Along-shelf connectivity and circumpolar gene flow in Antarctic silverfish (Pleuragramma antarctica)

The Antarctic silverfish (Pleuragramma antarctica) is a critically important forage species with a circumpolar distribution and is unique among other notothenioid species for its wholly pelagic life cycle. Previous studies have provided mixed evidence of population structure over regional and circumpolar scales. The aim of the present study was to test the recent population hypothesis for Antarctic silverfish, which emphasizes the interplay between life history and hydrography in shaping connectivity. A total of 1067 individuals were collected over 25 years from different locations on a circumpolar scale. Samples were genotyped at fifteen microsatellites to assess population differentiation and genetic structuring using clustering methods, F-statistics, and hierarchical analysis of variance. A lack of differentiation was found between locations connected by the Antarctic Slope Front Current (ASF), indicative of high levels of gene flow. However, gene flow was significantly reduced at the South Orkney Islands and the western Antarctic Peninsula where the ASF is absent. This pattern of gene flow emphasized the relevance of large-scale circulation as a mechanism for circumpolar connectivity. Chaotic genetic patchiness characterized population structure over time, with varying patterns of differentiation observed between years, accompanied by heterogeneous standard length distributions. The present study supports a more nuanced version of the genetic panmixia hypothesis that reflects physical-biological interactions over the life history.

Genetic structuring and gene flow around the Antarctic. Estimates of genetic differentiation are a principal tool for defining population structure, by assessing gene flow between populations, as well as the extent to which inbreeding and genetic drift contribute to population differentiation 27 . The first hypotheses concerning population structure in silverfish arose from research carried out in the 1980s in the eastern Weddell Sea and Antarctic Peninsula 22,23 . Recent advances in our understanding of hydrography and physical-biological interactions provide a contemporary life history framework in which the population genetic structure of silverfish needs to be considered (see Ashford et al. 6 ).
Highly connected populations resulting in genetic panmixia have previously been the null hypothesis to describe genetic structuring in notothenioids. This hypothesis is based on the expected influence of circumpolar currents on the transport of pelagic larvae, which would sustain high levels of gene flow 28 . This was supported by genetic homogeneity found among multiple species of the families Nototheniidae and Channichthyidae inhabiting the Scotia Arc 29,30 . At the same time however, this panmictic hypothesis was challenged by evidence of population structuring based on genetic heterogeneity in multiple other species, including Dissostichus mawsoni 31,32 , Chionodraco hamatus 33 , Chaenocephalus aceratus 34,35 , and Champsocephalus gunnari 35 .
Evidence for gene flow and genetic panmixia has been similarly variable for silverfish. Using mitochondrial DNA markers, Zane et al. 36 found little indication of population structuring in silverfish on a circumpolar scale, although differences were found between sampling years in the eastern Weddell Sea, and between one sample in the eastern Weddell Sea and one of two taken in the western Ross Sea 36 . A more recent study on silverfish connectivity around the Antarctic Peninsula employing microsatellite markers revealed that fish taken from Marguerite Bay and Charcot Island along the western Antarctic Peninsula (WAP) were significantly different from those along the northern Antarctic Peninsula (NAP) at Joinville Island and on the eastern side of the peninsula at Larsen Bay (LB) 19 (see Fig. 1, WAP, NAP, and LB for place names). Samples were collected over multiple years, and as in Zane et al. 36 , evidence for restrictions to gene flow varied between years 19 .
Variation in recruitment and dispersal, which could explain the observed changes in silverfish distribution, is predicted by the hypothesis of chaotic genetic patchiness 37 , developed to explain the fluctuations in gene flow between sampling years 19 . Current theory emphasizes the role of physical-biological interactions in structuring populations and the potential for extensive along-shelf connectivity. Yet to date, there has been no comprehensive investigation of silverfish gene flow and population structure at different spatial scales and over time 6 , nor re-examination of earlier genetic evidence in the context of recent theoretical advances in understanding.
