Sea-ice retreat may decrease carbon export and vertical microbial connectivity in the Eurasian Arctic basins

Arctic Ocean sea-ice cover is shrinking due to warming. Long-term sediment trap data show higher 19 export efficiency of particulate organic carbon in regions with seasonal sea-ice compared to regions 20 without sea-ice. To investigate this sea-ice enhanced export, we compared how different 21 phytoplankton communities in seasonally ice-free and ice-covered regions of the Fram Strait affect 22 carbon export and vertical dispersal of microbes. In situ collected aggregates, combined with 23 microbial source tracking revealed that larger aggregates from sea-ice and under-ice diatom blooms 24 were responsible for higher export efficiency and vertical microbial connectivity. During early 25 summer, Phaeocystis aggregates dominated the ice-free regions and exported two-fold less carbon 26 than diatom aggregates in ice-covered . that stronger export Fram Strait . Here we the effect of sea-ice distribution on phytoplankton community composition, carbon export efficiency and vertical microbial connectivity in the Fram Strait. We test the hypotheses 1) that settling aggregates formed in ice-covered regions sink faster than those formed in ice-free regions, which may result in lower export flux for ice-free regions. With regard to the effect on microbial community structure we postulated that 2) particle-associated microbes originate from surface waters and have little exchange with free-living microbes at depth, 3) that there is stronger vertical microbial connectivity in ice- covered compared to ice-free regions, and 4) that the stronger microbial connectivity leads to higher representation of surface-born microbes in the deep-sea including sediments. We found that large and fast-settling diatom aggregates in ice-covered regions resulted in higher carbon export efficiency near the ice and thus stronger vertical microbial connectivity compared to the seasonally ice-free region of the Fram Strait, which by smaller, slow-sinking Phaeocystis spp. aggregates. Our results suggest


Introduction 31
The Arctic Ocean is currently undergoing unprecedented changes due to ongoing climate warming. 32 Ice coverage is declining by ~13% per decade compared to the mean September extent for 1981-33 2010 1 , and climate models project that business-as-usual scenarios will result in seasonally ice-free 34 conditions by 2050 2 . Increasing temperature, combined with declining sea-ice extent, ice thickness 35 and multiyear ice 3 are impacting the composition of primary producers in the Arctic Ocean 4 . For 36 example, Atlantic phytoplankton species, such as Phaeocystis spp., are already seasonally 37 reoccurring in the Fram Strait 5 , and were also recently observed during a phytoplankton bloom in the 38 central Arctic Ocean 6 . It has been suggested that temperate phytoplankton species will become 39 resident in the Eurasian basins of the Arctic Ocean if the intrusion of warming Atlantic waters 40 continues 7 . As primary producers form the base of the food web, such shifts are likely to have drastic 41 consequences, not just in the pelagic realm, but also for pelagic-benthic coupling and biogeochemical 42 cycling in the Arctic Ocean 8 . However, the complexity of factors driving Arctic productivity 43 regionally makes it difficult to generalize to the future carbon flux in the entire Arctic Ocean 9 . 44 Remote sensing of ocean color in the Arctic shelf seas suggests an increase in net primary production 45 by 57% since 1998 10 , which likely enhances vertical carbon and nutrient fluxes 11 . Furthermore, the 46 transformation from thick multiyear to thin first year ice is increasing light transmission through the 47 ice 12 . Accordingly, field observations show an increased spatial and temporal extent of sea-ice algae 48 and under-ice pelagic phytoplankton in the Arctic basins, for which ocean color based PP 49 assessments are not available 6,13-15 . When the ice melts, such ice-algae and under-ice phytoplankton 50 blooms that are mostly composed of diatoms can also deliver substantial pulses of carbon and 51 nutrients to benthic ecosystems 8,16,17 . For example in 2012, the release of the fast-sinking ice-algae, 52 Melosira spp., from melting sea-ice delivered up to 9 g of carbon per square meter of seafloor, which 53 was more