interaction Model (AVIM) (a), Biosphere Energy Transfer Hydrology Scheme (BETHY) (b), Community Land Model (CLM) (long dashes, c) and Hybrid (short dashes, c). Gas exchange was measured in seven species located on the Barrow Environmental Observatory, Barrow, Alaska. Box plots show the interquartile range (box), median (solid line) and mean (broken line). The whiskers show the lowest and highest datum still within 1.5 × interquartile range of the lower and upper quartiles (= 8–44 plants).
Details are in the caption following the image
Apparent maximum electron transport rate measured at growth temperature and scaled to 25°C (apparent Jmax.25) using the temperature response functions, parameterization or the relationship with Na (Hybrid), associated with three of the four terrestrial biosphere models (TBMs: Biosphere Energy Transfer Hydrology Scheme (BETHY), a; Community Land Model (CLM), b; Hybrid, c) considered in this study. The broken lines with long dashes in each panel indicate the TBM parameterization for the Arctic plant functional type (PFT). Gas exchange was measured in seven species located on the Barrow Environmental Observatory, Barrow, Alaska. Box plots show the interquartile range (box), median (solid line) and mean (broken line). The whiskers show the lowest and highest datum still within 1.5 × interquartile range of the lower and upper quartiles (= 8–44 plants).

Model parameterization associated with leaf chemistry more closely matched the observations. For Na, the seven-species mean was not significantly different from the value used by Hybrid (Fig. 4a, short dashes), but was significantly (one-sample t(6) = 2.1, < 0.05) higher than the mean used by CLM (Fig. 4a, long dashes). CLM also uses CNratio as a model input and the seven-species mean was 23% lower than the CLM value (one-sample t(6) = 2.4, > 0.05). With the exception of E. angustifolium, all species had a higher investment in Rubisco than that prescribed by CLM, and the seven-species mean was c. 30% higher than the model input (one sample t(6) = 2.1, < 0.05, Fig. 5, long dashed line). The FLNR value calculated for Hybrid was 60% lower than the observed seven-species mean (Fig. 5, short dashed line).

Comparison of modeled and measured CO2 assimilation

To evaluate the effect of these new data on the modeled CO2 uptake in the Arctic, we compared model estimates for light-saturated leaf-level A at 5°C℃ modeled with the TBM-specific parameterization and TRFs (Table 1) with the data presented here (Figs 1, 2). The mean observed Vc,max.5 was more than twice the TBM model mean Vc,max.5 (t(5) = 4.6, < 0.01, Table 2), and the resulting species mean A modeled using our parameterization and TRF was 2.5 times greater than the A value modeled with current TBM parameterization (t(5) = 4.6, < 0.01, Table 2).

Table 2. Modeled and measured light-saturated CO2 assimilation at 5°C
Species/TBM Modeled photosynthesis Measured photosynthesis
Vc,max.5 (μmol m−2 s−1) A (μmol m−2 s−1) A (μmol m−2 s−1)
Arctagrostis latifolia 17.9 11.2 9.2 ± 2.5
Dupontia fisheri 14.4 9.0 ND
Arctophila fulva 23.5 14.8 10.1 ± 1.8
Carex aquatilis 18.3 11.5 8.8 ± 1.9
Eriophorum angustifolium 15.0 9.4 10.1 ± 1.6
Petasites frigidus 20.2 12.7 11.3 ± 2.7
Salix pulchra 21.4 13.5 9.1 ± 2.5
Species mean 18.6 ± 3.3 11.7 ± 2.1 9.8 ± 0.9
AVIM 9.5 6.0
BETHY 2.0 1.3
CLM4.5 11.8 7.4
Hybrid6.5 5.6 3.5
Model mean 7.2 ± 4.3 4.5 ± 2.7
  • Leaf-level CO2 assimilation (A) modeled for the seven Arctic species in this study and for the Arctic plant functional types (PFT) represented in four terrestrial biosphere models (Table 1). Leaf-level, light-saturated A was modeled using Eqns 13, where leaf temperature = 5°C, CO2 concentration = 390 μmol mol−1, O2 concentration = 210 mmol mol−1 and Ci : Ca ratio = 0.7. Under these conditions, photosynthesis was limited by carboxylation capacity. Vc,max.5 for the seven species was obtained by scaling Vc,max.25 (Fig. 2) to 5°C using Eqn 4 and the Ea values presented in Fig. 1. Vc,max.5 for the models was calculated by scaling Vc,max.25 for the Arctic PFT using the model-specific Vc,max.25 and temperature response function (TRF) (Table 1). Measured photosynthesis (n = 8–17 individual ramets per species, ± SD) was derived from a separate study on different plants in different locations on the Barrow Environmental Observatory (BEO), measured in a different year, where mean Tleaf = 5°C. Species and model means are shown ± SD. AVIM, Atmosphere–Vegetation Interaction Model; BETHY, Biosphere Energy Transfer Hydrology Scheme; CLM4.5, Community Land Model v.4.5; Hybrid6.5, Hybrid v.6.5; ND, no data; TBM, terrestrial biosphere model.

