Shifting stoichiometry: Long‐term trends in stream‐dissolved organic matter reveal altered C:N ratios due to history of atmospheric acid deposition

Abstract Dissolved organic carbon (DOC) and nitrogen (DON) are important energy and nutrient sources for aquatic ecosystems. In many northern temperate, freshwater systems DOC has increased in the past 50 years. Less is known about how changes in DOC may vary across latitudes, and whether changes in DON track those of DOC. Here, we present long‐term DOC and DON data from 74 streams distributed across seven sites in biomes ranging from the tropics to northern boreal forests with varying histories of atmospheric acid deposition. For each stream, we examined the temporal trends of DOC and DON concentrations and DOC:DON molar ratios. While some sites displayed consistent positive or negative trends in stream DOC and DON concentrations, changes in direction or magnitude were inconsistent at regional or local scales. DON trends did not always track those of DOC, though DOC:DON ratios increased over time for ~30% of streams. Our results indicate that the dissolved organic matter (DOM) pool is experiencing fundamental changes due to the recovery from atmospheric acid deposition. Changes in DOC:DON stoichiometry point to a shifting energy‐nutrient balance in many aquatic ecosystems. Sustained changes in the character of DOM can have major implications for stream metabolism, biogeochemical processes, food webs, and drinking water quality (including disinfection by‐products). Understanding regional and global variation in DOC and DON concentrations is important for developing realistic models and watershed management protocols to effectively target mitigation efforts aimed at bringing DOM flux and nutrient enrichment under control.


| INTRODUC TI ON
Dissolved organic matter (DOM) provides an essential energy and nutrient source to aquatic ecosystems (Webster & Meyer, 1997).
DOM varies in availability to biota along the hydrologic continuum (McArthur et al., 1985) and its composition and properties are closely linked to the surrounding landscape (Jaffé et al., 2008;Mattsson et al., 2005;Wymore et al., 2021c;Yates et al., 2019).
The DOM pool is a complex mixture of organic compounds mostly composed of dissolved organic carbon (DOC) and dissolved organic nitrogen (DON) to a minor extent (Pagano et al., 2014). Numerous studies have confirmed that increases in DOC concentrations in north temperate freshwater ecosystems have occurred over time.
For example, DOC concentrations have increased between 50% and 91% in streams and lakes of northern and central Europe, the United Kingdom, and eastern North America since the 1980s (Couture et al., 2012;de Wit et al., 2016;Driscoll et al., 2003;Evans et al., 2005;Gavin et al., 2018;Hall et al., 2021;Lawrence et al., 2011;Monteith et al., 2007;Worrall et al., 2004). Increased DOC concentration is often attributed to the recovery from acid deposition after the implementation of the Clean Air Act in the United States and similar legislation in Europe (Driscoll et al., 2003). The leading hypothesized mechanism is that a decrease in ionic strength and protonation in soil water following recovery from acid deposition leads to increases in solubility and the mobilization of DOC to adjacent water bodies (Borken et al., 2011;De Wit et al., 2007;Evans et al., 2005;Hruška et al., 2009;Lawrence & Roy, 2021).
A suite of different hypotheses has been put forward to explain the increasing trends in DOC concentrations, in addition to declines in atmospheric deposition, each associated with global change.
DOC concentrations are not increasing everywhere, however. Declines in DOC concentration over time have been associated with decreasing soil organic matter solubility (Clair et al., 2008), declines in carbon inputs from upstream acidified lakes (Schindler et al., 1997), increases in soil aluminum pools (Löfgren et al., 2010), and greater adsorption of DOM to the mineral layer and infiltration of DOM deeper into permafrost soils (Kendrick et al., 2018;Striegl et al., 2005). Long-term stability in stream DOC concentrations has also been observed (Chow et al., 2017;Clair et al., 2008;Monteith et al., 2007;Worrall et al., 2004), even in some of the longest existing records of stream chemistry (e.g., since 1975; Räike et al., 2012).
Trends in total organic carbon concentrations have even varied in direction within a continuous 35-year record (Erlandsson et al., 2008;Lepistö et al., 2008). Despite the evidence that a wide range of changes in DOC concentration can be expected, a broad multibiome assessment of global DOC trends is lacking. A spatially distributed analysis would allow for the examination of trends along multiple environmental gradients and for the testing of coherent cross biome patterns (e.g., Dodds et al., 2019).
Concentrations of dissolved organic nitrogen (DON) are rarely measured in long-term studies of DOM. Changes in DON concentration can have critical implications for freshwater ecosystems, especially when DON serves as a primary source of N for biota (Kissman et al., 2017;Mackay et al., 2020). While analytical challenges exist in the assessment of DON, researchers often assume that the concentrations of DON track those of DOC (i.e., concentrations are European Regional Development Fund, Grant/Award Number: RTI2018-094521-B-100 and RYC-2017-22643 negative trends in stream DOC and DON concentrations, changes in direction or magnitude were inconsistent at regional or local scales. DON trends did not always track those of DOC, though DOC:DON ratios increased over time for ~30% of streams. Our results indicate that the dissolved organic matter (DOM) pool is experiencing fundamental changes due to the recovery from atmospheric acid deposition. Changes in DOC:DON stoichiometry point to a shifting energynutrient balance in many aquatic ecosystems. Sustained changes in the character of DOM can have major implications for stream metabolism, biogeochemical processes, food webs, and drinking water quality (including disinfection by-products).
Understanding regional and global variation in DOC and DON concentrations is important for developing realistic models and watershed management protocols to effectively target mitigation efforts aimed at bringing DOM flux and nutrient enrichment under control.
Other lines of evidence, however, suggest that concentrations of DOC and DON can respond differently to environmental change such as changes in the concentrations of inorganic nutrients (Lutz et al., 2012;Wymore et al., 2015Wymore et al., , 2021cYates et al., 2019) and seasonal variability in precipitation and stream runoff (Bernal et al., 2005). Recent evidence has pointed to the stream DOC:DON ratio varying according to the extent of nutrient enrichment in catchments, diverging from the soil DOC:DON ratio as systems become more nutrient-enriched through land-use change and increasing human population density (Yates et al., 2019). Such divergent trends in DOC and DON concentrations will lead to changes in DOM stoichiometry (i.e., DOC:DON ratios). DOC:DON ratios provide a relatively simple quantification of bulk DOM characteristics, which serves as an indicator of bioavailability (del Giorgio & Cole, 1998) and of changing OM sources within catchments (Yates et al., 2019).
A broad assessment of how DOM stoichiometry changes concurrently with changes in concentrations of DOC and DON could provide insights into how the energy and nutrient balance of one of the larger pools of organic matter in freshwater ecosystems is changing with potential impacts on other biogeochemical reactions (e.g., Strauss & Lamberti, 2002;Wymore et al., 2019).
The objective of this study was to explore long-term trends in DOC and DON concentrations, and DOM stoichiometry in streams and rivers across biomes of the Northern Hemisphere. Our overarching hypothesis is that changes in concentrations of DON will track those of DOC and consequently the stoichiometry of DOM will remain consistent through time (Brookshire et al., 2007;Wymore et al., 2021c). We also hypothesize that sites historically affected by acid deposition will be associated with increases in concentrations of DOC and DON assuming the same external forces are acting on each of these components of the DOM pool (Deininger et al., 2020). A global assessment of how riverine DOM is responding to global change is essential for robust regional and global scale predictive ecosystem models and for future watershed management protocols.

