Biogeochemical evolution of ponded meltwater in a High Arctic subglacial tunnel

.


Introduction
Subglacial environments currently cover approximately 10% of Earth's land surface and have occupied most of these areas 30 for thousands of years. The broadly inaccessible nature of these systems has left them vastly understudied relative to other terrestrial ecosystems. Yet, over the last two decades efforts to explore subglacial sediment, water, and ice have consistently characterized them as active microbial and chemical systems that are important to global-scale biogeochemical cycles (e.g. Boyd et al., 2010;Hawkings et al., 2014;Wadham et al., 2019;Kellerman et al., 2021).
to interpret observations in the context of residence times, water histories, or solute sources, and thus fully resolve the processes that occur in-situ or the rates at which they occur. In this study, we extend our understanding of subglacial biogeochemistry and microbiology to include observations on the evolution of meltwater after it overwinters at the endpoint of a remnant subglacial channel that extends 467 m beneath a glacier. The nature of this system provides us with a unique opportunity to 65 observe the biogeochemical signatures of a relatively isolated subglacial system with strong constraints on the potential solute sources, water histories, and subglacial residence times.
We compared the biogeochemistry and microbiology of water samples from a pond at the terminus of the channel to their water and solute sources: late season runoff (represented by ice samples collected from frozen sections of the channel floor) and basal solute (represented by basal ice exposed along the tunnel walls). We used water isotopes (δ 2 H-δ 18 O) and a 70 conservative geochemical tracer (Cl -) to quantify the extent to which the late-season water froze and evaporated in situ. Using these results, a geochemical freeze-fractionation model was developed to 1) identify solutes that appear to behave conservatively and those in which there is an apparent in-situ source or sink, 2) evaluate the extent to which basal solutes may have contributed to the waterbody, and 3) in combination with 16S rRNA gene amplicon sequencing, explore evidence for in situ microbial activity and nutrient cycling. Collectively, the biogeochemical and microbial datasets provide evidence of in 75 situ microbial activity and show that a distinct microbial community can develop in a subglacial waterbody and deplete reservoirs of the most labile nutrients within months.

Field Site
The Sverdrup Glacier is a 25-km long tidewater glacier that overrides metasedimentary rocks and gneiss bedrock (Harrison et 80 al., 2016) while draining a ~800 km 2 northwest sector of the Devon Ice Cap, Canadian Arctic ( Figure 1). Ice velocities along the Sverdrup Glacier are moderate, ranging from ~30 m yr -1 near the glacier headwall to ~75 m yr -1 near the terminus (Van Wychen et al., 2017;Cress and Wyness, 1961). While some areas of the glacier are probably frozen to the bed (Van Wychen et al., 2017), flow rates along much of the glacier are influenced by basal sliding or enhanced deformation of basal ice, suggesting that basal ice temperatures, in at least some areas of the bed, approach the pressure-melting point (Van Wychen et 85 al., 2017;Burgess et al., 2005). A two-fold increase in surface velocity of the Sverdrup Glacier first measured during the summer of 1961 (Cress and Wyness, 1961) suggests meltwater reaches the bed and significantly reduces friction.
Consistent with other observations across the Canadian high Arctic, net mass balance of the Sverdrup glacier basin is controlled almost entirely by surface melt (Koerner, 2005). Surface temperatures measured on the Sverdrup Glacier from 2005-2021 indicate a daily average of 4.5°C over the summer melt season, which extends from early June to late August. Most of the melt 90 generated on the Sverdrup Glacier drains ice-marginally, with the remainder draining via supraglacial streams that reach the bed through moulins near the marine-terminating glacier snout (Keeler, 1964;Koerner, 1961). Winter temperatures average -26°C, suggesting that penetration of the winter cold wave likely leaves ice frozen to the bed along the lateral margins, as is the https://doi.org/10.5194/tc-2022-240 Preprint. Discussion started: 21 February 2023 c Author(s) 2023. CC BY 4.0 License.  case for many polythermal glaciers (Bingham et al., 2006;Irvine-Fynn et al., 2011). The area explored in this study focuses on ponded water located at the endpoint of a remnant subglacial tunnel along the eastern margin, ~3 km from the glacier 100 terminus (Figure 1).

