Decreasing Photoreactivity and Concurrent Change in Dissolved Organic Matter Composition With Increasing Inland Water Residence Time

Photochemical degradation of dissolved organic matter (DOM) has been the subject of numerous studies; however, its regulation along the inland water continuum is still unclear. We aimed to unravel the DOM photoreactivity and concurrent DOM compositional changes across 30 boreal aquatic ecosystems including peat waters, streams, rivers, and lakes distributed along a water residence time (WRT) gradient. Samples were subjected to a standardized exposure of simulated sunlight. We measured the apparent quantum yield (AQY), which corresponds to DOM photomineralization per photon absorbed, and the compositional change in DOM at bulk and individual compound levels in the original samples and after irradiation. AQY increased with the abundance of terrestrially derived DOM and decreased at higher WRT. Additionally, the photochemical changes in both DOM optical properties and molecular composition resembled changes along the natural boreal WRT gradient at low WRT (<3 years). Accordingly, mass spectrometry revealed that the abundance of photolabile and photoproduced molecules decreased with WRT along the boreal aquatic continuum. Our study highlights the tight link between DOM composition and DOM photodegradation. We suggest that photodegradation is an important driver of DOM composition change in waters with low WRT, where DOM is highly photoreactive.


Introduction
Organic matter processing in inland waters is a considerable component of the global carbon (C) cycle.Dissolved organic matter (DOM) entering from terrestrial ecosystems is gradually transformed and supplemented with new DOM that is produced in situ during transport via lakes, streams, and rivers toward the coastal ocean (Berggren et al., 2022;Cole et al., 2007;Drake et al., 2018;Weyhenmeyer et al., 2012).DOM is transformed by microbes (Amon & Benner, 1996;Tranvik, 1988), photochemical degradation (Granéli et al., 1996;Wetzel et al., 1995), sorption to mineral and organic particles, aggregation and sedimentation (Droppo & Ongley, 1994;Einarsdóttir et al., 2020;Groeneveld et al., 2020;von Wachenfeldt & Tranvik, 2008).Together, these processes either remove C from the active C cycle with long term storage of organic C in inland water sediments (Mendonça et al., 2017) or can lead to emissions of CO 2 and CH 4 to the atmosphere (Bastviken et al., 2011;Cole et al., 1994;Kling et al., 1992).In boreal landscapes, photochemical degradation may contribute to approximately 10% of CO 2 emissions from inland waters (Koehler et al., 2014), while in Arctic freshwaters, it may contribute to up to 40% of CO 2 emissions (Cory et al., 2014).Although photochemical degradation has been the subject of numerous studies, its regulation and effect on molecular DOM composition along the aquatic continuum are still unclear.
The intrinsic properties of DOM, and in particular the ability of DOM to absorb sunlight is a strong driver of DOM photomineralization (i.e., its complete photooxidization into CO 2 ; Bertilsson & Tranvik, 2000;Koehler et al., 2014;Lindell et al., 2000).Terrestrially derived DOM has typically higher photomineralization rates than in situ produced DOM because it is richer in chromophoric compounds such as phenolic and other aromatic functional groups that strongly absorb sunlight (Granéli et al., 1996;Lindell et al., 2000;Opsahl & Benner, 1998).However, it is uncertain if terrestrially derived DOM has a higher mineralization per photon absorbed, that is, a higher apparent quantum yield (AQY) (Gao & Zepp, 1998), and if it is thus more photoreactive than DOM produced in situ.Although DOM photomineralization occurs via direct and indirect photochemical reactions (Goldstone et al., 2002;Vähätalo et al., 1999;Xie et al., 2004) and thus can involve chromophoric and non-chromophoric DOM, aromatic DOM is generally considered the most efficiently photomineralized DOM pool (Koehler et al., 2016;Lindell et al., 2000).This is probably because chromophoric DOM acts as the primary chromophore, meaning that it needs to be present for photochemical reactions to take place, and direct photochemical reactions are dependent on the capacity of DOM to absorb light (Sulzberger & Durisch-Kaiser, 2009).As a result, higher AQY and photomineralization rates per unit of energy absorbed are found for the generally more aromatic, terrestrially derived DOM (Bertilsson & Tranvik, 2000;Koehler et al., 2016).DOM photoreactivity generally decreases with time of light exposure (Cory et al., 2007;Powers & Miller, 2015) due to the progressive mineralization or partial oxidation of the photoreactive fraction (Molot & Dillon, 1997;Stubbins et al., 2010), a process that naturally happens during transport along the aquatic continuum.Additionally, with accumulated water residence time (WRT), the contribution of DOM from in situ primary production increases, resulting in a decrease in the proportion of terrestrially derived DOM (Kothawala et al., 2014).Several studies have shown that DOM photomineralization rates were higher in headwater streams and wetlands than in lakes because of the high proportion of photoreactive terrestrially derived DOM (Cory et al., 2014;Lapierre & del Giorgio, 2014).However, processes other than photodegradation (i.e., biodegradation, adsorption, aggregation) and new lateral inputs of DOM also modify its composition and thus DOM photoreactivity along the aquatic continuum.It is therefore not clear if the AQY gradually decreases along the aquatic continuum and if a quantitative relationship between AQY and WRT exists across different types of inland waters.
A large part of DOM is not completely photomineralized into CO 2 but is only partially oxidized after photoexposure.Such "incomplete" transformations are nevertheless important because they are likely to increase DOM bioavailability and thus its bio-mineralization into CO 2 or incorporation into the food web (Bertilsson & Tranvik, 2000;Lindell et al., 1995;Wetzel et al., 1995).Because all processes that shape the DOM composition can act simultaneously, it is difficult to attribute specific DOM composition changes to photodegradation only.In a comparison of DOM in lakes across Sweden, Kellerman et al. (2014) and Kothawala et al. (2014) found a selective loss of vascular plant derived polyphenols and humic-like compounds and an increase in aliphatic and protein-like compounds with increasing WRT.Photodegradation preferentially removes aromatic and humic-like compounds (Gonsior et al., 2013;Hansen et al., 2016;Murphy et al., 2018;Stubbins et al., 2010) and has thus been suggested as one of the main processes driving the change in DOM molecular composition with WRT.However, this remains to be validated by concomitantly (a) observing the differences in DOM composition along WRT gradients (Kellerman et al., 2014;Kothawala et al., 2014), and (b) experimentally demonstrating the role of photodegradation.Therefore, we still lack knowledge on the qualitative changes in DOM composition along a WRT gradient that can be attributed to photodegradation.
Here, we aimed to unravel the DOM photoreactivity and concurrent DOM compositional changes along a boreal aquatic continuum.More specifically, because the abundance of photo-reactive terrestrially derived DOM decreases along the aquatic continuum, we aimed to test whether DOM photo-reactivity was correlated with WRT.We hypothesized that the AQY would be indeed positively correlated with the abundance of aromatic terrestrially derived DOM and negatively correlated with WRT across different types of inland waters.We also aimed to test qualitatively if the change in DOM composition along the aquatic continuum resembles the change in DOM composition during irradiation.We hypothesized that DOM photodegradation during irradiation selectively removes aromatic terrestrially derived compounds and reflects the natural change in DOM composition observed along the aquatic continuum.To test these hypotheses, we measured the AQY as a proxy of photoreactivity and the change in DOM composition during a standardized light incubation across 30 boreal inland waters, from peatlands to streams, rivers, and lakes.