As an additional step, we compared our modeled data with independent measurements of light-saturated A made in the field at 5°C in 2016. These data showed that A modeled with the mean TBM parameterization of Vc,max.25 resulted in a CO2 uptake rate that was less than half the measured CO2 assimilation rate (t(3) = 3.7, < 0.05). A modeled with the parameter values determined by this study was c. 20% higher than the measured values, but not significantly different (t(9) = 2.2, > 0.05) from A measured in 2016.


We have shown that TBM representation of photosynthetic capacity in Arctic vegetation markedly underestimates the capacity for CO2 assimilation in this globally important biome. Our data showed that the photosynthetic capacity was high (Fig. 2) – all seven species had values for Vc,max.25 that were comparable with, or higher than, those found in major C3 crops (Bernacchi et al., 2005; James et al., 2006; Zhu et al., 2012) – and up to five times higher than some model estimates (Fig. 7). Leaf-level modeling demonstrated that current TBM parameterization of photosynthetic capacity results in a two-fold underestimation of CO2 assimilation by Arctic vegetation (Table 2). This study provides one of the first datasets of the key photosynthetic parameters Vc,max and Jmax in Arctic vegetation and the first estimates of their TRFs (Figs 1, 2). Our data also showed that model underestimation of photosynthetic capacity is attributable to a combination of low estimates of the leaf N content (Fig. 4) and the fraction of N partitioned to Rubisco (Fig. 5).

The JVratio affects the CO2 responsiveness of A, such that a high JVratio will mean that A remains RuBP saturated at a high [CO2], and is therefore more responsive to increasing [CO2], than RuBP limited A in a plant with a lower JVratio (Rogers et al., 2017c). The result of the high JVratio in these Arctic species can be readily seen in Fig. 3b, where, even at 15°C, A does not become RuBP limited in any species until a Ci value of c. 550 μmol mol−1. This means that Arctic vegetation has the potential to respond maximally to rising [CO2] through most of this century. The high investment in Rubisco, coupled with the high JVratio.25, enables plants growing at current [CO2] to sustain high, RuBP-saturated, photosynthetic rates, even at low temperature and low light levels.

The high JVratio will also enable Arctic plants to continue CO2 assimilation during the continuous but lower irradiance of the Arctic midnight sun, and also to maximally exploit periods of high light on clear days during the short Arctic thaw season. In addition, many Arctic plants emerge from underneath the snow at the beginning of the growth season where temperatures are low, light levels are low and [CO2] is typically elevated (Starr & Oberbauer, 2003; Saarinen et al., 2016). Plants with a high investment in Rubisco and a high JVratio would be well adapted to exploit such conditions.

We observed a JVratio.25 (2.53) that is 28% higher than the JVratio.25 used by many models which do not account for temperature acclimation of photosynthesis (e.g. 1.97, Bonan et al., 2011), highlighting the need for TBMs to account for thermal acclimation. Kattge & Knorr (2007) developed an approach to account for the acclimation of JVratio.25 to growth temperature, but their work did not include any measurements at which the growth temperature was below 10°C, and the data from boreal species in their study appeared to be outliers, exhibiting a markedly lower JVratio.25 than other species. This study adds new data at the low-temperature end of the synthesis by Kattge & Knorr (2007) and supports the relationship they presented, which is implemented in BETHY and CLM. However, the JVratio.25 observed here was still markedly higher than the value used in CLM, because CLM includes a lower limit for growth temperature of 11°C. This study suggests that, to more accurately capture the JVratio.25 in Arctic species, this lower limit should be removed from the model formulation.