The present study builds upon the Matschiner et al. 28 hypothesis of genetic panmixia in the light of the physical-biological framework provided by Ashford et al. 6 . It aims to test for population structuring and along-shelf connectivity mediated by large-scale circulation. Employing microsatellite markers, we assessed gene flow between silverfish sampled over 25 years from six different geographic regions around Antarctica ( Table 1). The following regions were investigated based on observed and predicted silverfish larval assemblages: the western Ross Sea 4,12,24 , eastern Weddell Sea 22 , Larsen Bay 26 , northern Antarctic Peninsula 23 , South Orkney Islands 38 , and western Antarctic Peninsula 39 . Inclusion of samples from previous studies allowed for a greater number of spatial and temporal comparisons. The expansion of the Agostini et al. 19 analysis of Antarctic Peninsular connectivity enabled comparisons to be made on a circumpolar scale. Finally, the use of microsatellite markers increased  11 and Whitworth et al. 8 , respectively. Ross Sea (RS) Position of the ASF is approximated based on Ashford et al. 6 Fig. 10.11. Weddell Sea (WS) Position of the AACC and ASF is approximated based on Fig.  2b in Nicholls et al. 56 . Larsen Bay (LB) Putative position of the ASF according to Whitworth et al. 8 . Northern Antarctic Peninsula (NAP) Position of the ACC and AACC are based on Fig. 7 in Orsi et al. 11 and Fig. 3 in Thompson et al. 47 , respectively. Entry of Circumpolar Deep Water from the ACC into the Bransfield Strait through the Boyd Strait is shown according to Savidge and Amft 10 . Yellow and pink arrows within the Bransfield Strait and Trough area represent components of the Bransfield Gyre with Bellingshausen influence and Weddell influence respectively, based on Fig. 15 in Sangrà et al. 87 . South Orkney Islands (SOI) Schematic illustration of the circulation drawn from: ACC (Fig. 7 in Orsi et al. 11 ), ASF (Whitworth et al. 8 , Fig. 6 in Heywood et al. 9 40 . With the present approach, we were able to compare shelf areas located along the ASF from the Ross Sea to the tip of the Antarctic Peninsula, as well as areas located along the ACC from Charcot Island to the South Shetland Islands. In addition, we tested for potential gene flow downstream to the South Orkney Islands along the southern ACC and Weddell Gyre. Finally, standard length (SL) data from over half of the samples collected indicated differences in SL distributions between sampling locations, and we used these to test for chaotic genetic patchiness. Where more than one SL mode existed, sampling locations were further divided into groups of smaller and larger fish to examine for genetic differentiation between and among different cohorts. Although temporal replicates differed between sampling locations, and more locations were sampled in some geographic regions than others, the dataset allowed us to test for genetic panmixia on a circumpolar scale for the first time in silverfish.

Results
Genotyping. Genotypes 10,20 and N e = 1000, 5000, 10000 using both Fisher's and Chi-square tests. Thus, the present dataset has the power to detect differences associated with F ST higher than 0.001. In the current study, all significant comparisons obtained for F ST estimates ranged between 0.00388 and 0.0125, and all near-significant values (for which 0.1 > P > 0.03326, the threshold for significance after SGoF + correction) ranged from 0.00168 to 0.00572. This confirms that all non-significant comparisons can be presumed to be such for the present dataset, and are not non-significant due to the inability of the markers used to detect differences among samples. Hierarchical AMOVA (Table S3) indicated that partitioning the 19 sampling locations into k = 6 groups based on geographic region (Ross Sea (RS), Weddell Sea (WS), Larsen Bay (LB), northern Antarctic Peninsula (NAP), South Orkney Islands (SOI), and western Antarctic Peninsula (WAP), Table 1) maximized between group heterogeneity (global F CT = 0.00159; P = 0.00218 ± 0.00050) while minimizing within group heterogeneity (global F sc = 0.00002; P = 0.63050 ± 0.00429). Genetic differentiation was present (global F ST = 0.00161; P = 0.00644 ± 0.00087) but very low, with <1% of the total variation due to sampling location. Only 0.16% of variation was attributed to variation among groups. Locus-by-locus AMOVA confirmed that no particular locus had a significantly outsized impact on the calculation of global F-statistics.
