VOC emissions were measured near the 2020 Deep Purple¹ basecamp (∼ 61° 05′ N, 46° 50′ W), which was established on the southern Greenland ice sheet (Figure 1A), between 3 July and 18 July 2020, to study a variety of processes related to biological darkening of the Greenland ice sheet. VOC sampling sites were located approximately 700 m from the PROMICE² (Programme for Monitoring of the Greenland Ice Sheet) weather station QAS_M, from which we retrieved hourly data for atmospheric temperature, relative humidity, and incoming shortwave radiation (W m^–2) during sample collection. Melting of the snowpack was at an advanced stage during early July, with only a thin snow cover remaining over some areas. The southern Greenland ice sheet is experiencing increasing rainfall (Niwano et al., 2021) and the 2020 basecamp area received precipitation in the form of rain on 7 days of the campaign period. At the basecamp site, the average air temperature was 3.3 °C, the average incoming shortwave radiation was 247 W m^–2, and the average number of daylight hours was 18.6.

VOC emissions from red snow habitats (Figure 1C) were collected on 3 July and 5 July, after which VOC emissions from the only remaining snow patches that covered crevasses and moulins could not be collected safely. In contrast, microbial colonization of the bare ice surface became more widely distributed as the melt season progressed. VOC emissions from three distinct habitats, namely, “dark ice” (bare ice surfaces with visible particles or algal bloom, Figure 1D), “clean ice” (bare ice surfaces without macroscopically visible particles or algal bloom), and cryoconite holes (Figure 1B), were collected between 6 July 2020 and 15 July 2020.

VOC emissions were collected using a push–pull enclosure setup (Tholl et al., 2006; Valolahti et al., 2015) with transparent polycarbonate chambers (thickness 1.5 mm, 220 mm × 220 mm, height 200 mm; Vink Finland, Kerava, Finland). Metal chamber bases were placed on the sampling sites [red snow (n = 2), bare ice (n = 31), or cryoconite holes (n = 4)] and left to melt into the ice for approximately half an hour. In the case of cryoconite holes, larger holes were selected so that the chamber bases could be placed within the water column to avoid collection of emissions from the surrounding bare ice surface. The grooves in the chamber bases were filled with water before placing the chamber enclosures over the bases, creating an airtight seal. Chambers were flushed at a flow rate of 1,000 mL min^–1 for 10 min using battery-operated pumps connected via Teflon tubing. Incoming air was purified during flow through a PTFE filter and an activated charcoal filter to remove particles and VOCs, as well as by flow through a potassium iodide ozone scrubber (Ortega and Helmig, 2008). Following the 10-min flushing, a stainless steel sorbent tube containing 150 mg Tenax TA and 200 mg Carbograph 1TD (Markes International, C2-AXXX-5032) was inserted between the chamber and the outflow line. The outflow rate was set to 200 mL min^–1 for the duration of the sample collection (between 0.5 and 5.5 h), while the inflow rate was set to 210 mL min^–1 to maintain a slight overpressure in the chamber and prevent ambient air from entering the chamber. The chambers were equipped with a fan to ensure a well-mixed headspace. Temperature and relative humidity in the chamber headspace were logged once per minute using a shaded iButton sensor (Hygrochron DS 1923-F5, Maxim Integrated Products Inc., CA, United States).

After sample collection, cartridges were sealed with Teflon-coated brass caps and stored refrigerated, when possible, until analysis in Denmark. The chambers were removed, and, for bare ice samples, the top 2 cm of the ice within the chamber base was scraped off with a preconditioned ice axe and collected in sterile Whirl-Pak bags. Once melted, samples were homogenized and 2 mL of each sample was transferred to a 2-mL Eppendorf tube. The samples were fixed with glutaraldehyde (2% final concentration) and stored in the dark at 4 °C until counting. Microscope counts of glacier ice algae and snow algae were performed on a Fuchs-Rosenthal hemocytometer (3.2 μL, Lancing, United Kingdom) using an inverted light microscope (Olympus CK2). A minimum of 100 cells were counted where possible, but counts of the lowest cell concentrations contained fewer cells.

