Drivers of interannual variability in virioplankton abundance at the coastal western Antarctic peninsula and the potential effects of climate change

An 8-year time-series in the Western Antarctic Peninsula (WAP) with an approximately weekly sampling frequency was used to elucidate changes in virioplankton abundance and their drivers in this climatically sensitive region. Virioplankton abundances at the coastal WAP show a pronounced seasonal cycle with interannual variability in the timing and magnitude of the summer maxima. Bacterioplankton abundance is the most influential driving factor of the virioplankton, and exhibit closely coupled dynamics. Sea ice cover and duration predetermine levels of phytoplankton stock and thus, influence virioplankton by dictating the substrates available to the bacterioplankton. However, variations in the composition of the phytoplankton community and particularly the prominence of Diatoms inferred from silicate drawdown, drive interannual differences in the magnitude of the virioplankton bloom; likely again mediated through changes in the bacterioplankton. Their findings suggest that future warming within the WAP will cause changes in sea ice that will influence viruses and their microbial hosts through changes in the timing, magnitude and composition of the phytoplankton bloom. Thus, the flow of matter and energy through the viral shunt may be decreased with consequences for the Antarctic food web and element cycling.


Introduction
Viruses are abundant and active components of seawater that primarily infect the numerically dominant bacterioplankton and eukaryotic marine microbes (Brussaard, 2004;Weinbauer, 2004). As host-specific pathogens they exert control over assemblage composition (Brussaard, 2004;Suttle, 2005), and via gene transfer may serve as genetic reservoirs for marine microbes (Dinsdale et al., 2008). Virus-mediated lysis of host cells generates substrates for bacterioplankton, sequestering matter and energy in the microbial component of marine food webs (Middelboe, 2000;Suttle, 2007;Sheik et al., 2014). Thus, viral activity affects the productivity of higher tropic levels and the efficiency of organic carbon transfer out of the surface ocean (Suttle, 2007;Brussaard et al., 2008a).
As in many other marine environments, the virioplankton in the waters surrounding Antarctica have been shown to be an active component of the ecosystem (Guixa-Boixereu et al., 2002;Pearce et al., 2007;Thomson et al., 2010;Evans and Brussaard, 2012;Brum et al., 2016). Nevertheless, they are omitted from most current models of the Southern Ocean food web (e.g., Murphy et al., 2013) due to a lack of knowledge of the way virioplankton respond to the extreme seasonality in polar waters, and more importantly the pronounced year-to-year variations in the productivity of these systems. Intense seasonal phytoplankton blooms that are controlled by marked variations in irradiance and sea ice dynamics form the base of the productive Antarctic marine ecosystem (Ducklow et al., 2007;Clarke et al., 2007;. In turn, fluctuations in phytoplankton abundance and primary productivity are thought to control the activity and dynamics of the bacterioplankton community via the supply of photosynthate (Ducklow et al., 2012). Differences in the observed levels of coupling between primary and secondary production in polar ecosystems have led to the suggestion that photosynthate may have variable bioavailability (Moran et al., 2002;Bird and Karl, 1999), or that top-down control might have greater significance in controlling bacterioplankton dynamics (Ortega-Retuerta et al., 2008). Alternatively, the release of organic carbon by phage lysis of bacterial cells has been suggested as a substrate source for 4 bacterioplankton in Antarctic waters, which could weaken the link between primary and secondary producers (Evans and Brussaard, 2012).
The Western Antarctic Peninsula (WAP) is an important focus of research activity, due to the fact that during the instrumental period it has been one of the most rapidly warming regions on Earth (Smith et al., 1996;Vaughan et al., 2003;Meredith and King, 2005). The rapid warming has recently been observed to have paused (Turner et al. 2016), and the trajectory/timing of future climate change is not clear. Nonetheless, a warming due to the influence of increasing atmospheric greenhouse gases is expected, and a mechanistic understanding of WAP marine processes is required in order to obtain insight into the potential impacts of that warming. The regional effects of climate change have been observed at various trophic levels of the food web, from shifts in the composition of phytoplankton assemblages (Moline et al., 2004;Montes-Hugo et al., 2009) through to reductions in the numbers of Adélie Penguins (Pygoscelis adeliae; Fraser and Trivelpiece, 1995). Because of their short generation times and free-floating planktonic lifestyles, microbes are expected to react rapidly to environmental conditions (Hays et al., 2005), and their sensitivity to projected changes in polar systems has been highlighted (Kirchman et al., 2009;Ducklow et al., 2010).
Eulerian, multi-year time series are a powerful tool with which to dissect the web of factors controlling functional microbial groups in marine systems, thereby helping to identify the potential implications of ongoing climate change. This approach has been successfully applied to unraveling the mechanisms controlling bacterioplankton and phytoplankton dynamics adjacent to the WAP (Clarke et al., 2007;Ducklow et al., 2012;Venables et al., 2013). The importance of this endeavor to marine virus ecology has been highlighted (Breitbart, 2012), and multi-year time series have improved our understanding of the virioplankton from open ocean and coastal environments (Parsons et al., 2012;Pagarete et al., 2013;Hurwitz et al., 2014). However, the interannual dynamics of the virioplankton prevalent in polar coastal waters, and their driving forces are as yet undetermined. Here we present an eight-year virioplankton 5 dataset with unprecedented temporal resolution from a climatically-sensitive WAP site (Fig. 1). Our aims are to establish: (1) the seasonal and inter-annual dynamics of virioplankton; (2) what linkages exist between the physiochemical and biological parameters, with a view to unravel the factors driving virus dynamics, and how viruses might in turn influence ecosystem function; (3) a baseline against which future virus data might be compared; and finally (4) how future changes associated with predicted greenhouse-gas driven warming at the WAP might impact on the virioplankton, and in turn affect ecosystem functioning.