Sample Collection
Water samples were collected from 30 different freshwater ecosystems in four regions of Sweden: 3 peatlands, 5 streams, 2 rivers and 20 lakes between 28 August and 6 September 2016 (Table S1 in Supporting Information S1).The freshwaters covered a wide gradient of WRT (from 0 days for peatlands to 60 years for a large lake).WRT is interpreted as a measure of aging or time during which a water parcel was exposed to transformation via biogeochemical processes (i.e., adsorption, biodegradation, flocculation, photodegradation).Temperature, oxygen, conductivity, and pH were measured directly in the system and water samples were filtered with glass fiber filters (pre-combusted Whatman GF/F, approximate pore size 0.7 μm) within 72 hr and stored in the dark at 4°C before the photoincubation that took place ca. 2 months after sample collection.Water was sampled from just below the surface in rivers and streams and from small pools in peatlands (except for site 10, where water was collected from a previously dug hole) and from 0.5 m depth in lakes.Further details on the sampling sites, the biophysico-chemical parameters measured in the water samples (i.e., chlorophyll a [Chla], dissolved organic carbon [DOC], total phosphorus [TP], total nitrogen [TN], particulate organic matter [POM], anions, cations, organic acids, bacterial abundance and production), and on WRT calculation can be found in Attermeyer et al. (2018) and Groeneveld et al. (2020) who studied the adsorption of DOM to minerals and POM degradation, respectively, in parallel to this study.

Experimental Setup
The filtered water collected at each site was irradiated for 24 hr in a solar simulator (Q-Sun 1000 Xenon test chamber, Q-panel Lab Products Europe, Bolton) in six replicates in cylindrical quartz vessels with planar tops and bottoms (height 5 cm, volume 50 mL) without headspace.The cylindrical quartz vessels were standing vertically in a water bath to keep the temperature between 20 and 25°C and were covered by black tape on the side so that the light only entered from the top of the cylinder.The total solar simulator irradiance dose (approximately 575 W m 2 from 280 to 600 nm for 24 hr, Figure S1 in Supporting Information S1) was equivalent to roughly 2 days of solar radiation at the surface of the water in July in Sweden (day length of 20 hr; SMHI, URL https:// www.smhi.se/en/climate/climate-indicators/climate-indicators-global-radiation-1.91484).DOC and dissolved inorganic carbon (DIC) were measured before and after irradiation in triplicate to calculate the extent of DOC photomineralization.To assess DOM change in composition induced by photodegradation, fluorescence and absorbance were analyzed on all six replicates after irradiation, and on three replicates before irradiation.Mass spectrometry analyses were performed on one replicate before irradiation and three replicates after irradiation.
The Napierian absorption coefficient a (m 1 ) was calculated as With Abs the absorbance at a specific wavelength and L the pathlength (m).The Napierian absorption coefficient was used for the calculation of the AQY (see below).The absorption coefficient at 420 nm (α 420 in m 1 ) was calculated as the absorbance at 420 nm divided by the pathlength.The specific UV absorption coefficient at 254 nm (SUVA 254 in L mg C 1 m 1 ) was calculated as the absorbance at 254 nm divided by the pathlength and DOC concentration.The slopes of the absorption spectra between 275 and 295 nm (S 275-295 ) and between 350 and 400 nm (S 350-400 ) were calculated by linear regression of the natural log-transformed absorption spectra and the spectral slope ratio Sr was calculated as the ratio of S 275-295 and S 350-400 (Helms et al., 2008).