than 85% of the total carbon export that year 16 . Primary production models suggest that 54 with the northward propagation of ice-edge blooms along leads and polynyas, the impact of ice-algae 55 and under-ice phytoplankton blooms on productivity is likely to increase 18 , with ecological 56 consequences for pelagic and benthic ecosystems 16,19 . However, it remains unclear what ecological 57 effects an earlier and spatially substantial retreat of the ice-edge will have for carbon export on the 58 Arctic shelf seas and margins. Key factors could be timing of stratification, e.g. by meltwater, the 59 phytoplankton composition during the bloom, mismatches to grazers, and other effects on carbon 60 export efficiency, including the microbial loop in the euphotic zone and remineralization of 61 aggregates during sinking to deep-waters. 62 It is well established that sinking particles are an essential conduit of carbon and nutrients to 63 heterotrophic organisms in the deep ocean 20 . The microbial loop retains most carbon and nutrients in 64 the surface ocean, and can influence carbon export efficiency substantially, next to other factors such 65 as particle sizes and the activities of grazers 21 . Recent studies have also revealed that particles are 66 vectors that disperse microbes from the surface to the deep ocean some of which may carry specific 67 heterotrophic functions in the remineralization of sinking matter [22][23][24] . Hence, sinking particles can 68 play a key role in determining the structure and functioning of deep-sea microbial communities 69 beyond supply of energy and nutrients 22,23 . To date, such so-called vertical microbial connectivity has 70 mainly been demonstrated in temperate and tropical oceanic settings [22][23][24][25][26] . However, Rapp et al. 71 (2018) found that sedimentation of sea-ice algae influences microbial community composition at the 72 seafloor in the central Arctic. Considering the changes that the Arctic Ocean is currently undergoing, 73 it is critical to understand how microbial connectivity and carbon export will be impacted by 74 changing ice regimes and associated alterations in the composition and rates of primary producers. 75 Here, we assess the efficiency of particle export and vertical connectivity in the Fram Strait under 76 seasonally ice-covered and ice-free conditions, which were defined based on temporal duration of 77 sea-ice presence during the productive season. The Fram Strait represents the major deep-water  78  gateway to the Arctic Ocean basins. Warm Atlantic water flows northwards via the West Spitzbergen  79  Current (WSC) through the eastern part of Fram Strait, whereas cold Arctic water and sea-ice flows  80 southwards into the Atlantic via the East Greenland Current (EGC) in its western part (Fig. 1a). The 81 annual ice volume export through the western Fram Strait is currently increasing by 11% per decade 82 during spring and summer due to ice thinning and increasing drift speed 27 . If Atlantic warming of the 83 Arctic Ocean continues, it is projected that stronger ice-melt will occur in the western Eurasian 84 Basin, eventually reducing ice export through Fram Strait 28 . Here we studied the effect of sea-ice 85 distribution on phytoplankton community composition, carbon export efficiency and vertical 86 microbial connectivity in the Fram Strait. We test the hypotheses 1) that settling aggregates formed in 87 ice-covered regions sink faster than those formed in ice-free regions, which may result in lower 88 export flux for ice-free regions. With regard to the effect on microbial community structure we 89 postulated that 2) particle-associated microbes originate from surface waters and have little exchange 90 with free-living microbes at depth, 3) that there is stronger vertical microbial connectivity in ice-91 covered compared to ice-free regions, and 4) that the stronger microbial connectivity leads to higher 92 representation of surface-born microbes in the deep-sea including sediments. We found that large and 93 fast-settling diatom aggregates in ice-covered regions resulted in higher carbon export efficiency near 94 the ice and thus stronger vertical microbial connectivity compared to the seasonally ice-free region of 95 the Fram Strait, which was dominated by smaller, slow-sinking Phaeocystis spp. aggregates. Our 96 results suggest that with the ice-edge seasonally retreating from the Eurasian basins, carbon export 97 efficiency and vertical connectivity may decline in large regions of the Arctic Ocean. 98