Clustering methods failed to indicate a clear structure among the sampling locations analyzed. Clustering methods employed by DAPC and FLOCK (without pre-defined groups) randomly assigned individuals to groups, with no correlation to sampling location or geographic region (Fig. 2, S1). DAPC run without a priori parameters suggested the existence of nine clusters (Fig. 2a), with k = 9 selected based on cluster optimization plotted against Bayesian Information Criteria (Fig. S2a). The same DAPC based on the six geographic regions (RS, WS, LB, NAP, 14 in Azaneu et al. 51 ), AACC ( Fig. 3 in Thompson et al. 47 , Fig. 15 in Sangrà et al. 87 ), and WF ( Fig. 6 in Heywood et al. 9 , Fig. 3 in Thompson et al. 47 , Fig. 14 in Azaneu et al. 51 ). Putative position of the ASF according to Azaneu et al. 51 . Western Antarctic Peninsula (WAP) Position of the ACC and AACC is approximated based on both, Fig. 7 in Orsi et al. 11 and Fig. 14  SOI, and WAP), run in an attempt to maximize variation between localities, showed no evidence of cluster formation (Fig. 2b). PCoA on all 19 sampling locations provided limited support for geographic clustering: the first principal coordinate accounted for 39.21% of the variance and clearly distinguished the SOI and WAP samples from all others (Fig. 3a). PCoA on the six geographic regions (RS, WS, LB, NAP, SOI, and WAP) emphasized this delineation further: the first principal coordinate accounted for 81.16% of the variance, and separated the SOI and WAP samples from the rest (Fig. 3b). However, in both cases PCoA failed to distinguish NAP, WS, and LB from one another, and also from the other geographic regions. In contrast to clustering methods, pairwise F ST provided support for geographic population clusters. Within sampling locations, there were no significant differences between samples collected over multiple years (F ST ranged from −0.0025 to 0.0024, P > 0.05 for all comparisons, Table 3), suggestive of genetic stability over time. In addition, within geographic regions (RS, WS, LB, NAP, SOI, and WAP), there was no significant difference between sampling locations (F ST ranged from −0.0051 to 0.0024, P > 0.05 for all comparisons, Table 3), indicative of genetic homogeneity within these regions.
Pairwise comparisons between years and different sampling locations among the geographic groupings displayed limited structure. Patterns of significant differentiation previously found along the Antarctic Peninsula were maintained to some extent (5 significant comparisons out of 36 in the current study and 9 out of 36 in Agostini et al. 19 after SGoF + correction for multiple tests, threshold for significance with 171 comparisons P = 0.0332). Such discrepancies are due to the use of 15 loci instead of 16 loci as in Agostini et al. 19 as well as to changes in significance thresholds resulting from the increased number of comparisons in the present study (nineteen sampling locations versus nine). Indeed, when the dataset of nine sampling locations from Agostini et al. 19

Sampling location
Year Acronym Campaign Coordinates n was reanalyzed with the removal of the problematic locus Ch11230, three differences were found with respect to the original analysis (Table S5). Two of these disparities from the original analysis (the comparisons between CI10 and LB07, and between MB01 and JI12) were also seen in the larger analysis between nineteen sampling locations (Table 3). Variation in differentiation between years in comparisons from the same sampling location, as previously documented in the Ross Sea 36 and the Antarctic Peninsula 19 , was also obtained in the current study. For every location for which samples existed from multiple years, patterns of differentiation varied over sampling years (Table 3). This was especially evident at Marguerite Bay, Joinville Island, the South Orkney Islands, Larsen Bay, and the Ross Sea. Grouping sampling locations by geographic region provided evidence of weak genetic structuring (F ST ranged from 0.0001 to 0.0153, Table 4). Significant pairwise F ST values were found between all groups and the western Antarctic Peninsula group (also seen in Agostini et al. 19 ), as well as between the South Orkney Islands and all groups except for the Weddell Sea, where the comparison was not significant after SGoF + correction for multiple tests (threshold for significance with 15 comparisons P = 0.0142, Table 4). In addition, the comparison between Larsen Bay and the Ross Sea was not significant after correction (P = 0.0312, Table 4).

Comparison by length modes.
Of the 1067 samples analyzed in this study, standard length (SL) data were available for 671 (63%) samples. SL ranged from 3.7-21.5 cm. Post-larvae and juveniles (SL < 9 cm) 3 composed 15% of all fish for which SL data were available. To control for the possibility of bias presented by the wide range of SL represented in the present study, both DAPC and pairwise F ST analyses were performed on SL-derived groups independent of sampling location. Six groupings were tested, separating fish by SL into groups divided by 1, 2, 3, 4, 5, and 6 cm increments (Table S6a). No evidence of clustering was seen in the DAPC analysis of the length groupings (Fig. S3), nor was any pattern of genetic differentiation associated with life stage observed (Table S6b-g).
