A red tide in the pack ice of the Arctic Ocean

In the Arctic Ocean ice algae constitute a key ecosystem component and the ice algal spring bloom a critical event in the annual production cycle. The bulk of ice algal biomass is usually found in the bottom few cm of the sea ice and dominated by pennate diatoms attached to the ice matrix. Here we report a red tide of the phototrophic ciliate Mesodinium rubrum located at the ice-water interface of newly formed pack ice of the high Arctic in early spring. These planktonic ciliates are not able to attach to the ice. Based on observations and theory of fluid dynamics, we propose that convection caused by brine rejection in growing sea ice enabled M. rubrum to bloom at the ice-water interface despite the relative flow between water and ice. We argue that red tides of M. rubrum are more likely to occur under the thinning Arctic sea ice regime.

Mesodinium rubrum across sea ice habitats and water column. Mesodinium rubrum was detected in young ice (YI) of the refrozen lead throughout the study period with abundance in the range of 0.3 to 15.7 × 10 6 cells m −2 ( Table 1). The highest abundance (2.9 to 211 × 10 6 cells m −2 ) was observed in slurp gun samples taken by divers from the ice-water interface (Fig. 2). The bloom was visible as a faint coloring of the ice undersurface, and the sampling with the slurp gun indicated that the algal layer had a thickness of <1 mm and was stationary at the interface. Because the area sampled with the slurp gun was known we could calculate cells per area. The per volume cell concentration in the slurp samples was in the range 2.3 × 10 4 to 3.1 × 10 6 cells L −1 , but the slurp samples were diluted when surrounding seawater entered due to the suction. Free chloroplasts originating from   Table 1. Temporal development of chlorophyll a (Chl a) and alloxanthin (Allo) standing stocks (mg m −2 ) in the ice-water interface of YI (average ± SE, n = 3 sites) and in ice cores of YI (average ± SE, n = 5 sites), FYI and MYI (n = 1) on Floe 3 (4 May-4 Jun 2015). From ice-water interface and ice cores of YI abundance (average 10 6 cells m −2 ± SE, n = 3-5, 0 = below detecton limit) of cryptophytes (Crypto), Mesodinium rubrum (M. rub) and free M. rubrum chloroplasts (chloro) are shown. Cryptophytes are prey, supplying chloroplasts to M. rubrum 49 . Alloxanthin is a pigment produced by cryptophytes and is therefore also found in the chloroplasts of M. rubrum 26  www.nature.com/scientificreports www.nature.com/scientificreports/ 9.2 × 10 7 to 3.8 × 10 9 m −2 . Neither M. rubrum cells nor free chloroplasts were detected by microscopy in FYI or MYI (Table 1). In the water column the abundance of M. rubrum (Fig. 4a) peaked on 18 May with 3.1 × 10 8 cells m −2 , and a regional helicopter sampling revealed abundances between 1.0 and 6.0 × 10 8 cells m −2 in a larger area around R/V Lance (Fig. 4a). The per volume concentration of M. rubrum in the water column was in the range of 2 to 52 × 10 3 cells L −1 . M. rubrum constituted up to 40% of the total cell abundance of the protist community  (Table 2). In the sediment traps deployed 1 m below the ice, a vertical flux in the order of 10 6 cells m −2 d −1 was observed on 10 and 18 May, whereas no M. rubrum cells were found in the traps at this depth on 26 April, or after 18 May (Table 2). In Fig. 2 we show schematically our observations of M. rubrum in YI, at the YI ice-water interface, and in the water column.
Abundances of cryptophyte algae, the prey and source of chloroplasts for M. rubrum, in the YI ranged from 0.4 to 8.8 × 10 6 cells m −2 (Table 1). At the ice-water interface, cryptophytes were only detected once, on 14 May, and none were observed in FYI or MYI ( Table 1). The abundance of cryptophytes in the water column was in the range of 0.3 to 2.1 × 10 8 cells m −2 in May, but a higher abundance of 1.4 × 10 9 cells m −2 was observed in June (Fig. 4b).