Study site and physicochemical setting
All data for this study were collected off the Western Antarctic Peninsula at the Rothera Oceanographic and Biological Time Series (RaTS) sampling sites #1 (67.570ºS 68.225ºW) and occasionally #2 (67.581ºS 68.156ºW), both situated in Ryder Bay, Adelaide Island (Fig. 1).
[insert figure 1 -map] Weekly averages of the environmental conditions during the 8-year sampling period are presented in the composite plots shown in Figure 2, while more detailed time series plots of individual parameters to showcase inter-annual variations are shown in Figure 3. The strong seasonality in the physiochemical conditions at the RaTS site is due to its high latitude location (see additional discussion in the Supplementary Materials). A substantial ice cover (pancake or pack ice, with fast ice developing during most years), low water temperatures of -1.7±0.1°C, salinity levels of 33.7±0.2 and a deep but variable mixed layer at 65±30 m are characteristic of winter conditions which commence between April and June. Summer conditions return between early October and early January: the sea ice cover rapidly decreases (to brash or ice-free) and a stratified water column develops with a mixed layer depth of less than 20 m and water temperatures of up to 1.8°C (median 0.7°C). Nutrient concentrations are highest 6 during winter and decrease during the phytoplankton spring bloom, although this decrease is much less pronounced for silicate than it is for phosphorous and NOx.
[insert figure 2 -physicochemical averages] It should be noted that the situation described here is an average over the observed 8 year time period, and that significant inter-annual variations exist in the timing and magnitude of these changes (Fig. 3). chlorophyll a concentrations reached maxima of up to 27 µg L -1 , while for the remaining four summers the maxima were substantially lower at 8-18 µg L -1 (Fig. 5A). Each year the phytoplankton bloom declines sharply to low levels (<1 mg m -3 ) at the end of summer (median during the first week of April), and continues to slowly decrease to winter concentrations throughout the autumn.
[insert figure 4 -biology averages] As with the primary producers, the bacterioplankton (bacterial and archaeal cells) show a strong seasonal cycle in most years. Bacterioplankton numbers start to increase from winter minima of 2.0±0.8 x 10 8 cells L -1 around the same time as the phytoplankton, albeit not as rapidly and not reaching maximum values until 2-3 weeks later ( Fig. 4B; lags were calculated from average number of days 7 between first maxima of the spring bloom, as well as subsequent maxima later on in the season). These maxima vary strongly from year to year, being low in the summers of 2003-2005 (about 0.5 x 10 9 cells L -1 ), consistently high throughout the summers of 2005-2007 (about 2.0 x 10 9 cells L -1 ), and fairly high but more variable in the remaining summers (average 1.5 x 10 9 cells L -1 ) (Fig. 5B). Bacterioplankton numbers decline later and much less sharply than the phytoplankton, and generally do not reach winter values until June. The bacterioplankton community was subdivided into cells with high and low DNA content (HNA and LNA, respectively) according to their green fluorescence after staining with SYBR Green (Gasol et al., 1999). The HNA subgroup is on average more abundant throughout the year than the LNA subgroup, constituting around 70% of total bacterioplankton in spring and summer, and between 50 and 60% during winter (Fig. 4D). However, in some years the LNA subgroup was more abundant,  Figure S2), in accordance with previous results from the Southern Ocean (Brussaard et al., 2008b;Evans et al., 2009). Viral subgroup V1 (lowest fluorescence) is generally 8 the most abundant group (26-93%), followed by V2 (6-60%) and V3 (highest fluorescence; 1-14%) (Fig.   4E). Although the magnitudes of change between summer and winter, as well as between different years, are similar for all three viral groups, the dynamics of V1 are somewhat different from those of V2 and V3 (Fig. 5F-H). During summer the virioplankton community is dominated by V1, but during most winters the relative abundances of V2 and V3 increase to equal that of V1 (Fig. 4E), although this is not the case in every year.
The virus-to-bacteria ratio (VBR) is generally around 20-30, although in some years it increases up to ~90 during summer and coinciding with the increases in bacterio-and virioplankton numbers (Fig. 5D).

