Increasing dominance of terrigenous organic matter in circumpolar freshwaters due to permafrost thaw

Climate change and permafrost thaw are unlocking the vast storage of organic carbon held in northern frozen soils. Here, we evaluated the effects of thawing ice‐rich permafrost on dissolved organic matter (DOM) in freshwaters by optical analysis of 253 ponds across the circumpolar North. For a subset of waters in subarctic Quebec, we also quantified the contribution of terrestrial sources to the DOM pool by stable isotopes. The optical measurements showed a higher proportion of terrestrial carbon and a lower algal contribution to DOM in waters affected by thawing permafrost. DOM composition was largely dominated (mean of 93%) by terrestrial substances at sites influenced by thawing permafrost, while the terrestrial influence was much less in waterbodies located on bedrock (36%) or with tundra soils unaffected by thermokarst processes (42%) in the catchment. Our results demonstrate a strong terrestrial imprint on freshwater ecosystems in degrading ice‐rich permafrost catchments, and the likely shift toward increasing dominance of land‐derived organic carbon in waters with ongoing permafrost thaw.

DOM in waters affected by thawing permafrost. DOM composition was largely dominated (mean of 93%) by terrestrial substances at sites influenced by thawing permafrost, while the terrestrial influence was much less in waterbodies located on bedrock (36%) or with tundra soils unaffected by thermokarst processes (42%) in the catchment. Our results demonstrate a strong terrestrial imprint on freshwater ecosystems in degrading ice-rich permafrost catchments, and the likely shift toward increasing dominance of land-derived organic carbon in waters with ongoing permafrost thaw.
Northern permafrost regions contain one of the largest pools of carbon in the Earth system (Schuur et al. 2015), and increased attention is now focused on how these massive carbon stocks may be mobilized and converted to greenhouse gases with ongoing climate change. One trajectory for this conversion is via microbial metabolism in the lakes and ponds that receive dissolved and particulate organic carbon from eroding permafrost soils. Many of these waterbodies are created by thermokarst (erosion and collapse of ice-rich permafrost), resulting in thaw or thermokarst lakes and ponds (Vonk et al. 2015), hereafter referred to as thaw ponds. In some parts of the Arctic landscape, these ponds are disappearing by drainage, evaporation, or infilling, while in other northern regions they are becoming larger and more numerous . These waters represent one of the most abundant freshwater ecosystem types in northern regions, and collectively they have an estimated total surface area in the range 250,000-380,000 km 2 (Grosse et al. 2013); this accounts for approximately 25% of the estimated total area covered by lakes and ponds in the Arctic (1.4 3 10 6 km 2 ; Muster et al. 2017). They encompass a variety of transparencies and trophic conditions, and are likely to become more turbid and heterotrophic with increasing carbon inputs from thawing permafrost soils (Vonk et al. 2015). Thaw ponds are also known to be hotspots in the landscape for strong emissions of greenhouse gases to the atmosphere (Abnizova et al. 2012;Negandhi et al. 2013;Sepulveda-Jauregui et al. 2015;Matveev et al. 2016).
Although the effect of increasing export of terrestrial organic compounds from the catchment and shifts in aquatic metabolism are known to change the carbon pool composition and the proportion of carbon of terrestrial origin (which we refer to as allochthony) in temperate freshwaters (Solomon et al. 2015), little is known about these effects in circumpolar ponds, despite the abundance of these ecosystems across the northern landscape (Muster et al. 2017) and their key role in carbon cycling (Abnizova et al. 2012). In this study, we investigated the impact of permafrost thaw on the concentration and composition of dissolved organic matter (DOM) using a suite of chemical, biological, optical, and stable isotopic (d 13 C and d 2 H) measurements. We hypothesized that DOM concentrations are higher in ponds that have emerged due to thawing permafrost soils than in waterbodies not affected by thermokarst processes. We further hypothesized that terrestrial carbon compounds are most abundant in ponds draining thawing soils, while ponds in regions with rock outcrops and nondegrading permafrost soils are relatively more enriched in autochthonous DOM, the fraction originating from aquatic primary production. Specifically, we predicted that these relative proportions of terrestrial matter in the DOM pool of different pond types are expressed in the DOM d 13 C and d 2 H isotopic composition.