The highest chlorophyll a (Chl a) standing stock, a proxy for algal biomass, was measured at the ice-water interface of the refrozen lead with values in the range 0.9-15 mg m −2 for the period 6-14 May. The Chl a concentration in these slurp samples was in the range 7-117 mg m −3 . In the same period, the Chl a standing stock in the ice cores was in the range of 0.09-1 mg m −2 . The highest Chl a standing stock measured in ice was around 3 mg m −2 in the bottom 10 cm of the refrozen lead in late May and early June (Table 1). In the water column, the depth integrated (0-25 m) Chl a standing stock increased in early May from 1 mg m −2 to 9.7 mg m −2 on 18 May (Fig. 4c). The regional sampling by helicopter to the north (19 May), east, south and west (20 May) relative to R/V Lance revealed similar levels of 6-19 mg Chl a m −2 in a larger area at this time (Fig. 4c). The concentration of Chl a at this time never exceeded 0.5 mg m −3 . From 25 May onwards, Chl a standing stocks increased by a factor of 10-20 (Fig. 4c), which was due to an under-ice bloom of Phaeocystis pouchetii described by Assmy et al. 25 . According to the ice classification performed on the ALOS-2 radar satellite scene YI made up 10.2% of the total area, thicker FYI or MYI constituted 84.4%, and open water 5.4% ( Fig. S2.1). Upscaling to the 2800 km 2 ALOS-2 scene area ( Fig. 1) indicate that the Chl a in the thin (<1 mm) layer of the YI interface equals 4.3% of the total integrated amount found in the water column from 0 to 25 m depth.
Alloxanthin standing stocks, a proxy for cryptophyte and/or Mesodinium rubrum biomass 26 , were orders of magnitude higher at the ice-water interface of YI (0.2 to 1.5 mg m −2 ) than in the YI cores (0.004 and 0.08 mg m −2 with a peak of 0.4 mg m −2 on 20 May). In the FYI and MYI, the standing stocks of alloxanthin were even lower, in the range 0.001-0.008 mg m −2 ( Table 1). The alloxanthin: Chl a ratio (mg: mg) was in the range of 0.01-0.03 in YI, except for ratios of 0.07-0.13 coinciding with Chl a peaks on 4, 10 and 20 May 13 , and up to 0.36 at the ice-water interface. In the water column alloxanthin standing stocks increased from winter levels of 0.02-0.04 mg m −2 to approximately 0.1 mg m −2 in the period 11-18 May (Fig. 4d). From the spatial sampling campaign on 19-20 May, we only have the northern point for alloxanthin, which showed 0.25 mg m −2 , indicating that the increase was taking place over a larger area. The alloxanthin to Chl a ratio in the water column was between 0.02 and 0.06, and after 20 May <0.02.
In the slurp gun samples taken by divers at the ice-water interface M. rubrum and its chloroplasts dominated throughout the sampling period from 7-14 May, contributing 87-97% and 92-97% of the total protist abundance www.nature.com/scientificreports www.nature.com/scientificreports/ and carbon biomass, respectively ( Fig. S1.2). In the YI cores the free M. rubrum chloroplasts totally dominated in abundance and constituted a large part of the carbon biomass until 20 May 13 . In addition flagellates constituted a significant fraction in early May while pennate diatoms gradually increased and became the dominating group in late May. See Kauko et al. 13 for a detailed description of the succession of the protist community in the YI of the refrozen lead, and Olsen et al. 21 for the surrounding FYI and MYI. The protist community in the water column during early May was dominated by dinoflagellates, flagellates and the "Other" group ( Fig. S1.3) dominated by Phaeocystis pouchetii, ciliates other than Mesodinium rubrum and coccolithophorids (Supplementary Section 4). For most sediment trap samples, in which M. rubrum was found, the majority of the protist cells were M. rubrum, which constituted almost the entire carbon biomass (Fig S1.4). In Fig. 4 is shown a schematic summary of the observations in ice, ice-water interface and water column, with sampling methods indicated. www.nature.com/scientificreports www.nature.com/scientificreports/ Photosyntetic response to irradiance. The maximum quantum yield of fluorescence (Φ PSIImax ) measured in slurp samples from the ice-water interface of the refrozen lead on 5-14 May was in the range 0.40-0.64 ( Table 3). The range for the photosynthetic parameters derived from fitting the Webb equation to the rapid light curve measured were: photosynthetic efficiency (α) = 0.32-0.59 (µmol photons m −2 s −1 ) −1 , maximum relative electron transfer rate (rETR max ) = 72-198 (no unit), saturation irradiance (E k ) = rETR max /α = 153-549 µmol photons m −2 s −1 ( Table 3). The measured downwelling PAR irradiance at the ice-water interface of the refrozen lead reached a maximum of 114 µmol photons m −2 s −1 22 .