Inter-annual variation in plankton dynamics and correlation with environmental data
Multivariate analysis was performed to determine trends in the inter-annual variation in the planktonic abundances and community composition, and to find the environmental drivers behind these dynamics.
As detailed in the Experimental Procedures, the time series was split into 16 seasonal segments (summer/winter) and the integrated values for each segment were used for statistical analyses.
Principal coordinates ordination (PCO) of the biological data shows that 87.8% of its variation can be accounted for by two axes, which represent the seasonal and year-to-year differences at the site (Fig. 6).
The strong seasonal pattern in chlorophyll concentrations and bacterio-and virioplankton abundances (low in winter and high in summer accounts) already explains 72.5% of the variation. This pattern is confirmed by the vector plot overlays, and yields two significantly different clusters (i.e., summer and winter) along the horizontal axis (ANOSIM test: p<0.01, 9999 permutations). The vertical axis explains a further 15.3% of the variation and represents year-to-year differences in the composition of the plankton community, driving the abundances of viral group V1 and HNA bacteria on the one hand, and 9 phytoplankton and viral group V3 on the other, with viral group V2 and LNA bacteria being intermediate.

[insert figure 6 -PCO]
In order to identify which environmental parameters could be driving these dynamics at the WAP the biological clustering results were first correlated with the available physicochemical data (vector plot in Fig. 6). As expected, temperature and ice cover correlate with the seasonal trend (PCO axis 1), as does phosphate. The factors that also correlate with the secondary PCO axis are silicate and NOx (both positive), salinity and MLD (both negative). Distance-based linear models (distLM) were used to further interrogate these results. Forward selection in combination with the Akaike Information Criterion (AIC) criteria selected salinity, phosphate, silicate and to a lesser extent NOx as the most influential variables.
A distLM comprising these four variables was able to explain nearly 70% of the total variation in the biological community (Fig. 7).