Fluorescence Spectroscopy
Fluorescence excitation-emission matrices (EEMs) were obtained and analyzed as described in Groeneveld et al. (2020).Briefly, EEMs (from 250 to 445 nm at intervals of 5 nm for excitation and from 300 to 600 nm at increments of 4 nm for emission) were obtained using a SPEX Fluoromax-4 spectrofluorometer (Horiba Jobin Yvon, Kyoto, Japan), blank-subtracted and corrected for inner filter effect using the FDOMcorr toolbox for MATLAB (Mathworks) following Murphy et al. (2010).
Parallel Factor Analysis (PARAFAC) was applied using the drEEM 0.5.1 package for MATLAB following Murphy et al. (2013).As the present study was part of a larger project with different types of incubations (see sample collection section), we modeled the EEMs from three incubations together, thus n = 512 samples in total, but here we only present the data corresponding to the photoincubation samples (n = 264).EEMs were preprocessed as follows: six samples were considered outliers and in four samples faulty parts were removed, primary and secondary Rayleigh and Raman were removed and smoothed-over and the data was normalized to total fluorescence intensity of each sample.Non-negativity constraints were applied and the appropriate number of components was determined by visual inspection of the residual fluorescence and the spectral shape of the components according to organic fluorophores.The model was validated by split-half analysis and random initialization.We then tested the components (Tucker's congruence coefficient = 94%) against the ones previously published and available in the OpenFluor database (URL: https://openfluor.lablicate.com/;May 2020) in order to search for quantitative matches with previously published and validated PARAFAC models.We report the PARAFAC components in their relative intensity (e.g., ) in order to be able to compare qualitative changes in fluorescence (Figure S2 and Table S2 in Supporting Information S1).The model is shared publicly in the OpenFluor database (URL: https://openfluor.lablicate.com/).
In addition to the five PARAFAC components, six commonly used indices were calculated from the fluorescence measurements and represent proxies of either terrestrial/humic-like or fresh/biologically produced DOM.All indices and their calculation and interpretation are detailed in Table S2 in Supporting Information S1.

Mass Spectrometry
The composition of ionizable DOM was analyzed by direct infusion electrospray ionization Orbitrap mass spectrometry (Hawkes et al., 2016) as described in Groeneveld et al. (2020) and in Text S1 in Supporting Information S1.Mass lists were assigned to formulas allowing up to C 50 H 100 O 40 N 2 S 1 .Formulas were only allowed to contain N or S but not both.Only peaks that were present in three samples (over the 120 samples in total) were kept, decreasing the number of total peaks from 6,653 to 2,911.We used the term "molecule" in the rest of the manuscript to design individual peaks although one molecule may correspond to several peaks in the mass spectrum, and each peak sums the information from numerous isomers.Three categories of molecules were defined according to Stubbins et al. (2010) by comparing the mass spectra before and after irradiation: (a) photolabile molecules present only before irradiation, (b) photoproduced molecules present only after irradiation, and (c) photoresistant molecules present before and after irradiation.The abundances of the three molecule categories were calculated as the relative intensity (in percentage) of the peaks before and after irradiation for photolabile and photoproduced molecules, respectively.The abundance of photoresistant molecules before and after irradiation can be deduced from the abundance of photolabile and photoproduced molecules (i.e., abundance of photoresistant molecules before irradiation [%] = 100 abundance of photolabile molecules [%]).The average and standard deviation (SD) of the abundance of photolabile and photoproduced molecules were calculated for each site by comparing the spectrum of the sample before irradiation with each of the three replicates after irradiation.SD > 1 was used as an arbitrary threshold to exclude the post-irradiation samples that were highly different from the two other replicates.A total of 8 out of 90 post-irradiation spectra were thus excluded from further analyses.

Analysis and Calculation of Photomineralization
DOC and DIC were measured on a Total Carbon Analyzer (Sievers M9 Laboratory Analyzer, GE Analytical Instruments, Boulder, Colorado, USA).DIC was measured directly in the quartz vessels before and after irradiation to avoid equilibration with outside air.The water sampled for DIC analysis before irradiation was replaced by the water used to fill the quartz vessels to avoid the creation of a headspace.One site that had a negative DIC gain (site 1, Table S1 in Supporting Information S1) was removed for the calculation of the AQY.
The wavelength-integrated AQY (mmol DIC mol photon 1 ) was calculated by dividing the DIC gain (mmol DIC) by Q tot (mol photon), the total number of photons absorbed between 280 and 600 nm during irradiation.Q tot was calculated according to Hu et al. (2002), correcting for the inner filter effect (or "self-shading") inside the quartz vessels by assuming that the light traveling vertically in the vessels decreases non-linearly along the pathlength, L (m), according to the absorbance of the sample: With T the irradiation time (s), S the surface of the illuminated area, equivalent to the cross section of the cylindrical quartz vessel (m 2 ), E(0) λ , the photon irradiance of the lamp at L = 0 and at a specific wavelength λ (mol

Global Biogeochemical Cycles
10.1029/2023GB007989 photon m 2 s 1 ) and a CDOM,λ and a tot,λ the Napierian absorption coefficients of the chromophoric molecules and the water sample at a specific wavelength (m 1 ), respectively.
We assumed that a CDOM,λ a tot ,λ = 1 because the absorbance of the water molecules was supposed to be negligible compared to the absorbance of the chromophoric compounds at short wavelengths, where a tot is the highest (Hu et al., 2002), hence: (3) E(0) λ was calculated as the irradiance of the lamp I λ (J m 2 s 1 ) divided by the energy per mol of photons at a specific wavelength: With h the Plank's constant (J s), N a the number of photons in 1 mol, c the speed of the light (m s 1 ) and λ the wavelength (m).
Since photobleaching occurred during the 24 hr exposition in the solar simulator (Vachon et al., 2016;Zhang & Xie, 2015), Q tot was calculated using the average of a CDOM,λ obtained before and after irradiation.

