Carbon isotope compositions of terrestrial C3 plants as indicators of (paleo)ecology and (paleo)climate

Abstract

A broad compilation of modern carbon isotope compositions in all C3 plant types shows a monotonic increase in δ¹³C with decreasing mean annual precipitation (MAP) that differs from previous models. Corrections for temperature, altitude, or latitude are smaller than previously estimated. As corrected for altitude, latitude, and the δ¹³C of atmospheric CO₂, these data permit refined interpretation of MAP, paleodiet, and paleoecology of ecosystems dominated by C3 plants, either prior to 7–8 million years ago (Ma), or more recently at mid- to high latitudes. Twenty-nine published paleontological studies suggest preservational or scientific bias toward dry ecosystems, although wet ecosystems are also represented. Unambiguous isotopic evidence for C4 plants is lacking prior to 7–8 Ma, and hominid ecosystems at 4.4 Ma show no isotopic evidence for dense forests. Consideration of global plant biomass indicates that average δ¹³C of C3 plants is commonly overestimated by approximately 2‰.

Plants exhibiting C3 photosynthesis have dominated the history of terrestrial vegetation, as CAM plants occupy only a small percentage of typical ecosystems (e.g., 1), and C4 plants became abundant in grasslands only within the last 7–8 million years (2). C3 plants exhibit a large range of carbon isotope compositions (-20 to -37‰, V-PDB; Fig. 1), generally reflecting a physiological response to aridity (anomalously high δ¹³C) and a combination of low light levels plus leaf litter recycling (anomalously low δ¹³C; 3). Past studies have conflicted on the dependence of δ¹³C on mean annual precipitation (MAP). Most studies support a negative correlation (e.g., 4), but some have reported no correlation (5) or even a positive correlation (6).

Fig. 1.

(A) Histogram of MAP values for isotopically charact.erized C3 plants, showing emphasis on relatively arid ecosystems (MAP ≤ 500 mm/yr) and tropical rainforests (spikes at MAP ∼ 2,000, 3,000 mm/yr). (B) Histogram of δ¹³C values of modern C3 plants. Data compiled in this study average -27.0‰, excluding analyses from the understory of closed-canopy forests. Estimated global average composition, based on global trends in precipitation and vegetation, is approximately -28.5‰, significantly lower than typically assumed. An accurate average δ¹³C value for C3 plants is needed for accurate models of carbon fluxes, atmospheric CO₂ compositions, and soil organic matter. (C) δ¹³C values vs. MAP showing increasing δ¹³C with aridity. Data sources are listed in SI Text. White dots are average compositions of data from a large collection made in a single month during a wet year (35).

This study explores carbon isotope systematics of C3 plants through a comprehensive compilation of literature data, minimizing local differences that have given rise to disparate interpretations. This work directly benefits paleodietary and paleoecological studies by providing estimates of MAP in ancient environments as well as clearer boundaries for identifying the understory of closed-canopy forests and the contribution of non-C3 plants to diet, which is used to gain insights into C4 origins (2). The new compilation also helps refine models of the modern carbon cycle by providing a better estimate of global C3 δ¹³C values (7). The present analysis differs from a recent investigation of correlations between tree + shrub δ¹³C and MAP or plant functional type (8) in that it is several times larger on a site-by-site basis, spans the full range of plant growth forms (including trees, bushes, grasses, and herbs), averages data from each site to minimize sampling bias (4), and regresses an arguably more appropriate function to the data. Selection criteria of climate and isotopic data also differ slightly. This broader approach of considering all C3 plants and averaging compositions is important when evaluating herbivore paleodiets, which are not restricted to trees, and further reveals significant differences in modeled compositions at low MAP and in the estimated effect of altitude and latitude. A simple function allows MAP to be estimated from δ¹³C of fossil bone collagen and tooth enamel and used for inferring terrestrial climate change. Past studies of dietary ecology and global C3 plant δ¹³C values further indicate an analytical bias toward dry ecosystems.

Results

Data.

The dataset, provided in Dataset S1, encompasses all types of C3 plants, including trees, shrubs, herbs, and grasses from approximately 570 individual sites, and spans ranges of MAP, mean annual temperature (MAT), altitude, and latitude of 1 to 3,700 mm/yr, -13.5 to 28.4 °C, -391 to 4,900 m, and 54.9°S to 69.5°N. This dataset is ca. six times larger than any previous analysis of C3 isotopic systematics and covers more types of C3 plants.

