Stepping stones to isolation: Impacts of a changing climate on the connectivity of fragmented fish populations

Abstract In the marine environment, understanding the biophysical mechanisms that drive variability in larval dispersal and population connectivity is essential for estimating the potential impacts of climate change on the resilience and genetic structure of populations. Species whose populations are small, isolated and discontinuous in distribution will differ fundamentally in their response and resilience to environmental stress, compared with species that are broadly distributed, abundant and frequently exchange conspecifics. Here, we use an individual‐based modelling approach, combined with a population genetics projection model, to consider the impacts of a warming climate on the population connectivity of two contrasting Antarctic fish species, Notothenia rossii and Champsocephalus gunnari. Focussing on the Scotia Sea region, sea surface temperatures are predicted to increase significantly by the end of the 21st century, resulting in reduced planktonic duration and increased egg and larval mortality. With shorter planktonic durations, the results of our study predict reduced dispersal of both species across the Scotia Sea, from Antarctic Peninsula sites to islands in the north and east, and increased dispersal among neighbouring sites, such as around the Antarctic Peninsula. Increased mortality modified the magnitude of population connectivity but had little effect on the overall patterns. Whilst the predicted changes in connectivity had little impact on the projected regional population genetic structure of N. rossii, which remained broadly genetically homogeneous within distances of ~1,500 km, the genetic isolation of C. gunnari populations in the northern Scotia Sea was predicted to increase with rising sea temperatures. Our study highlights the potential for increased isolation of island populations in a warming world, with implications for the resilience of populations and their ability to adapt to ongoing environmental change, a matter of high relevance to fisheries and ecosystem‐level management.


| INTRODUC TI ON
Despite the plethora of studies focusing on marine connectivity (Kelley, Brown, Therkildsen, & Foote, 2016), there remains a need to better elucidate key environmental factors and life-history stages influencing population genetic structure. Such information is vital for effective conservation management, and for predicting the potential impacts of ongoing and future environmental change, both globally and regionally. A complexity of processes is at play, resulting often in an uncoupling of dispersal potential, realized gene flow and patterns of genetic differentiation (Selkoe, Henzler, & Gaines, 2008;Selkoe et al., 2016). Whilst in some cases, variance in life-history traits combined with general oceanographic patterns can predict connectivity (e.g., Pascual, Rives, Schunter, & Macpherson, 2017;Riginos, Douglas, Jin, Shanahan, & Treml, 2011;Young et al., 2015), interplay among oceanographic variability, mortality, and habitat availability, and potential differences in local adaptation, often obscure predictions based solely on the product of the two factors (e.g., Galarza et al., 2009;White et al., 2010).
Whilst marine connectivity is influenced by dispersal of juveniles and adults in mobile species, larval dispersal plays a primary role in determining population and species distributions, especially in highly fragmented habitats (Cowen & Sponaugle, 2009). Oceanic temperature will affect larval development, dispersal and recruitment in various ways by influencing, for example, spawning date, rates of feeding (planktotrophic) or nonfeeding (lecithotrophic) modes of development, levels of predation and the ability to reach suitable habitats and subsequent postsettlement survival (Cowen & Sponaugle, 2009). Theoretically, given sufficient prey availability in planktotrophic larvae, more rapid growth has the potential to improve survival of larval fish; larvae could more rapidly outgrow gape-limited predators. As basal metabolic rates, growth, development and energetic costs of larvae are determined in general by water temperature, increasing ocean temperatures will generally reduce the duration of the larval phase (PLD) (O'Connor et al., 2007), and thus decrease the cumulative mortality risk during this particularly vulnerable life phase. However, such effects are offset by the theoretical increase in instantaneous mortality at higher temperatures due to starvation if nonfeeding, or if food availability and feeding become uncoupled (Kristiansen, Drinkwater, Lough, & Sundby, 2011). A shorter PLD also implies that larval fish may no longer have sufficient time to disperse between scattered habitats, and a higher proportion die in the open ocean before they can reach a suitable settlement site (Kendall, Poti, Wynne, Kinlan, & Bauer, 2013;O'Connor et al., 2007).
In addition to biological effects, a warmer climate may change underlying ocean conditions. Increased vertical stratification caused by stronger surface heat fluxes and increased freshwater input from precipitation and ice melt may impact not only primary productivity, with consequences for the availability of prey, but also vertical mixing and patterns of circulation (Meijers, 2014). It has been theorized that warming in the Southern Ocean over recent decades is in part due to a poleward shift of the frontal features of the Antarctic Circumpolar Current (e.g., Boning, Dispert, Visbeck, Rintoul, & Schwarzkopf, 2008;Gille, 2008), though whether such a shift has occurred is subject to debate, with recent assessments casting doubt on this (e.g., Gille, 2014). The strength and direction of currents may also be influenced by large-scale modes of climate variability, such as El Niño-Southern Oscillation and the Southern Annular Mode, although the response is likely to be complex and involve interactions between the mean flow and eddy field (e.g., Hallberg & Gnanadesikan, 2006).
Overall, there is currently considerable uncertainty concerning the impact of climate change on circulation patterns in the Southern Ocean, and no clear consensus across the latest generation of coupled climate models (Meijers, 2014).
Identifying key drivers that influence marine connectivity is especially pertinent in Antarctic waters that are experiencing unprecedented rates of warming (Meredith & King, 2005;Whitehouse et al., 2008), and where historical localized collapses of exploited commercial fishes increase vulnerability to marked shifts in trophic relations (Kock, 1992;Layman, Quattrochi, Peyer, & Allgeier, 2007). Moreover, the link between dispersal and gene flow depends ultimately on successful reproduction of migrants in recipient populations and phenotypic match of the offspring to the local environment, and thus is not necessarily a function of immigrant larval recruitment. In particular, the inference of population connectivity from genetic data alone can be misleading, primarily because of variance in effective population size and nonequilibrium conditions on estimates of the fixation index, F ST (Faurby & Barber, 2012;Waples & Gaggiotti, 2006).
It follows, therefore, that a multidisciplinary approach to exploring marine connectivity is optimal to capture the diversity of processes underway.
Here, we employ data and inferences from life histories, physiology, population genetics and oceanography, supported through simulations, to examine the likely impact of predicted shifts in Antarctic water temperatures on dispersal, gene flow and population genetic structure. Specifically, we use an established modelling system (Young et al., 2015) to investigate how changes in the duration of the planktonic phases and mortality rates of two fish species with contrasting life histories may affect connectivity and population genetic structure within the Scotia Sea. We then consider the wider implications of our findings for the resilience of fragmented populations under climate change scenarios.