Procedural blank VOC collections were performed in the laboratory in Denmark to account for VOCs released from the sampling materials and analytical systems. The sampling setup (chamber base and chamber enclosure, with fan and iButton sensor) was placed on top of a VOC-free polyethylene terephthalate film (prebaked at 120 °C for 1 h) in a refrigerated chamber (0 °C) that was equipped with adjustable lights (Valoya C65 with NS12 spectrum, Helsinki, Finland) to mimic natural sunlight. Blanks were collected for 1 (n = 6), 2 (n = 2), and 4 (n = 2) h to account for potential background compounds, as well as compounds that were released continuously over time, using VOC-free air (Parker ChromGas Zero Air Generator) for inflow. The limitations of the sampling setup used in this study, such as the lack of steady state in the beginning of the sampling period and the possibility of compounds adhering to the chamber walls, have previously been discussed by Faubert et al. (2012).

Samples were analyzed by gas chromatography–mass spectrometry (GC–MS) (7890A Series GC coupled with a 5975C inert MSD/DS Performance Turbo EI System, Agilent Technologies, Santa Clara, CA, United States). Sorbent tubes were thermally desorbed (TD-100xr, Markes International, Llantrisant, United Kingdom) at 250 °C for 10 min, cryo-focused at −5 °C, and injected onto an HP-5 capillary column (length 50 m, diameter 0.2 mm, and film thickness 0.33 mm) with helium as carrier gas (1.2 mL min^–1). During analysis, the column was kept at 40 °C for 3 min, then raised to 210 °C at a rate of 5 °C min^–1, and finally to 250 °C at a rate of 20 °C min^–1. Duplicate standard solutions (Supplementary Table 1) were injected into sorbent tubes and analyzed alongside samples at the start and end of each sample batch.

Chromatograms were analyzed using PARADISe (version 3.88) software (Johnsen et al., 2017). A total of 92 compounds were identified using pure standards where available (Supplementary Table 1), or were tentatively identified using the NIST Mass Spectral Library (version 2.2, 2014). Tentative identities were assigned if the NIST match factor (MF) was higher than 800 (the NIST Mass Spectrometry Data Center refers to MFs >800 as a good match) and the probability was higher than 30%. A tentative compound class was assigned (e.g., unknown alkane 1) for database hits that did not meet these conditions, but where at least two of the top three NIST matches had the same molecular formula. Compounds were classified as “oxygenated VOC” (e.g., unknown OVOC 1) when the top three NIST matches suggested different molecular formulae but all formulae contained oxygen, or as “other VOC” (e.g., unknown other 1) if they did not. Compounds were grouped into one of the following classes: alkanes, alkenes, terpenoids, oxygenated benzenoids, oxygenated VOCs, nitrogen-containing VOCs, and other VOCs. Concentrations were obtained by comparing peak areas between samples and standards, using the known mass of the standard injected. The closest structurally related standard (Supplementary Table 1) was used for quantification purposes for compounds that did not have an exact match with any of the standards injected.

Emission rates were calculated according to Faubert et al. (2012). In short, the VOC mass in the sorbent tube was divided by the sampled volume (a function of inflow rate and sampling time) to obtain the VOC concentrations in μg L^–1 air. Average concentrations in the procedural blanks were subtracted from those detected in the samples, taking into account the sampling time for compounds that increased over time in the blank measurements. Compounds for which sample concentrations were less than two times higher than concentrations detected in the blanks were excluded from the dataset. After blank corrections, only the compounds with concentrations above zero in at least 50% of the samples for each habitat were included in the final analysis. The VOC concentrations were then divided by the sampling mid-time, multiplied by the total chamber volume, and normalized to the plot surface area to obtain emissions rates in μg m^–2 h^–1. All analyses were performed in R version 3.6.3 (R Core Team, 2015). Correlation analyses were performed using the Pearson correlation. Data on individual VOC emission rates were log(x + 1) transformed and unit variance scaled prior to principal component analysis (PCA).