C3 Data Distributions.

For the global C3 δ¹³C dataset, a histogram of corresponding MAP values (Fig. 1A) demonstrates research bias toward dry ecosystems (high δ¹³C) and tropical rain forests (low δ¹³C). Thus the histogram of δ¹³C values, which shows the well-known range from -20 to -37‰ (Fig. 1B), is broader than expected for global C3 biomass. Values above -23‰ are almost completely restricted to the Atacama Desert (9), the driest desert on Earth, and to Pinus in dry settings (10). Values below -31.5‰ reflect canopy effects in low-light tropical forests. The compilation from O’Leary (11; downward corrected by -0.25‰ for fossil fuel burning) and from this study average about -27.25‰ and -27.0‰, respectively, omitting understory analyses below -31.5‰.

Correlation of MAP with δ¹³C.

Carbon isotope compositions exhibit a systematic change over the range of MAP, excepting low δ¹³C values associated with the understory of dense forests (Figs. 1 and 2). High δ¹³C values (above -25.5‰) are essentially restricted to environments with MAP < 500 mm/yr. The trend noticeably “flattens” at high MAP, indicating nearly constant isotopic discrimination in wet environments. Other attempts to quantify this correlation have resulted in widely disparate results (Fig. 2), largely because they were based on limited datasets. The results of Stewart et al. (4) and Diefendorf et al. (8) most closely match the new compilation, but the Stewart et al. model deviates from observations at MAP > 1,000 mm/yr, whereas the Diefendorf et al. model predicts unusually high δ¹³C values at low MAP (Fig. 2). The preferred equation for MAP as a function of δ¹³C is:

[1]

or alternatively for Δ:

[2]

where:

[3]

(3). Both regressions have an R² of 0.59, a significant improvement over the functional form used by Diefendorf et al. (8; R² = 0.34, including their altitude and latitude coefficients). These equations allow evaluation of the effects of MAP, altitude, and latitude on δ¹³C values or Δ, or alternatively estimation of MAP from altitude, latitude, and δ¹³C or Δ. Note that the altitude coefficient is smaller by a factor of 3 to 5,000 than indicated in other studies (8, 12). Latitudinal effects have not been estimated for data that simultaneously account for altitude and MAP. Uncertainties of ± 0.5‰ in mean δ¹³C or Δ propagate to uncertainties in MAP of as little as ± 100 mm/yr at MAP = 100–500 mm/yr, to ± 500 mm/yr at MAP ∼ 2,000 mm/yr.

Fig. 2.

(A) Averaged δ¹³C (Table S1) vs. MAP and models of carbon isotope compositions. Averaged data are provided for clarity; thin lines are published models; thick curve is preferred regression from this study. Solid portions of lines represent MAP range over which models were developed; dotted lines show extrapolations. D = Diefendorf et al. (8); G1 = Gouveia and Freitas (54); G2a,b = Guo and Xie (6); H = Hatté et al. (55); L1 = Leffler and Enquist (56); L2 = Liu et al. (57); M1 = Miller et al. (58); M2 = Macfarlane et al. (59); R = Roden et al. (60, averaged from two similar regressions); S = Song et al. (61); Y = Youfeng et al. (62). Original line of Stewart et al. (4) is best linear model, but deviates from data at large values of MAP; model of Diefendorf et al. (8) fits high MAP data but does not predict low MAP data well. (B) Carbon isotope residuals for a model that omits latitude and altitude, showing significant correlation with altitude. “D” indicates model of Diefendorf et al. (8). Inset shows raw δ¹³C values (uncorrected for any parameter) vs. altitude. (C) Carbon isotope residuals for a model that includes altitude but omits latitude, showing small but significant correlation. Inset shows raw δ¹³C values (uncorrected for any parameter) vs. latitude; high values centered at approximately 30° latitude reflect dry ecosystems on Earth. (D) Carbon isotope residuals for a model that includes latitude and altitude vs. MAT showing no significant trend. Inset shows raw δ¹³C values (uncorrected for any parameter) vs. MAT. Numbers are values for slopes, and errors are ± 2σ. X-axes on insets are same as in encompassing panel.