Results 99
Settling aggregates in ice-free and ice-covered regions 100 We studied microbial communities associated with settling aggregates in contrasting sea-ice 101 conditions between June 24 th and July 16 th 2016 at the Long-Term Ecological Research (LTER) 102 observatory HAUSGARTEN in the Fram Strait (Expedition PS99.2 with RV POLARSTERN). We 103 classified ice concentration >15% as ice-covered conditions and ice concentration <15% as ice-free 104 conditions. According to this classification, the stations in the West Spitsbergen Current ('HG';  105 seafloor depths ~2500 m) were seasonally ice-free stations, as the majority of the productive season 106 in 2016 (March -July) they had no sea-ice ( Fig. 1a; Table 1). On the other hand, in the East 107 Greenland Current ('EG'; seafloor depth: ~1000-2700 m) and the northern ('N'; ~2500-2800 m) 108 stations, sea-ice was present during most of the productive season, and these sites were thus defined 109 as ice-covered stations ( Fig. 1a; Table 1). Long-term sediment trap collections of particulate organic 110 carbon (POC) fluxes at 200 m depth in the seasonally ice-free HG4 station have shown that the POC 111 flux peaks early in the season due to ice-associated carbon export and again later in the season during 112 the ice-free period due to POC export of pelagic production 29 . The POC export during the ice-113 associated flux peak (February to April) between 2001 and 2013 showed an inverse relationship 114 between POC flux and distance to the ice-edge within 0-80 km, i.e. the zone most influenced by melt 115 water (R 2 =0.39, p<0.01; Fig. 1b). This suggests that sea ice proximity can enhance POC fluxes. 116 When the ice edge was beyond 80 km from the sediment traps, we observed no spatial effects 117 between sea ice and POC export. To further test the hypotheses of sea-ice effects on carbon export 118 efficiency, we assessed the exported organic matter and the vertical microbial connectivity at stations 119 with contrasting ice-conditions in the Fram Strait during summer 2016. 120 First we used a Lagrangian Particle Tracking algorithm based on the observed aggregate sinking 121 velocities (Table 1), to test whether our spatial classification scheme permitted the differentiation of 122 particle origins. It showed less horizontal transport in the ice-covered regions compared to the ice-123 free regions ( Supplementary Fig. 1), and differences in the origin of the particles. Particles in the ice-124 free region ('HG' stations) were primarily from the Atlantic waters south of the investigated region, 125 and the majority of aggregates (82%) reaching the deep ocean (>1000 m) and seafloor originated 126 from ice-free surface waters ( Supplementary Fig. 1). In the ice-covered region ('EG' and 'N' 127 stations) below 1000 m, a 60% of the aggregates originated from the ice-covered surface waters 128 (Table 1). 129 All sampled stations were at the later stage of the phytoplankton bloom, based on the rate of 130 consumed nitrate, silica and phosphate above the seasonal pycnocline (50 m depth, Supplementary 131 Table 1). Microscopic analyses of water samples revealed that phyto-and protozooplankton 132 communities in the chlorophyll a maximum (10 -28 m depth) in the ice-free regions were dominated 133 by Phaeocystis spp., heterotrophic dinoflagellates and ciliates, while the ice-covered regions were 134 dominated by planktonic diatoms and Phaeocystis spp. (Supplementary Table 1). This was reflected 135 in the composition of in situ formed aggregates collected using a marine snow catcher (MSC) 136 directly below the chlorophyll a maximum (60 m depth), where Phaeocystis spp. dominated 137 aggregates of the ice-free region and planktonic diatoms those of the ice-covered regions (Fig. 2). 138 The aggregates from the ice-covered regions were two-fold larger (Wilcoxon Signed-Ranks Test; 139 p<0.01; Table 2) and sank two-fold faster than the aggregates collected in the ice-free regions 140 (Wilcoxon Signed-Rank Test; p<0.01; Table 2). Half of the aggregates collected in the ice-free 141 region (13 out of 24) were smaller than 512 µm in diameter, while almost all (33 out of 36) collected 142 aggregates in the ice-covered regions were larger than 512 µm (Fig. 2). 