Weather conditions during the sampling campaign were highly variable, with bad weather conditions making sample collection impossible on 7 out of 16 possible sampling days. This was because of the risk of water damage to the electronics in the sampling system during periods of heavy rain, or persistent cloud cover causing a lack of charge on the solar panels that supplied the camp, making it impossible to charge the battery-operated pumps. Samples were collected whenever possible between 10 AM and 8.30 PM, covering most of the natural variation in the environmental conditions at the sampling site (Table 1). The temperature and relative humidity inside the sampling chambers during a single sample collection varied by, on average, 2.3 °C and 6.3%, respectively. The average temperature inside the sampling chambers did not correlate with the air temperature during sample collection (r = −0.15, p = 0.50, n = 23), but correlated positively with the average incoming shortwave radiation at the time of sample collection (r = 0.63, p = 0.0012, n = 23).

Total VOC emission rates from bare ice had a low but statistically significant positive correlation (r = 0.43, p = 0.015, n = 31) with the incoming shortwave radiation averaged over the sample collection period for each sample (Figure 2). A similar significant positive correlation between emission rates and incoming shortwave radiation averaged over the sample collection time was found for all compound classes into which detected VOCs were grouped (Table 2). Total VOC emission rates from bare ice did not correlate significantly with chamber temperature (r = 0.37, p = 0.084, n = 23).

Bare ice surfaces, cryoconite holes, and red snow surfaces presented distinct VOC profiles as analyzed by PCA (Figure 3). The three habitats were separated from each other along PC2, driven by compounds most characteristic of the red snow and cryoconite hole VOC profiles (Supplementary Figure 1).

Mean VOC emission rates [± standard error (SE)] were 3.97 ± 0.70 μg m^–2 h^–1 from bare ice surfaces (n = 31), 1.63 ± 0.13 μg m^–2 h^–1 from cryoconite holes (n = 4), and 0.92 ± 0.08 μg m^–2 h^–1 from red snow (n = 2), indicating that bare ice surface VOC emission rates were on average 2.4-fold and 4.3-fold higher than cryoconite hole and red snow emission rates, respectively. Oxygenated benzenoids accounted for just over half of the total emissions from bare ice surfaces (Figure 4 and Supplementary Table 2), with the three oxygenated benzenoids phenyl maleic anhydride, acetophenone and benzaldehyde accounting for 51.1 ± 11.7% of total emissions. The contributions of oxygenated benzenoids to red snow (2.8 ± 2.4%) and cryoconite hole (0.7 ± 0.2%) emissions were low. Cryoconite hole emissions were dominated by alkanes, which accounted for 74.6 ± 13.2% of emissions (Table 3 and Supplementary Table 2), while the majority of emissions from red snow consisted of alkenes (47.5 ± 6.0%, Table 4) and alkanes (36.3 ± 1.3%) (Table 3).

VOC emissions from bare ice surfaces did not correlate with the number (cells mL^–1) of glacier ice algae (A. alaskanum and A. nordenskiöldii) (r = 0.13, p = 0.54, n = 26) or red snow algae (r = 0.34, p = 0.16, n = 18). The limited number of samples collected from red snow and cryoconite holes prevented a similar analysis for these habitats.

The average chamber headspace temperature and relative humidity did not correlate with either air temperature or air relative humidity. However, the range of relative humidity in the chambers was similar to the range of relative humidity in the ambient air, as recorded at the PROMICE weather station QAS_M, giving confidence that representative relative humidity was prevalent during sample collection. In some cases, chamber temperatures were up to five times higher than air temperatures during the time of sampling, likely due to the greenhouse effect inside the chambers. This effect cannot be mitigated in the current experimental setup. We note that the parameters recorded at the PROMICE station, namely, air temperature and air relative humidity, were measured at approximately 2.6 m above the bare ice surface. This may explain some of the temperature variation, as near-surface conditions are highly variable and are affected by near-surface meteorology and heat exchange with the glacier surface (Bravo et al., 2019). Chamber temperature and incoming shortwave radiation correlated positively (r = 0.63, p = 0.0012, n = 23), and the potential effect of these two parameters could not be separated in this study.