| Study region
We focus on the Scotia Sea (Figure 1), which is one of the most productive regions in the Southern Ocean supporting high abundances of zooplankton, fish, seabirds and marine mammals . This region has recently experienced rapid increases in ocean temperatures, with surface summer temperatures near the Western Antarctic Peninsula increasing by >1°C in the latter half of the 20th century (Meredith & King, 2005), and increases in wintertime nearsurface temperatures near South Georgia of ~2.3°C for the period 1925-2006 (Whitehouse et al., 2008). Further significant warming of up to 2.5°C by the end of the 21st century is predicted ( Figure S1).
Thus, it is a prime location for considering the response of population connectivity and resilience to climate change. Horizontal flow in this region is dominated by the Antarctic Circumpolar Current (ACC), which is known to have been broadly stable since the Last Glacial Maximum (McCave, Crowhurst, Kuhn, Hillenbrand, & Meredith, 2014), with only relatively low levels of shorter period variability superposed (e.g., Meredith, Woodworth, Hughes, & Stepanov, 2004).
Such long-term stability suggests that populations are likely to have reached migration-drift equilibria (Wright, 1931), which lends credence to the baseline present-day genetic projection modelling utilized in our study. The ACC has a banded structure consisting of several fast-moving current jets separated by relatively quiescent zones of water (Orsi, Whitworth, & Nowlin, 1995). Each of these jets is associated with a frontal region characterized by marked meridional gradients in temperature and/or salinity. Within the Scotia Sea, the ACC flows in a predominantly north-eastward direction, but is also characterized by strong mesoscale activity, whereby instabilities in the current cores lead to frequent generation of eddies, with spatial scales up to tens of kilometres. Thus, whilst the dominant northeastward flows have a major role in connecting populations and food webs, transporting organisms from areas around the Antarctic Peninsula north and east across the Scotia Sea (Matschiner, Hanel, & Salzburger, 2009;Murphy, Thorpe, Watkins, & Hewitt, 2004), previous modelling work has demonstrated spatial and inter-annual variability in larval dispersal (Young et al., 2015).