Ice-ocean boundary layer dynamics. The average relative velocity between ice and water was 11 cm s −1 ( Fig. S1.1). Figure 5 shows the velocity profiles from the sea ice boundary down to 2 m depth, including both the laminar sub-layer and the turbulent logarithmic layer. A zoom of the upper 0.15 cm right below the sea ice gives details of the laminar sub-layer velocity structure and thickness. For a free-stream velocity of U ∞ = 10 cm s −1 (Fig. 5b), the thickness of the laminar sub-layer is δ lsl = 0.06 cm. During times with free-stream velocities lower than average (Fig. 5a), the thickness of the laminar sub-layer is larger (δ lsl = 0.12 cm for U ∞ = 5 cm s −1 ), whereas during events of stronger free-stream velocities (Fig. 5c,d), the thickness of the laminar sub-layer is much smaller (δ lsl = 0.03 cm and δ lsl = 0.02 cm for U ∞ = 20 cm s −1 and U ∞ = 30 cm s −1 , respectively). In Supplementary Material Section 4 we describe how under-surface roughness can affect the boundary layer dynamics.

Discussion
The 7-117 mg Chl a m −3 we measured in the slurp samples from the YI interface layer is 14-234 times higher than the concentration in the water column (<0.5 mg Chl a m −3 ), which was at the level of non-bloom concentrations reported from the North Atlantic 27 . The interface bloom can be likened to the red tides of M. rubrum often observed at lower latitudes, where the Chl a concentration can be >100 mg m −3 and abundance up to 10 6 cells L −1 28 . To our knowledge, ours is the first observation of a pack ice associated red water bloom. The only published observation of a red tide in the Arctic Ocean is from ice-free, coastal waters near Barrow, Alaska in September 1968, caused by an unidentified ciliate similar to, but not identical with M. rubrum 29 . The abundance in the bottom 10 cm of YI (0.3 to 15.7 × 10 6 cells m −2 ) is comparable to some other observations. Up to 1.6 × 10 6 cells m −2 of M. rubrum was observed in the bottom 2-4 cm of 30-40 cm thick FYI in the Saroma-ko lagoon in Hokkaido, with an integrated abundance in the water column under the ice from 0 to 1 m depth of 3.4 × 10 6 cells m −2 30 . Likewise, when M. rubrum was observed at abundance 2 × 10 5 cells m −2 in the bottom 2-4 cm of 1.5-2 m thick FYI in the Canadian Arctic, the average abundance in the water column below the ice down to 8 m was 0.14 × 10 3 cells L −1 31 . The maximum abundance we measured in the YI ice-water interface (211 × 10 6 cells m −2 ) was considerably higher than these observations. M. rubrum cells are known to be fragile and difficult to preserve. The mix of glutaraldehyde and formaldehyde used during the N-ICE campaign was chosen in order to preserve the highest possible fraction of the entire protist community, but might not be the best method to preserve M. rubrum 18 . The high abundance of chloroplasts originating in M. rubrum in the ice core samples ( Table 1), indicate that cells had disintegrated. The melting of the ice cores could also have caused ciliates to burst 32 Use of the free chloroplasts as a tracer of M. rubrum relies on the accurate identification of them. The morphology of the free chloroplast was very similar to those we observed inside of M. rubrum cells (Fig. 3), and agrees well with previous descriptions 33 .   www.nature.com/scientificreports www.nature.com/scientificreports/ According to the quantum yield of fluorescence (0.40 to 0.64) the protists, mainly M. rubrum, were in a good physiological condition 34 , with active photosynthesis at the YI ice-water interface. Under the FYI and MYI with 20-50 cm of snow the irradiance was only 1-10 µmol photons m −2 s −1 21 . This ice type covered most of our study area (Fig. S2.1). The M. rubrum bloom was confined to the ice-water interface of the YI in the refrozen lead, where the irradiance was higher, on average 114 µmol photons m −2 s −1 22 . Moeller et al. 35 showed that M. rubrum acclimates to the irradiance level so that the saturation irradiance for photosynthesis (E k ) is similar to the irradiance they grow under. We measured E k > 153 µmol photons m −2 s −1 , indicating they were growing stationary at the YI ice-water interface. This E k is similar to what Stoecker et al. 36  Under YI is a near optimal place with regards to light, but can M. rubrum cells actively position themselves under the thin ice, or is there some external physical mechanism keeping them there? The swimming speed of M. rubrum is approximately 0.16 mm s −1 19 , whereas the average relative velocity of ice vs. water was 0.11 m s −1 (Fig. S1.1), so they could certainly not outswim the ice. Previously proposed mechanisms like scavenging of cells by frazil ice in the water column and waves pushing the cells into the ice 7 seems unlikely here because both processes are most active when there is open water or still unconsolidated ice, whereas our bloom took place at the ice-water interface of consolidated YI. Sieving of the water column by the protruding ice crystals of the skeletal layer 5 seems more likely to work for sticky algae like diatoms 38 than for the fast swimming/jumping 19,39 and, to our knowledge, non-sticky M. rubrum.