143 In addition, drifting sediment traps were equipped with a viscous gel to capture and preserve the size 144 and structure of intact settling aggregates. The gel traps confirmed the MSC observations of 145 Phaeocystis spp. dominated aggregates in the ice-free regions and planktonic diatom-dominated 146 aggregates in the ice-covered regions. The gel traps showed similar numbers of particles exported in 147 the ice-free and ice-covered regions, but confirmed that the aggregates in the ice-covered regions had 148 on average two-fold larger diameters than those collected in the ice-free regions (Wilcoxon Signed-149 Ranks Test; p<0.01; Table 2). The larger diameters in the ice-covered regions translated into an order 150 of magnitude larger average volume compared to aggregates of the ice-free regions (Table 2). Hence, 151 in the period assessed in early summer, larger and faster-settling aggregates in the ice-covered 152 regions caused a two-fold higher carbon export compared to the ice-free regions ( Table 2). The 153 carbon to nitrogen ratios (C:N, mol:mol) were 11 in the ice-free regions and 8 in the ice-covered 154 regions, indicating export of fresher material from under the ice (Table 2). Furthermore, at the ice-155 stations macroscopic strands of the sea-ice diatom Melosira arctica were observed by sea ice 156 sampling ( Table 2). Rarefaction curves did not reach a plateau in any of the sampled 166 communities, however, estimated asymptotic extrapolation to double amount of sequences showed 167 only few additional ASVs (Supplementary Figure 2). Thus, our sequencing depth was satisfactory to 168 represent most of the bacterial and archaeal diversity in all sampled microbial communities 31 . The 169 classes Alphaproteobacteria, Bacteroidia and Gammaproteobacteria dominated the microbial 170 communities in both FL and PA fractions, with no differences between ice-free and ice-covered 171 regions ( Supplementary Fig. 3). Each of these classes comprised more than 15% of the sequences 172 and more than 10% of the ASVs in the entire dataset ( Supplementary Fig. 3). In the deep ocean 173 communities (>1000 m) there was an increasing abundance of the clades SAR202 (class 174 Dehalococcoidia), SAR324 (Marine group B), SAR406 (Marinimicrobia), and the archaeal class 175 Nitrososphaeria, each comprising 1-6% of the sequences and 3-6% of the ASVs in the entire dataset 176 ( Supplementary Fig. 3 p>0.01). However, in the PA fraction the communities of the ice-free region had significantly higher 197 dissimilarity along the water column, compared to the PA communities of the ice-covered region 198 ( Fig. 3; Wilcoxon Signed-Ranks Test; p<0.01). 199

Vertical connectivity and shifts in particle-associated communities 200
Many free-living microbes are adapted to colonize particles in the water column. Thus, the observed 201 vertical dissimilarity pattern of the PA communities could be associated with the changing diversity 202 of the FL communities. In order to test this hypothesis and to estimate the extent of colonization, we 203 applied a microbial source tracking (MST) Bayesian algorithm 'SourceTracker'. This MST approach 204 assumes that ASVs diversity in various 'source' (i.e. FL) and corresponding 'sink' (i.e. PA) 205 communities allows identification of statistically probable links between them (for detailed 206 explanation see Methods section). The MST analysis showed a strong effect of the surface and 207 epipelagic FL microbes on the composition of PA communities along the entire water column (Fig.  208 4). Within the surface and epipelagic layers, a particularly high proportion (84±5%) of the PA 209 communities was associated with surface and epipelagic FL communities. In contrast, at meso-and 210 bathypelagic depths the PA communities showed only a weak link to meso-and bathypelagic FL 211 communities (ca. 2 and 8% of the communities, respectively), and a large fraction (72±5%) was not 212 linked to any FL community. However, at meso-and bathypelagic depths, 27±6% of the PA 213 communities in ice-covered and 11±2% of PA communities in ice-free regions were linked to surface 214 and epipelagic FL communities ( Fig. 4; Supplementary Table 3). 