Standard length (SL) distributions varied considerably between sampling locations (Fig. 4). Agostini et al. 19 observed a pronounced bimodal distribution in the sample taken from Joinville Island 2010 (JI10), with modes at SL = 5 cm and SL = 10 cm. After dividing the JI10 group into small and large fish representative of more and less recent cohorts, pairwise F ST analysis revealed that the small fish (n = 68) and large fish (n = 72) were significantly different. Seven of eight comparisons were significant between all groups and JI10 small, whereas three of eight comparisons were significant between all groups and JI10 large after SGoF + correction for multiple tests 19 (threshold for significance with 190 comparisons P = 0.0453). When this analysis was extended to the sampling locations considered in the current study, this result was largely replicated: six of eight comparisons were significant between the original AP groups and JI10 small, and two of eight comparisons were significant between the original AP groups and JI10 large after SGoF + correction for multiple tests, (Table S7). However, SOI was significantly different from JI10 small (F ST = 0.0045, P = 0.0265) and JI10 large (F ST = 0.0042, P = 0.0312), in contrast to  Table S7 for details). Spatially recurring length modes among sampling locations collected in the eastern Weddell Sea in 2014 provided evidence for episodic connectivity (Fig. 4). Thus, a separate pairwise F ST analysis was performed including only sampling locations collected in 2014 within the eastern Weddell Sea. No significant differences were seen between any of the sampling areas, which included west Filchner Trough (n = 23), east Filchner Trough (n = 27), Coats Land (n = 29), Halley Bay (n = 52), and Atka Bay (n = 25) (F ST ranged from −0.0062 to 0.0027, P > 0.05 for all comparisons, Table S8). With such low levels of differentiation between areas despite the diverse SL distributions, a final pairwise F ST analysis was performed on groupings of fish from the eastern Weddell Sea that were based solely on SL range, independent of sampling location. SL among specimens from the eastern Weddell Sea ranged from 7.0-21.5 cm. First, fish were divided into six length-based groups (<10 cm, 10-12 cm, 12-14 cm, 14-16 cm, 16-18 cm, >18 cm), with group n ranging from 14-43 individuals. When no significant differences were found between length-based groups after SGoF + correction for multiple tests (F ST ranged from −0.0081 to 0.01019), an exploratory analysis of various pools of the six original length-based groups was conducted. The smallest length-based pool was defined as SL ≤ 12 cm and then subsequently as SL ≤ 14 cm. The largest length-based pool was defined as SL < 16 cm and various combinations of the intermediate groups were created. Testing these length-based pools allowed for the control of the influence of group size, and none of the pairwise comparisons were significant after SGoF + correction for multiple tests (F ST ranged from −0.0081 to 0.0103).

Discussion
The present study tests the hypothesis of genetic panmixia 28 in the light of the physical-biological framework provided by Ashford et al. 6 for Antarctic silverfish. Evidence of high gene flow between regions connected by the Antarctic Slope Front (ASF) and associated Antarctic Slope Front Current on a circumpolar scale was accompanied by indications of reduced gene flow in regions not reached by the ASF. The Discriminant Analysis of Principle Components (DAPC), revealed evidence of nine clusters, but these were independent of geographic region. Instead, they may have arisen due to historical extinction and recolonization events predicted by Ashford et al. 6 , related to glacial periods that recurrently structured past populations of silverfish. Overall levels of genetic differentiation were low (maximum F ST = 0.0153), and no significant differences were found from the Ross Sea westward to the northern Antarctic Peninsula. However, limited evidence of genetic structuring was found. We corroborated the findings from Agostini et al. 19 that Charcot Island and Marguerite Bay represent a homogeneous population with significantly reduced gene flow from populations around the AP and beyond, while also discovering significant reductions in gene flow to the population at the South Orkney Islands. These results were broadly consistent with the Matschiner et al. 28 prediction of gene flow facilitated by the large-scale circulation. However, the genetic structuring we found corresponded to discontinuities between major current systems along the shelf. Accordingly, we argue for a more nuanced version of the genetic panmixia hypothesis that incorporates physical-biological interactions over silverfish life history.  I  CI+ I  CI I I I  CI CI CI CI CI CI CI I I I I I I  CI+ CI CI+ CI CI+ CI CI+ CI CI+ I  CI+ I I I I I  CI CI CI CI CI CI CI CI I I I I I I I I  CI+ CI CI CI+ CI CI+ CI I I I  CI I I I  CI CI CI CI CI CI I I I I I I I  CI+ CI CI+ CI CI+ CI CI+ CI I I I I I I  CI CI CI CI CI CI CI CI I I I I I I Fig. 1, WAP for place names) despite having previously been the dominant fish species in these areas since the 1980s 14 . In addition, the South Orkney Islands differentiated from all other regions except for the eastern Weddell Sea. Differentiation between years within certain sampling locations echoed evidence for chaotic genetic patchiness seen in the Ross Sea by Zane et al. 36 and in the Antarctic Peninsula by Agostini et al. 19 . This was further indicated by the variability in SL distributions between years in particular sampling locations, providing evidence for variation in recruitment and dispersal of different cohorts 37 .