Field Methods
We visited the Sverdrup subglacial tunnel in May 2019 and mapped the system using a TruPulse®360 (Laser Tech (LTI), Colorado USA) rangefinder relative to a known reference point at the glacier margin for which coordinates were collected using a handheld GPS. The rangefinder device was mounted on a tripod with a known height, and the azimuth, inclination, 105 horizontal distance, and vertical distance were collected for a back-sight and fore-sight targets at intervals (of <30 m) along the channel. Absolute distances and azimuths were validated using a tape measure and protractor, respectively. Ice temperature of the channel wall was measured at ~50 m intervals along the survey track using a BIOS digital contact thermometer.
Samples were collected from the subglacial pond for water isotope, biogeochemical, and microbial analyses at four locations (S6-S9; Figure 2). The first location (S6) was at the pond edge and was comprised of wet ice, which was collected into a whirl-110 pak® bag using a flame-sterilized ice axe. Though this sample was ice, we consider it to be affiliated with the pond water since it was situated in a depression that would receive continual seepage of water from other areas of the pond, and where temperatures would be sufficiently cold to freeze it relatively fast and completely. The other three sample sites were collected at intervals (~ 25 m apart) into the pond. At each location, pond slush was scooped into a glass beaker that was acid-washed, rinsed with 18.2 MΩ cm -1 deionized water (DIW), and furnaced (450°C for 4 hrs). Pond ice and water were separated by 115 decanting the two fractions into separate whirl-pak® bags. Water samples were immediately frozen, and all samples were kept at -20°C until analysis.
Basal ice from the channel walls and ice from frozen sections of the channel floor were collected at five locations at 50-100 m intervals (S1-S5; Figure 2). Ice along the channel floor was generally clear and had the appearance of frozen meltwater, in contrast to the banded opaque glacier and debris-rich basal ice exposed in the tunnel walls. We expect that stagnant late-season 120 runoff occupied the subglacial channel floor at the end of the melt season, and while some of this water remained liquid near the endpoint (pond), it froze along the outer section of the channel at the end of the melt season, preserving its microbial and biogeochemical characteristics. Since freezing would have propagated from the surface to the bottom of the water column introducing a biogeochemical gradient in the ice, bulk samples were collected along the vertical profile of the channel ice at each site. Channel ice samples were collected into whirl-pak® bags using a flame-sterilized ice axe. Figure 2: Ice temperature (top) and relative elevation, extent, and slope of the Sverdrup subglacial channel including sample locations of basal ice and channel ice (S1-S5), pond edge ice (S6) and pond slush water and ice (S7-S9) (bottom). Note vertical exaggeration of 50:1.
 A 10 ml sample was collected in a 25 ml glass beaker for immediate measurement of pH using a Fisherbrand TM accumet TM liquid-filled pH/ATC electrode (13-620-530A). One ml was also collected for immediate measurement of electrical conductivity (Oakton®).
 A borosilicate glass filter tower (Fisherbrand TM ) and 0.7 µm 25 mm GF/F filter paper were used to filter aliquots of sample into two 15 ml centrifuge tubes (Cole-Palmer) that were frozen immediately for NH4 + and PO4 3analysis, one 140 20 ml amber glass EPA vial (Fisherbrand TM ), that was kept at 4°C with no headspace, was used for spectrofluorescence analysis, and one 40 ml amber glass EPA vial that was amended with 5 M HCl to pH = 2, and stored at 4°C until dissolved organic carbon (DOC) analysis.
 A 10 ml sterile BD Luer Lok syringe and 0.2 µm PES syringe filter (Whatman TM Puradisc) was used to filter aliquots of sample into a 15 ml centrifuge tube (Cole-Palmer), which was stored at 4°C until major ion analysis, a 2 ml 145 Eppendorf TM tube, which was stored at 4°C until Si analysis, and a 2 ml Fisherbrand TM cryogenic storage vial, which was frozen until water isotope analysis.
 For debris-rich (basal ice) samples, 10 ml of sample was collected in a 15 ml Falcon® tube, 5 ml of a cell-sediment separation detergent (Morono et al., 2013) was added and the mixture was vortexed for 1 minute, then centrifuged (500 xg) until the sediment separated. The supernatant (for basal ice samples) or 15 ml of water (for all other samples) 150 was transferred to a new 15 ml Falcon® tube, stained with SYBR TM Gold Nucleic Acid Gel Stain (Invitrogen), and then filtered using a borosilicate glass filter tower (Fisherbrand TM ) and 0.2 µm MF-Millipore TM filter membrane.
Filter papers were mounted on a microscope slide and 20 fields of view were counted at 100x magnification using a Nikon Eclipse 80i microscope.
 Remaining sample was filtered through Pall Supor ® 47 mm filter papers using a Thermo Scientific TM Nalgene 155 filtration kit. DNA was extracted from these filter papers using a FastDNA SPIN Kit for Soil (MP Biomedicals, Irvine, CA, USA) by loading the filter paper directly into the lysis tube then following manufacturer's instructions. The DNA extracts were quantified using a Qubit HS DNA kit (Invitrogen, Waltham, MA, USA). The 16S rRNA gene was amplified in triplicate from each sample with 30 cycles of PCR using modified universal primers (515F (Parada et al., 2016) and 806R (Apprill et al., 2015)). The triplicate reactions were pooled, and Illumina sequencing adapters 160 were added with an additional 8 cycles of PCR, in triplicate. The triplicate adapter reactions were pooled for each sample and purified using a Wizard PCR clean-up system (Promega, Madison, WI, USA) then sequenced at the University of Wisconsin Genomics Core Facility using Illumina MiSeq technology.