Drivers of virioplankton at the WAP
Virioplankton abundances at the coastal WAP showed a pronounced seasonal cycle with interannual variability in the timing and magnitude of the summer maxima. Viral abundances are primarily linked to the type and availability of their microbial hosts, although they may also be influenced directly by shifts in environmental conditions (Mojica and Brussaard, 2014). As the numerically dominant hosts, bacterioplankton abundance was the most influential factor in virioplankton dynamics (Weinbauer, 2004), with shifts in the former generally mirrored by the latter following a lag of approximately one week. Bacterioplankton production in coastal Antarctic waters is regulated indirectly by the levels of phytoplankton biomass accumulation and rates of primary production, as this determines substrate 10 availability (Ducklow et al., 2012). In turn, the temporal variability of primary production, and thus phytoplankton biomass, at high latitudes is constrained by sea ice cover and seasonal duration during winter. This in turn influences the timing and magnitude of the spring-summer blooms, the stability of the water column, and the availability of light and iron (Vernet et al., 2008;Venables et al., 2013Saba et al., (2014). Consequently, one would expect to see a strong relationship between chlorophyll, bacterioplankton and in turn viral dynamics, driven primarily by the sea ice conditions of the preceding winter and subsequent summer mixing. Our data revealed that the coupling of bacterioplankton and virioplankton to chlorophyll is variable, with the magnitude of the chlorophyll for a given year not necessarily predicting the magnitude of bacterioplankton yield for that same year. Thus, factors other than those which determine phytoplankton accumulation are implicated as indirect drivers of virioplankton via their influence on bacterioplankton and these are discussed below.
Decoupling between primary and secondary production in the coastal Antarctic has been previously reported (Bird and Karl, 1999;Moran et al., 2001;Duarte et al., 2005), although as the time series confirms, it is not the general state for this ecosystem (Ortega-Retuerta et al., 2008;Ducklow et al., 2012). Low temperature preclusion of optimal substrate assimilation has been proposed as an explanation for low microbial rates, and thereby decoupling, in high latitude ecosystems (Pomeroy and Wiebe, 2001), but we found temperature not to be a significant driver of the microbial system. Instead, its variability appears to relate to levels of the macronutrients NOx, phosphate and silicate. In coastal Antarctic waters, macronutrients are considered to be replete; however drawdown to potentially limiting levels can occur during periods of high primary productivity (Ducklow et al., 2007;Alderkamp et al., 2012). It is interesting to note that whilst levels of dissolved inorganic nitrogen and phosphorous strongly co-varied, silicate levels were more distinct (Figs 3 and 6). The level and stoichiometry of nutrient drawdown is controlled by the productivity and composition of the phytoplankton community, which in coastal Antarctic waters varies annually and is made up primarily of diatoms and small 11 flagellated cells, particularly the Prymnesiophyte Phaeocystis antarctica (Garibotti et al., 2005;Clarke et al., 2008;Annett et al., 2010). The distinction in silicate's influence from that of the other macronutrients could indicate that the phytoplankton composition, i.e. the relative contribution of diatoms (Kim et al., 2016), affects bacterioplankton dynamics, and thereby indirectly the virioplankton.
This influence is most likely mediated by variations in the quantity and/or bioavailability of organic matter generated by the primary producers, which is proposed to vary according to phytoplankton taxa (Moline and Prezelin, 1996), and may affect bacterioplankton biomass (Ducklow et al., 2012) andcomposition (Tada et al., 2013). Indeed, the bioavailability of photosynthate to secondary producers has been proposed as an alternative explanation for the decoupling of primary and secondary producers in Antarctic waters (Fiala and Delille, 1992). Drawdown of macronutrients by phytoplankton requires iron to be available, and interannual variation in summer macronutrient minima could be indicative of variable iron supply from melting glaciers (Gerringa et al., 2012). Iron is known to be a limiting factor to Southern Ocean bacterioplankton communities (Church et al., 2000), and variations in its availability may exert indirect influence over virioplankton dynamics.
Throughout the seasonal cycle, the highest VBR values were observed in the summer, most likely as a result of higher burst sizes (yield of viruses per host cell lysed) and rates of virally-mediated mortality in the productive summer months (Brum et al., 2016). Maximal substrate inputs by primary producers during the summer months stimulates metabolic activity in bacterioplankton (Ducklow et al., 2012), a parameter predictive of burst size (Lenski, 1988;Weinbauer and Holfe, 1998;Middelboe, 2000).
Bacterioplankton diversity is also known to decrease from winter to summer (Ghiglione and Murray, 2012;Grzymski et al., 2012), which will increase the likelihood of successful infection in the summer (Murray and Jackson, 1992;Weinbauer and Holfe, 1998). Hence, water column stability reduces bacterioplankton diversity and in turn, enhances viral production rates (Winget et al., 2011); thereby explaining the lower viral yields observed in the deeply mixed years of the time-series (2008-2011).
12 Furthermore, lower host abundances in winter promote a shift towards lysogenic over lytic viral infection (Brum et al., 2016), reducing levels of viral production and VBR, as has been shown for the Southern Ocean (Evans and Brussaard, 2012).
Tight control of bacterioplankton by top-down factors has been suggested as an alternative explanation for the lack of coupling between primary and secondary production in coastal Antarctic waters (Ortega-Retuerta et al., 2008). Typically, top-down control is thought to be mediated by grazers (Bird and Karl, 1999;Duarte et al., 2005), although viruses have also been suggested to fulfill this role (Guixa-Boixereu et al., 2002). Indeed high levels of phage activity are indicated by the close coupling of the bacterio-and virioplankton dynamics throughout the RaTS time series (Figs 4, 5 and 6) and supported by reports of virus-mediated mortality in the waters surrounding Antarctica (Guixa-Boixereu et al., 2002, Evans andBrussaard, 2012;Brum et al., 2016). Furthermore, as lysates (the hosts cellular contents released during viral lysis) serve as bacterial substrates (Middelboe et al., 1996;Wilhelm and Suttle, 1999;Suttle, 2007) high levels of viral infection may reduce bacterioplankton's reliance upon phytoplankton derived substrates. This may further serve to decouple primary and secondary production in the waters around Antarctica (Evans and Brussaard, 2012).