The clearest example of interannual variability was seen in the western Antarctic Peninsula (WAP). Specimens from MB01 and MB02 exhibited a bimodal distribution of SL representative of more and less recent cohorts. In contrast, specimens from MB10 and CI10 exhibited a much tighter unimodal distribution around SL = 15 cm, as well as MB11, which had a similarly tight unimodal distribution around SL = 10 cm (Fig. 4). This disparity between cohorts was reflected in patterns of genetic differentiation within the WAP area between years, with MB01 and MB02 differentiating from two of the nineteen other sampling locations, while MB11 differed from six of the nineteen other sampling locations (Table 3). However, position with respect to local hydrographic features must also be considered. Despite being collected only a year before the strongly differentiated MB11 sample, MB10 shares the weak differentiation pattern of nearby MB01 and MB02, which were all collected in the Labeuf Fjord closer to the coast (Fig. 1, WAP, Table 3). This greater connectivity with populations outside of the WAP is consistent with gene flow down the Antarctic Peninsula promoted by the AACC 19 . Furthermore, the position of MB11 along the trough inflow in Marguerite Trough (Fig. 1, WAP) raises the possibility that the sample consisted of fish resulting from connectivity along the slope, where entrainment in the southern ACC potentially transports fish from the Bellingshausen and Amundsen Seas 23 . The influx of migrants from populations further upstream may contribute to the stronger pattern of differentiation found between MB11 and sampling locations beyond the Antarctic Peninsula.  Differences between the previously published Antarctic Peninsula study 19 and the current results are likely attributable to changes in significance thresholds resulting from the increased number of comparisons in the present study, as well as the use of 15 instead of 16 loci due to the removal of one problematic locus. In the original Antarctic Peninsula analysis, Agostini et al. 19 found 9 out of 36 significant comparisons around the Antarctic Peninsula, while in the current analysis, only four out of the thirty-six Antarctic Peninsula comparisons were significant. This was mainly due to weaker differentiation between WAP locations and JI10 and JI12. Discrepancies in patterns of differentiation between the current study and the original investigation of population connectivity in Zane et al. 36 can be attributed to their use of mtDNA as opposed to microsatellite markers to resolve patterns of genetic heterogeneity 40 , in addition to differences in sample size (Table 3). Given the small differences geographically, variation associated with reproduction, differential survival, and recruitment can lead to changes in resolution in subsequent investigations 6 . These results highlight the importance of continued, targeted sampling that incorporates hydrography in order to monitor changes in population structure over time. The integration of multidisciplinary techniques, notably circulation and movement modelling 43 , otolith chemistry 44 , and genomics 45 , will also help in resolving population structure and connectivity over the life history.
Population structure and life history connectivity. The silverfish physical-biological population hypothesis 6 provides a mechanistic explanation consistent with the patterns of gene flow and genetic structuring we observed. When fish in outflows from local trough systems, for instance in the Ross Sea 20 and eastern Weddell Sea 22 , reach the shelf break, entrainment in the ASF and westward transport has the potential to introduce fish into trough systems downstream. Samples from Terra Nova Bay in the Ross Sea collected in 1996 and 1997 were genetically very similar to those collected in the Weddell Sea and Antarctic Peninsula, only differentiating from samples taken along the western Antarctic Peninsula and the South Orkney Islands. This high level of connectivity between regions on opposite sides of the continent may be attributed to two main factors. The first is related to the extent and importance of the Terra Nova Bay silverfish nursery, where silverfish comprise 98% of the ichthyoplankton 4,13 . The second is related to potential silverfish populations along East Antarctica. Despite the extensive geographic purview of this study, significant gaps remain for this circumpolar species, especially along East Antarctica. Substantial distributions of silverfish have been found across the continental shelf along East Antarctica 3 , as well as in association with the Ninnis and Adélie Troughs off Wilkes Land 46 (Fig. S4, East Antarctica). Other trough systems along East Antarctica in Lützow-Holm Bay, Iceberg Alley, Nielsen Basin, and Prydz Bay may also provide suitable habitats for coherent silverfish populations 6 (Fig. S4, East Antarctica). The existence of a network of local populations of silverfish associated with trough systems along East Antarctica provides a mechanism for the exchange of individuals via the westward-flowing ASF. Such a mechanism would   26,47 ; or (2) the ASF, which could transport enough individuals from the eastern Weddell Sea and beyond to render the Joinville population genetically indistinguishable from upstream populations 6,9,48 . Although the path of the AACC south of Joinville Island is undefined, buoyancy forcing is likely to generate connectivity that is shallow and close to the coast 7 . This could result in larval transport north from Larsen Bay which would explain the differentiation between size modes in the Antarctic Sound ( Fig. 1., NAP).
