Metabolism is a major driver of hydrogen isotope fractionation recorded in tree‐ring glucose of Pinus nigra

Summary Stable isotope abundances convey valuable information about plant physiological processes and underlying environmental controls. Central gaps in our mechanistic understanding of hydrogen isotope abundances impede their widespread application within the plant and biogeosciences. To address these gaps, we analysed intramolecular deuterium abundances in glucose of Pinus nigra extracted from an annually resolved tree‐ring series (1961–1995). We found fractionation signals (i.e. temporal variability in deuterium abundance) at glucose H1 and H2 introduced by closely related metabolic processes. Regression analysis indicates that these signals (and thus metabolism) respond to drought and atmospheric CO2 concentration beyond a response change point. They explain ≈ 60% of the whole‐molecule deuterium variability. Altered metabolism is associated with below‐average yet not exceptionally low growth. We propose the signals are introduced at the leaf level by changes in sucrose‐to‐starch carbon partitioning and anaplerotic carbon flux into the Calvin–Benson cycle. In conclusion, metabolism can be the main driver of hydrogen isotope variation in plant glucose.


Supporting information -Metabolism is a major driver of hydrogen isotope fractionation recorded in tree-ring glucose of Pinus nigra
Thomas Wieloch, Michael Grabner, Angela Augusti, Henrik Serk, Ina Ehlers, Jun Yu, Jürgen Schleucher (Accepted: 24 January 2022) Notes S1. Statistical analyses Hierarchical cluster analysis was performed using the R function hclust() of the STATS package on z-scores of δDi using Euclidean distances and Ward's fusion criterion for cluster formation (n=7*31). Mood's median tests were performed using the R function median_test() of the COIN package. Batch change point analysis based on the non-parametric Mann-Whitney-Wilcoxon test was performed using the R function detectChangePointBatch() of the CPM package (Ross, 2015). Multiple linear regression modelling was performed using the R function lm() of the STATS package. Change point regression modelling (step type) was performed using the R function chngptm() of the CHNGPT package (Fong et al., 2017). Statistical significances of the change point and the change point model were calculated using the R functions chngpt.test() of the CHNGPT package and lrtest() of the LMTEST package, respectively.

Notes S2. Grouping of annual δDi patterns of tree-ring glucose by HCA
Metabolic fractionations affect specific intramolecular H-C positions and thus introduce intramolecular δDi patterns (Figs. 1,2b,and S2b). To identify annual patterns that were similarly/differently modified by metabolic fractionation, we performed a Hierarchical Cluster Analysis (HCA) on annual δDi patterns (Fig. S2a). We found two groups of δDi patterns: a high-value and a low-value group according to δD1 and δD2 values (Figs. 2b,and S2b). Please note that the population to which our data belong must be continuously distributed despite the apparent grouping and the gap in δD2 in the ≈200 to 250‰ range (Figs. S2a,and S2b). This is because tree rings record ecophysiological information continuously over the course of growing seasons, i.e., the temporal impact of metabolic fractionations at H 1 and H 2 principally varies in a continuous way. Nevertheless, the apparent grouping in our data is convenient to investigate causes of high and low δD1 and δD2 values. Please note that if our dataset had been continuously distributed, we could still have arbitrarily separated it into low-and high-value groups and investigated underlying causes. However, in the present case, HCA separates the data for us. Annual δDi patterns. δDi denotes D abundances at intramolecular H-C positions in tree-ring glucose. Data were acquired for tree-ring glucose of Pinus nigra laid down from 1961 to 1995 at a site in the Vienna basin (±SE=5.4‰, n≥3). Prior to HCA, outliers were replaced by timeseries averages. Data reference: Average D abundance of the methyl-group hydrogens of the glucose derivative used for NMR measurements. Figure S2b shows discrete data. Lines used to guide the eye.  Note that low εmet values can occur under low groundwater storage (Fig. 3b, 1987(Fig. 3b, , 1991(Fig. 3b, , 1994(Fig. 3b, , and 1995. This is explained by high precipitation during spring and summer and high atmospheric CO2 concentrations (Figs. S4a, and b, filled circles). δDg. To assess this weight exclusively for years with upregulated fractionating metabolic processes, we repeated the variance partitioning on corresponding data but exclude 1988 and 1990 because of data gaps as result of the outlier analysis (1983-1986, 1989, 1992, and 1993Fig. 2c). We found that δD1 and δD2 together account for 86.8% of the variance in δDg. By contrast, δD3 to δD5 each account for 6.3% on average. Interestingly, δD6S and δD6R reduce the variability of δDg by -2.8% on average. Assuming the variability in δD3 to δD5 reflects the combined influence of known fractionation processes affecting all δDi, such as leaf water D enrichment, metabolic fractionations in δD1 and δD2 together account for 74.2% of the variance in δDg (86.8%-2*6.3%).