Composition of the viral community
The V1-V3 virus groups all accumulate and decline in line with the wax and wane of primary and secondary producers (Figs. 4 and 5), although their dynamics, particularly those of V1 and V3, appear to be independent of one another (Fig. 5G-H) and their relative contribution to the total viral assemblage varies from year to year. V1 is the most abundant, closely following the dynamics of the bacterioplankton (HNA and LNA), whereas V3 partitions more closely with chlorophyll (Fig. 6). Marine viral assemblages are diverse, but are presumably dominated by those viruses pathogenic to the most abundant microbial hosts (Angly et al., 2006). Our results align with previous findings from polar oceans that suggest the numerically dominant, low fluorescence group V1 is most likely to  (Payet and Suttle, 2007;Brussaard et al., 2008b;Evans et al., 2011). During the three summers where bacterioplankton reached their highest levels (2005-2006, 2006-2007 and 2009-2010 [ Fig. 5B]), VBR was highest when compared to the remaining years (Fig. 5D), and V1 appeared to make up a far greater proportion of the total viral assemblage (Figs 5F-H), supporting the hypothesis that V1 may be bacteriophages. Interestingly, during the summer months when bacterioplankton exhibited lowest abundances (summers 2003-2004), V1's contribution to the total virioplankton was also at a minimum. In addition, the VBR was at its highest, suggesting that virus production was from other (eukaryotic) sources. This variability in the availability of potential microbial hosts directly influenced both the abundance and composition of the virioplankton in the coastal WAP waters.