Sample measurements
Water isotopes were measured using a Los Gatos Research Liquid Water Isotope Analyzer (LWIA-45-EP), and calibrated with 165 USGS reference standards. Water isotopes are reported by reference to the Vienna standard mean ocean water (VSMOW), in δ notation (Pinti, 2011). Analytical reproducibility for δ 2 H and δ 18 O was 0.6‰ and 0.4‰, respectively. Samples were analyzed for major ions (F -, Cl -, Br -, NO3 -, SO4 2-, Na + , Mg 2+ , and Ca 2+ ) by ion chromatography using a Metrohm Compact IC Flex ion chromatograph equipped with a C4 cation column and an aSupp5 anion column. Determination of reactive phosphorus and ammonium were measured using principles of spectroscopy following methods outlined by 170 (Strickland and Parsons, 1972) and a Thermo Scientific TM Genesys TM 150 UV-Visible Spectrophotometer with a 10 cm path length quartz cuvette. Silica was also determined using this spectrometer, but with 1 cm path length disposable cuvettes, following methods for heteropoly blue (APHA et al., 2017). Precision and accuracy were better than 5% for all analyses. Nonpurgeable DOC was measured by high temperature combustion with a Shimadzu TOC-V (CPH) analyzer equipped with a high sensitivity platinum catalyst. We used fluorescence spectroscopy to characterize fluorescent dissolved organic matter (FDOM) 175 using a HORIBA Aqualog with a quartz cuvette (1 cm path length) and a xenon lamp as an excitation source. Excitation was measured at 5 nm intervals from 230 nm to 600 nm and emission was measured every 2.33 nm from 245 nm to 826 nm, using an integration time of 10 seconds to produce excitation emission matrices (EEMs) for samples and blanks.

Parallel factor analysis 180
We used parallel factor analysis (PARAFAC) to decompose the EEMs and identify/quantify FDOM characteristics (Murphy et al., 2013). PARAFAC was completed using the N-way and drEEM toolboxes in Matlab (2021), following methods described by Murphy et al (2013). EEMs were corrected for instrument bias, inner filter effects, and regions of scatter were excised after a DIW blank was subtracted from the measured sample. Components derived from PARAFAC modeling were compared to those identified in other studies using the online spectral library of auto-fluorescence by organic compounds in the 185 environment, OpenFluor (Murphy et al., 2014).

Bioinformatics
The 16S rRNA gene amplicon sequences were processed using mothur (Kozich et al., 2013) to filter reads that did not meet minimum quality control thresholds (maxambig=0, maxlength =315, maxhomop=8), join paired reads, cluster sequences into operational taxonomic units (OTUs) using 97% sequence identity, and assign taxonomic identification to OTUs using the 190 SILVA SSU database v138. A total of 2868 singleton OTUs, or those that appeared once across the entire dataset, were removed prior to further analysis. Diversity calculations were completed using the phyloseq package (v 1.28.0) in R (v 3.6.0).

Water isotope model
Using the principles of isotopic fractionation, a model was developed to estimate the isotopic composition of incremental ice, incremental vapor, and residual water as a hydraulically isolated waterbody progressively freezes and evaporates 195 (Supplementary Methods S1). The degree to which isotopes fractionate during freezing depends on the rate of freezing and can be described by Eq. (1): Where δ and δ are the isotopic composition of the ice and water, respectively, ε − is the isotopic difference between ice and water, and is the respective equilibrium fractionation factor. The degree to which isotopes fractionate during evaporation 200 (ε − ) depends on the temperature of the water during evaporation and can also be described by Eq. (1) using the isotopic composition of water and vapor instead of ice and water, respectively.
The isotope model equations are described in detail in Supplementary Methods S1. Fixed variables included: 1) the initial isotopic composition of the waterbody, which was set to the average isotopic composition of channel ice samples, 2) equilibrium values for evaporation, which were set to experimental values obtained for evaporation at 0°C (Majoube, 1971), 205 3) equilibrium values for freezing, which were set to the average of the isotopic fractionation between ice and water in pond slush, and 4) the step interval, which was set to simulate 0.1% of the waterbody freezing in each step of the model. We adjusted the rate of evaporation vs freezing, to optimize the model's fit with the observed δ 2 H-δ 18 O of channel ice and pond ice along the incremental ice line, and the pond water samples along the residual water line ( Figure S2). We then fixed the evaporation vs freezing rate and ran the model to estimate the stage of freezing (i.e. the fraction of residual water vs frozen and evaporated 210 water) at each sample site in the pond. To do so, the observed δ 2 H-δ 18 O of ice and water samples at each site in the pond were compared to the modeled incremental ice line and the modeled residual water line, respectively. The step with the most similar δ 2 H-δ 18 O was identified for each site and the corresponding stage of freezing was determined.