Notes S6. Contributions of δDi to the variance in δDg after excluding data affected by the fractionating metabolic processes
Metabolic processes have strong effects on δD1, δD2, and δDg (Figs. 1, and 2). After excluding years affected by these processes from the variance partitioning analysis (1983 to 1995), all δDi exhibit similar degrees of variance and contribute similarly to δDg (Fig. S6).

Notes S7. Metabolic fractionation at the whole-molecule level
Within this paragraph, the term 'metabolic fractionation' refers to metabolic fractionation at glucose H 1 and H 2 .
Variability in δDg is predominantly controlled by metabolic fractionation (Fig. 2d). Since δDg can be measured by high-throughput isotope ratio mass spectrometry (a technique accessible to numerous laboratories), we will now investigate possibilities to (i) identify δDg datasets affected 6 by metabolic fractionation, (ii) separate δDg datapoints affected by metabolic fractionation from other datapoints, and (iii) retrieve information from δDg about metabolic fractionation.
(i) Metabolic fractionation caused occasional δDg increases above normal δDg values but never δDg decreases (green dots in Fig. S7a). Consequently, the δDg distribution is asymmetrical with a moderate positive skew, has increased variability, and is nearly significantly different from normality ( Fig. S7c; skewness=0. 55,range=83.2‰,SD=23.8‰;p=0.12). After excluding data affected by metabolic fractionation, the δDg distribution is approximately symmetrical, has lower variability, and is not significantly different from normality ( Fig. S7d; skewness=0.32 Our δDg dataset is relatively small (n=25) and, therefore, not an ideal approximation of the underlying probability distribution. Theoretically, we would expect a bimodal distribution. Data not affected by metabolic fractionation (black dots in Fig. S7a) would be represented by a lowvalue peak in the histogram. Data affected by metabolic fractionation (green dots in Fig. S7a) would be represented by a high-value peak adjacent to the low-value peak. The relative height of these peaks would depend on the relative frequency of long-term drought events (groundwater depletions below the critical level).  (ii) Figure S7a shows δDg as function of time with green dots representing data affected by metabolic fractionation (cf. Fig. 2c). Without colour coding, a clear separation between datapoints affected by metabolic fractionation and other datapoints is not feasible. Figure S7b shows the same data ranked by value from low to high. Data affected by metabolic fractionation have the highest ranks and are sitting neatly on a line (green line, R 2 =0.99, n=7). The slope of this line is 2.4 times steeper than the slope of the line pertaining to data not affected by metabolic fractionation (black line, R 2 =0.94, n=18). This may enable δDg data separation yet not with high confidence. For instance, without colour coding, it is unclear whether the four datapoints before the green datapoints were also affected by metabolic fractionation. Furthermore, if the number of data affected by metabolic fractionation was 2.4 times higher (n≈17), both lines would have the same slope.
(iii) A change point model explains most of the variance in εmet (main text, 'Model 2', R 2 =0.94, p<10 -15 , n=31, Eq. 7), and all explanatory variables contribute significantly to this model (Table   3). While the same change point model explains a significant fraction of the variance in δDg (R 2 =0.76, p<10 -5 , n=25, Eq. 7), most explanatory variables do not contribute significantly (Table S7). Thus, at the level of δDg, the environmental dependences of the fractionating metabolic processes are insufficiently constraint for interpretation. A change point model was fitted to measured whole-molecule deuterium abundances, δDg (R 2 =0.76, p<10 -5 , n=25, Eq. 7) (Fong et al., 2017). Data were acquired for tree-ring glucose of Pinus nigra laid down from 1961 to 1995 at a site in the Vienna basin (±SE=3.4‰, n≥3). β1 to β6, and e denote model coefficients (Eq. 7). SE and CI denote the standard error and confidence interval, respectively. Asterisks mark estimations which assume that bootstrap sampling followed a normal distribution.