Viral control of microbial populations
Multivariate statistical analysis revealed stronger association of the V1 group with the HNA compared to the LNA, consistent with reports that suggest viral infection of HNA results in larger burst sizes than LNA, and likely accounts for the majority of viral production (Payet and Suttle, 2008;Bouvier and Maurice, 2011). HNA have been proposed to have a higher metabolic activity relative to LNA (Lebaron et al., 2001), although this is not always the case (Zubkov et al., 2001;Longnecker et al., 2005). Differences between the two categories have been attributed to their growth states, as studies indicate they are not always phylogenetically distinct (Lebaron et al., 2002, Longnecker et al., 2005). Yet, Zubkov and coworkers (2001) found that flow cytometric subgroups of prokaryotes were affiliated with different phylogenetic clusters.
Throughout the time series we saw a seasonal shift in the relative abundance of the two bacterioplankton groups: while they were more or less equally abundant in winter, HNA dominated 14 during summer, at times comprising >90% of bacterioplankton (Fig. 5C). This result could be consistent with the idea that HNA are more metabolically active, as they appear to dominate in the months when more substrates would be available via either primary producers, or mortality of secondary producers.
However, along the Antarctic Peninsula LNA cells have been shown to relate more closely to secondary production than HNA, indicating nucleic acid content is an unreliable proxy for metabolic activity (Ortega-Retuerta et al., 2008). Thus, it seems more likely that the seasonal shifts in H/LNA ratio reflect changes in the composition of the bacterioplankton community. Metagenomic assessment of the prokaryotes from the WAP supports this theory, as community composition differs from summer to winter (Grzymski et al., 2012). Hence, the weaker relationship between V1 and LNA relative to HNA could be interpreted to indicate that bacterioplankton clades with smaller genomes were subject to less viral infection than those with larger genomes. While we do not have molecular identification of the bacterioplankton groups, we speculate that the primary phylotype of the LNA group may belong to the SAR11 clade since it is identified as a dominant component of the bacterioplankton assemblage at the WAP (Grzymski et al., 2012), is characterized by a small genome (Giovannoni et al., 2005) and has been shown to make up the vast majority of LNA in marine waters (Gómez-Pereira et al., 2013). SAR11 has been proposed to be a defense specialist, allowing it to co-exist with its phage and maintain its abundance (Suttle, 2007;Zhao et al., 2013), and evidence from a time series in the Sargasso Sea suggests its viruses comprise a disproportionally small percentage of the total virioplankton (Parsons et al., 2012). Our results would then support the hypothesis that viral infection rates of this clade are low.
In this case the weaker correlation between V1 and LNA (Fig. 6) may be more indicative of a codependency of both on HNA, as a host for the former and potentially as substrate source for the latter.
Alternatively, the greater abundance of HNA relative to LNA potentially masked viral production from the latter, a phenomenon that would be compounded by lower burst sizes (Bouvier and Maurice, 2011).

15
Conversely a tight control of HNA by viruses, and particularly V1, is apparent from their pronounced covariation, the most striking example of which occurred during February to March 2010. A spike in HNA abundance was observed in concert with a spike in V1 abundance (offset by approximately one week), indicative of a viral-induced bacterioplankton bloom crash (Fig. 8). Oscillations such as these could be interpreted as the rapid multiplication of a specific host strain, followed by the proliferation of a virus pathogenic to that strain by infection and death of the host, making the niche available to resistant strains. This model of predator-prey interactions has been proposed as a theory which explains the diversity in microbial communities and is termed the 'Kill the Winner' hypothesis (Thingstad and Lignell, 1997;Thingstad, 2000). By inference, the HNA is likely composed of a diverse bacterioplankton assemblage, containing some or all competition strategists (Suttle, 2007), and whose dominance is variable and determined, at least in part, by viruses. Interestingly, though perhaps unsurprisingly, viral and prokaryote communities' coexistence appeared more stable throughout the winter months when bacterioplankton diversity (Grzymski et al., 2012), and thus host diversity is greater. However, as the Kill the Winner model of virus-host dynamics has also been linked to the maintenance of susceptible hosts at lower levels of abundance (Bouvier and del Giorgio, 2007), it is plausible that under the less dynamic conditions of winter, viruses aid in maintaining bacterioplankton diversity.