Statistical Analyses
All statistical analyses were performed using the R software (R Core Team, 2020).The statistical significance was defined as p < 0.05.To test the correlations between AQY, WRT and DOM quality assessed by optical spectroscopy, we selected a set of variables that maximized the variance in DOM quality between the different samples with a principal component analysis (PCA).The first axis of the PCA (PC1) explained 58% of the variability and was negatively correlated to indices related to autotrophic production of organic matter within the water body, and the abundance of low molecular weight, protein-like and aliphatic compounds (i.e., correlation between PC1 and %C 5 , Sr, BIX, FrI, H:C between 0.79 and 0.88), and positively correlated to the indices related to the abundance of humic-like and aromatic compounds (i.e., correlation between PC1 and HIX, SUVA 254 , A:T, C:T, α 420 , %C 3-4 between 0.81 and 0.96) (Figure S3 and Table S2 in Supporting Information S1).We selected, as the main descriptors of DOM quality, the humic-like component %C 3 (excitation peaks at 275/ 405 nm and emission peak at 508 nm), reflecting aromatic and terrestrially derived DOM (Ishii & Boyer, 2012;Kothawala et al., 2014) and the variables SUVA 254 and α 420 describing aromatic and colored DOM which we consider to be largely of terrestrial origin.Additionally, the protein-like component %C 5 (excitation peak at 275 nm and emission peak at 324 nm) was selected as the main descriptor of DOM in situ production, as previously shown (equivalent to %C 6 in Kothawala et al. (2014)), and Sr as a main proxy of low molecular weight compounds (Hansen et al., 2016;Helms et al., 2008).
To test the first hypothesis (i.e., photoreactivity is positively correlated to the abundance of aromatic terrestrially derived DOM and negatively correlated to WRT), the relationships between AQY, the initial DOC concentration, DOM quality at the start of the incubation and WRT were tested with linear regressions.In order to assess the effect of DOC and DOM quality (SUVA 254, α 420, %C 3, %C 5 , Sr) and the effect of other environmental variables (WRT, pH, TN, TP, POM, Chla) on AQY, we performed a stepwise regression (function step; R core Team, 2020) after transforming, scaling and removing collinear variables according to Feld et al. (2016).Collinear variables were either removed according to their correlation coefficients (Pearson |r| > 0.7) or automatically with the function vifstep (package usdm; Naimi et al., 2013).The variance explained by the variables of the best models returned by the stepwise function was obtained from the function calc.relimp(package relaimpo, lmg metric; Groemping, 2006).This allows to estimate the contribution of each predictor to the total model variance, even if the predictors are correlated.Additionally, the AQY and the abundance of individual molecules before irradiation as assessed by mass spectrometry were correlated using the non-parametric test of Spearman.
To test the second hypothesis (i.e., DOM photodegradation resembles the natural change in DOM composition along the aquatic continuum), we assessed how the changes in DOC concentration and DOM quality measured by Global Biogeochemical Cycles 10.1029/2023GB007989 optical spectroscopy after irradiation relate to the natural change in DOM quality along WRT.To do so, we (a) performed linear regressions between initial DOC concentration, DOM quality at the start of the incubation and WRT and (b) tested the change in DOC concentration and in the variables describing DOM quality following irradiation for all sites pooled with a Wilcoxon test for paired samples.
All variables except SUVA 254 and %C 5 were logged (natural log) to reach a normal distribution.%C 3 was transformed using the logit function.As the minimum was approximated to 0 for WRT, 0.001 was added before logging to give the best data normalization (according to Shapiro tests and histogram plots of the data).The models were checked by visualizing the observed against predicted data and the residual distribution.Other nonlinear (exponential) models were also tested and the best model was selected according to the Akaike information criterion (AIC).
The natural modification of DOM molecular composition along the WRT gradient was also compared with the modification of DOM molecular composition after irradiation through mass spectrometry.The modification of the DOM molecular composition along the WRT gradient was obtained by correlating each molecule abundance before irradiation in the 30 sites with WRT using the non-parametric test of Spearman, as in Groeneveld et al. (2020) and Kellerman et al. (2014).The modification of the DOM molecular composition after irradiation was assessed with the relative change in molecule abundance after irradiation.The relative change in molecule abundance was calculated as the difference in the intensity of peaks before and after irradiation and divided by the sum of the intensity of peaks before and after irradiation to obtain values between 1 and 1.The change in molecule abundance was calculated separately for sites with WRT below and above 3 years.All sites with WRT ≥ 4.3 years (at the exception of site 6) were clustered toward low PC1 and high PC2 values (Figure S3c in Supporting Information S1, FactoMineR package; Lê et al., 2008) indicating that they had less aromatic DOM than sites with WRT ≤ 2.3 years.A 3-year threshold was thus chosen as an intermediate WRT value between those two clusters.Only one of the three replicates measured after irradiation was used to calculate the change in molecule abundance since doing an average of the spectra for the three replicates can artificially increase the number of photoproduced molecules, but the results using the two other post-irradiation replicates are shown in Supporting Information S1 (see below).In addition, the correlations between the relative abundance of photolabile and photoproduced molecules obtained using mass spectrometry and WRT were tested with Pearson's correlation coefficient.The change in molecular mass before and after irradiation was tested with a Wilcoxon test for paired samples.