215 By statistical tests of comparative sequence enrichment, we identified the microbial taxonomic 216 groups that became significantly more abundant on sinking particles as a function of depth. The 217 ASVs within the PA communities were defined as enriched when they had a log 2 fold change of 218 absolute value higher than 1 (i.e., double the amount of sequences) and an adjusted p value lower 219 than 0.1 (Fig. 5). This test looked at consecutive pelagic layers: surface-epipelagic, epipelagic-220 mesopelagic and mesopelagic-bathypelagic. In both ice-free and ice-covered regions PA 221 communities became enriched with increasing depth in the classes Gammaproteobacteria (with 40  222 and 18 enriched ASVs, respectively), Planctomycetes (with 37 and 27 enriched ASVs, respectively), 223 Bacteroidia ( Bacteroidia were present also in the FL communities, the enriched ASVs of the classes 228 Planctomycetes and OM190 were absent from the FL fraction (<0.5% of the sequences). Overall, we 229 observed larger changes with depth in the PA communities of the ice-free region (where sinking 230 speed was lower), resulting in more than double the amount of PA-enriched ASVs, in comparison to 231 the ice-covered regions (348 and 158 ASVs, respectively; Supplementary Table 4). 232 Transport of surface water-originating microbes to the bathypelagic: water column vs. seafloor 233 Some of the vertically PA-enriched ASVs were also present in the FL communities along the water 234 column (Fig. 6). In the bathypelagic, the PA-enriched ASVs comprised 17±2% of the sequences in 235 the FL communities of the ice-covered region, and 47±4% of the sequences in the FL communities of 236 the ice-free region. The most abundant family that consisted of such ASVs was the archaeal family 237 Nitrosopumilacea, which comprised 3-4% and 6-19% of sequences in FL communities of the ice-238 covered and ice-free regions, respectively. 239 The seeding of the deep-sea sediment by microbes on sinking particles was tested using 7 deep-sea 240 sediment samples (uppermost centimeter) collected at the same stations as the water column 241 communities across the Fram Strait. This dataset consisted of 1,209,785 sequences that were assigned 242 to 11,145 ASVs associated with bacterial and archaeal lineages (Supplementary Table S2; 243 Supplementary Fig. S2). The sediment microbial communities were mainly affiliated to the classes 244 Alphaproteobacteria, Gammaproteobacteria, and Nitrososphaeria ( Supplementary Fig. S3). 245 The vertically PA-enriched ASVs were also identified in the deep-sea sediment communities of both 246 ice-covered and ice-free regions. These shared ASVs between the PA and the sediment communities 247 were associated mainly with the archaeal family Nitrosopumilaceae (17 ASVs) and the bacterial 248 family Woeseiaceae (8 ASVs; class Gammaproteobacteria), each comprised ca. 2-4% of the 249 sequences in the sediment communities (Fig. 6). Interestingly, in contrast to the PA-enriched ASVs 250 of the family Nitrosopumilaceae that were also abundant in the FL communities of the bathypelagic, 251 the shared ASVs of the family Woeseiaceae were absent from the FL communities (<0.3% of 252 sequences in all FL communities). Overall, in the ice-free region, 31% of the PA-enriched ASVs 253 were present in the sediment and comprised ca. 17% of the sequences in the sediment communities. 254 In contrast, in the ice-covered regions 39% of the PA-enriched ASVs were present in the sediment 255 and comprised ca. 11% of the sequences in the sediment communities (Fig. 6) Here, we studied the role of sea ice on settling particle characteristics and vertical microbial 273 connectivity, and postulate links to carbon export efficiency in the Fram Strait. Our long-term 274 assessment of the role of ice-coverage on particle export during periods with sea ice near the HG4 275 station suggested an important function of sea ice-distance on export fluxes early in spring during the 276 ice-influenced phytoplankton bloom period. This encouraged us to assess the underlying principles of 277 this connection between ice-associated export and the fate of microbial communities attached to 278 particles close to the ice margin. At the time of sampling in June-July 2016, the late stage of the 279 ongoing phytoplankton bloom was dominated by diatoms in the ice-covered region while in the 280 adjacent ice-free region it was dominated by Phaeocystis spp. To test how ice-coverage