| Study species
We focus on populations of two Antarctic fish species within the Scotia Sea, Champsocephalus gunnari and Notothenia rossii. These species were chosen based on their relatively isolated shelf habitats and salient differences in respective early life histories. The two species share similar distribution patterns, with populations around sub-Antarctic islands and at the northern end of the Antarctic Peninsula (e.g., Gon & Heemstra, 1990;Kock & Everson, 2003), although Shag Rocks does not have a spawning population of N. rossii. However, whilst C. gunnari spawns benthic adhesive eggs (Everson et al., 2001) in deep (100-350 m) inshore waters and fjords (Frolkina, 2002;Kock, 2005), N. rossii spawns planktonic eggs in offshore shelf areas with depths of around 200-360 m (Kock, Belchier, & Jones, 2004).
The incubation period of C. gunnari eggs is approximately 4 months F I G U R E 1 Map of the Scotia Sea region. Population boxes used in the connectivity analyses are outlined in red. Thin black lines indicate the mean positions of the Polar Front (PF) following Moore, Abbott, and Richman (1997), the Southern ACC Front (SACCF) and the southern boundary of the ACC (SB), both from Thorpe (2001). Bathymetry is from the GEBCO_2014 grid, version 20150318 (www.gebco.net) at South Georgia and Shag Rocks (North, 2001), increasing to around 6 months at more southerly locations (Kock, 2005). After hatching, the larvae spend approximately 3 months dispersing passively with the ocean currents (Duhamel, 1995), vertically distributed within the upper 50 m of the water column (North & Murray, 1992), before developing into juveniles. At this stage, they are able to undergo considerable diurnal vertical migration and can no longer be considered planktonic. For the first two years of life, C. gunnari use the shallow shelf areas as a nursery ground (Duhamel, 1995). N. rossii eggs develop into larvae after approximately 4 months in the Scotia Sea (Kock & Kellermann, 1991), and then into blue fingerlings after about a further 3 months, at which stage they migrate inshore to kelp beds (Gon & Heemstra, 1990). N. rossii eggs are found in the top 100 m of the water column, and their larvae in the upper 50 m (North, pers. comm.). The total planktonic phase of N. rossii is thus more than double that of C. gunnari, and it is this key characteristic of their early life histories that drives differences in patterns of population differentiation; with a shorter planktonic phase and reduced dispersal capability, C. gunnari exhibits greater genetic structuring across the Scotia Sea (Young et al., 2015).

| Individual-based model
The underlying modelling techniques used in this study are described in detail by Young et al. (2015); here we provide a summary of the numerical models and highlight the changes made to the parameterization and implementation of the models for the climate change simulations.
Five-day mean velocity fields from a state-of-the-art oceanographic modelling framework, NEMO (Nucleus for European Modelling of the Ocean), provided by the National Oceanography Centre, Southampton, were used to advect Lagrangian particles representing the early life stages of C. gunnari and N. rossii. NEMO has been widely used over a range of spatial scales and resolutions and has been shown to provide a good representation of the dominant oceanography of the Antarctic Peninsula and Scotia Sea region (Renner, Heywood, & Thorpe, 2009;Renner et al., 2012). Full details may be found at http://www.nemo-ocean.eu/About-NEMO with the specific NEMO application used in this study described by Young et al. (2015). The Lagrangian model has been parameterized for the simulation of C. gunnari and N. rossii eggs and larvae (Young et al., 2012), and its application to the study of the present-day population connectivity of these fish species in the Scotia Sea has been demonstrated (Young et al., 2015). In summary, particles are advected at each model time step (5 min) according to the imposed three-dimensional velocity field, using a second-order Runge-Kutta method. Additional horizontal and vertical diffusions are included using a random-walk approach (Dyke, 2001).
The reference present-day simulations are described by Young et al. (2015). The representative study period, 1996-2001, in-cludes years with relative extremes in atmospheric forcing over the Southern Ocean, associated with extreme phases of large-scale coupled modes of inter-annual climate variability (see Meredith, Murphy, Hawker, King, & Wallace, 2008; for full discussion). This choice of study period allows robust inferences concerning larval dispersal whilst accounting for variability in present-day physical forcing (Young et al., 2015). For each of the five study years, model particles representing the early life stages of the two fish species were released at the locations of known spawning populations described in published literature (population boxes; Figure 1; Barrera-Oro & Casaux, 1992;Everson et al., 2001;Frolkina, 2002;Kock et al., 2004;Parkes, 2000). 1,000 particles were released per day at each site for the duration of the spawning periods, with species-specific characteristics assigned to each particle (for full details, see Young et al., 2015). Pairwise transport matrices TM for each species were then constructed, describing the proportion of individuals arriving in a destination population (rows) from a given source population (columns), and allowing a 4-week recruitment window at the end of the prescribed planktonic period. These were converted to connectivity matrices M by incorporating egg and larval mortality, assuming temperature-dependent mortality rates from Pepin (1991), with an assumed present-day temperature (T) of 1°C, representative of annual mean near-surface sea temperatures across the Scotia Sea (Mapping and Geographic Information, British Antarctic Survey, pers. comm.).