Is it possible that in the boundary layer close to the ice undersurface the relative motion of water and ice is so slow that M. rubrum cells can remain stationary there? Boundary layer shear is known to affect algal colonization of the benthic environment 40 , and although the boundary layer under sea ice is well studied in other physical contexts 8,41 , there seems to be no studies on how it affects algal colonization. Figure 6 shows a schematic compilation of the various forces acting on a cell of M. rubrum to modify its position relative to the ice. In a simplified scenario of a smooth sea ice bottom and no sea-ice melt or growth, and with the range of relative velocities between ice and water that we observed (Fig. S1.1), the thickness of the laminar part of the boundary layer would theoretically be 0.2-1.2 mm (Fig. 5). M. rubrum cells have a maximum width of 20 µm and length of 40 µm 27 , and move in jumps of 0.16 mm. A typical jumping rate is 1 s −1 , and thus, the effective swimming velocity is 0.16 mm s −1 19 . This implies that the laminar layer was 3-4 jumps thick at the average ice-water relative velocity. The shear in the layer is in the range 8-283 s −1 , from lowest to highest free stream velocity (Fig. 5). In addition to being phototactic, M. rubrum is also rheotactic, and a shear of 1-3 s −1 is enough to trigger an escape response according to Fenchel and www.nature.com/scientificreports www.nature.com/scientificreports/ Hansen 19 . In addition to the thinness of the layer and the high shear, the velocity reached within the laminar layer was 1 cm s −1 or higher (Fig. 5), i.e. above the swimming speed of M. rubrum. Thus, it seems unlikely that a bloom at the ice-water interface can be maintained within the laminar boundary layer.
Macroscopically the cores appeared relatively smooth at the bottom, with roughness on the mm scale. If the roughness created hydrodynamically rough conditions, i.e. allowed turbulence to reach the ice surface, it is unlikely that it helped M. rubrum cells to stay in the ice-water interface because the laminar layer would become even thinner or disappear completely (Supplementary Section 3). Roughness did not help sticky diatoms to colonize benthic surfaces 9 , then it might be even more unlikely to help non-sticky ciliates.
The bloom at the YI interface disappeared abruptly over a few days after 20 May (Fig. 2). At this time there was still a surplus of inorganic nutrients in the water below the ice, and ice diatoms continued to grow at the interface for two weeks until the floe broke up 13,21 , i.e. nutrient limitation was not causing the disappearance. M. rubrum was at saturating abundance for copepods in the interface if they were able to exploit this food source, but we observed no response in grazer abundance or aggregation of grazers at the interface. Thus, it is unlikely that grazing terminated the bloom.
Noteworthy, over the entire period we observed the interface bloom the YI was growing. The disappearance of the bloom coincided with the cessation of ice growth (Fig. 4), suggesting that ice growth might create physical conditions favorable to keep M. rubrum at the interface. Almost no ice growth was observed in FYI and MYI in early May 42 due to the insulating effect of the thick snow cover 43 , and M. rubrum was not found there (Table 1). During ice growth a porous skeletal layer is formed at the bottom of the ice with pockets and tubes, which can be 1-3 cm long and with a diameter up to 0.5 mm 44 . In this process brine is rejected by gravity drainage 45 . The decrease in bulk ice salinity observed indicates that this happened as predicted when the refrozen lead ice formed 42 . Brine drainage from the ice is compensated by an inflow of seawater, forming convection cells [46][47][48] . It is possible that this skeletal layer convection disrupts the laminar boundary layer (J. Morison personal communication) and helps M. rubrum cells to remain in the skeletal layer, maybe assisted by their own upwards, phototactic swimming (Fig. 6). In addition convection renews the water in the skeletal layer, supplying nutrients from the water column 4 . This might also supply cryptophyte prey and thus new chloroplasts to M. rubrum 49 .
