The carbon isotope (δ13C) value of modern and fossil wood is widely used as a proxy for environmental and climatic change. Many researchers who study stable carbon isotopes in modern and recently deceased trees chemically extract cellulose (δ13Ccell) rather than analyzing whole wood (δ13Cwood) due to concerns that molecular variability across tree rings could influence δ13Cwood values, and that diagenesis may preferentially degrade cellulose over lignin. However, the majority of deep-time researchers analyze δ13Cwood without correcting for possible diagenetic effects due to cellulose loss. We measured δ13Ccell, δ13Cwood, and cellulose content of 38 wood fossils that span ∼50 m.y. in age from early Eocene to late Miocene, using variability across such a large range of geologic ages and settings as a natural laboratory in diagenesis. For comparison with our measurements, we produced a literature compilation of 1210 paired δ13Ccell and δ13Cwood values made on fossil and modern trees. We report that, on average, the apparent enrichment factor (ε) between δ13Ccell and δ13Cwood (ε = δ13Ccell – δ13Cwood) is 1.4‰ ± 0.4‰ larger in deep-time samples than Holocene wood, and this can be explained by loss of cellulose during degradation, independent of atmospheric chemistry or climate conditions during growth. A strong linear correlation exists between δ13Cwood and δ13Ccell in both deep-time (r2 = 0.92) and Holocene (r2 = 0.87) samples, suggesting that either substrate can provide a reliable record of environmental conditions during growth. However, diagenetic effects must be corrected if δ13Cwood values are compared to extant trees or across long time scales, where cellulose content may vary.

The stable carbon isotope composition (δ13C) of terrestrial plants is one of the primary means for tracking changes in Earth’s carbon cycle and climate before the instrument record (Nordt et al., 2016; Strauss and Peters-Kottig, 2003). Diagenetic alteration of organic substrates presents a fundamental limitation for interpreting δ13C signals from deep-time (i.e., pre-Quaternary) records (Jones, 1994; Tu et al., 2004; Baczynski et al., 2016). This challenge has been mitigated through the measurement of compound-specific isotope ratios using biomolecular substrates selected for their recalcitrance to diagenetic modification (e.g., Goni et al., 2000; Ververis et al., 2004). In the past 30 yr, at least 229 δ13C records have been developed for reconstructing past climate using annual growth rings in modern trees; 83% of these measured δ13C of cellulose (δ13Ccell), a polysaccharide that can be chemically extracted from whole wood (Table DR1 in the GSA Data Repository1). However, for deep-time applications that utilized fossil wood across a wide range of preservation states (e.g., lignified [Bechtel et al., 2003] to mummified [Jahren and Sternberg, 2002]), only 23% of studies measured δ13Ccell, compared to 77% that reported δ13C values of whole wood (δ13Cwood; Table DR1).

Despite clear differences in the use of whole wood versus compound-specific analyses, measurement of δ13C values of living, buried, and fossilized trees is routinely used to discern past hydroclimate (Livingston and Spittlehouse, 1996; Barber et al., 2000; Edvardsson et al., 2014), precipitation patterns (Schubert et al., 2012; Schubert and Timmermann, 2015; Schubert et al., 2017), plant life strategy and taxonomy (Bechtel et al., 2007; Jahren and Sternberg, 2008), and atmospheric chemistry (Hesselbo et al., 2000; Schubert and Jahren, 2013). As yet, the effect of decomposition on δ13C value is poorly constrained, but it is necessary for the quantification of atmospheric and climatic change within the deep-time geologic record. For example, recent work linking organic matter δ13C values to changes in atmospheric pCO2 have the potential to greatly increase the resolution of paleoclimate and paleoatmospheric reconstructions (Schubert and Jahren, 2012; Cui and Schubert, 2018). Correction of δ13C values for diagenesis will better constrain the nature and magnitude of pCO2 change and help to reduce uncertainty in δ13C-based paleoclimatic and paleoenvironmental proxies.