Study regions
During the summer periods from 2002 to 2016, we sampled a total of 253 ponds distributed in 14 circumpolar regions ( Fig. 1), for a total of 356 samples, including from a subset of ponds (55) that were sampled more than one time during the 15 yr. The regions span over a wide geographic area, covering around 200 degrees of longitude (from Alaska to Russia) and 30 degrees of latitude (from Subarctic to High Arctic), and encompassing a large range of temperatures (mean annual from 22.68C to 2188C), vegetation types (from spruce or birch forest and shrub tundra to polar desert), and permafrost cover (from sporadic to continuous coverage; further details in Supporting Information Table  S1). We divided the ponds into three categories according to their catchment characteristics and exposure to permafrost thaw, following Rautio et al. (2011): (1) bedrock ponds (21 ponds), characterized by rocky surroundings and with little terrestrial vegetation, and no effect of thawing permafrost soils; (2) tundra ponds (88 ponds), not directly formed or significantly impacted by degrading permafrost soils, but surrounded by grass-, shrub-, or forest-tundra vegetation; and (3) thaw ponds (144 ponds), which are thermokarst waterbodies formed by thawing and collapse of ice-rich permafrost (Vonk et al. 2015). Photographs of the ponds in each category are shown in Supporting Information Fig. S1. The sampled ponds were smaller than 10 ha, with the exception of 17 larger but shallow (< 3.5 m) waterbodies.

Chemical and biological analyses
We collected surface-water samples to measure pH, total phosphorus (TP), total nitrogen (TN), and total dissolved iron (Fe) concentrations, and filtered subsamples through pre-rinsed cellulose acetate filters (0.2 lm) in order to analyze dissolved organic carbon (DOC) concentrations and perform optical analyses on the chromophoric DOM (CDOM; see below). This filter pore size removes small-size inorganic soil particles common in thaw ponds (Watanabe et al. 2011), and was chosen to be consistent with earlier studies on circumpolar ponds (Breton et al. 2009;Laurion et al. 2010;Roiha et al. 2015). The samples were stored in acid-washed and combusted glass vials at 48C in the dark, and DOC quantification was carried out using a carbon analyzer (TOC-5000A or TOC-VCPH, Shimadzu, Kyoto, Japan). Seston in the surface water was filtered onto GF/F glass fiber filters to determine phytoplankton chlorophyll a (Chl a) concentrations, as in Nusch (1980).

Optical analyses
CDOM absorbance was measured between 250 nm and 800 nm using a UV-visible Cary 100 (Agilent, Santa Clara, California), Cary 300 (Agilent, Santa Clara, California), or LAMBDA 650 (PerkinElmer, Waltham, Massachusetts) spectrophotometer, depending on sample origin. After subtracting the blank spectrum, we applied a null-point adjustment, using the mean value from 750 nm to 800 nm, and report CDOM as the absorption coefficient at 320 nm (a 320 ) and 440 nm (a 440 ) according to the equation: a k 52:303 3 A k =L; (1) where a k is the absorption coefficient (m 21 ) at wavelength k, A k the absorbance corrected at wavelength k, and L the path length of the cuvette (m) (Blough and Del Vecchio 2002). The specific ultraviolet absorbance at 254 nm (SUVA 254 ) was determined from DOC normalized A 254 as an index of aromaticity and the relative proportion of terrestrial vs. algal carbon sources in DOM (Weishaar et al. 2003). Iron can complex humic substances and increase DOM absorbance in elevated concentrations, inducing an overestimation of SUVA 254 (Xiao et al. 2013), and we therefore applied the following equation when the Fe concentration was higher than 2 mg L 21 (Poulin et al. 2014): A 254 corrected 5 A 254 measured 2 0:0653 3 Fe ½ ð Þ The Fe concentration was only higher than 2 mg L 21 in some ponds of Bylot Island (region 6), but this variable was not available for the regions of Zackenberg (region 12) and Seida (region 14). Therefore, the interpretation of the optical indices should be made with caution, especially in systems influenced by active permafrost erosion and containing high iron concentrations. We also determined spectral slopes (S) following Loiselle et al. (2009) Enterprises 2015). We used S 289 to estimate the importance of fulvic and humic acids related to algal production (Loiselle et al. 2009), and the slope ratio (S R ) S 285 /S 375 was calculated as an index of CDOM molecular weight (Helms et al. 2008). For a subsample of 100 ponds, we also recorded fluorescence intensity on a Cary Eclipse spectrofluorometer (Agilent,  Santa Clara, California) across the excitation waveband from 250 nm to 450 nm (10 nm increments) and emission waveband of 300-560 nm (2 nm increments) in order to construct excitation-emission matrices (EEMs). We calculated the fluorescence index (FI) as the ratio of fluorescence emission intensities at 450 nm and 500 nm at the excitation wavelength of 370 nm to investigate the origin of fulvic acids (McKnight et al. 2001). To identify and quantify the main DOM components, we ran a parallel factor (PARAFAC) model on 129 samples from 95 ponds in MATLAB v R2013a (MathWorks, Natick, Massachusetts), as in Murphy et al. (2013). We corrected EEMs for Raman and Rayleigh scattering and inner filter effects, and standardized the fluorescence to Raman units using the FDOMcorr 1.4 toolbox (Murphy et al. 2010). The model was performed on corrected EEMs and validated by split-half analysis (Supporting Information Fig. S2) using the drEEM toolbox from Murphy et al. (2013). For each sample, we summed the maximum fluorescence [Cx] of the different components x to determine the total fluorescence (F T ) and calculated the relative abundance of any component x, according to the following equation: To identify the components of the model, we compared the excitation and emission spectra to published components from more than 70 papers available in the OpenFluor database, following Murphy et al. (2014).