When the refrozen lead ice stopped growing around 20 May (Fig. 2), it follows that brine drainage, and thus also convection stopped 45 . At this point the physical factors at the ice-water interface were presumably dominated by the boundary layer dynamics, which, as discussed above, did not help M. rubrum to stay in the interface. In contrast, the interstitial ice diatom community continued to grow and reached maximal biomass in late May after ice melt had started 13,21 , illustrating the benefit of being adhered to the ice 14,15,50 .
The highest abundance of M. rubrum in the water column was observed on 18 May (Fig. 4a) coinciding with its highest vertical flux (Table 2). This could be related to a mass release of M. rubrum from the ice due to the cessation of ice growth, as discussed above. The tendency of M. rubrum to migrate and aggregate in the water column makes it difficult to get an accurate measure of the abundance with the fixed sampling depths during the CTD casts 18 . The sediment traps capture cells during 1-2 days and therefore might be a more reliable device for detecting M. rubrum. According to the vertical flux (Table 2) about 10% of the standing stock from 0-25 m was captured per day on 18 May, and the apparent sinking velocity was 0.28 m d −1 at 25 m. The sinking speed of a resting M. rubrum cell is 0.7 m d −1 according to Fenchel and Hansen 19 . It is reasonable that these motile, phototactic ciliates had a low sinking velocity, whereas the migratory behavior might lead M. rubrum cells to swim into the traps 51 . www.nature.com/scientificreports www.nature.com/scientificreports/ The regional sampling showed similar abundances in a larger area surrounding R/V Lance on 19 and 20 May (Fig. 4a), suggesting that the M. rubrum red tide was not restricted to the refrozen lead we studied but was a regional phenomenon. The drift track of many ice-tethered buoys in the area around R/V Lance for the same time period indicated that the wind speed and direction was the same in the entire area covered by the ALOS-2 scene 52 . Thus, it is reasonable to assume that the temperature conditions were similar, and therefore that all YI was growing at this time, facilitating a large scale ice-water interface red tide of M. rubrum in an area exceeding 2800 km 2 (Fig. 1). According to the ice type classification in the ALOS-2 satellite radar scene from 18 May (Fig. S2.1), which partly covered the regional sampling campaign by helicopter (Fig. 1), YI made up 10.2% of the total area. The Chl a in the YI ice-water interface equaled 4.3% of the total amount found in the water column from 0 to 25 m depth, which is considerable considering the huge volume of water and the thinness of this layer (<1 mm).
The ongoing regime shift towards a thinner, more dynamic ice cover in the Arctic Ocean, with more lead formation 25 can promote ephemeral blooms of M. rubrum below growing young ice. It is important to improve our understanding of the mechanisms enabling ice-associated blooms of different algal taxa. Shifts in species composition at the base of the ice-associated ecosystem is an indicator of change, and are likely to have cascading effects on the Arctic marine food web and the biological carbon pump of the Arctic Ocean.

Current measurement.
A medium-range vessel-mounted broadband 150 kHz acoustic Doppler current profiler (ADCP) from Teledyne RD Instruments was used to measure current speed and direction below the ice. The profiles were hourly-averaged in 8-m vertical bins and the first bin was centered at 23 m 24 . The current speed and direction at 23 m depth were used to calculate the current relative to the ice floe based on ship navigation data. The relative current speed measured between the water column and the sea ice was in the range 0-0.3 m s −1 , with an average of 0.11 m s −1 (Fig. S1.1). The apparent northward direction of the water was mainly due to the faster southward movement of the ice driven by prevailing northerly winds 52 . To study the variability of water column properties in a larger area around R/V Lance, samples were taken at the end of helicopter transects about 60 km north of the ship on 19 May, and about 50 km east, west, and south of the ship on 20 May (Fig. 3).