impacts 281 vertical connectivity and export of organic matter, we compared characteristics of sinking marine 282 aggregates from ice-covered and ice-free regions of the Fram Strait during the productive period. In 283 the ice-covered region we found larger diatom aggregates, with two-fold higher size-specific sinking 284 velocities compared to the smaller Phaeocystis spp. aggregates that dominated ice-free regions. This 285 caused almost two-fold higher carbon export rates under the ice, compared to adjacent ice-free waters 286 during the same period. The long-term record in the Fram Strait also shows that annual particle flux 287 is lower during warm water phases with less ice 38 , and characterized by a shift from diatom to 288 coccolithophorid and Phaeocystis spp. dominated phytoplankton during summertime at HG4 289 station 39 . This is similar to observations north of Svalbard where ice-associated diatom production 290 resulted in higher export than that observed for ice-free regions dominated by Phaeocystis spp. 37 . 291 Taken together, this suggests that in the early Arctic summer, fast settling diatom aggregates drive 292 export in ice-covered regions, whereas in warming, Atlantic-water influenced regions, the slower 293 settling Phaeocystis spp. aggregates dominate and will lead to more pelagic recycling. In this study, 294 this also affected carbon to nitrogen ratios of the sinking matter, which were lower for the settling 295 particles collected by the drifting traps in the ice-covered regions compared to the ice-free regions of 296 the Fram Strait (Table 1). Hence, a potential future shift to Atlantification of the Eurasian Arctic 297 basins 40 , with larger areas of thermally stratified open waters, flagellate-dominated phytoplankton 298 blooms, slower settling aggregates and stronger grazing pressure may lead to higher degradation and 299 transformation of organic matter during its journey through the water column, thus, resulting in lower 300 amounts and less labile organic matter reaching the seafloor. 301 In this study we tested, for the first time in Arctic deep waters, the previously established hypothesis 302 that vertical microbial connectivity is stronger in ecosystems dominated by fast-settling 303 aggregates 22,23 , due to the shorter transit time through the water column. In both ice-covered and ice-304 free regions of the Fram Strait, free-living (FL) pelagic microbial communities from different depths 305 had greater dissimilarities to each other than the particle-associated (PA) communities from the same 306 depths. This suggests a stratified water column with distinct microbial communities in the different 307 water layers, as well as a vertical dispersal of microbial communities between surface ocean and 308 deep-sea via sinking particles. In this context, settling particles are not only important for the export 309 of organic matter to the deep ocean, but they also promote microbial heterotrophic activity and 310 seeding 22,23,33 , and thereby shape microbial biogeography and biogeochemical functioning in meso-311 and bathypelagic realms. 312 The surface water-originating microbial families that were significantly enriched on particles 313 collected at depth, such as various members of the class Bacteroidia, are associated with 314 phytoplankton blooms in the region 32,41 , and are known to be highly active organic matter 315 degraders 42 . Furthermore, it has recently been shown that there is a dominance of enzymatic activity 316 phylogenetically linked to these taxonomic groups in the bathypelagic 43 and that this enzymatic 317 activity is predominantly linked to a particle-associated lifestyles 44 . This indicates that active 318 microbes originating from surface waters and associated with sinking particles continue to process 319 organic matter while they sink to the deep ocean, and thus may remain key players in the 320 biogeochemical cycling in the deep ocean.  51,52 . Common to all these processes is that slower sinking will enhance selection of some 346 taxonomic groups and lead to the demise of others, potentially allowing rare taxa to become abundant 347 at depth while those that were abundant at shallower depths become rarer. Furthermore, a large 348 fraction of sinking particles remains suspended in the bathypelagic 53 . In this way, settling aggregates 349 should be viewed as constantly changing microcosms that have some exchange with their 350 surroundings in the deep ocean 26 , but where particle sinking speed is an important driver of 351 succession. In Arctic waters, it seems that fast aggregate sinking speed is strongly related to 352 ecological impact from sea-ice cover. 