| Projected changes to early life histories
The baseline present-day simulations were repeated for a future scenario of increased temperatures. A rise in sea surface temperature of 2.5°C was assumed, which is within the range of the forecast increase in temperature for the Scotia Sea region by the end of the 21st century from the latest generation of coupled climate model simulations ( Figure S1).
Notothenia rossii has pelagic egg and larval stages, and thus, the same temperature increase was assumed to impact both the eggs and larvae. However, C. gunnari lays benthic eggs and it is difficult to predict with certainty how climate change may impact near-bed shelf temperatures, in part due to the poor resolution of global models used for climate change simulations, and their inadequate representation of shelf regions. However, analyses of recent temperature changes at the Antarctic Peninsula suggest that near-bed temperatures may well increase less than near-surface temperatures (Meredith & King, 2005). Thus, for C. gunnari, the increase in sea surface temperatures was assumed to impact only the larval phase. The effect of a possible increase in near-bed temperatures on the egg phase was investigated through a series of sensitivity experiments (see Model sensitivity study).
We considered two physiological responses to increased temperatures in this study: egg and larval stage durations and mortality rates. Relative to the present-day simulations, a reduction in the duration of the egg phase was determined using equations detailed by Hirst and Lopez-Urrutia (2006), and data on marine fish egg (1) z e = 0.030e 0.18T , for eggs z l = 0.044e 0.077T , for larvae development in Pauly and Pullin (1988) (Appendix A). The duration of the larval phases was shortened following the universal model for the temperature dependency of larval development of marine animals developed by O'Connor et al. (2007). The effect of increased temperature on egg and larval mortality rates was estimated using the generalized temperature-dependent mortality rates from Pepin (1991; Equation 1). The subsequent present-day and climate change model parameterizations are detailed in Table 1. Changes to the species-specific pairwise transport and connectivity matrices were calculated, firstly considering solely the effect of shortened planktonic phase and secondly considering changes to both planktonic phase and mortality rates.

| Population genetic structure projection model
The methodology used to project population genetic structure is described in detail by Kool, Paris, Andrefouet, and Cowen (2010), and Kool, Paris, Barber, and Cowen (2011), and the application of the methodology to understand observed present-day patterns of population genetic structure in C. gunnari and N. rossii populations is described by Young et al. (2015). Briefly, a connectivity matrix M is applied to a state matrix Q t containing information on the frequency of alleles of each type in each population at time t, to yield the ex- Data on population sizes of the fish species at each spawning site are sparse, thus the carrying capacities of each site were assumed to be the same. Mean birth rates were derived from empirical fecundity data (Kock & Kellermann, 1991), with 2,800 larvae per individual C. gunnari and 20,000 larvae per individual N. rossii.

| Model sensitivity study
The effect of a possible increase in near-bed temperatures on the egg phase of C. gunnari was investigated through a series of sensitivity experiments, with assumed near-bed temperature increases of 0.5, 1 and 1.5°C. For each, larval phase was assumed to experience a rise of 2.5°C in near-surface temperatures. The range of model parameterizations for these sensitivity studies is detailed in Table 2.