DOM Photomineralization and AQY
DOC loss and DIC gain correlated well but DOC loss was slightly higher than DIC gain for peatland samples (R 2 = 0.96, p value < 0.0001, Figure S4 in Supporting Information S1).For a few sites with low DOC values and with a low change in DOC (≤0.33 mg L 1 ; Figure S4 in Supporting Information S1), DOC increased after irradiation (5 sites out of 30 sites) and DIC decreased (0.18 mg L 1 ; site 1, Table S1 in Supporting Information S1), contrary to the expectation.These low changes in DOC and DIC concentrations were very close to the measurement accuracy, which was 2% of the measured value.When the site with a negative DIC production was excluded, the mean DIC produced was 0.04 (from 0 to 0.13) mg L 1 h 1 , and was the highest for a peatland (site 4) and the lowest for a lake with a high WRT (site 21, WRT = 4.3 years).In the following analyses and graphs dealing with the AQY, the site with the negative change in DIC was excluded.The mean AQY was 0.08 (from 0.02 to 0.22) mmol DIC mol photon 1 , and was the highest for a lake with a relatively low WRT and colonized by floating macrophytes (site 2, WRT = 0.12 years), and the lowest for a lake with a moderate WRT (site 11, WRT = 1.1 year; Table S1 in Supporting Information S1).

Relationships Between AQY, DOC Concentration, DOM Quality, and Environmental Variables
AQY increased significantly with ln(DOC) and DOM aromaticity (ln(α 420 ), SUVA 254 , logit(%C 3 )) and decreased significantly with the abundance of low molecular weight and in situ-produced compounds (ln(Sr), %C 5 ; Figure 1, Figure S5 in Supporting Information S1).Two multivariate models that explain AQY with DOM quality and environmental variables were obtained with the stepwise function after selecting the variables according to their correlation coefficients (model 1, Table S4 in Supporting Information S1) and automatically with the function vifstep (model 2, Table S4 in Supporting Information S1).In both models, after DOC and DOM quality (α 420 , Global Biogeochemical Cycles 10.1029/2023GB007989 SUVA 254 ), pH was the predictor that contributed the most to the model R 2 .In both cases, DOM quality parameters selected are related with color and aromaticity and show a positive relationship with AQY, and pH showed a negative relationship with AQY (Table S4 in Supporting Information S1).At the individual compound level, as assessed by mass spectrometry, aromatic and unsaturated molecules with high O:C ratios were positively correlated with AQY, while aliphatic and unsaturated molecules with low O:C ratios were negatively correlated (Figure 2).AQY also decreased significantly with increasing WRT (Figure 3).

Change in DOM Absorbance and Fluorescence Properties With Irradiation and WRT
The indices related to the degree of aromaticity of DOM (i.e., SUVA 254 , α 420 , %C 3 ) significantly decreased after irradiation (p < 0.01, Wilcoxon test) and with increasing WRT, while the indices related to DOM low molecular weight and in situ-produced compounds (i.e., Sr, %C 5 ) significantly increased after irradiation (p < 0.0001, Wilcoxon test) and with increasing WRT (Figure 4, Figure S6 and Table S2 in Supporting Information S1).The exponential model between %C 5 and WRT was preferred over the linear model because it had a lower AIC.The  S3 in Supporting Information S1 whereas the R 2 and p value are given in each panel.

Global Biogeochemical Cycles
10.1029/2023GB007989 R 2 and p-value of the exponential model were obtained by comparing observed versus predicted values (Figure 4f).

Changes in DOM Molecular Composition With Irradiation and WRT Assessed by Mass Spectrometry
As reported previously in Groeneveld et al. (2020) and Kellerman et al. (2014), the abundance of molecules with H:C < 1 decreased with WRT while molecules with H:C > 1 increased significantly (Figure 5a).Our study additionally shows that for inland waters with WRT < 3 years (n = 24, Figure 5b), the abundance of molecules with H:C < 1 also generally decreased with irradiation with the exception of molecules with intermediate O:C ratios.In addition, molecules with H:C > 1 increased in relative abundance after irradiation except for low O:C unsaturated molecules.However, no clear pattern was visible for inland waters with WRT > 3 years (n = 6, Figure 5c).Only one out of the three replicates measured after irradiation was used to calculate the change in molecular abundance in Figures 5b and 5c, but using the two other replicates for the calculations returned very similar figures (Figures S8 and S9 in Supporting Information S1).Furthermore, the abundance of photoproduced and photolabile molecules decreased significantly with WRT (Pearson correlation coefficient r = 0.65, p value < 0.0001, and r = 0.64, p < 0.001, for photoproduced and photolabile molecules, respectively; Figure 6).The average mass (±SD) of photo-labile molecules (m/z = 453 ± 18) was higher than that of photoproduced molecules (424 ± 13) (Table S5 in Supporting S1).The average mass of all molecules decreased significantly after irradiation (399 ± 12 vs.396 ± 13 before and after irradiation, respectively; p < 0.01, Wilcoxon test).