Clmodel
Cldoes not readily precipitate as salts, interact with rocks, or become assimilated in significant quantities by microorganisms 215 (Davis et al., 1998). Since there were no identifiable sources or sinks for Clin the system, we used the principles of solute fractionation to develop a model that estimates the [Cl -] in incremental ice and residual water as an isolated waterbody progressively freezes and evaporates. Clis largely rejected from the ice crystal lattice during freezing (Killawee et al., 1998;Clayton et al., 1990) and from water vapor during evaporation, and thus it concentrates in the residual water. Though effectively all non-volatile solutes are excluded from water vapor, the degree to which solutes are excluded from the ice is described by 220 the effective segregation coefficient (Keff) in Eq. (2): using Eq. (2), 3) the step interval, which was set to simulate 0.1% of the waterbody freezing in each step of the model, and 4) the rate of evaporation vs freezing, which was set to the ratio derived from isotope modeling. 230 Results from the Clmodel were used to estimate the stage of freezing (i.e. the fraction of residual water vs frozen and evaporated water) at each sample site in the pond ( Figure S2). To do so, the observed [Cl -] in ice and water at each pond sample site were compared to modeled [Cl -] in incremental ice and residual water, respectively. The step with the most similar [Cl -] was identified for each site and the corresponding stage of freezing was determined.

Geochemical model 235
Similar to Cl -, other dissolved impurities are excluded from water vapor during evaporation and the extent to which they are also rejected from the ice crystal lattice during freezing can be quantified using Equation 2 (Killawee et al., 1998;Clayton et al., 1990). We therefore used the same theory and equations as the Clmodel to produce a geochemical model that simulates the concentration of other solutes and impurities in residual water and incremental ice (Supplementary Methods S2; Figure   S2) at the stage of freezing affiliated with each sample site (determined from the Clmodel). We then compared measured 240 concentrations at each pond sampling site to modeled concentrations. Since the model only simulates the effects of freezing and evaporation, we interpreted [X]measured > [X]modeled as evidence for potential net sources of chemical species 'X' and [X]measured < [X]modeled as evidence for potential net sinks of chemical species 'X'.

Physical system 245
The ponded water from which samples were retrieved was located at the endpoint of a remnant subglacial tunnel that extended 467 m horizontally into the glacier at an angle of ~18 o relative to the direction of ice flow (NNW; Figure 1). The location of the ponded water was therefore ~150 m in perpendicular distance from the margin and ~79 m below the glacier surface ( Figure   1). Along its 467 m length, the channel floor dropped 5.3 m in elevation ( Figure 2) and terminated at an ice-wall that is located

Water isotopes 260
The isotopic composition of ice samples from the channel floor (S1-S5) lie below the local meteoric water line (LMWL) for the Sverdrup catchment (Copland et al., 2021), which is consistent with the isotopic composition of meltwater originating from local snow/glacier ice. Channel pond and basal ice samples fall above the LMWL and exhibit more negative δ 18 O and δ 2 H values relative to the channel ice. The difference between respective ice and water samples at sites 8 and 9 was 1.8-1.9 ‰ for δ 18 O and 10.9 ‰ for δ 2 H, which is consistent with equilibrium values measured in other slow-freezing environments (Table  265 S1) (Jouzel et al., 1999;Ferrick et al., 2002).
Our isotope modelling experiments indicate that a freeze:evaporation rate of ~40:1 was required to best fit the model to the observed water and ice data points (Figure 3a). This model also indicates that ~10% of the initial water remained residual at

Major ion chemistry 325
Channel ice samples from S1-S5 were generally dilute (x̄ = 0.91 mg L -1 ) but displayed considerable variability in the concentration of total measured solutes. Pond ice samples (at sites 8-9) had similar total measured solute concentrations (1.1 and 1.2 mg L -1 ) but coincident pond water samples had concentrations ~7-fold higher (7.0 mg L -1 and 7.2 mg L -1 , respectively).
Pond edge ice (site 6) had the highest total measured solute load of all channel samples (11.6 mg L -1 ). Basal ice had highly simulations that also incorporated basal solute contributions to the pond (comprising 5-15% of the initial solute load; Simulations 2 and 3) generally yielded more accurate concentrations of these solutes (Figure 5b).