| Transport and connectivity
Present-day patterns of connectivity indicated wider dispersal of N. rossii than of C. gunnari, with more consistent and higher mean   Instantaneous egg and larval mortality rates were predicted to increase with higher sea temperatures (Equation 1; Table 1).
However, the reduction in the length of the planktonic phases modulated the impact of this increase on the integrated mortality.
Thus, whilst integrated mortality during the egg phase of N. rossii increased, integrated mortality of the larval phases of both species decreased slightly. Consequently, overall survival of N. rossii decreased, but for C. gunnari it increased. The new mortality rates were used to derive connectivity matrices from the simulated pairwise transport patterns ( Figure 2). For N. rossii, the reduced connectivity across the Scotia Sea due to a shorter planktonic period was further reduced by the increased mortality rates in the climate change simulation. Over shorter geographic distances, where connectivity and local retention were broadly predicted to increase with the shorter planktonic period, the increased mortality rates were sufficient to negate this effect, with the exception of South Sandwich Islands.
Although C. gunnari was predicted to have a slight increase in inte-  connectivity reduced by more than 90% for most sites separated by greater than ~1,000 km ( Figure 5), implying that increased temperatures had a relatively greater impact on connectivity between more distant sites. Significant negative correlations were predicted for the percentage change in connectivity with distance between sites for both species, with correlation coefficients of r = −.82 (p = 5 × 10 −5 ; C. gunnari) and r = −.89 (p = 3 × 10 −8 ; N. rossii).

| Population genetic structure
Projections of present-day population genetic structure derived from the connectivity matrices differed considerably between the two species ( Figure 6, above diagonals; Figure 7a,b) and were in good agreement with empirically derived patterns of genetic structuring from prior microsatellite analyses (Young et al., 2015). with the population at South Orkneys becoming genetically more similar to that at the Antarctic Peninsula ( Figure 6c). However, the significant increase in the number of time steps required to reach this point suggests that increased regional isolation is more likely.

| Model sensitivity study
The control run for comparison with the sensitivity experiments was the climate change simulation described above, with larval duration and mortality for an assumed temperature increase of +2.5°C. In the first series of experiments, the effect of increasing temperature on egg duration was considered, with the egg mortality rate held at present-day levels. In the second series of experiments, the effects of increasing temperature on both egg duration and mortality rate were included.
Theoretically, egg duration decreases as temperature increases (Appendix A;

| Shifts in population connectivity
With a rise in sea temperatures, the theoretical shortening of the planktonic phases of C. gunnari and N. rossii is predicted to reduce transport between southern and northern populations in the Scotia Sea. Over smaller scales, the two species are predicted to respond somewhat differently to increased temperatures. Broadly, there is a tendency towards a decrease in pairwise transport between sites Percentage change in connectivity Distance between sites (km) C. gunnari N. rossii F I G U R E 6 Projected genetic differentiation between populations (G" ST ) for Champsocephalus gunnari (a, c) and Notothenia rossii (b, d); present day above the diagonal, increased temperature scenario below the diagonal. Model projections were stopped after the same number of time steps as the present-day simulations (a, b), or once the maximum projected genetic differentiation reached present-day levels (c, d) maintenance of ecosystem-level processes such as productivity and food webs. Compared to tropical and temperate waters, polar seas are frequently dominated by relatively few top predators and somewhat simpler trophic relationships. When krill, Euphausia superba, is abundant, it dominates the diet of both C. gunnari and N. rossii, although they can switch to alternative prey types during periods of low krill abundance (Main, Collins, Mitchell, & Belchier, 2009). As trophic niche width of key predators is predicted to reduce with habitat fragmentation (Layman et al., 2007;Timpane-Padgham et al., 2017), driven by the lower diversity in prey items, such shifts can exert consequences at the ecosystem level. Reduced trophic niche width can destabilize local food webs and correspondingly increase susceptibility of predators to extinction through time. Here, such effects are likely to be amplified by reductions in fish population size and ongoing habitat change. sink populations, key transport pathways, and understanding of the potential for climate change to restructure these systems, needs to be considered for fisheries and ecosystem management, and in the design of MPAs (Andrello, Mouillot, Somot, Thuiller, & Manel, 2015;Kough, Paris, & Butler, 2013). For example, protecting fish populations at South Georgia will have no impact on upstream populations at South Orkney Islands or around the Antarctic Peninsula. Similarly, it may be invalid to assume that populations at the Antarctic Peninsula will continue to seed fish populations at South Georgia as the climate warms. Thus, MPAs in scenarios of reduced connectivity may need to be more abundant and closer together in the future to achieve the same network effects expected today (Andrello et al., 2015).