Outlook: virioplankton at the WAP in a changing climate
Understanding the dynamic aspects of ocean microbial communities in response to seasonal processes can inform predictions of their response and resilience to environmental perturbation (Giovannoni and Vergin, 2012). Our findings imply linkages between the physiochemical conditions and virioplankton activity, mediated by effects on the functioning of the microbial food web. Interestingly, viral activity, as mediated by bacterioplankton dynamics and likely also diversity, was not directly coupled to 16 phytoplankton biomass. Instead, our results are suggestive of the influence of phytoplankton assemblage composition, as indicated by differences in inter-annual nutrient stoichiometery. The longterm warming along the WAP (which has recently weakened however [Turner et al., 2016]) is causing a shift from a maritime-Antarctic climate to a warmer sub-Antarctic-type climate. The resultant reductions in sea ice extent, concentration (e.g., Holland, 2014) and seasonal duration (Stammerjohn et al., 2012) have implications on the composition of phytoplankton assemblages (Moline et al., 2004;Montes-Hugo et al., 2009), as well as reductions in the magnitude of phytoplankton blooms (Venables et al., 2013).
Phytoplankton species are predicted to shift from large diatoms to smaller forms (Falkowski and Oliver, 2007) with consequences for the structure and activity of Antarctic bacterioplankton assemblages (Ghiglione and Murray, 2012), likely mediated by the supply of bacterial substrates (Kirchman et al., 2009). Our findings suggest these shifts will have implications for the activity and degree of virioplankton influence. This will have consequences for the flow of matter and energy through the viral shunt and their availability to the Antarctic food web (Suttle, 2005), as well as the cycling of elements such as iron and carbon (Poorvin et al.,2004;Evans et al., 2009). For example, viral infection has been proposed to either reduce the efficiency of the biological pump by converting particulate to dissolved organic matter, or to promote carbon sinking by stimulating particle aggregation (Brussaard et al., 2008a). Thus, future shifts in viral activity may influence the WAP's ability to act as a CO 2 sink. Viruses are dynamic and active components of the coastal waters of the WAP and wider marine Antarctic environment and we recommend their inclusion in models of Antarctic ecosystem functioning. Given the apparent sensitivity of viral activity to potential future changes at the WAP (a region noted already for strong climatic sensitivity), further study is warranted to quantify viral-mediated mortality and determine its role in element cycling and the functioning of the Antarctic food web. Furthermore, our study highlights the need for future work to assign identities to the main players within the viral and Page 16 of 76

Wiley-Blackwell and Society for Applied Microbiology
This article is protected by copyright. All rights reserved. microbial communities studied, and in particular to confirm that SAR11 is the dominant component of the LNA at the WAP, in order to better understand their ecological and biogeochemical role.

Study site and sampling
The Rothera Oceanographic and Biological Time Series (RaTS) sampling site is situated at the southeastern point of Adelaide Island, adjacent to the Western Antarctic Peninsula (WAP; Fig. 1). The primary RaTS site (Site 1) is located at 67.570ºS 68.225ºW in Ryder Bay, approximately 4 km from the British Antarctic Survey's Rothera Research Station, and has a water depth of 520 m. At times when (partial) sea ice cover precludes access to Site 1 by either boat or skidoo, sampling is performed at RaTS site 2, which is located at 67.581ºS 68.156ºW and has a water depth of 300 m. The two sites exhibit no significant difference in seasonal cycle or water column characteristics and are therefore considered

Bacterio-and virioplankton abundances
For enumeration of viruses and bacteria, 2 mL seawater samples were fixed with a final concentration of 0.5% glutaraldehyde (EM grade, Sigma Aldrich, USA) and stored at -80ºC. Viral and prokaryote 18 abundances were determined according to the methods of Mojica et al. (2014) and Marie et al. (1999), respectively. Briefly, samples were defrosted immediately prior to analysis, diluted with TE-buffer (pH 8.2) and stained with SYBR-Green I (Molecular Probes, Invitrogen, USA) at a final concentration of 0.5 x 10 -4 (viruses) and 1 x 10 -4 (bacteria) of the commercial stock. Staining was completed in the dark for 10 min at 80ºC for viral samples, and at room temperature for 15 min for bacterioplankton samples. Counts were performed on a Becton-Dickinson FACSCalibur flow cytometer (BD Biosciences, USA). Groups were determined in bivariate scatter plots of green fluorescence of stained nucleic acids versus side scatter.

Data analysis
All data analyses were performed in parallel in the dedicated software package PRIMER 6 (PRIMER-E, Lutton;Clarke and Gorley, 2006;Anderson et al., 2008) and the R package VEGAN (Dixon, 2003;Oksanen et al., 2013). Our aims were first to describe the interannual variability in abundances of bacterio-, virio-and phytoplankton (expressed as chlorophyll levels), and second to determine which physicochemical parameters are driving this variability. To reduce noise levels in the high resolution data and omit the problems associated with missing data, integrated values (area under curve) were calculated for each variable. Furthermore, to facilitate year-by-year comparison of the biological variability, the time series was subdivided into "summer" and "winter" time segments, using the level of local sea ice cover in Ryder Bay (visual assessment) and in the wider Marguerite Bay area (satellite data; Stammerjohn et al. 2008;2012) as an independent criterion (see Supplementary Material a full discussion).
To visualize the similarity between the different summer and winter time segments, PCO plots were constructed on the biological data using the Euclidian distance of square root transformed and

Wiley-Blackwell and Society for Applied Microbiology
This article is protected by copyright. All rights reserved.