FDOM components
PARAFAC modeling resolved two fluorescent components that comprised 92 % of the variability in the dataset ( Figure S5). Component 1 (C1) overlaps with ''humic-like'' fluorescence (Coble, 1996) and has been associated with fulvic acid DOM 340 fractions (Ohno and Bro, 2006). While this fluorescence feature is not consistently identified in glacier ice, similar fluorescence signatures are ubiquitous in a wide range of terrestrial (Ohno and Bro, 2006;Yamashita et al., 2011) and marine (Jørgensen et al., 2011;Retelletti Brogi et al., 2018;Walker et al., 2009) environments. Further, this fluorescence component has been identified in Antarctic watersheds without higher plants (Cory and McKnight, 2005;Barker et al., 2013) and has been correlated with microbial activity in ocean waters (Jørgensen et al., 2011) indicating that it may also be produced by microbial 345 reworking of organic matter (OM). Further, this fluorophore is persistent in many environments suggesting that, once formed, it may not be readily altered (Yamashita et al., 2010).
Component 2 (C2) overlaps with the fluorescence of amino acids, in particular as either free or bound tyrosine (Peak B, Coble, 1996). Similar FDOM is widely detected in glacial ice, including englacial and basal ice from Greenland, Antarctica, and across the Canadian Arctic from the supraglacial snowpack to ice dating back to the Last Glacial Maximum (D' Andrilli and 350 McConnell, 2021;D'Andrilli et al., 2017;Pautler et al., 2012;Dubnick et al., 2010Dubnick et al., , 2020Barker et al., 2006Barker et al., , 2009Fellman et al., 2010). Tyrosine-like fluorescence is commonly associated with OM of low molecular weight and aromaticity and chemical species that are readily degraded by microorganisms (Coble, 2014). Unlike C1, C2 is rapidly transformed in proglacial and downstream aquatic systems (Wu et al., 2003;Saadi et al., 2006;Fellman et al., 2008;Barker et al., 2013).

Microbiology
Cell concentrations in channel ice averaged 5.2 x 10 4 cells mL -1 while concentrations in the pond water were 6.3 -11.9 x 10 4 cells mL -1 , comparable to the range found in other subglacial waters (Christner et al., 2014;Gaidos et al., 2004;Mikucki et al., 2009). Over 85% of the 16S rRNA gene sequences in pond water samples were affiliated with one of six OTUs that fell in the class Gammaproteobacteria: Massilia (28-40% of total reads), two Polaromonas OTUs (25-27%), an unclassified 375 Oxalobacteraceae (14-18%), Rhodoferax (5-15%), and Undibacterium (3%) (Figure 6). While these OTUs (except the unclassified Oxalobacteraceae) were all detected in basal ice, they were more abundant in channel ice samples (including the unclassified Oxalobacteraceae), comprising 23-32% of the assemblage (Figure 6). Their higher relative abundance in channel ice suggests these organisms probably originated from late season runoff, rather than from subglacial sources. Chao1, Shannon's (H) and Simpsons (D) diversity indices all indicate taxonomic diversity was lower in pond water than in channel 380 ice or basal ice ( Figure 6).