| Predicting response to climate change
Model simulations have highlighted the species-dependence of the spatial scale over which changes in connectivity due to increasing temperatures impact the projected population genetic structure; whilst N. rossii populations remain broadly homogeneous except at the larger scale (> ~1,500 km), C. gunnari populations become more isolated, with the exception of the most proximate populations, such as those around the Antarctic Peninsula. These results suggest that the impact of climate change on the connectivity of Antarctic fish populations, and fragmented fish populations in general, cannot be assumed to be the same for all species. In particular, the species response likely depends on species-specific early lifehistory characteristics including planktonic duration and the temperature dependence of growth and mortality rates. In addition, the impacts of modified connectivity will likely be exacerbated by other biological, ecological and evolutionary factors such as larval growth and subsequent juvenile viability, and adult physiological and adaptive capabilities, including changes to reproductive output and the timing of spawning, with subsequent consequences for food availability (the match-mismatch hypothesis; Edwards & Richardson, 2004;Hoegh-Guldberg & Bruno, 2010). Such biological controls may be particularly relevant for Antarctic species that have evolved in a stable cold environment (Clarke et al., 2007).
Polar stenothermal species may have increased physiological sensitivity and reduced adaptive capability to tolerate warmer water.
Icefish, in particular, may be unable to adapt to warmer temperatures as they possess no haemoglobin and have developed adaptations that allow them to thrive in the cold, oxygen-rich waters of the Southern Ocean (Eastman, 1993). Hence, although N. rossii and C. gunnari have remarkably similar distributions, C. gunnari may be particularly sensitive to any rise in temperature due to their lack of haemoglobin (Beers & Sidell, 2011). However, for both species increases in temperature would be expected to have greatest impact on the growth, development and resilience of populations in the northern parts of their range where temperatures are already highest.
The existence of locally adapted populations in these two species has not yet been assessed. Under the climate change scenario, the projected reduced influx of Antarctic Peninsula genotypes to northerly populations (breakdown of connectivity), combined with an increase in self-recruitment and accentuation of differences in local environments could, hypothetically in evolutionary time, allow the retention and proliferation of adaptations that may allow local populations to survive (Lenormand, 2002). Nevertheless, the potential for further development of such adaptations in range margin populations may already be limited (Hoffmann & Sgro, 2011), whilst our view of the role of gene flow on local adaptation is rapidly changing (Tigano & Friesen, 2016). Nonrandom gene flow combined with spatial heterogeneity and habitat matching can have positive effects on the rate of local adaptation (Edelaar & Bolnick, 2012;Jacob et al., 2017), and intermediate levels of gene flow maximize local adaptation in temporally variable environments (Blanquart, Kaltz, Nuismer, & Gandon, 2013). Overall, the breakdown in connectivity among populations of C. gunnari and N. rossii may thus further hinder the adaptive potential of isolated populations. To fully appreciate the potential impacts of climate change on range margin populations, we need improved understanding of the adaptive structuring of populations and the mechanisms underpinning such structuring.
Changing rates and patterns of ocean circulation are a key influence on the connectivity of marine populations, and hence, there is a clear need to understand how such phenomena may change in future as natural and anthropogenic climate change progress.
Unfortunately, our ability to reliably predict future changes in ocean circulation is limited, not least because key processes that control circulation occur on spatial scales significantly smaller than the resolution of most IPCC-class climate models (e.g., Meijers, 2014). For example, mesoscale eddies have horizontal scales of just a few tens of km or less in the Southern Ocean and are commonly represented in climate models in parameterized form (e.g., Gent & Mcwilliams, 1990). Despite these restrictions, fine-scale model analyses and our understanding of the underlying ocean dynamics allows at least informed speculation concerning how patterns of circulation may change in future. On the large (circumpolar) scale, IPCC models generally agree that Southern Ocean winds will intensify and shift polewards in coming decades, in response to continued anthropogenic forcing (Meijers, 2014). The dynamics of the response of the ACC to changes in wind forcing are complex and nonlinear, but it is reasonably well established that extra energy imparted from winds will likely generate an increase in mesoscale eddy activity, with little or no long-term change in overall ACC transport expected (Meredith & Hogg, 2006). Current understanding also suggests that wholesale southward shifts in the ACC current cores would be unlikely (Gille, 2014), although smaller scale changes to circulation patterns may occur. For example, Renner et al. (2012) studied the response of regional-scale transport pathways in the vicinity of the Antarctic Peninsula under different wind forcing conditions, and found greater northward transport under stronger wind forcing, and changed pathways of mean circulation.
Such studies suggest that whilst the mean direction and strength of transport pathways over basin scales are likely to remain broadly similar to the present day under future climate change, over smaller scales, such as around the Antarctic Peninsula, changes are likely to occur. Increased eddy activity may also enhance the horizontal dispersion of upper ocean biota, potentially modifying the patterns of inter-island connectivity.
To progress this area of connectivity research, it will be necessary to utilize ocean models that reliably resolve the small scales known to be important for controlling circulation pathways and rates (mesoscale eddies, boundary currents, shelf-edge circulation features etc.), whilst simultaneously being able to run for the decades over which climatic change (both anthropogenic and natural) is relevant. Such models are in development, but are only now beginning to become widely available. In addition, there is a lack of data on the response of Antarctic fish species to changes in temperature, in part due to the logistical complexities of sampling and undertaking experimental work in the Southern Ocean. More detailed representation of oceanographic processes, and speciesspecific evolutionary and physiological capabilities, is essential for furthering our understanding of the response of fragmented fish populations to a changing climate and will be key for the evaluation of fisheries management options (Cheung et al., 2010).