Stable isotope analyses
To quantify the relative contribution of terrestrial carbon among pond types, we carried out stable isotope analyses (SIA) on surface-water DOM and its potential sources in a subsample of 10 ponds in the vicinity of Kuujjuarapik, subarctic Quebec (region 7 in Fig. 1). Given the presence of the three pond types in this same region sharing similar environmental conditions, Kuujjuarapik was a convenient place to perform such analyses. In addition to collecting soils surrounding ponds, DOM, benthic bulk material and seston were sampled in bedrock and tundra waterbodies for d 13 C and d 2 H analyses. Thaw ponds in Kuujjuarapik are located in a peatland with abundant semi-aquatic macrophytes. Because of their high turbidity, no light reaches the bottom in these ponds (as measured with an underwater radiometer; Li-Cor BioSciences, Lincoln, Nebraska), and strong thermal stratification limits the exchange between bottom and surface waters (Matveev et al. 2016). Therefore, we considered the contribution of benthic material to surface-water DOM to be negligible and took the wetland macrophytes into account as a potential source to DOM in these thaw ponds. DOM samples for stable isotopes were collected and stored as indicated in "Chemical and biological analyses" section. Soil samples were collected from the top layer (0-5 cm) close to the ponds. For bedrock and tundra ponds, this layer represents the shallow soils around the ponds, while for the thaw ponds, this was surface soils from the organic-rich palsas that are collapsing into the ponds (see Fig. 7 in Vincent et al. 2017). In order to remove the carbonate, we applied an acid fumigation to the soil samples during 96 h prior the d 13 C analyses, as described in Ramnarine et al. (2011). Decaying submerged macrophytes (Carex sp.) were sampled from the edge of thaw ponds. In bedrock and tundra ponds, we sampled the surface of submerged rocks by scraping with a spatula to collect the epibenthic material. All samples were freeze-dried before SIA. DOM samples were analyzed for d 13 C using an Aurora 1030W TOC Analyzer ( Doucett et al. (2007), using a CONFLO II coupled to a Delta Plus XL MS (Thermo Fisher Scientific, Waltham, Massachusetts). To determine the phytoplankton d 13 C signature, we used specific algal fatty acids (FAs) extracted from bulk seston as a proxy, performing SIA on 16:1n7, 18:2n6, 18:3n3, and 20:5n3 fractions (Taipale et al. 2015;Grosbois et al. 2017a,b). The FAs were transmethylated according to a protocol adapted from Lepage and Roy (1984). The d 13 C analyses on FAs were carried out in the Stable Isotope Laboratory of Memorial University (Memorial University of Newfoundland, Newfoundland and Labrador), using a 6890N GS (Agilent, Santa Clara, California) linked to a Delta V Plus MS (Thermo Fisher Scientific, Waltham, Massachusetts). We estimated the phytoplankton d 2 H signature from filtered water as in Grosbois et al. (2017a). As for the phytoplankton, we extracted the FAs from the benthic bulk material and performed d 13 C analyses on the 14:0, 16:0, and 18:0 fractions, which are saturated FAs specific to most organisms (Napolitano 1999), and were thus considered representative of the d 13 C isotopic composition of the benthic organic matter.