Evolution of the subglacial tunnel and pond
Subglacial channels develop when inflowing waters have sufficient heat and pressure to melt ice at a rate that exceeds that of 390 creep closure from the overlying ice (Röthlisberger, 1972;Nye, 1976). These channels grow during spring-summer to deliver meltwater to the glacier terminus and are closed by ice creep as runoff rates recede in the fall. Closure rates are typically lower near the ice margin where overlying ice is thin and overburden pressure is low and are faster at internal locations where overlying ice is thick and overburden pressure is high (Röthlisberger, 1972). At the time of the field survey, the subglacial tunnel terminated at approximately the upper edge of a 'cliff', or rapid deepening along the glacier bed profile (Figure 1). 395 Rates of creep closure would be much higher at internal locations beyond this subglacial 'cliff' due to the ice overburden pressure ( Figure S3). Creep closure near the pond was likely responsible for squeezing the subglacial pond slush into a mound, resulting in the positive slope of the pond surface in this area at the end of winter ( Figure 2).
As ambient air temperatures fell towards the end of the melt season of 2018 (remaining below 0°C from September 2018 -June 2019; Figure S1) drainage into the subglacial channel from the ice margin would have ceased, leaving ponded water 400 along the channel floor in areas with low slope. The cold, dense ambient air drained into the tunnel, progressively freezing the stagnant water on the channel floor. Although the freezing process along the channel floor may have released some latent heat, additional heat source(s) would have been required to maintain liquid water and the near 0°C ice and air temperatures that were observed at the tunnel endpoint in May. The Sverdrup Glacier is marine-terminating ~3 km downstream of the subglacial tunnel and displays a retrograde subglacial slope (Paden et al., 2019), resulting in ice that is grounded below sea level across 405 most of the glacier's width near the subglacial tunnel ( Figure 1). This geometry allows heat from the ocean to more effectively warm areas of the glacier bed. Additional heat may also originate from friction near the glacier bed. Prior studies suggest regions of the Sverdrup Glacier have flow regimes that involve enhanced deformation of basal ice (Van Wychen et al., 2017), which produce frictional heat. Geothermal heat may also contribute to the subglacial heat flux but the regional geothermal heat flux is relatively low (65 +/-5 m W m -2 ; (Grasby et al., 2012)). 410

Concentration effects
The water isotope and Clmodels produced consistent estimates of the extent of freezing at the pond sampling sites ( Figure   3c). The portion of residual water increased with distance into the pond, from 6% remaining as residual water at the pond edge (S6) to 9% remaining as residual water at S9. Sites S6 and S7 are at lower elevations than S8 and S9 (Figure 2), and would have experienced colder overlying air in this depression. Freezing processes (and to a lesser degree, evaporation) were, 415 therefore, the dominant controls on the concentration of solutes in the pond, resulting in concentrations up to 7 times those observed in late-season runoff.
The temperature gradient along the downward-sloping subglacial tunnel (Figure 2) facilitates the drainage of cold, dense ambient air into the tunnel and drainage of warm buoyant air out of the tunnel. Regional air masses are cold and have low https://doi.org/10.5194/tc-2022-240 Preprint. Discussion started: 21 February 2023 c Author(s) 2023. CC BY 4.0 License. relative (65%) and specific (0.2 g kg -1 ) humidity during the winter (Vincent et al., 2007). The convection and warming of this 420 ambient air through the tunnel would therefore facilitate continual evaporation from the subglacial pond throughout the cold season. The tunnel showed evidence of evaporation; the roof/upper section of the walls were covered in a blanket of intricate ice crystals formed by condensation of water vapor as the buoyant/warm/moist air mass traveled out of the tunnel and cooled.
During evaporation, solutes are preferentially excluded from the vapor phase, so evaporation at the pond surface increased solute concentrations in the residual water. 425 The drainage of cold, dense air along the tunnel floor would also have frozen the late-season runoff pooled on the channel floor, from the surface to the bed, and from the channel entrance towards the endpoint. The pond was comprised of a slushy mixture of ice and water, indicating a slow freezing rate (Michel and Ramseier, 1971) near the tunnel endpoint, and isotope modeling suggests that freezing occurred at a rate ~40 times that of evaporation. Suspended and dissolved impurities and light water isotopes are preferentially rejected from ice crystals during the freezing process, resulting in residual water that becomes 430 increasingly concentrated in impurities (Clayton et al., 1990;Killawee et al., 1998) and isotopically light.