| Data set compilation
We compiled long-term data on DOC and DON concentrations for 74 individual streams from 7 different sites (Table 1; Figure S1) in the Northern Hemisphere spanning 42 degrees of latitude (Tables   S1 and S2). For each stream, DOC and DON data were collected at either weekly or monthly intervals, except for streams in a tallgrass prairie ecosystem (Konza Prairie: KNZ), for which we have limited DON data. For consistency across sites, we set minimum detection limits (MDL) for each solute: DOC (0.1 mg C/L), TDN (0.05 mg N/L), DON (0.01 mg N/L), NO 3 − (0.005 mg NO 3 -N/L), and NH 4 + (0.004 mg NH 4 -N/L). In addition, we only used DON values that were 5% or more of the TDN pool, to account for analytical uncertainty (Lloyd et al., 2016). For data points that were below the MDL, values were replaced with half the MDL. To estimate DOC and TDN from the Finnish data set, we multiplied TOC and TN by 0.95 (Kortelainen et al., 2006;Mattsson et al., 2005 Table S4. Concentrations used in this study were not flow-weighted as discharge data were not available for the same time frame as the chemistry time series nor available for all streams. Past work found that long-term data collection can account for the variety of discharge values that occur at a site, and in at least one of our sites DOC concentrations were not correlated with discharge (Coble et al., 2018;Rüegg et al., 2015).