More than 60% of the tentatively identified compounds detected in samples collected for bare ice surfaces have been reported as microbial volatiles in the literature (Tables 3–9). This suggests that the detected volatiles may be of microbial origin. Yet, no correlation was found between VOC emissions and algal counts in the top 2 cm of surface ice at the sampling sites. This lack of correlation may be explained by the fact that only the top layer of the ice surface was sampled for counts, while the bare ice surface is characterized by the presence of a porous ice weathering crust of variable depth (up to 30 cm or more) that forms due to subsurface shortwave radiation penetration, melting, and percolation. This deeper weathering crust may host microbial life (Irvine-Fynn et al., 2021) that contributes to VOC emissions but was not accounted for in the 2-cm deep surface layer used for the cell counts. Similarly, the cell counts only accounted for the presence of algae, while ice surface microbial communities are known to also include bacteria, archaea, and fungi, which likely also contributed to the measured VOC emission rates. In addition, bare ice surface microbial communities live in association with liquid water at the ice surface, indicating that volatiles are likely released into the water-phase rather than directly into the air. Microbial volatiles typically diffuse quickly through both the gas and water phases (Weisskopf et al., 2021). However, air-liquid partition coefficients are directly dependent on temperature, meaning that slower volatilization of VOCs from cold ice sheet surface waters may be anticipated. Hence, microbial volatiles may be distributed away from the source microorganisms through the hydrologically connected weathering crust (Cook et al., 2016). This is further supported by similar emission rates being detected in samples collected simultaneously (Figure 2, samples with identical incoming shortwave radiation were collected at the same time), as the distance between the sampling locations was limited to 5 m by the length of tubing on the pump system. The positive correlation between chamber temperature and incoming shortwave radiation may further interfere with a possible linear relationship between algal abundance and VOC emissions, as warming inside the chamber may lead to an increase in the volatilization of VOCs from surface water.

Kramshøj et al. (2019) suggested that the depth of the water table in permafrost soils following a thaw event influences the soil VOC emission rates, as compounds diffuse slower through water than through air. The water table in the Greenland ice sheet weathering crust is highly variable (Cooper, 2018) and data on water table depth during sample collection for this study is not available. Hence, we speculate that variations in near-surface hydrology may explain some of the variations in the measured emissions. Furthermore, a range of photochemical reactions are known to occur in snow and ice (Grannas et al., 2007). To the best of our knowledge, volatile concentrations have been measured in melted snow (Herbert et al., 2006; Kos and Ariya, 2006), but no in situ measurements of VOC emission rates for supraglacial microbial habitats have been reported. Ariya et al. (2011) reported (semi-)volatile concentrations ranging from 0.007 to 7.4 μg L^–1 in melted snow samples but found no link with bacterial or fungal counts from those samples. The majority of the most abundant VOCs in our study have previously been reported as microbial VOCs, but we cannot disregard the possibility that snow or ice melt may provide non-microbial sources of volatiles, hence further confounding simple correlations of cell counts with VOC emissions. Volatile release from melting ice needs to be assessed to better quantify potential microbial contributions to bare ice surface VOC emissions.

Irradiance is one of the main drivers of ice sheet surface melt, and so any volatiles originating from the ice, as a result of melting out of recent and/or past deposition, will likely peak on sampling days with high irradiance. However, microbial contributions to the total VOC emissions from bare ice surfaces are also expected to peak at high irradiances for several reasons. First, Williamson et al. (2020) showed that glacier ice algae were more stressed when exposed to 100% ambient irradiance compared to those exposed to 50 or 0% ambient irradiance. Stress has previously been shown to induce VOC production in algae (Zuo, 2019). Second, changes in algal fatty acid biosynthesis have been shown in response to abiotic stresses (Teoh et al., 2004; Holzinger and Karsten, 2013), such as high levels of irradiance and/or freeze-thaw cycles, both of which supraglacial microbial communities are subjected to. Alkanes (Table 3) and alkenes (Table 4), which account for 38.5, 86.2, and 83.8% of emissions from bare ice, cryoconite holes, and red snow, respectively, (Figure 4 and Supplementary Table 2) are known volatile (by-)products of fatty acid biosynthesis in microorganisms (Schmidt et al., 2015; Sorigué et al., 2016). Recently, an enzyme belonging to an algae-specific subgroup of the glucose-methanol-choline oxidoreductase family, fatty acid photodecarboxylase (FAP), was identified (Sorigué et al., 2017). FAP is one of the few known enzymes that requires light for their catalytic cycle and operates independently of photosystem II (Moulin et al., 2021). Hence, both the stress caused by high levels of irradiance and the fact that a light-driven enzyme plays a role in hydrocarbon production point toward higher microbial volatile (hydrocarbon) production when irradiance is high. High light conditions have also been shown to increase emissions of isoprene and monoterpenes in cyanobacteria (Shaw et al., 2003) and algae (Meskhidze et al., 2015) and of halogenated hydrocarbons in marine algae (Bondu et al., 2008).