Wood offers a unique opportunity to study diagenetic alteration of organic matter, because it is primarily composed of two compounds: lignin (25%–35%) and cellulose (40%–50%), with lesser amounts of hemicellulose and extractives (Parham and Gray, 1984; Pettersen, 1984; Scott, 2009). Lignin is more recalcitrant than cellulose; therefore, the preservation state of wood can be modeled using these two end members (Loader et al., 2003), where the δ13C value of lignin is typically 0.5‰–2‰ lower than that of cellulose (Harlow et al., 2006). Several researchers have concluded that as cellulose preferentially degrades in fossil wood, δ13Cwood will trend toward the δ13C value of lignin and result in up to 2‰ lower values (Benner et al., 1987; Spiker and Hatcher, 1987; Schleser et al., 1999; van Bergen and Poole, 2002). Consequently, the apparent enrichment (ε) between δ13Ccell and δ13Cwood (i.e., ε = δ13Ccell – δ13Cwood) should show a negative correlation with cellulose content of wood. In contrast, incubation experiments (Schleser et al., 1999) and some deep-time fossil wood assemblages (Bechtel et al., 2007) possibly indicate that higher ε values may result from enrichment of 13C during cellulose degradation.

Numerous studies have reported a 1:1 correlation between δ13Ccell and δ13Cwood for living and exceptionally well-preserved, subfossil trees (Schleser, 1990; Leavitt and Long, 1991; Livingston and Spittlehouse, 1996; Borella and Leuenberger, 1998; Macfarlane et al., 1999; Loader et al., 2003; Edvardsson et al., 2014). Testing of this correlation for deep-time wood samples has been inhibited by the paucity of well-preserved wood in the stratigraphic record relative to Quaternary strata, and by differences in analytical approaches between Quaternary and deep-time workers. Whereas δ13Ccell and δ13Cwood are widely measured on living and recently felled trees (see review in McCarroll and Loader, 2004), only δ13Cwood is routinely measured on deep-time wood fossils (e.g., Gröcke et al., 1999; Hesselbo et al., 2000, 2002, 2003, 2007; Pearce et al., 2005; Yans et al., 2010; Table DR1). We sought to rectify this discrepancy and quantify the effects of diagenesis on δ13C values by measuring δ13Ccell, δ13Cwood, and cellulose content on 38 fossil wood specimens that ranged in age from early Eocene to late Miocene. We supplemented these analyses with a new compilation of published paired δ13Ccell and δ13Cwood values from modern trees and deep-time wood fossils to test the following hypotheses: (1) δ13Cwood correlates with δ13Ccell regardless of geologic age, and (2) deep-time wood fossils that have lost cellulose will have greater ε values.

The fossil wood samples analyzed in this study were recovered from deltaic and lacustrine deposits in the Eureka Sound Formation on Banks Island, Northwest Territories, Canada; the Yongning Formation in Nanning, China; the Xiaolongtan Formation in Yunnan Province, China; and the Khapchansky locality in northeast Siberia (Fig. 1; Table DR1). Descriptions of these localities and details of laboratory sampling and analytical methods are provided in the Data Repository.

Typically, δ13C values of tree-ring tissue are corrected for changes in the δ13C value of atmospheric CO2 and pCO2 prior to interpretation of climatic signals (McCarroll et al., 2009; Treydte et al., 2009; Wang et al., 2011; Schubert and Timmermann, 2015; Trahan and Schubert, 2016). Biosynthetic fractionation between different plant substrates, however, is not affected by atmospheric chemistry (e.g., Loader et al., 2003; Schubert and Jahren, 2012; Diefendorf et al., 2015); therefore, determination of net carbon isotope discrimination was unnecessary for interpretation of the carbon isotopic difference between cellulose and whole wood.

Effect of Diagenesis on Fossil Wood δ13C Values

Our new analyses extend the age range and more than double the number of reported deep-time fossil wood δ13Ccell and δ13Cwood pairs. The δ13Cwood values of the 38 fossil samples ranged from −31.8‰ to −22.6‰ (δ13Ccell = −30.2‰ to − 19.8‰), consistent with the wide range of environments, taxa, climates, and atmospheric compositions represented by these fossils. The δ13C values of paired wood and cellulose samples were highly correlated between substrates (Spearman’s ρ = 0.97), and δ13Ccell values were higher than δ13Cwood values in every pair (Fig. DR2). Cellulose content ranged from 0.4% to 44.5%.