Mixing model
We considered three potential sources contributing to DOM in each pond type: soils, phytoplankton, and benthic bulk for bedrock and tundra ponds, and soils, phytoplankton, and macrophytes for thaw ponds. We performed a dual Bayesian mixing model adapted from Wilkinson et al. (2014), using d 13 C and d 2 H as end-members. The model was run in R v 3.3.2. (R Development Core Team 2016).

Statistical analyses
To determine how the chemical, biological, and optical properties were individually influenced by permafrost thaw, the data were analyzed by Kruskal-Wallis rank tests, with pair-wise comparisons using a post hoc test (Bonferroni). We carried out principal component analyses (PCAs) and permutational multivariate analyses of variance (PERMANOVAs) on log 10 (x 1 1)-transformed data to illustrate and test the influence of pond type on optical properties. The data were centered and standardized before applying the PCAs. In the PERMANOVAs, Euclidean distance was used as the dissimilarity index, and the number of permutations was fixed at 999. Pair-wise comparisons were performed using Bonferroni correction to identify differences among the types of ponds. The multivariate homogeneity of group dispersions was verified by performing a permutational analysis of multivariate dispersions (PERMDISP). All statistical analyses were performed on R v 3.3.2. (R Development Core Team 2016). The data from the 55 ponds that were sampled more than once but in different years were considered independent based on the assumption that in these rapidly changing landscapes, the biological, chemical, and optical properties of the ponds can vary greatly from one sampling year to another.

Chemical, biological, and optical properties
The overall data set (archived in Wauthy et al. 2017) showed a strong effect of permafrost thaw (Fig. 2), with highly significant differences between thaw and non-thaw ponds for all variables, except for pH and %C4 (p < 0.01, Kruskal-Wallis by rank test). Chl a, TP, and TN had highest values in thaw ponds, indicating a more enriched trophic state. There were wide variations in DOC concentration, from 1.0 mg L 21 to 116.8 mg L 21 , with highest values in the thaw ponds. CDOM followed the same trend, as indicated by a 320 and a 440 values. SUVA 254 was also higher in thaw ponds, suggesting a larger proportion of terrestrial DOM in these ponds as compared to bedrock and tundra ponds. Consistent with this interpretation, S 289 , S R , and FI values were lowest in thaw ponds, suggesting smaller amounts of carbon derived from aquatic primary production, higher DOM molecular weights, and a terrestrial origin for fulvic acids, respectively.
The PARAFAC model identified five fluorescence components (Supporting Information Fig. S3), of which four shared fluorescence characteristics with humic materials from terrestrial (C1-C3) and microbial origin (C4). The last component (C5) presented spectra similar to amino acids or proteins, and was attributed to algal production (Stedmon and Markager 2005) (more details in Supporting Information Table  S2). The percentage of terrestrial humic components (%C1-%C3) showed different patterns among the pond types ( Fig.  2): while %C1 and %C2 were significantly higher in thaw ponds, %C3 was greater in the ponds not affected by permafrost thaw. Although terrestrial humic components dominated in all pond types, thaw ponds showed higher proportions of C1-C3 (73.2% 6 10.2%) than bedrock (63.4% 6 19.7%) and tundra ponds (64.9% 6 15.5%). Moreover, thaw ponds had the highest mean value for %C4 (17.6% 6 5.1%), further indicating the presence of humicdominated microbial material in these ponds. However, the proportion was not different from bedrock ponds (16.6% 6 5.6%). Finally, the proportion of DOM associated with algal production was small in the thaw ponds, as indicated by its significantly lower %C5 (9.2% 6 7.0%).
The two PCAs showed the tendency of pond types to group according to optical properties (Fig. 3). Figure 3A is for the PCA performed on the subsample of 95 ponds for which we had both spectrophotometric and fluorometric properties; the first two components explained 74.8% of the total variability (PC1 52.4% and PC2 22.4%). We observed a significant effect of catchment type (PERMANOVA, F 2,126 5 19.1, p 5 0.001), with the three pond types forming three distinct clusters. However, only the thaw pond cluster was significantly different from the bedrock and tundra groups (pair-wise PERMANOVA comparisons, p-adjusted < 0.05). We applied the second PCA on the complete dataset, but only for DOC and DOM spectrophotometric proxies; the first two components explained 89.0% of the variance (PC1 72.6% and PC2 16.4%) (Fig. 3B). The effect of catchment type was again significant (PERMANOVA, F 2,353 5 65.52, p 5 0.001) and the three pond types formed three significantly different clusters (pair-wise PERMANOVA comparisons, p-adjusted < 0.05). Bedrock and tundra ponds were mainly defined by their positive correlation with S R , S 289 , and %C5, but were also positively correlated to FI, %C3, and