Subdivision of the time series into seasonal segments for data exploration and statistical analyses
Whilst attempting to find correlations and trends in the high-resolution RaTS time series data set, two issues were initially identified. The first was the relatively high prevalence of 'missing' and 'misaligned' data. Reasons for this varied, but included measurements and/or samples not being taken, lost, or not recorded, or for example water samples for nutrient concentrations and plankton abundances being collected on different days. This results in incomplete rows of data, which need to be either filled in using any of a number of data imputation strategies, or deleted before statistical comparison. The second issue was the magnitude of change between summer and winter conditions, which for all measured parameters was much higher than the year-to-year variation, as is typical for high latitude marine ecosystems. This trend causes the seasonality to mask any underlying functional correlations within the data. To omit the seasonality issue and simultaneously preserve the entire data set, we therefore decided to subdivide the time series into a smaller number of discrete time segments (reflecting "summer" and "winter" conditions), and for each parameter calculate a single integrated value for each time segment (area under curve). This condensed data set was then used to describe the interannual variability in plankton abundances and to model which physicochemical parameters are driving this variability.
To define the boundaries between the "summer" and "winter" time segments, an independent criterion was sought. Previous studies by Stammerjohn et al. (2008;2012) used satellite observations to define the dates of sea ice advance and retreat around Antarctica. We compared sea ice data from the wider region (Marguerite Bay) as measured by satellite, to the visual assessment of local sea ice cover in Ryder Bay (on a scale of 1 to 6, 1 being ice-free and 6 being fast ice). A local sea ice cover score of 4 or higher for more than one week corresponded very well with a regional sea ice onset threshold of more than 50% cover derived from the satellite image, whereas a local cover score of 3 or lower corresponded with Page 31 of 76

Wiley-Blackwell and Society for Applied Microbiology
This article is protected by copyright. All rights reserved. a regional retreat threshold of less than 50% cover. As the dates correspond very well in both cases, the local sea ice score was chosen as the independent criterion for subdivision of the time series.
[insert figure S1 -sea ice assessment] Figure S1: Day of (A) sea ice onset using a visual assessment score for Ryder Bay of 4 or higher for more than one week (red line), and a value of more than 50% cover in the Marguerite Bay satellite image (blue line); and B) sea ice retreat a visual assessment score for Ryder Bay of 3 or less (red line), and a value of less than 50% cover in the Marguerite Bay satellite image (blue line). In both cases the dates were normalized to the 1 st of January as day 1.

Flow cytometry
Flow cytometry has gained popularity as a method to quantify viruses from aquatic systems due to its high sample throughput and ability to distinguish several viral subpopulations. Comparison of flow cytometric viral counts with those generated by epifluorescence microscopy generally yield complete agreement indicating the method is accurate and standard deviations for replicate samples are typically below 5% confirming high levels of precision (Brussaard et al., 2010). Furthermore, virioplankton counts generated by flow cytometry are unlikely to be perturbed by bacterial vesicles present in seawater since vesicles have comparatively much smaller associated DNA fragments, typically around 3 kilobase pairs (Biller et al., 2014), thus generating insufficient green fluorescence after staining to be detected by flow cytometry (Brussaard et al., 2010). N.B. typically even the small genome marine viruses have a genome size an order of magnitude larger than the vesicles (e.g. Steward et al., 2000). Therefore, it is unlikely that vesicles contributed significantly to LNA prokaryotes or the viral subgroups detected during our study.
[ Distance-based Redundancy Analysis (dbRDA) plot of the distance-based linear mixed model (distLM in PRIMER) fitted to the variation in plankton abundances. The model was constructed using a combination of the Akaike Information Criterion (AIC) and a forward selection procedure to select the most significant and useful predictor variables (in red): salinity (Sal), phosphorous (P), silicate (Si) and NOx.