| Concluding remarks
At the global level, biodiversity is experiencing fundamental shifts in distribution patterns, with range-restricted species, such as those from polar habitats, affected the most, predominantly by range contractions (Parmesan, 2006;Urban, Tewksbury, & Sheldon, 2012). Whilst interconnected locally adapted regional populations may be expected to modulate the severity of climate change effects by migrating, or through evolution of novel adaptations, the relative isolation of species restricted to the Southern Ocean makes range shifts and the emergence of adaptive novelty problematic: they have "nowhere left to go." Indeed, marked adaptation is unlikely as these species are living near physiological limits (Hoffmann & Sgro, 2011), especially when exposed to predicted rates of climate change. It is well recognized that predicting the impacts of climate change on connectivity is complex, and depends upon such factors as taxon-specific variance in response to shifts in temperature or localized impacts of island topography and alignment. Nevertheless, our results indicate that connectivity among populations will be markedly reduced by accelerating temperatures, likely leading to a potential loss of ecological resilience, increased localized extinctions and further range contractions. In such circumstances, it follows that those islands that presently export high volumes of larvae, and are predicted to continue to do so in future climate change scenarios, should be the focus of conservation measures that promote long-term resilience of larval supply.
Recently, it has been shown that at the macroevolutionary scale, the Antarctic Peninsula and neighbouring islands have acted as the evolutionary source of notothenioid species diversity, and the repeated export of colonizers to nearshore Antarctic continental regions (Dornburg, Federman, Lamb, Jones, & Near, 2017). Here, we show that these source-sink dynamics are mirrored at the microevolutionary scale in C. gunnari and N. rossii. Critically, as predicted by Dornburg et al. (2017), we demonstrate, using population genetics and biologically relevant model simulations, that increasing temperatures will likely lead to a restriction, or even complete breakdown, in connectivity, and increased isolation of source locations for the regional species pool (Ricklefs, 1987).  Figure S1. We thank two anonymous reviewers, whose comments helped us improve the original manuscript.

DATA A R C H I V I N G S TAT E M E N T
Model data are archived at the British Antarctic Survey and will be made available on request through contacting Dr Emma Young.

S U PP O RTI N G I N FO R M ATI O N
Additional Supporting Information may be found online in the supporting information tab for this article.

Relationship between egg duration and temperature
The dependence of egg duration on temperature was calculated using a theoretical relationship described by Gillooly, Charnov, West, Savage, and Brown (2002) and further discussed by Hirst and Lopez-Urrutia (2006): where t is time to hatch (days), m is mass at hatch (g), T c is temperature (°C), T 0 is 273K, a(T 0 ) is a normalization factor for development, independent of body size and temperature, and variable across taxa, Ẽ is the average activation energy for metabolic reaction, and k is Boltzmann's constant. This may be rewritten and by plotting ln t∕m 1∕4 against T c ∕(1 + T c ∕T 0 ) , solutions for the intercept, ln 4∕aT 0 and gradient, − ∼ E ∕kT 0 , may be found (Hirst & Lopez-Urrutia, 2006). However, as there are no data for Antarctic fish species, neither the intercept nor the gradient are known. Therefore, a value for the gradient term of −0.105 was assumed, based on data on marine fish egg development (Pauly & Pullin, 1988