Photoreactivity Increases With Increasing Terrestrially Derived DOM and Decreases Along the Aquatic Continuum
Our results show that AQY increases with DOM aromaticity (Figures 1 and  2), and decreases with increasing WRT (Figure 3), as hypothesized.This means that for the same amount of photons absorbed, a higher quantity of CO 2 is likely to be produced in waters with low WRT and high DOM aromaticity.Since the majority of the photochemically active radiation is generally absorbed in the upper decimeters of boreal inland waters (Koehler et al., 2014), the total quantity of photons absorbed can be assumed to be similar for different boreal water bodies (i.e., whether clear or brown water).Our results consequently indicate that photoinduced CO 2 emissions are likely to increase with increasing terrestrial inputs and gradually decrease over the course of the aquatic continuum, with a gradual loss in aromaticity.Several studies have found an overall decrease in DOM degradation rate or turnover with increasing WRT (Catalán et al., 2016;Evans et al., 2017) and our study suggests that photomineralization may contribute to this pattern.The decrease in photoreactivity along the continuum is consistent with earlier studies that have shown higher mineralization rates in headwater streams and wetlands in comparison to lakes in Arctic (Cory et al., 2014) and boreal systems (Lapierre & del Giorgio, 2014).This is, however, the first time that a gradual decrease in AQY is demonstrated along a broad aquatic continuum from the interface with the terrestrial biome in peatlands to lakes with WRT of years to decades (Figure 3).Additionally, our study suggests that browning, resulting from increased terrestrial inputs in boreal systems, is likely to increase photomineralization rates, although it may not increase the proportion of  photoinduced CO 2 emissions in comparison to CO 2 emissions coming from respiration and DIC lateral input (Allesson et al., 2021).
We hypothesized that AQY gradually decreases with WRT because photoreactive terrestrially derived compounds decrease along the aquatic continuum (Evans et al., 2017;Kellerman et al., 2014;Weyhenmeyer et al., 2012).AQY was indeed positively correlated to molecules that decreased in abundance after irradiation (aromatic and unsaturated molecules with high O:C ratios; Figures 2 and 5b) and with increasing WRT (Figures 2  and 5a).In addition, we found a negative correlation between the abundance of photolabile and photoproduced molecules and WRT (Figure 6) that supports that photoreactive compounds are progressively lost along the WRT gradient.Recent research indicates that a large portion of high molecular weight, colored, terrestrial DOM is not ionized, and thus overlooked in the ESI-mass spectrometry analysis (Hawkes et al., 2019;Patriarca et al., 2020).Since these supposedly photoreactive compounds are likely to be more present in the low WRT sites, the decrease of photolabile and photoproduced molecules with WRT is likely to be much more pronounced than shown in this ) (e) and %C 5 (f) after irradiation, and correlation with ln WRT + 0.001.All linear regressions were performed on the variable measured before irradiation.The change between the start (start of the arrow) and the end (head of the arrow) of the irradiation was significant for all variables ( p < 0.01, Wilcoxon test; Table S2 in Supporting Information S1).Red arrows depict an increase in the variable with irradiation, while blue arrows depict a decrease.The gray area represents the 95% confidence interval for the linear models.The equations are given in Table S3 in Supporting Information S1, whereas the R 2 and p values are given in each panel for the linear models.

Global Biogeochemical Cycles
10.1029/2023GB007989 study (Figure 6).In addition, in more aromatic samples that absorbed more photons, the photon irradiance decreased more strongly along the pathlength.As a consequence, the entire sample was not exposed to the same irradiation, the surface of the sample being exposed to surface irradiation, while the bottom of the sample was exposed to a irradiation.We can thus assume that the number of photolabile and photoproduced molecules in the samples with short residence times would have been higher if the entire sample had been exposed to surface irradiation.This is also expected to accentuate the decrease of photolabile and photoproduced molecules with WRT shown in this study (Figure 6).Similarly, the change in DOM composition with WRT corresponded to the change we found due to irradiation (Figures 4 and 5), and some of the effect of irradiation was attenuated toward   S1 in Supporting Information S1).Sites 21, 25, 7, 5, 6, and 1 on the right correspond to lakes with WRT > 3 years.

Global Biogeochemical Cycles
10.1029/2023GB007989 longer WRT (Figure 5c).These results show that a large share of the DOM characteristics that are preferentially lost with irradiation also decrease along the WRT gradient.Hence, our results support our hypothesis that photoreactivity declines with increasing WRT, and that the decrease in AQY can be attributed to a loss in photoreactive terrestrially derived DOM.It is difficult to specifically attribute the decrease in photoreactivity to a decreasing terrestrial contribution to DOM or to attenuation of photoreactivity of DOM per se, since both occur concomitantly.Our results however directly show that the decrease in photoreactive compounds plays a significant role in the decrease in DOM photoreactivity along the aquatic continuum.

Comparison of the Drivers of Photoreactivity
WRT only explained a limited part of AQY variability in comparison to SUVA 254 and α 420 as shown when comparing the different univariate linear models (R 2 = 0.20, 0.33, and 0.42 for WRT, SUVA 254 and α 420 , respectively, Figures 1 and 3) and when comparing the partial variance explained by the different predictors of the multivariate models (Table S4 in Supporting Information S1).Koehler et al. (2016) also found strong correlations between AQY, SUVA 254 and α 420 (R 2 between 0.26 and 0.64 for both SUVA 254 and α 420 ).WRT may only partially reflect DOM photoreactivity because it does not account for new DOM inputs or for the "pause" or "bumps" in DOM processing along the aquatic continuum (Casas-Ruiz et al., 2020;Kayler et al., 2019;Kothawala et al., 2020).Riparian terrestrial inputs, light exposure and quenching, and DOM transformation by processes other than photodegradation (e.g., flocculation, adsorption and biodegradation) may indeed vary noncontinuously along the aquatic continuum.
Extrinsic factors and inorganic variables (e.g., iron, pH, nutrients) that also affect the sensitivity of DOM to photodegradation (Berggren et al., 2022;Gao & Zepp, 1998;Pace et al., 2012;Panneer Selvam et al., 2019) may explain why some studies did not find a correlation between DOM aromaticity and the AQY or photomineralization per unit of energy absorbed (Cory et al., 2013;Panneer Selvam et al., 2019).In our study, pH was negatively correlated with DOM aromaticity (Pearson r = 0.54 for α 420 ).Although the effect of pH was lower than that of DOM quality (α 420 or SUVA 254 ), a low pH had a positive effect on photoreactivity, that could not be directly attributed to DOM quality (Table S4 in Supporting Information S1).This is possibly because a low pH can enhance DOM photoreactivity by stabilizing double bonds or by interacting with iron (Panneer Selvam et al., 2019).Our results thus indicate that pH also drives DOM photoreactivity in line with previous studies (Gao & Zepp, 1998;Pace et al., 2012;Panneer Selvam et al., 2019) but to a lesser extent than DOM aromaticity.
The range of the AQY values was similar to the range of values reported in the literature (Table S6 in Supporting Information S1), but the site with the highest AQY (site 2, Table S1 in Supporting Information S1), colonized by floating macrophytes, consistently fell out of the range of the predictions of AQY (outlier in Figures 1 and 3 and Figure S5 in Supporting Information S1).It is well known that macrophytes release DOM leachates that can be highly photoreactive (Anesio et al., 1999;Vachon et al., 2016;Wetzel et al., 1995) and the commonly used indicators to describe DOM quality might not reflect the high photoreactivity of DOM originating from macrophytes.Additionally, we want to highlight that the pH decreased by 1.4 in the samples from this site during incubation.A low pH could increase the photoreactivity of the samples, since pH is known to affect photoreactivity (Panneer Selvam et al., 2019).This change could be caused by contaminated vials from acid washing.However, a DIC increase of ca. 3 mg L 1 in those poorly buffered samples (initial DIC of 1 mg L 1 ) also leads to a significant decrease in pH (ca.0.7, assuming a constant alkalinity over the incubation (Ouisse et al., 2014) using the initial pH of 5.6 and the AquaEnv function (Hofmann et al., 2010)).