In-situ geochemical sources
Unlike the waters contained beneath many other polythermal glaciers, our geochemical modeling simulations suggest that the pond only received small (~5-15%) contributions of solutes from basal sources. As discussed above, the channel floor ice and pond water originated as late-season runoff that pooled along the channel floor in areas of low slope. Compared to geochemical 435 modeling that simulates late-season runoff as the sole solute and water source to the pond, pond water samples were enriched in total solutes, including SO4 2-, K + , Ca 2+ , Mg 2+ (Figure 5a) and Si. In glacial environments, these elements are commonly derived from rock-water reactions (Tranter et al., 2002) and are found at high concentrations in basal ice (Dubnick et al., 2020) and subglacial discharge (Wadham et al., 2010;Li et al., 2022). Importantly, these solutes were enriched in basal ice samples in this study (Figure 5b), suggesting they are liberated from basal processes. 440 Basal solute contributions to the pond most likely originated as seepage from basal ice along the pond perimeter. The interstitial water content of basal ice scales with temperature, so the relatively warm ice in this area (Figure 2) could promote drainage of solute-rich water into the pond. Though the pond could also receive basal solutes from a distributed drainage system at the icebed interface, hydraulic modeling suggests that if waters occupied a distributed drainage system, they were more likely to drain away from the pond than towards the pond ( Figure S3). The pond could also acquire basal solutes by rock-water 445 interactions at its base. However, this channel receives high volumes of seasonal meltwater over consecutive years, so reactive components of the bedrock would be depleted (including those with slow reaction kinetics).
Simulations that incorporated basal solutes improved the accuracy of modeled solute concentrations, though the accuracy of the models were inconsistent among geochemically-relevant compounds (Figure 5a). The solute composition of basal ice, subglacial water, and species liberated from in situ rock weathering, can vary dramatically within a subglacial catchment 450 (Dubnick et al., 2020;Yde et al., 2010;Tranter et al., 2002) due to variation in the composition of the underlying bedrock, redox conditions, ice/water temperatures, and freeze-thaw histories. The five basal ice samples explored in this study also showed high variability in solute chemistry (Figure 4), confirming that potential basal solute sources in the subglacial tunnel are heterogeneous and that our model may not accurately represent the precise composition of the basal solutes entering the pond. 455

Microbiology and biogeochemical nutrient sinks
Sequencing of 16S rRNA gene amplicons indicates that the microbial assemblage in the subglacial pond was dominated by organisms in the order Burkholderiales, which are commonly found in subglacial systems (e.g. Cheng and Foght, 2007;Achberger et al., 2016;Dubnick et al., 2020). More specifically, the most dominant OTUs, Massilia, Polaromonas, Rhodoferax and Undibacterium ( Figure 6) are all related to microbes detected at high abundance in other glacial environments 460 (Foght et al., 2004;Lanoil et al., 2009;Darcy et al., 2011;Mitchell et al., 2013;Perini et al., 2019;Dubnick et al., 2020;Dunham et al., 2021) including supraglacial, proglacial, and/or subglacial aquatic systems worldwide. Moreover, these OTUs are related to psychrophilic or psychrotolerant species (e.g. Foght et al., 2004;Darcy et al., 2011;Wang et al., 2018;Perini et al., 2019), and many of these taxa have been shown to have a range of structural and functional adaptations for survival in cold temperatures (Margesin and Miteva, 2011). 465 All six dominant OTUs in the pond were closely related (>98%) to globally distributed environmental sequences and cultured representatives, suggesting these organisms are habitat generalists with broad environmental tolerances. For example, Polaromonas are dominant in polar and high-elevation environments (Darcy et al., 2011), are thought to rapidly evolve to new environments through extensive horizontal gene transfer (Yagi et al., 2009), and are considered metabolically diverse "opportunitrophs" (Meyer et al., 2004;Polz et al., 2006). The meltwaters that drained into the subglacial system towards the 470 end of the melt season originated from cryoconite holes, supraglacial streams, englacial ice, precipitation, and extraglacial aquatic or terrestrial sources. These environments are exposed to distinct nutrient pools, redox conditions, and, in some cases, high levels of solar radiation and/or warmer temperatures. The shift in environmental conditions to those in the cold, dark, oligotrophic subglacial pond may have decimated populations not capable of survival in this very different system, selecting for generalist organisms that are better adapted to those conditions (Xu et al., 2021). 475 Consistent with the notion of being habitat generalists, the six dominant OTUs in the pond water are closely related to genera with extreme metabolic versatility. For example, despite sharing nearly identical 16S rRNA genes, Polaromonas strains from different glaciers can have very different phenotypes (Gawor et al., 2016), with some strains variably showing evidence of an ability to oxidize H2 (Sizova and Panikov, 2007;Dunham et al., 2021), arsenite (Osborne et al., 2010), and/or organic matter (Jeon et al., 2003;Mattes et al., 2008;Osborne et al., 2010). Furthermore, several cold-tolerant species of Rhodoferax (e.g. R. 480 antarcticus and R. fermentans) have been shown to be facultative anoxygenic phototrophs that can grow photoheterotrophically (using a variety of organic and fatty acids or glucose) and photoautotrophically (using CO2) or that can grow via aerobic respiration in dark environments (Madigan et al., 2000). R. fermentans has also been shown to be capable of nitrate reduction (Hougardy and Klemme, 1995) and a close relative of R. ferrireducens from glacial sediments in Iceland has been shown to fix CO2 with energy derived from the H2-ferric iron redox couple (Dunham et al., 2021 in the pond may be due to its metabolic flexibility which allows it to compete for nutrients despite their changing availabilities as they are transported from the supraglacial/extraglacial to the subglacial system, and in the subglacial pond as it becomes increasingly cold, concentrated in solutes, and depleted in labile nutrients over the winter. Our microbial and biogeochemical datasets both suggest the pond functioned as a hotspot for C-cycling over winter. We measured ~43% less DOC than our modeling predicted and a more depleted reservoir of the labile tyrosine-like (C2) 490 fluorophore, which is often rapidly transformed in proglacial and downstream aquatic systems (Wu et al., 2003;Saadi et al., 2006;Barker et al., 2013;Fellman et al., 2010) (Figure 5). Previous studies of Polaromonas, Rhodoferax, and Massilia strains, which comprised ~70% of the pond microbial assemblage, suggest they are fuel their metabolisms by coupling the oxidation of organic matter with reduction of O2 (Hougardy and Klemme, 1995;Jeon et al., 2003;Mattes et al., 2008;Osborne et al., 2010), and thus could be responsible for depletion of DOM. Assuming 70% of the cells in the pond (x̄ = 7.5 x 10 4 cells mL -1 ) 495 were active heterotrophs, and that the size of this population was relatively stable throughout the ~240 days over winter, the observed DOC deficit (3.9 µM) equates to an average rate of C-remineralization of 0.2 fmol C cell -1 day -1 . This rate is comparable to metabolic rates sufficient for microbial growth at ~0°C and is orders of magnitude higher than metabolic rates required for cellular maintenance or survival at 0°C (Price and Sowers, 2004). Though this is only an estimate for Cremineralization in the pond, rates could be higher if less than 70% of the microbial assemblage was engaged in heterotrophic 500 energy metabolisms, or if additional OC pools were also degraded, such as particulate OM, basally-derived OM, or autochthonous OM (our calculation above only accounts for DOC supplied in the late-season runoff).
Microbial activity in the subglacial pond could also explain the depleted reservoir of other inorganic nutrients, including PO4 3and NH4 + ( Figure 5). Past work has shown that heterotrophic microbial biomass is C-poor yet P-and N-rich (Makino et al., 2003;Godwin and Cotner, 2015) relative to many terrestrial DOM sources, and that microbial heterotrophic activity has been 505 linked to a simultaneous assimilation of mineral nutrients (Fenchel and Blackburn, 1979;Martinussen and Thingstad, 1987).
While PO4 3is considered to be a preferred and universal source of phosphorus to microbes (Björkman and Karl, 1994), N can be assimilated as NO3 -/NO2or as the preferred reduced state, NH4 + /NH3 (Paul and Clark, 1996;Nyyssönen et al., 2014).