| Time series and trend analyses
We examined time series from mean monthly DOC, DON, and DOC:DON values for each stream using the longest record possible from each site (  (Hirsch et al., 1982) obtained from the trend package (Pohlert, 2018)  . Although we recognize that length of the record can be an important factor in trends over time (Argerich et al., 2013), we found no clear relationships between length of the record and trends in DOC, DON, and DOC:DON across our sites ( Figure S2).
We used mutual information (MI) to determine the degree to which DOC and DON covary in each stream over time with the muti package (Scheuerell, 2017) in R (R Core Team, 2016). Mutual information is a non-parametric method that characterizes the mutual dependence of two time series (Ardón et al., 2017;Cazelles, 2004 Figure S3). Sites showing a decline in atmospheric NO 3 − and SO 4 2− fluxes over time were classified as affected by atmospheric deposition and those that showed constant atmospheric NO 3 − and SO 4 2− fluxes were classified as sites that were not affected by acid deposition ( Figure S3). We corroborated this approach with the expert knowledge of authors for their respective research sites. We used NADP data from HBF for LMP as these sites are in the same region. The sites across We determined potential predictor variables of DOC, DON, and DOC:DON trends via an elastic net analysis, which is a form of penalized regression that shrinks variables that do not influence the model (Zou & Hastie, 2005). Elastic net produces a parsimonious model with the most influential variables and is minimally influenced by collinearity among predictor variables (Finlay et al., 2015). Lambda and alpha values were determined by cross-validation and choosing the lowest mean squared error (Finlay et al., 2015). Lambda controls the shrinkage of variables while alpha selects the type of penalty where alpha values between 0 and 1 denote elastic net regression (Friedman et al., 2010

| Acid deposition history
We did not find strong evidence to confirm our second hypothesis that DOC and DON would both increase in sites historically affected by atmospheric deposition. There was no difference in longterm trends in DOC concentration (p = .38; Figure 6a)    Trends in DOC:DON ratios did not differ in their response to acid deposition history (p = .56). DOC:DON Sens slopes for sites affected by acid deposition were not different from 0 (t-test p = .33), whereas DOC:DON Sens slopes were significantly different from zero in sites unaffected by acid deposition (t-test p = .008, Figure 6c).

| Predictor variables of DOM trends
The elastic net models for chemistry and watershed characteristics identified several predictor variables for DOC, DON, and DOC:DON trends. For DOC trends, the ambient stream chemistry model accounted for a large percentage of variability, followed by acid deposition, and watershed characteristics: r 2 = .66, r 2 = .34, and r 2 = .32, respectively (Table 2) trends were more constrained in volcaniclastic areas ( Figure S4b).   , Ca 2+ , Na + ), watershed characteristics (mean annual temperature (MAT °C), mean annual precipitations (MAP, mm), mean watershed elevation, and watershed area (km 2 )), and atmospheric acid deposition (mean and peak NO 3 − and SO 4 2− deposition (kg/ha)) on DOC, DON, and DOC:DON trends (for streams with significant Sen slopes) that were considered as response variables  (Figure 4), the majority of sites exhibited no significant longterm trends (Figure 4; Arvola et al., 2004;Clair et al., 2008;Coble et al., 2018;Räike et al., 2012;Rodríguez-Murillo et al., 2015), suggesting that increasing DOC is not ubiquitous across the landscape and that local context influences these long-term trends. Increasing DOC concentrations were also not exclusive to sites affected by acid deposi-  (Veach et al., 2014) and increased drying in intermittent streams (Dodds et al., 2012) which could lead to changes in instream C concentrations (Rüegg et al., 2015).

| Cross biome patterns in DOC and DON concentration trends
Many studies examining the response of DOC over time are reported from regions exposed to significant amounts of acidic deposition (Driscoll et al., 2003;Hall et al., 2021;Hruška et al., 2009;Monteith et al., 2007;Worrall et al., 2004). Whereas these studies have in-  (Clair et al., 2008;Coble et al., 2018;De Wit et al., 2007;Driscoll et al., 2003;Evans et al., 2005). And while the DON Sen slopes presented here are within the range of those reported earlier (0.0027-0.003 mg N/L per year in northern latitudes [Clair et al., 2008;Lepistö et al., 2008]), we also present negative DON trends. Studies addressing the long-term trends in DON are rarer than those of DOC, necessitating a broader assessment of DON trends. Our results suggest that changes in DOM composition may have the greatest impact in ecosystems with the lowest DOM concentrations such as tall grass prairies (KNZ) and tropical rainforest (LUQ). In these ecosystems with low DOM concentration, small changes in DOC and DON can create a large proportional change with potentially meaningful ramifications for stream metabolic regimes (Bernhardt et al., 2018) and biogeochemical reaction rates that are often limited by the availability of energy (Brailsford, Glanville, Golyshin, Johnes, et al., 2019;Brailsford, Glanville, Golyshin, Marshall, et al., 2019;Rodríguez-Cardona et al., 2021). ing that the DOM pool as a whole is highly dynamic and that the different constituents of DOM do not always have the same ecological and biogeochemical sources and roles (e.g., Bernal et al., 2005;Brookshire et al., 2007;Lutz et al., 2011;McDowell et al., 2004;Wymore et al., 2015Wymore et al., , 2018Yates et al., 2019). For example, we found streams increasing in DOC but decreasing in DON ( Figure 5) as well as sites that changed in either DOC or DON, but not in the other constituent. These scenarios suggest a biogeochemical decoupling of the C-rich and N-rich fractions of the DOM pool where DON cycling has little effect on the overall DOC pool. Changes in concentrations of DON with no significant trend in concentrations of DOC may be the result of DON being more mobile and reactive along flow paths relative to DOC due to its hydrophilic nature (Aiken et al., 1992;Hood et al., 2003;Inamdar et al., 2012;). Scenarios in which no significant trend in DOC concentrations occurs but DON concentrations decline could also occur in the nutrient limited systems where both terrestrial and aquatic biota mine the N contained within DOM (Brailsford, Glanville, Golyshin, Marshall, et al Wymore et al., 2015). We did not detect any instances where DOC concentration is declining but DON concentration is increasing, suggesting that autochthonous contributions to the DON pool are small compared to heterogeneous terrestrial inputs from the watershed, at least for the streams included in this study.