Sample collection and analysis. Seawater samples for chlorophyll a (Chl a), pigment composition and protist counts were collected with a rosette water sampler with 8 L Niskin bottles deployed from the ship or with 3.5 L Niskin bottles on a rosette deployed from the ice. Samples were taken at 5, 25, 50 and 100 m depths from the ship, and from 2, 5 and 15 m with the on-ice system. During a regional sampling campaign by helicopter on 19 and 20 May water samples were taken manually with a Limnos water collection bottle closed with a messenger (Limnos. pl) at 5, 15 and 25 m depth. To obtain depth integrated values of Chl a or abundance we used the trapezoid method. Because we had no measurements from 0 m we set the values to equal those at 5 m depth.
Three ice-tethered sediment traps (KC Denmark) were deployed on a rope stretched under the YI horizontally at 1 m depth. In addition, four sediment traps were deployed vertically at 5, 25, 50, and 100 m depth, respectively, along a mooring attached to the ice. The deployment time was between 36 and 72 h, usually 48 h. To avoid loosing sample water from the traps during deployment and recovery, the trap cylinders were filled with filtered seawater, made hypersaline (i.e. more dense) by adding sodium chloride, before deployment. Each trap had two cylinders with internal diameter 7.2 cm and height 45 cm, with no baffle at the top. At sampling the water from both cylinders were combined into one sample. Copepods and other zooplankton were removed before taking samples for algal taxonomy. Sinking flux for protists was calculated from cell concentration in the traps, trap volume and area, and trap deployment time.
Samples from the sea ice were taken with 9 and 14 cm diameter ice corers (Mark II coring system, KOVACS enterprise, Roseburg, USA). The cores were cut into 10 or 20 cm sections, put in cleaned opaque plastic containers and melted during 18-24 hours at room temperature without seawater buffer on board the ship, according to Rintala et al. 31 .
Samples from the ice-water interface under the refrozen lead were taken by scuba divers using a modified 3.5 L Trident ® suction gun (slurp gun). The front nozzle was oblique so that it was possible to fill the gun while moving it along the undersurface of the sea ice. The surface area sampled was 5 × 54 cm for a full slurp gun and this area was used to transform cells per volume in the sample to cells per area. For protist taxonomy analysis and cell counts, 190 mL from Niskin bottles, slurp gun, sediment traps, or melted ice cores were transferred into 200 mL brown glass bottles and fixed with an aldehyde mixture consisting of glutaraldehyde at a final concentration of 0.1% and hexamethylenetetramine-buffered formaldehyde at a final concentration of 1% (vol:vol). The samples were stored dark and cool until analysis. Protists were counted with an inverted Nikon Ti-U light microscope (Nikon TE300 and Ti-S, Tokyo, Japan) using the sedimentation chamber method of Utermöhl 53 . In most cases 50 ml of the samples was settled, in some cases 10 ml. 20, 40 and 60X magnification was used and the number of view fields counted varied to obtain a minimum of 50 cells of the dominating species, i.e., with a maximum count error of ±28% according to Edler and Elbrächter 54 . Carbon biomass was determined by calculating volume from cell size 55 , which was converted to carbon using published conversion factors 56 . With this method and maximal magnification of 600X we detected mainly protists with cell diameter >2 µm.
Samples for Chl a were collected on 25 mm diameter GF/F filters (Whatman, GE Healthcare, Little Chalfont, UK). The volume filtered was noted. Chl a was extracted in 100% methanol for 12 h at 5 °C in the dark and subsequently measured using a Turner Fluorometer 10-AU (Turner Design, Inc.). Phaeopigments were measured by acidifying the sample with 5% HCl before measuring the fluorescence 57 . Samples to measure algal pigment composition were collected by filtering 10-1000 mL of sample through 25 mm GF/F filters, which were snap frozen in liquid nitrogen and then kept frozen at −80 °C until analysis. After an extraction step the pigments were measured using a Waters photodiode array detector (2996), Waters fluorescence detector (2475), and the (2019) 9:9536 | https://doi.org/10.1038/s41598-019-45935-0 www.nature.com/scientificreports www.nature.com/scientificreports/ EMPOWER software. The pigments were separated by reverse-phase high-performance liquid chromatography (HPLC) in a VARIAN Microsorb-MV3 C8 column (4.6 × 100 mm) using HPLC-grade solvents (Merck). For further details see Tran et al. 58 . Chl a was measured by both Fluorometer