Source contributions to DOM
The isotopic signatures of DOM showed similar d 13 C values between the different pond types in the Kuujjuarapik region, within the range 229.5& to 227.3& (Fig. 4). However, the DOM deuterium signature d 2 H was more depleted in thaw ponds (2153.9 6 11.0&) compared to other ponds (2131.2 6 8.2&). Similarly, the phytoplankton d 13 C signature was more negative in thaw ponds than in non-thaw ponds (237.1 6 2.5& vs. 230.5 6 0.8&). According to its location in biplots (Fig. 4A,B), DOM in bedrock and tundra ponds fitted well within the polygon of source end-members, and appeared to be mainly composed of benthic and terrestrial materials. The DOM in thaw ponds was positioned at the limit of the polygon, with closest affinity to terrestrial sources (Fig. 4C).
The mixing model showed a large contribution of benthic (median of 55%) and terrestrial (median of 22%) material to the DOM of bedrock ponds, although the range of the 95% highest contribution probability showed high variabilities (0-97% and 0-99%, respectively) (Fig. 5A). The terrestrial sources were more substantial in tundra ponds (median of 42%), but the benthic material remained the principal contributor to DOM (median of 47%) (Fig. 5B). However, the range was also high for the two main sources (2-90% for the benthic source, and 2-82% for the terrestrial source). Phytoplankton was a significant but minor source of DOM in both pond types not affected by thawing permafrost, with a median of 11% and 6% in bedrock and tundra ponds, respectively. In the thaw ponds, there was a major shift to a higher proportion of carbon from terrestrial origin for DOM, with terrestrial sources contributing 96% (median, Fig. 5C) and negligible contributions from phytoplankton and macrophytes (medians less than 2%).

Nutrients and primary producers
The studied 253 ponds spanned a wide gradient of environmental conditions, from transparent oligotrophic waterbodies in areas not affected by thawing permafrost, to humic-and nutrient-rich ponds exposed to thermal erosion (Fig. 2). Nutrients in bedrock and tundra ponds were typical of oligotrophic freshwater systems, and comparable to values usually found in clear-water high-latitude ponds (Rautio et al. 2011). In thaw ponds, they showed values more characteristic of mesotrophic and eutrophic systems, and likely originated from the eroding catchment (Larsen et al. 2017). Following the low nutrient concentrations in the water column, Chl a values in bedrock and tundra ponds indicated low phytoplankton biomass. Consequently, most primary production in clear-water ponds is produced by the benthic mat and biofilm communities, with phytoplankton often representing less than 2% of the total photosynthetic biomass in these systems (Bonilla et al. 2005;Rautio et al. 2011). In thaw ponds, more elevated Chl a values suggested a higher planktonic primary production, more probably supported by the higher nutrient concentrations (Vonk et al. 2015). However, the higher CDOM concentrations and suspended solids in thaw ponds efficiently attenuate the solar radiation (Watanabe et al. 2011), limiting benthic and therefore overall primary production (Vadeboncoeur et al. 2008). Increased terrestrial DOM in circumpolar surface waters could therefore lead to a considerable decrease in the light availability for photosynthesis, resulting in a shift toward a heterotrophic production-based food web Bent. Terr.