Discussion

Average C3 Composition and Isotopic Bounds on C3 δ¹³C.

A commonly quoted “average” C3 composition () is -26 to -27‰ (e.g., 2, 13–15) similar to or slightly higher than compilation means (Fig. 1). Such high values, however, are strongly biased toward dry ecosystems (Fig. 1A). For example, the global , as estimated from Eq. 3 and distributions of plant biomass and precipitation, is approximately -28.5‰, or approximately 2‰ lower than commonly assumed. This low mainly reflects the importance of equatorial and midlatitude northern hemisphere C3 biomass, which is less well represented in the literature. The value is important for models of carbon fluxes, atmospheric CO₂ compositions, and soil organic matter (e.g., 7, 15–17). Seasonal changes to depend on C3–C4 biomass ratios and their compositions, and differences in modeled C3 biomass discrimination perturb predicted δ¹³C of atmospheric CO₂ most strongly (16). The dataset and equations developed here could be used to test and calibrate these models. Considering that δ¹³C > -23‰ is restricted to MAP < 10 mm/yr, non-C3 vegetation (e.g., C4 plants, lichen, or CAM plants) or the genus Pinus in some settings, this value is recommended as a likely maximum for typical C3 plants. Likewise, δ¹³C < -31.5 is recommended as a cutoff indicating the understory of closed-canopy forests.

Paleoprecipitation Reconstructions.

The regressed curve allows prediction of MAP from the average modern equivalent of diet composition (), which can be estimated from fossil tooth or collagen δ¹³C values and (see SI Text). Most studies that inferred mixed C3-C4 ecosystems were omitted from consideration because obvious C4 consumption ( values for individual species well above -22‰) would otherwise imply low or negative MAP. Corrections for are key for predicting accurate C3 composition and MAP. For example, two high values for tooth enamel δ¹³C at approximately 15.5 Ma in East Africa (c. -8‰; 18) were interpreted as approximately 2‰ higher than the range of C3 compositions, requiring a C4 dietary component. A high at that time (c. -5.25, or a 2.75‰ downward correction), however, implies of ∼-24.8‰, well within the range of a pure C3 diet in dry ecosystems (Fig. 1B and Fig. 3, “m1”). Similarly large corrections apply to several other studies (19–21) and indicate wetter conditions than suggested by corrections for modern fossil fuel burning alone (c. -1.5‰ correction).

Fig. 3.

Paleodietary compositions corrected for altitude and latitude contoured for MAP; this plot permits interpretation of paleoenvironments from carbon isotope compositions of fossil tooth enamel or collagen. Tooth enamel and collagen compositions are averaged across species and corrected for δ¹³C of atmospheric CO₂, physiological fractionations, altitude, and latitude. Most data plot at MAP ≤ 800 mm/yr, i.e., relatively dry environments, although wetter environments are also represented. b1 = Bibi (51); b2 = Bocherens and Drucker (63); c1 = Chritz et al. (30); c2, c3 = Coltrain et al. (32, 20–25 ka and 12 ka); c4 = Cerling et al. (2); d = DeSantis and Wallace (26); e = Eberle et al. (21); f1, f2 = Fox-Dobbs et al. (31, caribou, equid); f3 = France et al. (64); i = Iacumin et al. (65); k = Koch et al. (19); m1 = Morgan et al. (18 at 15.5 Ma); m2 = Merceron et al. (52); m3 = MacFadden and Higgins (20); n = Nelson (66); p1 = Passey et al. (49); p2 = Palmqvist et al. (27); s1 = Ségalen and Lee-Thorp (67); s2 = Secord et al. (25); t1 = Tütken et al. (68); t2 = Tütken et al. (69); v = van Dam and Reichart (34); w = Wang et al. (29); z = Zanazzi and Kohn (33).