Caveats of Photodegradation Incubations
Several aspects are likely to affect the AQY values in our study, and accordingly, a recent study has evidenced that there can be a high variability in AQY measurements between different laboratories (Koehler et al., 2022).In our study, the irradiance of the lamp used in the solar incubator varied with time and irradiance varied between the different vessel positions.First, I λ differed significantly according to the position of the vessels within the solar incubator (e.g., ranging between 1.07 and 1.65 J m 2 s at λ = 350 nm).The mean of the irradiance measured at the different vessel positions was used for the calculations and the replicates were placed randomly within the solar incubator.This likely led to an overestimation or underestimation of the AQY for some replicates, but we can consider the mean AQY of the three replicates robust.Second, the lamp irradiance was not measured during the experiment in 2016 but inferred from the most recent measurements made in November 2014 (irradiance dose Global Biogeochemical Cycles 10.1029/2023GB007989 of 575 W m 2 , Figure S1 in Supporting Information S1).The irradiance of the solar incubator was measured at six occasions during 2014, including before and after changing the lamp, and varied significantly with values ranging between 552 and 668 W m 2 (coefficient of variation of 10%).Furthermore, the samples still contained microorganisms because they were filtered at 0.7 μm before incubation.However, under such a strong light and during a short interval of 24 hr, microbial consumption of DOC and production of DIC in the samples was likely negligible, also considering that filtration prior to incubation substantially lowered microbial abundance.Indeed, a parallel study on the same water samples measured a biological DOC consumption after 24 hr under dark conditions of 0.10 ± 0.12 mgC L 1 (average for all sites pooled ± SD).Developments should be continued in order to obtain more reproducible AQY measurements.However, uncertainties of absolute AQY values do not affect the relative comparison of AQY between the different sites, and thus the findings of this study.

The Role of Photodegradation in Shaping DOM Composition Along the Aquatic Continuum
Our results show that changes in both DOM optical properties and molecular composition due to photodegradation were similar to changes along the WRT gradient, as hypothesized (Figures 4 and 5).This indicates that photodegradation partially shapes the DOM molecular composition of inland waters.Changes in DOM composition following irradiation (Gonsior et al., 2013;Hansen et al., 2016;Stubbins et al., 2010) and along the WRT gradient (Kellerman et al., 2014;Kothawala et al., 2014) have already been assessed separately.However, our study is the first to concomitantly measure how DOM composition changes via photodegradation and along a WRT, and demonstrate that they act similarly on DOM.Irradiation caused an overall loss of more unsaturated and aromatic molecules and an increase in more aliphatic and low molecular weight molecules (Figures 4 and 5, Table S5 in Supporting Information S1), in agreement with previous studies at bulk (Hansen et al., 2016;Helms et al., 2014;Murphy et al., 2018) and individual compound levels (Gonsior et al., 2013;Stubbins et al., 2010).Fluorescence and absorbance analyses showed that DOM aromaticity decreased and in situ produced compounds increased with increasing WRT in the same direction as seen after irradiation (Figure 4) and in agreement with previous studies (Cory et al., 2007;Kothawala et al., 2014).Mass spectrometry results also indicated that changes in DOM composition after irradiation were generally similar to the transformation of DOM with increasing WRT, but only for inland waters with WRT < 3 years (Figure 5a vs. 5b).We found, as in Groeneveld et al. (2020) and Kellerman et al. (2014), an overall decrease in the abundance of more unsaturated molecules (H:C < 1) and an increase in more aliphatic molecules (H:C > 1) with WRT.For inland waters with WRT < 3 years, irradiation affected DOM similarly with the exception of more unsaturated molecules with an intermediate degree of oxidation (H:C < 1 and 0.35 < O:C < 0.60) that increased after irradiation.Additionally, while they generally increased in relative abundance with increasing WRT, more aliphatic molecules with a low degree of oxidation (1 < H:C < 1.5 and O:C < 0.4) decreased after irradiation.This decrease was clearly visible in previous studies (Gonsior et al., 2013;Stubbins et al., 2010) and nuances the view that more aliphatic molecules (H:C > 1) always tend to be photoresistant and/or photoproduced while more unsaturated molecules (H:C < 1) are photolabile.Most importantly, this discrepancy between changes in composition with irradiation and with WRT for some DOM pools suggests that other processes obscure the effect of photodegradation.For example, the aliphatic DOM pool could be replenished by DOM derived from in situ production (Patriarca et al., 2021), while the aromatic pool with an intermediate degree of oxidation could be lost by adsorption (Groeneveld et al., 2020).For inland waters with WRT > 3 years (n = 6 sites), a lower number of molecules were identified as photoproduced and/or photolabile (Figure 6).Moreover, in waters with WRT > 3 years, there was no clear pattern in how different types of molecules changed with irradiation (Figure 5c).Hence, at long WRT, the change in DOM composition was not clearly attributable to light exposure.A similar threshold was found by Evans et al. (2017), where light-driven removal of DOM processes appeared negligible at WRT > 1.4 years.This suggests that the importance of photodegradation for shaping DOM composition is the highest for waters situated at the beginning of the aquatic continuum.This is also consistent with our finding that photoreactivity (i.e., the AQY) decreases with increasing WRT and reinforces the view that DOM composition and DOM photodegradation are tightly linked (Kellerman et al., 2015;Kothawala et al., 2020).Our application of radiation experiments along a natural WRT gradient suggests that for inland waters with low WRT, photomineralization rates are likely to be high and photodegradation is an ecologically relevant process that shapes DOM composition.