Calculated ε values were negatively correlated with cellulose content (Pearson’s r = −0.49, p = 0.003; Fig. 2). A linear regression model estimated that pure cellulose (i.e., δ13Cwood = δ13Ccell) would have an ε value (δ13Ccell – δ13Cwood) within error of zero (−0.4‰ ± 0.4‰). The difference in ε between a sample that has 45% cellulose—a typical value for living trees—and one that has no cellulose is 1.4‰ ± 0.4‰. Cellulose content broadly corresponds to geologic age, with older wood fossils tending to have less cellulose remaining (Fig. 2B). However, individual deposits (e.g., the Oligocene Nanning Lagerstätte; Table DR2) can contain a large range of cellulose content among fossil samples. These findings indicate that diagenetic alteration of wood, which preferentially removes cellulose, can significantly bias deep-time δ13Cwood values.

Comparison of the δ13C Value of Whole Wood Versus Cellulose

We augmented our data set (n = 38) with paired δ13Ccell and δ13Cwood data from 18 published studies, yielding a total of 1248 pairs (Table DR2). This comprehensive data set was stratified into three age categories: post-industrial Holocene (post–1850 C.E.), pre-industrial Holocene (1850 C.E.–12 ka), and deep-time (pre-Quaternary). Together, this data set includes paired δ13C values from both field and herbarium collections and includes at least 56 genera sampled across 80 degrees of latitude. Deep-time fossil samples are represented from across the Cenozoic (early Eocene, middle Eocene, Oligocene, early Miocene, middle Miocene, late Miocene, and Pliocene).

We report a strong linear correlation between δ13Ccell and δ13Cwood (p ≤ 0.001) for Holocene (r2 = 0.87, m = 0.94) and deep-time samples (r2 = 0.92, m = 1.08; Fig. 3A). The median ε values between cellulose and whole wood for post-industrial (ε = 1.3‰) and pre-industrial (ε = 1.2‰) Holocene samples are within analytical uncertainty of each other, but the value is significantly larger (paired Wilcoxon rank-sum test, p < 0.001) for deep-time samples (ε = 2.9‰; Fig. 3B), consistent with lower cellulose content in the deep-time samples compared with modern trees.

Although deep-time samples have, on average, larger ε values than Holocene trees, ε values do not systematically vary across deep-time age bins (Fig. 4). For example, the Oligocene Nanning, China, locality contained 19 samples recovered from a single stratum with ε = 1.0‰–3.3‰, which spans the 23rd to 90th percentiles of post-industrial Holocene ε values. These samples also have a large range of cellulose contents that vary between extremely low values (1.6%) and those similar to living trees (44.5%). The large ranges in ε values and cellulose content across 19 samples of the same age support our finding that cellulose content—rather than age—is a better predictor of ε in fossil wood.

These results have two significant implications. First, the correlation between δ13Ccell and δ13Cwood values in modern and well-preserved, subfossil Holocene wood indicates that isotopic signals are similarly represented by both substrates, and whole wood can therefore be analyzed rather than cellulose when higher sample throughput is desired. Researchers who previously reported a lack of correlation between δ13Ccell and δ13Cwood were observing samples from a limited isotopic range, across which the strong correlation between substrates was not evident (e.g., Schleser et al., 2015). In rare cases, in situ decay of standing trees can lead to moderate cellulose loss and consequent bias in δ13Cwood values, which suggests a need for analysis of δ13Ccell for some species (e.g., Populus deltoides, eastern cottonwood; Friedman et al., 2019).

The second implication of our findings is that the pervasive analysis of δ13Cwood in deep-time geologic archives has, in many cases, included a negative bias in δ13C values due to cellulose loss. We note that the presence of exquisitely preserved fossil wood is itself a rarity in the geologic record. Due to substantial taphonomic bias against wood preservation, we do not consider geologic age to be a predictive variable for estimating fossil wood cellulose content, even though our older samples tended to have less average cellulose content (Fig. 3).