and HPLC for most samples in this study. A linear regression of all data from YI cores (n = 87) gave the relationship Chl a HPLC = 0.65 Chl a Fluorometer + 0.11, R 2 = 0.83. All our reported Chl a concentrations were measured by fluorometer, whereas all alloxanthin: Chl a ratios were calculated from alloxanthin and Chl a measured by HPLC. Photosyntetic response measured by fluorescence kinetics. The physiological status and light response of the photosynthetic apparatus of M. rubrum in samples from the refrozen lead ice-water interface were assessed using in vivo Chl a fluorescence kinetics measured with a Pulse Amplitude Modulation (PAM) fluorometer (Phyto-PAM, Walz, Germany). Samples were kept in a fridge with temperature in the range 1-2 °C and dark-acclimated for 30 min prior to measurement. The maximum quantum yield of fluorescence of photosystem II (Φ PSIImax ) was measured with the saturation pulse method 31 . Rapid Light Curves (RLC) in which the quantum yield of fluorescence in the light (Φ PSII ) was measured by illuminating the sample with actinic light increasing stepwise from 1 to 900 µmol photons m −2 s −1 in 13 steps at 20-second intervals, were used to assess the light response of the algae. The first measurement was after dark-acclimation, i.e. Φ PSIImax . Relative electron transfer rate (rETR) was calculated by Φ PSII × E, where E is the actinic irradiance. The photosynthesis-light function of Webb et al. 59 was fitted to the rETR data as a function of the incident actinic light: where α is the initial slope of the curve, i.e. the photosynthetic efficiency, and rETR max is the curve asymptote, i.e. the maximal rETR. We did not observe inhibition of rETR at high irradiance so we did not add an inhibition term to the equation.
Ice-ocean boundary layer dynamics. The ice-ocean boundary layer may be thought of as three different vertical zones 8 : (1) a laminar, molecular sub-layer (~0-1 mm thick) close to the sea ice-ocean interface, where the velocity varies linearly with depth; (2) a logarithmic turbulent layer (~1-3 m thick) below the laminar sub-layer, with constant stress and where the velocity varies logarithmically with depth; (3) a turbulent, thicker outer layer (~10 m thick), where the velocity is affected by the Coriolis effect. In this study we focus on the first two layers closest to the ice-water interface: the laminar sub-layer and the logarithmic turbulent layer. The surface shear stress in the ice-water interface 60 is defined as: where ρ is the density of the water and u * is the frictional or shear velocity at the boundary layer, which provides a scale for turbulence strength and for the laminar boundary layer thickness. The surface stress τ in the vicinity of the sea ice, which is dominated by viscous (as opposed to inertial) forces, may also be given by White 61 : where µ is the dynamic viscosity of seawater, which at 0 °C is µ = 1.8 × 10 -2 g cm −1 s −1 . Additionally, the magnitude of the surface stress is related to the drag force of the geostrophic fluid under a boundary (e.g., further from the sea ice) as: where C d is the dimensionless geostrophic drag coefficient, taken here as 5.5 × 10 −3 and u g is the geostrophic flow away from the boundary, also referred to as the free stream velocity relative to the sea ice velocity. From (2) and (4), the frictional velocity may be estimated as: with the thickness of the laminar sub-layer being 8 : where ϑ is the kinematic viscosity coefficient equal to the dynamic viscosity, µ, divided by the density of seawater, ρ, taken here as ρ = 1028 kg m −3 , yielding ϑ = / µ ρ = 1.78 × 10 −2 cm 2 s −1 ; and k is the dimensionless Von Kármán's constant, equal to 0.41. Following Eq. 3, the velocity structure within the laminar sub-layer (from z = 0 to z = δ lsl ) varies with depth as: for a very smooth sea ice surface. The free-stream velocity U ∞ measured at 20 m, relative to the sea ice, was on average 11 cm s −1 , often weaker, and with a few events reaching up to approximately 30 cm s −1 (Fig. S1). We start with the simplest scenario of a smooth sea ice bottom, and assume that no sea-ice melt/growth was occurring when these samples were taken. We select 4 observed values of U ∞ : (a) 5 cm s −1 ; (b) 10 cm s −1 (representative of the mean value 11 cm s −1 ); (c) 20 cm s −1 and (d) 30 cm s −1 . Using Eqs 6 and 7, we solve for the frictional velocity u * and the thickness of the laminar sub-layer δ lsl . We then solve for the velocity profiles both at the laminar sub-layer (linear) and at the logarithmic layer.

Data Availability
The data sets used in this study are publicly available from the Norwegian Polar Data Centre (https://data.npolar.