The results of the current study raise the possibility of a coherent silverfish population at the South Orkney Islands. Surveys carried out between 1967-1981 revealed the presence of silverfish larvae at the South Orkney Islands 38 , consistent with a local breeding population. Furthermore, a recent bathymetric analysis of the South Orkney Islands microcontinent found seven trough systems on the northern and southern sides of the islands 49 . Features supportive of a locally breeding silverfish population were found in Signy Trough between Coronation and Signy Island 49 , including active glaciers at the trough head, thought to be necessary for early-life stages 6 ( Fig. 1, SOI). Evidence from otolith chemistry analyses has revealed that fish from the northern Antarctic Peninsula significantly differed from those at the South Orkney Islands, indicating that fish in these areas were sourced from different spawning grounds 50 . This is supported by our findings of genetic differentiation between the northern Antarctic Peninsula and South Orkney Islands populations.
However, the lack of differentiation between specimens from the South Orkney Islands and specimens from the eastern Weddell Sea means connectivity between these two regions cannot be discounted based on these data. The existence of migrants or non-breeding vagrants at the South Orkney Islands would be predicated by a clear and consistent transport system from the eastern Weddell Sea to the South Orkney Islands microcontinent. Although fish carried to the shelf slope along trough outflows in the eastern Weddell Sea can come into contact with the ASF 6 , subsequent transport to the South Orkney Islands microcontinent remains unclear. In addition to the distance along the ASF, fish are potentially exposed to inflows at the mouths of the Hughes, Ronne (Fig. 1,  WS), Jason, and Robertson (Fig. 1, LB) Troughs that can draw them away from the slope towards the inner shelf. Continued entrainment in the ASF north of Larsen Bay carries fish around Joinville Island, where the AACC merges with the ASF. Subsequently, the currents separate again, the AACC forming the southern branch of the Bransfield Gyre 10 (Fig. 1, NAP) and the ASF moving north along the South Scotia Ridge 9,47,51 . However, upon entering the Hesperides Trough, increased influence by Weddell-Scotia Confluence waters combined with local bathymetric complexity impede continued identification of the ASF 51 . Thus, the ASF does not provide a direct pathway to the South Orkney Islands. Alternatively, the Weddell Front, which forms east of the Joinville Ridge and rounds the Powell Basin to reach the South Orkney Islands microcontinent, may represent a path from the Weddell Sea to the South Orkney Islands. Future studies will require an interdisciplinary approach to assess the impact of episodic connectivity on genetic similarity between the South Orkney Islands and the eastern Weddell Sea. The hydrography between these areas needs to be elaborated in order to determine the eventual fate of the ASF. Otolith chemistry analyses 52 and dispersal modelling 53 of fish from the South Orkney Islands and the eastern Weddell Sea will be able to clarify how much the genetic similarity between these regions is due to episodic connectivity into the local population.
Hydrography may also explain the chaotic genetic patchiness observed in the differentiation patterns between the two Larsen Bay samples and the eastern Weddell Sea and South Orkney Islands (Table 3). It is important to note that the LB07 and LB11 samples were collected from different parts of Larsen Bay (Fig. 1, LB). LB07 was collected in the Larsen B embayment off Scar Inlet, near one of the heads of the Larsen Inlet Trough. The LB11 samples were collected in the Larsen A embayment, just north of the Seal Nunataks Ice Shelf, at the head of the Robertson Trough. Since the disintegration of the Larsen A and B shelves, runoff into the Larsen embayments has been derived from smaller glaciers along the peninsula that contribute to disparate patterns of freshwater influx between the Larsen A and B areas 54,55 . Differences in seasonal melting patterns between the two areas are likely to directly impact coastal buoyancy and connectivity between trough systems in Larsen Bay. Considering the evidence of gene flow between the eastern Weddell Sea and both the Larsen Bay and South Orkney Islands, the lack of differentiation between LB11 and SOI11 may be due to changes in the influx of eastern Weddell Sea migrants into these areas. This is in contrast to the hypothesis that the observed gene flow results from the direct  Table 4. Genetic differentiation among Pleuragramma antarctica samples, pooling sampling locations from common geographic regions. Pairwise F ST estimates (above the diagonal) and corresponding P-values (below the diagonal) are shown. Values in bold were significant after correction for multiple tests as implemented in SGoF + (threshold for significance with 15 comparisons P = 0.0142). Numbers in parentheses adjacent to group names indicate the sample size (n). Group acronyms are as in Table 1. The pattern of homogeneity observed among fish across all sampling locations within the eastern Weddell Sea fails to explain the variation in SL distributions. This suggests population mixing between local and migrant fish, seen from Atka Bay in the northeast, to Halley Bay and Coats Land, and finally to the eastern and western sides of the Filchner Trough in the southeast. Aspects of local hydrography help to explain these observations. In the southeast Weddell Sea, the AACC and the ASF are functionally equivalent given the narrowness of the continental shelf in this area 9 . Seasonal and interannual variability in the AACC in the southeast Weddell Sea 56 may promote and restrict connectivity along the shelf from Atka Bay to Filchner Trough. Fish that become entrained in the ASF upstream are introduced to the Weddell Sea east of Atka Bay, as it becomes the southern limb of the Weddell Gyre 8,57 . Having formed in the Amundsen Sea, the ASF has the potential to connect local trough systems in the Ross Sea and East Antarctica 6,8 . While it remains a more consistent feature than the AACC, the position and strength of the ASF is nevertheless sensitive to wind stress forcing on seasonal and interannual time scales in the northwest Weddell Sea 51,58,59 . Such variations in the intensity of the ASF could explain fluctuations in the influx of migrants into eastern Weddell Sea populations and provide a general mechanism for the chaotic genetic patchiness observed among silverfish populations.