Results correlate generally with previous interpretations regarding “dry” vs. “wet” environments and with the possibility of alternative food sources. Few independent measures of MAP are available but include estimates of 300–700 mm/yr for late Cenozoic Spain (22) vs. ∼500 mm/yr (Fig. 3, “v”), approximately 1,200 mm/yr for Paleocene-Eocene strata in Wyoming (23) vs. ∼1,000 mm/yr (Fig. 3, “k”), and 740 ± 280 mm/yr for Eocene-Oligocene strata in Nebraska (24) vs. approximately 200 mm/yr (Fig. 3, “z”). In the latter case, only a few taxa were analyzed, probably compromising estimates of average compositions. Several other paleoenvironments were viewed as particularly wet, either from paleobotanical and paleofaunal observations or because of geographic location (20, 21, 25, 26). For three of these studies, estimates of average MAP for these localities exceed 2,000 mm/yr. For data from the Eocene Arctic (21), the relatively high δ¹³C value results from the latitude correction, and a smaller correction would result in a higher estimated MAP. Only two studies have reported sufficiently low δ¹³C for any fossil species to indicate closed-canopy conditions (20, 27). Taxonomically extensive isotopic data do not directly support the conclusion that the early hominid Ardipithecus ramidus occupied a closed forest (28); the lowest inferred is approximately -30.5‰ (for a colobine monkey), and the assumed isotopic boundary for closed-canopy forests ) was unrealistically high. Overall most data from most studies appear to support low MAP, typically < 800 mm/yr, below mean global MAP between 60°S and 75°N (∼850 mm/yr). This probably reflects overall preservation or research bias toward drier ecosystems.

Two studies indicated average above the range of average C3 δ¹³C. Wang et al. (29) inferred consumption of high δ¹³C C4 plants, which is consistent with the new compilation (Fig. 3, “w”) and with individual δ values for several taxa above -22‰ (i.e., generally inaccessible to C3 plants). Pleistocene Irish deer data imply either dietary specialization on high C3 plants (Fig. 3, point “c1”; 30), or lichen consumption (31).

Several qualifications apply to estimating MAP. First, environments with C4 plants cannot be interpreted because high may reflect C4 consumption rather than aridity, although closed-canopy occupancy or C4 consumption can be evaluated. Second, C3 plant isotope compositions within a single locality show significant variation (e.g., Fig. 1B), and different taxa prefer different microhabitats and foods. Robust estimates of MAP thus require averaging over multiple taxa in a single locality, just as strong correlations between global plant δ¹³C and MAP require averaging (4; this study). Some studies analyze numerous taxa (27, 32) and are well suited for estimating MAP, whereas others focus on specific ecological or climatic questions with only a few taxa or even just one taxon (19, 30, 33, 34), and MAP estimates are more tentative. Dry environments may contain wet microhabitats, e.g., along rivers or at springs, and flora may exhibit relatively low δ¹³C values either seasonally or in an unusually wet year. For example, δ¹³C of plants from the dry environments in one study (35) might be interpreted as higher MAP. In the most arid environments, plants may preferentially grow in cracks or declivities where precipitation accumulates, effectively increasing MAP (36), and other sources of precipitation, such as fog, may contribute significantly to total moisture (37, 38). These processes provide greater moisture than implied by MAP alone, and in these cases paleo-MAP estimates will be maxima. Last, it has been argued that higher past pCO₂ resulted in lower δ¹³C values for C3 plants, with a dependence of approximately 2‰/100 ppm (e.g., 39). Eocene pCO₂ estimates of 1,000–1,500 ppm (40), of -5.5 to -6 (41) and fossil tooth enamel δ¹³C (19, 21, 25, 33) would then imply of -5 to -10‰. Such high values exceed the range of even C4 plants today (e.g., 11, 42), further supporting a negligible pCO₂ correction (43).

Conclusions

Carbon isotope compositions show a distinct but nonlinear increase in δ¹³C values with decreasing MAP. A regressed expression provides a new basis for estimating MAP from carbon isotope compositions of fossil tooth enamel or collagen, after first correcting for changes to , altitude, latitude, and physiological fractionations. Research bias toward dry ecosystems appears in analysis of paleoecologies. A downward revision in modern for C3 plants may improve models of carbon fluxes and soil δ¹³C values.