Unifying Relationship to Explain ε in the Fossil Record

The difference in median ε value between deep-time and post-industrial Holocene samples in the comprehensive data set is 1.6‰, which is similar to the value of 1.4‰ predicted by cellulose degradation in our fossil wood samples (Fig. 2). These values are also similar to the value of 1.3‰ found in artificial aging of wood via carbonization experiments (Turney et al., 2006). From this, we conclude that the δ13Cwood value of deep-time fossil trees is, on average, biased toward lower values. We note that the opposite trend is observed in soil organic matter, which experiences an increase in δ13C value of 1‰–8‰ during diagenesis (Wynn, 2007), caused by respiration of 12C and fixation of 13C by microbes.

These data serve as an empirical confirmation that the δ13Cwood value of fossil specimens varies as a function of cellulose content, which is in turn an indicator of preservation state. The empirical relationship between ε and cellulose content from our fossil wood samples (Fig. 2A) can be used as a correction for comparing δ13Ccell and δ13Cwood values among fossil wood samples in various preservation states:
where x is cellulose content in weight percent, and the root mean square error is 0.4‰. However, direct analysis of δ13Ccell is likely more reliable for interpreting paleoclimatic and paleoenvironmental change from wood across myriad preservation states.

According to Equation 1, fossil wood δ13Cwood values can be biased by up to 1.4‰ as a result of cellulose loss during diagenesis (i.e., 45% vs. 0% cellulose). If unaccounted for, this would lead to the erroneous interpretation of climatic and environmental change. A 1.4‰ decrease in δ13C value due to diagenesis may be incorrectly interpreted as a large increase in precipitation (Kohn, 2010) or pCO2 (Schubert and Jahren, 2012), or it may lead to misinterpretation of plant taxa when using δ13Cwood values as a chemotaxonomic tool (Bechtel et al., 2007). We illustrate the importance of accounting for diagenetic alteration through calculation of pCO2 using the methods described in Schubert and Jahren (2015) for a published δ13Cwood record across the Early Jurassic (Toarcian) carbon isotope excursion (CIE; Hesselbo et al., 2007). We found that using uncorrected δ13Cwood values results in an overestimation of peak pCO2 at the CIE by ∼100% (i.e., 2154 vs. 1039 ppm; see the Data Repository). The pCO2 estimate of 1039 ppm at the CIE calculated after correcting for cellulose loss overlaps with an estimate of 1200 ± 400 ppm based on fossil leaf stomatal frequency (McElwain et al., 2005).

Our laboratory analyses and literature compilation demonstrate that apparent enrichment between δ13Ccell and δ13Cwood is consistent across a wide range of environments, climates, and taxonomic groups. The 1:1 correlation between δ13Ccell and δ13Cwood in modern and Holocene trees should unburden researchers in these areas from performing time- and resource-intensive cellulose extraction procedures. Further, the correlation between δ13Ccell and δ13Cwood across deep-time samples demonstrates that each substrate provides similar paleoenvironmental information, but it can only be used for detecting relative changes in δ13C value. The increased ε values determined for deep-time samples relative to modern trees suggests a need for workers to correct δ13Cwood values based on cellulose content when comparing modern and fossil δ13Cwood values.

This research was supported by U.S. National Science Foundation (NSF) award EAR-1603051. We thank Yingfeng Xu for laboratory assistance and Cheng Quan for providing wood fossils from Nanning and Yunnan, China. Fossil wood samples from Aulavik National Park on northern Banks Island, Northwest Territories, Canada, were collected by Schubert in July 2012 on a field expedition led by Jaelyn Eberle (University of Colorado at Boulder, USA), which was financially supported by NSF grant ARC-0804627 to Eberle. A permit to conduct field research in Aulavik National Park was provided by Parks Canada, Western Arctic Field Unit. Logistical support was provided by the Polar Continental Shelf Program (Natural Resources Canada), the Aurora Research Institute (Inuvik, Northwest Territories, Canada), the people of Sachs Harbour on Banks Island, and Canadian Helicopters. The Banks Island wood specimens were curated into the collection of the Canadian Museum of Nature in Ottawa, Canada, and made available for study through the assistance of K. Shepherd and M. Currie.

1GSA Data Repository item 2019354, descriptions of fossil wood localities and analytical methods, Figures DR1–DR3, and Tables DR1 and DR2, is available online at, or on request from
Gold Open Access: This paper is published under the terms of the CC-BY license.