Conclusions
This research documents the evolution of ponded meltwater and its resident microbial community at the endpoint of a 467 m long remnant subglacial channel through a Canadian Arctic winter. Solute concentrations in the pond were controlled by: 1) freezing processes, which functioned to cryo-concentrate solutes in the residual water by up to 7 times; 2) seepage of small 515 amounts of basal solutes (comprising <15% of the total solutes) into the pond; and 3) microbial activity, which functioned to deplete the pond's reservoir of the most labile and biogeochemically-relevant compounds, including NH4 + , PO4 3-, and DOM. Sequencing of the 16S rRNA gene revealed decreasing taxonomic diversity among microbial communities with distance into the channel. Six OTUs dominated the microbial community in the pond. These microorganisms likely originated from the extraglacial or supraglacial (rather than subglacial) environment and were related to taxa that are 520 psychrophilic/psychrotolerant, exhibit extreme metabolic diversity, and have broad habitat ranges. Collectively, our findings suggest that generalist microorganisms from the extraglacial or supraglacial environments can become established in subglacial aquatic systems and deplete reservoirs of nutrients over a period of months. The inferred generalist lifestyle of these microorganisms may help them survive the extreme selection pressures imposed by the environment, allowing for not only their persistence but activity. These findings extend our understanding of the microbiology and biogeochemistry of subglacial 525 ecosystems.

Data Availability
Sequence data has been deposited in NCBI SRA under BioProject ID PRJNA907039 and geochemical data are available via Zenodo DOI 10.5281/zenodo.7384156.

Author contribution 530
AJD designed and carried out the fieldwork, laboratory work, data analysis, developed the model codes and performed the simulations. RLS extracted DNA from field samples and conducted the bioinformatics, BDD prepared temperature data record, DB prepared radar and laser altimeter transect data, and CD conducted laboratory DOC analysis. AJD, RLS and BDD prepared the manuscript with contribution from all co-authors. MJS, MLS, ESB funded and supervised the work.

Competing Interests 535
The authors declare that they have no conflict of interest.