| A changing stoichiometry of the DOM pool
DOC and DON concentrations represent two ways to measure the composition of the DOM pool, yet few studies have used both bulk elemental analyses as a way to describe the heterogenous DOM pool . Although a stoichiometric approach to understanding nutrient and elemental cycling has a rich history (Elser et al., 2000;Redfield, 1958), the principles have seldom been applied to understanding changes in bulk DOM composition over time. For those sites where a significant change in DOC:DON stoichiometry was detected, the predominant direction of change reflected the C-enrichment or N-depletion of DOM. The exception to this general pattern was in streams at the LMP site, located in the temperate deciduous forests of New England (Wymore et al., 2021a), where DOC:DON ratios are decreasing, indicating the relative N-enrichment of DOM. These sites have a high percentage of wetlands (Flint & McDowell, 2015), which are likely contributing to these changing stoichiometric ratios (Coble et al., 2019). N-enriched DOM may provide additional nutrients to microbial communities making more NH 4 + available through mineralization. In turn, competition for dissolved inorganic N may decline with higher rates of nitrification and increased NO 3 − production and export (Wymore et al., 2019), while increasing DON concentrations instream may provide an alternative nutrient resource for uptake by the primary producers (Mackay et al., 2020). In contrast, streams with increasing DOC:DON ratios may reflect increasing watershed N demand from greater retention in soils and increasing vegetative growth, possibly from CO 2 enrichment (Craine et al., 2018;Groffman et al., 2018;Huang et al., 2015). Just as instream primary producers can take up DON compounds directly as a nutrient resource, trees can bypass microbial symbionts taking up labile forms of DON directly from soils (Neff et al., 2003), which in turn would decrease DON exports to streams leading to increases in DOC:DON ratios. Changes in DOC:DON ratios can alter rates of N transformations including nitrification (Strauss & Lamberti, 2002), and NO 3 − concentrations (Bernhardt & McDowell, 2008). While the ecosystem and biogeochemical consequences of changes in DOC:DON is a relatively understudied topic, stoichiometric shifts in this particular compartment of organic matter will likely influence other biogeochemical cycles (Wymore et al., 2019;Yates et al., 2019), driving changes in the aquatic ecosystem and downstream, creating nutrient export regimes that can affect trophic assemblages in receiving bodies of water (Schade et al., 2005).

| Predictors of DOM trends
The ambient stream chemistry models for all DOM trend models se-  Figure S4). The type of soil also played a role in DOM trends being streams in silty and sandy loam landscapes the ones showing the lowest DOC and DON trends ( Figure S5). This finding suggests that the adsorption to silt particles can influence DOM availability by controlling the long-term storage and export of DOC and DON (Dosskey & Bertsch, 1997;Kaiser & Guggenberger, 2000). The higher trends of DOC and DON in moraine sites ( Figure S5) could be due to greater OM availability and associated microbial decomposition activity (Bruhn et al., 2021).
Collectively, the results of these models support the hypothesis that regional state factors such as geology and soil type are important controls of stream long-term DOM trends. Continued monitoring of these long-term trends in DOM concentration and stoichiometry in response to climatic and landscape attributes is important to better understand the ultimate fate of DOM and nutrients in freshwater ecosystems in the face of global change.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data used in these analyses represent a synthesis of multiple data sets. The individual data sets and their associated repositories and references can be found in Tables S4 and S5. The archived data set (Wymore et al., 2021b)