Sources:
Phyt. as has been documented in experimental conditions (For-sstr€ om et al. 2015), and a high production of CO 2 and CH 4 (Roiha et al. 2015).

DOC and CDOM properties
Highest DOC and CDOM values were observed in the thaw ponds, with several CDOM proxies indicating higher terrestrial inputs from the catchment (Fig. 2). Similar accumulation of DOM has also been reported previously at circumpolar sites with high terrestrial inputs (Vonk et al. 2013;Abbott et al. 2014;Roiha et al. 2015). The elevated values of a 320 and a 440 in thaw ponds indicate high concentrations of CDOM, inducing more light attenuation in the water column. Consistent with earlier studies, S R , S 289 , and SUVA 254 indicated fresher aromatic compounds of higher molecular weight and a large proportion of terrestrial vs. algal carbon sources in thaw ponds (e.g., Roiha et al. 2015). This high degree of DOM allochthony in thaw ponds was also supported by low values of FI, an indicator of large inputs of fulvic acids from terrestrial sources and a function of carbon storage in the catchment (Rantala et al. 2016).
The association of [C1] and [C2] with SUVA 254 and other CDOM proxies in ponds influenced by thawing permafrost also indicates strong DOM allochthony in these waters (Fig.  3A), with high inputs of DOM from the catchment. C3 also had fluorescence characteristics of humic materials from a terrestrial origin, but showed an opposite association with C1 and C2. This suggests that C3 could be the product of biological transformation of C1 and C2 in the water column (Jørgensen et al. 2011). C4 matched well with humic materials of microbial origin, and therefore can be linked to the degradation of both algal and terrestrial sources. Finally, the amino acid-or protein-like algal component C5 showed higher values in bedrock and tundra ponds, and negative relationships with SUVA 254 and thawing permafrost, indicating a greater algal origin of DOM in non-thaw ponds. This algal signature likely reflects the benthic primary production in these ponds, given its dominant contribution to overall algal biomass in clear-water circumpolar ponds (Rautio et al. 2011).
It is important to note that we focused on thermokarst ponds, with permafrost thaw and degradation along pond banks, as is commonly found across the North. There are other modes of permafrost thaw that can have different consequences for surface-water DOM concentrations and composition. For instance, in certain hydrological conditions, catchment-scale permafrost thaw via active layer thickening or permafrost loss can reduce DOC concentrations or SUVA 254 values (Cory et al. 2013;O'Donnell et al. 2014). The highly variable organic carbon content of thawing permafrost soils ) may also influence the DOM properties in the receiving waterbodies. In this study, we did not measure the organic carbon content in the watershed, but the existing information from the study regions (Bouchard et al. 2015;Vincent et al. 2017) as well as our observations of the bank morphology and benthic substrates indicate that the studied thaw ponds were predominantly located in organic-rich sites. Therefore, substantial impacts of permafrost thaw on pond DOM were expected.
Thawing of ice-rich permafrost appears to have a strong effect on the ratio of allochthonous to autochthonous DOM in surface waters, resulting from direct inputs of allochthonous DOM from eroding permafrost soils, and from its effect on DOM age (O'Donnell et al. 2014), in situ transformations and respiration (Laurion and Mladenov 2013;Cory et al. 2014). In particular, the level of bacterial and photochemical transformation of terrigenous DOM is likely to vary between the turbid and bacteria-rich thaw ponds as compared to clear and oligotrophic waters of non-thaw ponds, with important consequences on DOM allochthony. Moreover, drivers that are independent of permafrost thaw could potentially explain part of the differences observed between thaw and non-thaw ponds. These drivers include the composition, size and slopes of the catchment (Olefeldt et al. 2014;Vonk et al. 2015), influence of groundwater and precipitation (Olefeldt et al. 2013), water retention time (Catal an et al. 2016) and temperature (Porcal et al. 2015). However, our sampling covered a great variety of environments, mitigating the influence of these other factors as drivers of the observed DOM differences among pond types. Furthermore, although some of these other factors likely differed considerably between bedrock and tundra ponds, the overall water chemistry as well as the DOM concentrations and optical characteristics differed less between these two pond types than for the thaw ponds. Additionally, the DOM sourcing by stable isotope analysis for ponds in the Kuujjuarapik region indicated that these other drivers were less likely relative to allochthony. Overall, our results point to the importance of thawing permafrost for the biogeochemistry of circumpolar surface waters.