While the present dataset is unique in both its temporal and spatial extents, as well as in its inclusion of life history data, it is important to keep in mind its limitations and what types of analyses and sampling could fill the present gaps. Variation in collection gear by research vessel between sampling locations (Table 1) may have introduced bias into the allele frequencies measured between location. However, the high levels of homogeneity found between locations sampled by different research vessels, coincident with differences between locations sampled by the same research vessels, imply that any gear biases did not have a consequential impact on the present results. Subsequent studies analyzing otolith nucleus chemistry in silverfish along the eastern Weddell Sea will allow us to distinguish between areas more heavily composed of locally spawned fish and migrants. Furthermore, enhanced oceanographic data would improve dispersal modelling, allowing for a more targeted approach in defining connectivity hypotheses. Contiguous sampling from regions that could contribute migrating fish to the ACC and the ASF are needed in order to better define in which ways the various current systems influence eastern Weddell Sea population mixing.
Nevertheless our results show no basis for rejecting along-shelf connectivity predicted by Ashford et al. 6   Drygalski Trough. Moreover, the implications of observed structure and gene flow in Antarctic silverfish is seen in other Antarctic species. Krill (Euphasia superba), a keystone species in waters beyond the continental shelf 60 , exhibit a similar lack of population structure, as shown in studies using mtDNA and microsatellites 61 , as well as restriction-site-associated DNA sequencing 62 . Such panmixia observed in Antarctic krill was explained by very large population sizes and ongoing gene flow 61,62 . However, it was stressed that genetically indistinguishable populations do not imply ecologically independent populations 61,63 . Evidence for a recent population expansion observed in analysis of mtDNA in Bortolotto et al. 61 emphasized the differential impact of glaciations on the population structure of pelagic versus benthic Antarctic species. Thus, impacts of climate change may result in range restrictions that will have the opposite effect of glacial expansions on pelagic species, cutting off intermediate populations which can result in cascading effects on the connectivity of ecologically interdependent populations 64,65 . Antarctic silverfish are a pillar of the Southern Ocean ecosystem 2 . Understanding their population structure and connectivity in the context of their life history and the impact of hydrography is critical to assessments of Southern Ocean ecosystem health.

Materials and Methods
Sample collection and DNA extraction. A total of 1067 individuals from 19 sampling locations collected by multiple institutions between 1989 and 2014 were included in the analysis (Fig. 1, Table 1). Of these, 249 individuals collected between 1989 and 1997 from the Ross Sea (RS96, RS97), Halley Bay (HB89, HB91), and South Shetland Islands (SSI96) were included from a previous study on population structure using mitochondrial DNA 36   Standard length (SL) data were available from all individuals except from those collected from RS96, RS97, and HB91. SL data from HB89 and SSI96 represent the length distribution of fish collected at the sampling stations, but lack identification numbers to correspond them to the tissue samples collected for genetic analysis. The distribution of SL was used to identify modes for consideration in the eventual genetics analysis.
Muscle tissue or fin clips were preserved in 95-100% ethanol at the time of sampling. Genomic DNA was extracted from specimens using a standard salting-out procedure 67 . Genomic DNA was extracted from all previously unexamined samples, as well as from tissue samples remaining from individuals on which the mitochondrial DNA analysis was performed in Zane et al. 36 . Concentration and quality of the extracted DNA (260/280 nm and 260/230 nm) was checked using a NanoDrop UV-Vis spectrophotometer (Thermo Scientific) prior to PCR amplification. All extracted DNA was of high enough quality to use in subsequent PCR reactions. DNA amplification and microsatellite genotyping. Individuals were genotyped using 16 published EST-linked microsatellites developed in Chionodroco hamatus 68,69 , that had been previously shown to cross-amplify successfully in P. antarctica 19 . The 16 loci were amplified in two multiplex PCR reactions (see Table S2 for primer sequences and final conditions for all loci). Multiplex PCR reaction volume was 10 µL, containing 1x QIAGEN Multiplex PCR Master mix (HotStartTaq DNA Polymerase, Multiplex PCR Buffer, dNTP Mix; QIAGEN, Hilden, Germany), 0.2 µM primer mix and 100 ng of genomic DNA. The PCR amplification profile for all loci consisted of: (1) an initial activation step of 15 min at 95 °C; (2) 30 cycles of denaturation at 94 °C for 30 s, annealing at 57 °C for 90 s, and extension at 72 °C for 60 s; and (3) a final extension of 30 min at 60 °C.