353 Since the seafloor is the final destination for those particles that make the journey through the water 354 column, we tested whether the vertical microbial connectivity extends to deep-sea sediment. We present throughout the entire water column, comprising <1% of the community. We found that 367 pelagic members of the Woeseiaceae were associated with sinking particles, but not free-living, 368 suggesting that this important benthic heterotroph is one of the few types of bacteria that cover all 369 water depths by a particle-associated life style. 370 In conclusion, our study supports the notion that sea-ice retreat can have an important ecological 371 impact on carbon flux characteristics, and on long term potentially affect the deep-ocean microbial 372 diversity. Fast settling ice-associated diatom aggregates drive higher export efficiency and cause 373 stronger pelagic-benthic coupling including the transport of functionally important microbial groups, 374 whereas slow settling Phaeocystis spp. aggregates associated with seasonally ice-free regions may 375 lead to more pelagic recycling and less connectivity. These changes may substantially alter deep 376 water and seafloor communities in the Arctic. 377

Methods 378
Water sampling and metadata collection 379 respectively. The distance to the ice-edge was defined at the position with 15 % sea-ice 418 concentration. The ice-edge nearest the HG4 position was used to calculate the daily ice-distance and 419 averaged for each opening time of the collection cups (~14 days) on the long-term moored sediment 420 traps. 421

Microscopic analysis of phyto-and protozooplankton 422
The plankton community composition at the chlorophyll a maximum was identified and 423 the phytoplankton abundance was counted using light microscopy. Seawater samples were 424 preserved in hexamethylenetetramine-buffered formalin (final concentration 0.5-1%) and stored in 425 brown glass bottles. For microscopic analyses an aliquot of 50 mL was transferred to Utermöhl 426 settling chambers where the cells were allowed to settle for 48 hours. At least 500 cells of the 427 dominant phytoplankton species or groups were counted with an inverted microscope at three 428 different magnifications using phase contrast according to Utermöhl (1958) and Edler (1979). 429

On-board characterization of marine aggregates and sinking velocity measurements 430
Using a marine snow catcher (MSC, OSIL, United Kingdom) we sampled intact aggregates from 60 431 m at ice-free and ice-covered regions, and measured their size, composition, and sinking velocities. 432 The aggregates were individually transferred to a vertical flow chamber 58 filled with Whatman GF/F 433 filtered seawater collected from the same MSC and kept at in situ temperature. The x-, y-, and z-axes 434 of each aggregate were measured in the vertical flow system using a horizontal dissection microscope 435 and an ocular with a scale. The aggregate volume was thereafter calculated assuming an ellipsoidal 436 shape and the equivalent spherical diameter (ESD) was calculated from the aggregate's volume. The 437 sinking velocity was measured by increasing the upward flow in the flow-chamber until the 438 aggregate was floating one diameter above the net. The sinking velocity was thereafter calculated by 439 determining the volumetric flow rate three times, and dividing the average of these measurements by 440 the area of the flow chamber. The composition of the aggregates was determined with an inverted 441 light microscope using Utermöhl chambers (Fig. 2). 442

Aggregate and carbon export to 100 m 443
Aggregate and carbon export to 100 m depth was measured using the free-drifting surface-tethered 444 sediment traps in the ice-free and ice-covered regions 26  Raw paired-end, primer-trimmed reads were deposited in the European Nucleotide Archive (ENA) 60 474 under accession number PRJEB30254. The data were archived using the brokerage service of the 475 German Federation for Biological Data (GFBio) 61 . 476

Bioinformatics and statistical analyses 477