Conclusions
Our study directly shows for the first time that the observed compositional changes of DOM during sunlight exposure resemble the DOM compositional changes along a natural boreal WRT gradient, with the aromatic terrestrial-derived DOM being highly photoreactive.As a result, along a continuum where the ratio of terrestrial to in situ-produced DOM decreases and the time of DOM photoexposure increases, photoreactivity decreases concomitantly.Hence, we show, for the first time, that photoreactivity can be quantitatively related to the WRT of inland waters.The complete photochemical mineralization may account for a limited fraction of the total loss of DOC in most water bodies (Allesson et al., 2021;Dempsey et al., 2020;Maisonneuve et al., 2022), estimated to correspond to about 10% of the total emission of CO 2 from lakes in the region studied here (Koehler et al., 2014).However, we demonstrate that photochemical reactions are likely important in shaping the composition of DOM in inland waters with low WRT.

Figure 2 .
Figure 2. Correlation between the wavelength integrated apparent quantum yield (AQY) (mmol dissolved inorganic carbon [DIC] mol photon 1 ) and the abundance of individual molecules before irradiation.Each molecule is represented by a dot according to its H:C versus O:C ratio.The color scale indicates the significant Spearman correlations r (|r| > 0.368 for p value < 0.05 and 29 sites) of AQY with the abundance (i.e., the relative intensity) of individual molecules (n = 529).Molecules represented by red dots are positively correlated with AQY while blue dots are negatively correlated.Several lines are drawn to define categories of molecules based on their H:C, O:C and AI mod according toKoch and Dittmar (2006).

Figure 3 .
Figure3.Relationship between the wavelength integrated apparent quantum yield (AQY) and ln WRT + 0.001.The solid line represents the linear regression with all sites included (R 2 = 0.20, p value < 0.05, slope = 0.006 intercept = 0.065).The gray area represents the 95% confidence interval associated with the solid line.The dashed line represents the linear regression excluding the site with the highest AQY (site 2) (R 2 = 0.35, p value < 0.001, slope = 0.006 intercept = 0.060).

Figure 4 .
Figure 4. Changes in dissolved organic carbon (DOC) concentration (a) and dissolved organic matter quality indices, SUVA 254 (b), ln(α 420 ) (c), ln(Sr) (d), logit (%C 3 ) (e) and %C 5 (f) after irradiation, and correlation with ln WRT + 0.001.All linear regressions were performed on the variable measured before irradiation.The change between the start (start of the arrow) and the end (head of the arrow) of the irradiation was significant for all variables ( p < 0.01, Wilcoxon test; TableS2in Supporting Information S1).Red arrows depict an increase in the variable with irradiation, while blue arrows depict a decrease.The gray area represents the 95% confidence interval for the linear models.The equations are given in TableS3in Supporting Information S1, whereas the R 2 and p values are given in each panel for the linear models.

Figure 5 .
Figure 5. Van Krevelen plots of the change in the abundance of individual molecules along the WRT gradient (a) and after irradiation for waters with WRT < 3 years (b) and WRT > 3 years (c).Each molecule is represented by a dot according to its H:C versus O:C ratio.Panel (a) originates from the same water samples presented in Groeneveld et al. (2020) but on different measurements taken on the sub-samples used for this experiment.In panel (a) the color scale indicates the significant Spearman correlations r (|r| > 0.362 for p value < 0.05 and 30 sites) of WRT with the abundance of individual molecules (n = 938) before irradiation.In panels (b) and (c), the color scale indicates the average of the change in the abundance of molecules for sites with WRT < 3 years (24 sites) (b) and for WRT > 3 years (6 sites) (c).Only the molecules with more than 10% of relative change in abundance are shown (i.e., |average change in abundance| > 0.1, n = 1,209 for WRT < 3 years and n = 1,292 for WRT > 3 years).Red dots indicate molecules increasing in relative abundance with increasing WRT or with irradiation, while blue dots indicate a decrease.Several lines are drawn to define categories of molecules according to their H:C, O:C and AI mod according to Koch and Dittmar (2006) and as detailed in panel (c).The changes in the abundance of molecules for each site separately are presented in Figure S7 in Supporting Information S1.

Figure 6 .
Figure6.Abundance of photolabile and photoproduced molecules (%) for each site ±SD.The sites are ordered by category of inland water and by increasing WRT, except for peats where WRT is zero (TableS1in Supporting Information S1).Sites 21, 25, 7, 5, 6, and 1 on the right correspond to lakes with WRT > 3 years.