PCR products were prepared for microsatellite genotyping and sent to an external service (BMR Genomics, http://www.bmr-genomics.com/), where they were sequenced using an ABI 3730xl automated sequencer (LIZ 500 as internal size ladder, Applied Biosystems, Waltham, MA, USA). Microsatellites were analyzed using GeneMarker ver. 2.6 (SoftGenetics). Genotypes of individuals included in the microsatellite analysis published by Agostini et al. 19 were integrated into this analysis. To ensure that the datasets were comparable, two individual samples from Agostini et al. 19 were processed alongside samples from the present study as positive controls during each amplification and sequencing run. This allowed the comparison of the raw microsatellite data between the previous and current study and confirmed that there was no change in microsatellite sizing. Binning was automated with FlexiBin 2 70 and refined by eye to assure accuracy with the corresponding binnings established in Agostini et al. 19 .

Data analysis.
A suite of programs was used to assess population differentiation and genetic structuring based on the microsatellite genotypes. Input files were created in the appropriate software-specific formats using CREATE ver 1.37 71 . Linkage Disequilibrium (LD) probability was tested with GENEPOP online 72 . Deviations from Hardy-Weinberg equilibrium were calculated with the R package diveRsity ver. 1.9.89 73 . The software Micro-Checker 2.2.3 was used to detect the presence of null alleles and scoring errors due to large allele drop-out and stuttering 74 . Significance levels for multiple comparisons were adjusted using the program SGoF+ 75 .
Detection of loci under selection was carried out using the methods implemented in LOSITAN ver. (SMM) 19 . In ARLEQUIN, genetic structure was assessed by running 50,000 simulations, simulating 19 groups (for each sampling location) and 6 groups (corresponding to the number of geographic regions), and assuming 100 demes per group under the hierarchical island model 78 .
Within-sampling location genetic variability was assessed by calculating observed (H O ) and expected (H e ) heterozygosity, number of alleles (N A ), and allelic richness (A R ) using the R package diveRsity ver. 1.9.89 73 . Genetic variability was compared between sampling locations using a one-way ANOVA performed in R 79 .
The extent to which sampling locations could be clustered into the same group was assessed using various methods. Discriminant analysis of principal components (DAPC), a multivariate method in which fish are assigned to known groups, and Principle Coordinate Analysis (PCoA), a multidimensional scaling method in which the matrix of dissimilarities was built from F ST comparisons, were carried out using the R package ADEGENET 2.0.1 80 and GenAlEx 6.501 81,82 , respectively, to provide a visual assessment of differentiation between sampling locations. Estimation of the number of groups into which genotypes from the various sampling locations cluster, ignoring a priori information about provenance, was tested using FLOCK 3.1 83 . FLOCK employs a non Markov Chain Monte Carlo (MCMC) algorithm that uses an iterative method to divide genotypes into k clusters. This method has been shown to allocate clusters more accurately than common Bayesian methods when population structure is weak 84 .
The power of microsatellite loci to detect differences in the sample sizes analyzed was assessed applying Fisher's and Chi-square tests in POWSIM ver 4.1 85 . Observed sample sizes and allele frequencies taken from the 19 sampling locations were used to gauge the statistical power of the 15 microsatellites to detect F ST values ≤ 0.01 using 1000 replicates. In accordance with the POWSIM manual, several combinations of N e (1000-10,000) and generations (t = 2-201) were tested to account for variability in the estimation of N e .
Population differentiation was determined by calculating pairwise genetic distances represented by F ST 86 between sampling locations as well as between SL-defined groups. Hierarchical analysis of molecular variance (AMOVA) was also used to assess population differentiation. Pairwise genetic differences and AMOVA were calculated in ARLEQUIN 3.5.1.3 77 . Significant differentiation was determined with 10,000 permutation tests. The significance threshold was adjusted according to SGoF + correction for multiple tests 75 .

Data Availability
The datasets generated and analyzed during the current study are available from the corresponding author on request.