Our biochemical data revealed markedly different strategies for N partitioning in these seven Arctic species. Some species (A. latifolia, D. fisheri and A. fulva) had a tightly constrained Na, whereas others, notably S. pulchra and P. frigidus, had a large range in Na (Fig. 4), suggesting that Na may drive the greater plasticity in photosynthetic capacity (Fig. 2). Most of the species had a low CNratio in comparison with the value used by CLM (Fig. 4c). This indicates that the high Na in some Arctic species is caused, in part, by a higher investment in N per unit C, and not just a high N density per unit leaf area (Fig. 4). Our dataset afforded the opportunity to calculate apparent FLNR for these Arctic species (a key model input for CLM), thus providing models with a parameter associated with N partitioning that could be used as a PFT-specific model input, or to evaluate prognostic N allocation models, e.g. Ali et al. (2016). It is notable that species with a lower Na (A. latifolia, D. fisheri and A. fulva, Fig. 4) also had a high FLNR (Fig. 5), suggesting that the high partitioning of N to Rubisco in these species enabled them to produce leaves with a lower Na (Fig. 4), but comparable photosynthetic capacity (Fig. 2), which, at the whole-plant level, may enable a higher productivity for a given N supply.

One approach used by TBMs to derive Vc,max.25 is to use PFT-specific slopes and intercepts from the linear relationship with Na (Medlyn et al., 1999; Kattge et al., 2009; Walker et al., 2014). Previously, there was insufficient data to examine this relationship in Arctic species (Kattge et al., 2009; Ziehn et al., 2011). We did not observe a Vc,max.25Na relationship (Fig. 6). The different N allocation strategies outlined above, and the high diversity of leaf morphology in these species, may explain why we did not observe a Vc,max.25Na relationship at our field site, a finding also observed in Betula nana and Eriophorum vaginatum (van der Weg et al. 2013). Although data at the global or biome scale often demonstrate a strong relationship between Vc,max.25 and Na (Kattge et al., 2009; Walker et al., 2014), these relationships can change markedly and even fall apart at finer scales (Feng & Dietze, 2013; Bahar et al., 2016; Croft et al., 2017), which has led to the use of other variables, including phosphorus, to explain global variation in Vc,max.25 (Walker et al., 2014; Ali et al., 2015; Croft et al., 2017). There is evidence that phosphorus plays a role in limiting the productivity of Arctic tundra ecosystems, and may be a limiting nutrient on the coastal tundra in Barrow (Chapin et al., 1975, 1995; Shaver & Chapin, 1980). However, the examination of N : phosphorus ratios across major biomes suggests that direct phosphorus limitation of photosynthesis in the Arctic is unlikely (Reich et al., 2009).

As recently pointed out, it is critically important for models to make consistent use of kinetic constants and temperature response functions (Rogers et al., 2017c). Therefore, the data presented here should be used in conjunction with the kinetic constants and temperature response functions associated with Kc, Ko and Γ* that are provided by Bernacchi et al. (2001). Similarly, use of the Arctic Vc,max.25 and Jmax.25 data (Fig. 2) will require temperature scaling with the Ea values and the TRFs presented here (Fig. 1, Eqn 4). We also recognize that there are many alternative approaches to the analysis of photosynthetic CO2 response curves, including, for example, those that use different TRFs, correct data for potential chamber leaks or account for mesophyll conductance (Ethier & Livingston, 2004; Flexas et al., 2007; Sharkey et al., 2007; Gu et al., 2010; Bernacchi et al., 2013). To allow for future reanalysis of our data, and to maximize their further use by the modeling community, all of our data – including our raw gas exchange data – are available online (Rogers et al., 2017a,b). In addition, we have submitted calculated photosynthetic parameters and biochemical trait data to the TRY database (Kattge et al., 2011) and the database (www.BETYdb.org) associated with the Predictive Ecosystem Analyzer (PEcAn) project (Lebauer et al., 2013).

In addition to advancing our understanding of photosynthesis in the Arctic, these data clearly indicate that CO2 assimilation in Arctic vegetation is poorly represented by current TBMs. Although we caution that these data and insights are only from one site in the high Arctic, this study represents a significant advance, and we hope that the TBM community will improve the representation of CO2 assimilation in the Arctic by using these data.


We are grateful for the technical assistance provided by Stefanie Lasota. This work was supported by the Next-Generation Ecosystem Experiments (NGEE Arctic) project which is supported by the Office of Biological and Environmental Research in the Department of Energy, Office of Science, and through the United States Department of Energy contract no. DE-SC0012704 to Brookhaven National Laboratory. The authors are grateful to UIC Science for logistical support.

    Author contributions

    A.R. designed the study with input from V.L.S., S.P.S. and S.D.W. The data were collected and analyzed by A.R., K.S.E. and S.P.S. The initial draft of the manuscript was prepared by A.R. All authors contributed to the writing.