DOM contribution quantification
In the subarctic region of Kuujjuarapik (region 7), the SIA data and the mixing model supported the optical analyses and confirmed a high terrestrial DOM input in thaw ponds (Fig. 5). However, terrestrial sources were also important in non-thaw ponds, particularly in tundra ponds where soils and benthic material contributed equally to DOM. This proportional influence of terrestrial inputs likely reflects the limitation of autochthonous sources under oligotrophic nutrient conditions, as well as the large perimeter length per unit area of these small waterbodies that would favor interactions with the surrounding tundra soils.
The DOM contribution by phytoplankton was low in bedrock and tundra ponds of the region, as to be expected given the low algal biomass and primary production in the water column compared to the benthos in these ponds (Rautio and Vincent 2006), and supporting studies that have shown a low contribution of phytoplankton to the food webs of circumpolar ponds (Cazzanelli et al. 2012;Mariash et al. 2014). In the thaw ponds, phytoplankton biomass was higher, possibly from more enriched nutrient conditions, but nevertheless its contribution to DOM was insignificant and overwhelmed by the terrestrial signal.
The d 13 C signatures of phytoplankton were more depleted in the thaw ponds vs. non-thaw ponds (Fig. 4), and an explanation of this may lie in their elevated concentrations of methane (Matveev et al. 2016). Due to the fractionation during methanogenesis, methane has more negative d 13 C values and can be metabolized by methanotrophic bacteria, which then enter the food web (Bastviken et al. 2003). Mixotrophic algae constitute a major fraction of the total phytoplankton community growing in the thaw ponds of the region (B egin and Vincent 2017) and could rely on the methanotrophic bacteria as an energy source, explaining the more depleted d 13 C signature of the phytoplankton. Given the extreme shallowness (< 1 m) and the absence of stratification in bedrock and tundra ponds, the benthic source was an important contributor to DOM, likely resulting from the resuspension of decomposed benthic materials (Evans 1994) and by diffusion of benthic carbon exudates to the water column (Rautio et al. 2011;Rodr ıguez et al. 2013).
In the thaw ponds, the DOM isotopic signature was close to the outside of the polygon defined by our three selected sources (Fig. 4C). One explanation could be that we missed a source contributing to DOM. Since the studied thaw ponds are located in a peatland, mosses (Sphagnum spp.) are abundant in the catchment and could be a candidate for this missing source. However, we sampled Sphagnum mosses as a potential endmember, and found that their isotopic signature was similar to macrophytes (d 13 C of 226.6& vs. 227.2&, d 2 H of 2180.9& vs. 2177.8&); hence, Sphagnum effect was considered to be included in the macrophyte values in the mixing model. A more plausible explanation may be that labile, 13 C-enriched terrestrial matter is rapidly metabolized by bacteria once in the ponds, leaving the more depleted and recalcitrant fraction dissolved in the water column, and causing more negative d 13 C DOM values (Biasi et al. 2005;Bianchi and Canuel 2011).

Conclusions
Consistent with previous syntheses (Vonk et al. 2015;Wrona et al. 2016), our observations underscore the large variations in limnological properties among northern highlatitude waters. Despite this variability, there was a clear effect of permafrost thaw, resulting in more allochthony, higher DOC concentrations, and DOM with a predominantly landderived signature. To our knowledge, this is the first time the DOM has been reported in this extent to waterbodies in the circumpolar North. The increasing terrestrial influence on arctic and subarctic ponds results in decreased water column transparency, which in turn will affect biogeochemical processes, energy supply to benthic primary producers, and the balance of autochthonous and allochthonous production. These impacts may be viewed together as an example of "browning" that is being observed throughout the world with increasing concern (Graneli 2012; Williamson et al. 2015). Because of the erodible and DOM-rich nature of the permafrost landscape, northern browning is an extreme version of this global phenomenon, and may be compounded in the future by increased rainfall and extreme weather events ). This extreme browning will likely shift northern freshwaters more toward net heterotrophic conditions, increased DOM flocculation and oxygen depletion, and greater production of greenhouse gases across the circumpolar North.