VOC production is a common defense strategy in plants (Baldwin et al., 2006), as well as in other organisms, including microorganisms (Weisskopf et al., 2021). Four oxygenated benzenoids (Table 8) were identified in bare ice surface VOC emissions, of which the three most emitted compounds (phenylmaleic anhydride, benzaldehyde, and acetophenone) have been reported to have antifungal activity (Chen et al., 2011; Boukaew et al., 2018; Calvo et al., 2020). The presence of fungi on bare ice surfaces (Perini et al., 2019) and recent evidence that parasitic fungi can penetrate the cells of snow and glacier ice algae, leading to their destruction (Fiołka et al., 2021), suggests that algal production of antifungal VOCs may have an ecological advantage. The data presented here do not allow us to identify specific sources of measured VOCs, but it is a strong first indication that supraglacial microbial habitats are a source of microbial volatile emissions that should be investigated further in more controlled and long-term field sampling campaigns, as well studies in laboratory cultures of glacier surface species, as they become available.

Cryoconite holes are considered hotspots for biogeochemical cycling in supraglacial systems (Anesio et al., 2009; Cameron et al., 2012). Thus, it may initially appear surprising that the first estimates of VOC emission rates from cryoconite holes reported here are lower than those reported for bare ice surfaces. However, the microbial communities in cryoconite holes are located at the bottom of the water column, which is typically between 5 and 10 cm deep, rather than on the exposed ice surface. This means that cryoconite hole microbial communities are likely to be less impacted by abiotic stresses, such as high irradiance and rapid temperature changes. Stress is known to induce microbial VOC production in cyanobacteria (Koksharova, 2020), which are often the dominant primary producers in cryoconite hole communities. Hence, lower stress levels in cryoconite hole microbial communities compared to bare ice surface microbial communities may explain the lower emission rates detected from cryoconite holes. In addition, cryoconite hole microorganisms have been shown to recycle carbon produced by other members of the community (Sanyal et al., 2018). Volatiles present a potential carbon source (Junker and Tholl, 2013) and may thus be partially recycled within the community, rather than emitted to the atmosphere. One-third of the compounds detected in cryoconite hole emissions have previously been reported as VOCs associated with algae, bacteria, fungi, or cyanobacteria (Tables 3–6, 8, and references therein). The limited number of samples collected in this study point toward little-to-no effect of incoming shortwave radiation on cryoconite hole total VOC emissions. However, more samples collected under controlled conditions are needed to assess any possible light dependence, as well as to better constrain the estimate of VOC emission rates from cryoconite holes and their potential recycling as an in situ carbon source.

Red snow patches near the 2020 basecamp were dominated by uncultured Chlamydomonadaceae (18S sequencing data, unpublished). One-third of compounds detected in red snow VOC emissions have been reported as microbial volatiles (Tables 3, 4, 6, 8, and references therein). Of these, 1-octene, 1-dodecene, and 1-tetradecene have been reported as volatiles from C. reinhardtii (Renukadevi et al., 2011; Zuo et al., 2015), a well-studied model organism that also belongs to the Chlamydomonadaceae family. We noted that 1-octene and 1-dodecene were detected in normal C. reinhardtii cells, as well as in C. reinhardtii cells subjected to programmed cell death (Zuo et al., 2015). Programmed cell death is hypothesized to be a pro-survival mechanism at the population level (de Carpentier et al., 2019) and may be an interesting future area of study in terms of controls on snow algal communities.