The raw paired-end reads were primer-trimmed using cutadapt 62  Based on the assumption that the particle-associated microbial communities (i.e., 'sink' 496 communities) are the result of various events of colonization of marine aggregates by free-living 497 microbes (i.e., 'source' communities); a Bayesian microbial source tracking algorithm 498 'SourceTracker' (v1.0) 67 was applied on the ASV abundance table. The algorithm performance was 499 validated using a 'leave-one-out' approach, in which each 'source' (i.e., FL) community was hidden, 500 in turn, from the training dataset, and its origin was predicted based on the rest of the source samples 501 in the dataset. The entire analysis was conducted under default conditions: burn-in period -100, 502 restarts -10, dirichlet hyperparameters (ɑ, β) -0.001. All samples were randomly sub-sampled to 503 5,000 sequences. Scripts for the molecular data processing and statistical analyses can be accessed at 504 https://github.com/edfadeev/Vertical_connectivity_Arctic_Ocean. 505 Modeled aggregates sinking trajectories 506 A Lagrangian particle tracking algorithm was used to back-track particles from the sampling depth to 507 the surface. A detailed description of the model can be found Wekerle et al, 2018 68 . Briefly, the 508 backward particle computation is done by reversing the flow field, i.e. particles are treated as if they 509 were rising from the sampling depth to the surface with a negative sinking speed, being horizontally 510 displaced with the reversed horizontal velocity. Particles were advected with daily averaged 3D 511 model velocities from the ocean general circulation model FESOM (an ocean-sea ice model based on 512 unstructured meshes) 69 . The particle sinking speed was computed by adding a constant sinking speed 513 to the modelled vertical velocity. In this study, we used a FESOM configuration optimized for the 514 Fram Strait, applying a mesh resolution of 1 km 70 . The performance of the model was validated for 515 the sampling time period by oceanographic observations (Supplementary Fig. 5). 516 The backward trajectory calculation was performed for all three sampled regions (ice-free 'HG' 517 stations, and ice-covered 'EG' and 'N' stations), using on-board measurements of aggregate sinking 518 velocities (Table 1). Trajectories were released around 300 m above the seafloor once per day during 519 the year March -July 2016, however we restricted the analysis to particles that reached the ocean 520 surface between March and July 2016. A time step of 30 min was used for the trajectory calculation, 521 and bi-hourly positions and corresponding temperature and salinity values were stored. To quantify 522 the vertical distribution of particles, their positions were binned into a grid with bin sizes of 25 m 523 depth x 0.05° Longitude/Latitude and then divided by the total number of particles to determine the 524 fraction of particles originating from each grid box (  April, plotted as a function of the distance to the ice-edge. The regression shows that there was a 725 significantly negative relationship between the distance to the ice-edge and the magnitude of POC 726 flux (R 2 =0.39, p <0.01). 727   community was conducted estimated using leave-one-out approach (i.e., based on all other FL 750 communities; see methods), and the sources of the particle-associated (PA) communities were 751 estimated based on the FL communities. The ice-covered stations are marked with an asterisk. 752 consecutive depths (surface-epipelagic, epipelagic-mesopelagic and mesopelagic-bathypelagic), 756 ordered according to labels between the panels. The y-axis represents the mean log 2 fold change for 757 microbial families with more than 3 ASVs with log 2 fold change absolute value higher than 1 758 (standard error is smaller than the point). Positive value represent enrichment in deeper water layers 759 and negative value represents enrichment in shallower water layer. The numbers near the symbols 760 represent the number of ASVs enriched in the depth. The x-axis is ordered according to the different 761 taxonomic classes, represented by the color code. 762 represented by colors according to the legend, all classes with sequence proportion below 2% were 766 classified as "Other classes". The ice-covered stations are marked with an asterisk. 767