Methods

Data and methods are described in further detail in SI Text. In brief, data were taken from the literature, preferring large natural datasets that included date of collection, location, MAP, and mean annual temperature. In some instances, specific locations were not provided, and an estimated or average location was assigned based on descriptions in the primary source. Compositions were corrected for secular changes to the composition of atmospheric CO₂ () to a common of -8.00‰ based on modern secular trends (0.023‰/yr; 44). If not reported, the date of collection was assumed to be 2 yr prior to the date of publication. Nearly all climate data were taken directly from the original publications, and any gaps in climate data were obtained from online and published local meteorological tabulations, or, in the few instances where local data were unavailable, from models. Most data are for whole leaves, and for nonleaf data, isotopic offsets were applied as recommended by the authors. Where no recommendation was made, offsets of -2‰ and -0.5‰ were assumed for soil organic matter and leaf litter, respectively (15, 45). It may be argued that environmental parameters other than MAP correlate better with leaf δ¹³C, such as potential evapotranspiration, water deficit, or growing season precipitation. Although such parameters may be calculated in modern settings, they involve additional variables (e.g., seasonal temperature or precipitation) that may be difficult to constrain in paleoenvironments. Modern data show sufficiently strong correlations with MAP to allow MAP inferences.

Prior to regressions, data were averaged over all C3 plant species at an individual site. This averaging approach differs markedly from all other studies except Stewart et al. (4), whose dataset was over 40 times smaller. Assignment of sites was based on how authors reported their data, i.e., if authors distinguished one set of analyses from another then they were treated as separate sites. Understory, midlevel canopy, and upper-level canopy compositions were distinguished in tropical forests, and analyses from different years were considered as different sites. The total dataset has approximately 570 sites (∼95 usable sites from ref. 8) and is especially dense at low MAP (Fig. 1C). Many different regression approaches were tested to relate carbon isotope composition to MAP. Averaging data over small MAP ranges (e.g., < 100, 100–200, 250–300 mm/yr, etc.) resulted in the highest correlation coefficients (c. 0.95), but this approach is compromised by requiring a priori corrections for altitude and latitude. Instead, the preferred and simpler approach involved regressing δ¹³C vs. altitude, latitude, and log ₁₀(MAP + m_o), where m_o is an offset that is determined iteratively and ensures that the intercept of the regression is finite. An alternative regression with Δ as the dependent variable was also calculated. Outliers at ± 3σ from local means were removed iteratively and represent approximately 4% of sites. Understory compositions from dense forests were also omitted (∼2% of sites). Data from Schulze et al. (35) for MAP = 130–250 mm/yr deviate significantly from global data trends. These data represent > 50% of data in that precipitation range and were collected in a single month during a wet year. To avoid bias compared to other datasets, the Schulze et al. data for < 200 and for 200–250 mm/yr were averaged to separate values.

Global distributions of precipitation between 60°S and 75°N latitudes were estimated from 2.5° grids obtained from the Global Precipitation Climatology Center (GPCC) and averaged over 20 yr (1986 to 2005) (http://gpcc.dwd.de). Global plant biomass was either assumed to increase linearly with precipitation or taken from compilations of C3 plant biomass vs. latitude (46), with precipitation vs. latitude for vegetated areas determined from GPCC. The global average C3 value was then predicted by using the fitted curve to model δ¹³C vs. MAP, ignoring low δ¹³C understory, which represents a negligible fraction of total leaf biomass in the tropics (47, 48). Both methods give comparable results (-28.4 vs. -28.5‰, respectively). Note that these calculations ignore corrections for altitude but include a latitude correction of approximately -0.3‰.

For comparison to herbivore tooth enamel and collagen compositions, correction for changes to over geologic time was made based on the δ¹³C of benthic foraminifera as adjusted for calcification temperature (41; see also refs. 25 and 49). No dependence of plant δ¹³C on atmospheric CO₂ concentrations was considered because experimental data show no consistent resolvable relationship for pCO₂ between 200 and 1,300 ppm (43). The effect of this assumption is further discussed. Compositions of fossil teeth and collagen were converted to paleodietary plant composition by subtracting 14‰ (42) and 5‰ (50), respectively, except for three studies that focused exclusively on tooth enamel from bovids younger than 10 Ma (30, 51, 52); for these a larger tooth-diet offset of 14.5‰ was used (53). All paleodiet compositions were converted to Δ, and modern equivalent compositions () were calculated for .