Population connectivity and genetic offset in the spawning coral Acropora digitifera in Western Australia

Abstract Anthropogenic climate change has caused widespread loss of species biodiversity and ecosystem productivity across the globe, particularly on tropical coral reefs. Predicting the future vulnerability of reef‐building corals, the foundation species of coral reef ecosystems, is crucial for cost‐effective conservation planning in the Anthropocene. In this study, we combine regional population genetic connectivity and seascape analyses to explore patterns of genetic offset (the mismatch of gene–environmental associations under future climate conditions) in Acropora digitifera across 12 degrees of latitude in Western Australia. Our data revealed a pattern of restricted gene flow and limited genetic connectivity among geographically distant reef systems. Environmental association analyses identified a suite of loci strongly associated with the regional temperature variation. These loci helped forecast future genetic offset in gradient forest and generalized dissimilarity models. These analyses predicted pronounced differences in the response of different reef systems in Western Australia to rising temperatures. Under the most optimistic future warming scenario (RCP 2.6), we predicted a general pattern of increasing genetic offset with latitude. Under the extreme climate scenario (RCP 8.5 in 2090–2100), coral populations at the Ningaloo World Heritage Area were predicted to experience a higher mismatch between current allele frequencies and those required to cope with local environmental change, compared to populations in the inshore Kimberley region. The study suggests complex and spatially heterogeneous patterns of climate‐change vulnerability in coral populations across Western Australia, reinforcing the notion that regionally tailored conservation efforts will be most effective at managing coral reef resilience into the future.


| INTRODUC TI ON
The impacts of climate change are intensifying across ecosystems on multiple levels (Malhi et al., 2020), affecting not only species demography and dispersal, both of which underlie short term recovery, but also the genetic diversity and metapopulation structure that determine longer-term recovery and adaptation (Osman et al., 2018;Pauls et al., 2013). Recurrent disturbances affect reproductive output (Hughes et al., 2019), the strength of connectivity networks and threaten to erode population resilience (Thomas et al., 2017(Thomas et al., , 2020 which could lead to local extinction events (Hoffmann & Sgro, 2011;Matz et al., 2020;Richards et al., 2021) and jeopardize ecosystem functioning (Benkwitt et al., 2020;Dietzel et al., 2021). In marine systems, monitoring changes in connectivity and genetic diversity among local populations at different spatio-temporal scales are central to assessing their vulnerability to a warming planet (Kleypas et al., 2016;Oscar, 2017;Veron et al., 2009). For this reason, it is critical to integrate genetic data into conservation planning and protected area management (Gaitán-Espitia & Hobday, 2021; Underwood et al., 2013).
A complex array of environmental and biological processes influence marine metapopulations (Guan et al., 2020;Suggett & Smith, 2020), so it can be difficult to extrapolate connectivity patterns from genetic variation (Oscar, 2017). Seascape genomic studies seek to investigate how (a)biotic factors such as environmental and biological parameters, as well as demographic processes are associated with genetic variation to identify potential drivers of population structure in the marine realm (Balkenhol et al., 2017;Riginos et al., 2016;Selmoni et al., 2020). Seascape analyses have revealed the role of the environment in shaping patterns of larval dispersal and coral population connectivity (Riginos et al., 2016;Riginos & Liggins, 2013;Selkoe et al., 2016;Thomas et al., 2015;Treml et al., 2012;Underwood et al., 2020Underwood et al., , 2009Underwood et al., , 2013. Geneenvironment association analyses (GEAs) provide a means to explore the influence of the physical environment on the genetic structure of populations (Duruz et al., 2019;Rellstab et al., 2015;Selmoni et al., 2020). Additionally, random forest (gradient forest [GF]) and generalized dissimilarity models (GDM) of individual single nucleotide polymorphisms (SNPs) with environmental variables are valuable tools to investigate the adaptive capacity at broader spatial and temporal scales by evaluating the goodness of fit for the response of specific variant sites (in this case, SNPs) to specific environmental conditions (Fitzpatrick & Keller, 2015). Good performing models are then used to estimate spatial variation in the existing GEAs, and to determine if present-day GEAs can be maintained under changing climate conditions (Fitzpatrick & Keller, 2015).
Genetic offset (Fitzpatrick & Keller, 2015) is the difference in the genetic composition of a population under present-day versus projected future climate conditions. Therefore, estimates of genetic offset can be used to evaluate the level of allelic shift or adaptation required to avoid disrupting present-day gene-environmental relationships (Fitzpatrick & Keller, 2015). Based on this analysis, a large genetic offset could lead to the reduced likelihood that a given population can adapt rapidly enough to survive future climate conditions. Several studies have investigated the potential link between environmental conditions and loci under selection in coral populations, using genetic markers and environmental parameters for coral growth and survival, such as tidal height, sea surface temperature and water clarity (Selmoni et al., 2020;Underwood et al., 2020Underwood et al., , 2018. However, few studies have integrated GEAs to predict the local adaptive potential over time (Bay et al., 2017;Wood et al., 2021), population connectivity beyond the study area (Selmoni et al., 2020), or to examine how these associations affect the species' genetic composition and the adaptive potential of populations more generally (Gervais et al., 2021).
In Western Australia, large-scale population connectivity studies have combined genotype data with environmental variables into integrated seascape analyses (Thomas et al., 2017(Thomas et al., , 2020Underwood, 2009;Underwood et al., 2017Underwood et al., , 2020Underwood et al., , 2018Underwood et al., , 2007Underwood et al., , 2006Underwood et al., , 2013. However, no study has explored how genetic structure patterns in these populations translate to climate change vulnerability. Here, we explore patterns of genetic offset in the ubiquitous broadcast spawning coral, Acropora digitifera Dana 1846, across north Western Australia by combining genotyping by sequencing (GBS) data with random forest and generalized dissimilarity models.
First, we used the GBS approach to measure levels of spatial genetic structure across Western Australia to infer levels of reproductive isolation within and among separate geographic populations.
Secondly, we utilized gene-environment association analyses to identify putatively adaptive variants likely to be under directional selection. Finally, we used these loci to predict mismatches in GEAs under future climate scenarios.

| Sample collection and genotype-bysequencing filtering
Population samples were collected from five reef systems ( Figure 1): (1) The oceanic reef systems of Ashmore Reef and (2)  A. digitifera samples (~1-6 cm 3 ) were collected from 31 sites across the four aforementioned reef systems in Western Australia ( Figure 1 and Table S1), along with an additional 33 samples collected from Pelorus Island (GBR). Samples were identified in the field according to the morphological description provided by Wallace (1999).
Samples were stored in 100% ethanol, subsampled and sent to Diversity Array Technology Pty Ltd (DArT P/L) for DNA extraction, library preparation, sequencing and SNP calling using the same protocol as in Thomas et al. (2020). Furthermore, sequencing tags were blasted against the available Acropora digitifera genome (Shinzato et al., 2011) to confirm they belonged to the coral host and not the symbiont. Before quality control filtering (QC), raw loci sequences, averaging 1,283,302 (±151,230 SD) reads per sample (Table 1), were aligned to available Symbiodinium symbiont genomes (Aranda et al., 2016;Lin et al., 2015;Liu et al., 2018;Shoguchi et al., 2018Shoguchi et al., , 2013 and any sequences with a blastn e-value below 10 −3 were discarded. Furthermore, a Euclidean distance matrix was generated based on replicate genotype data for a subset of samples. Only unique multilocus genotypes with a distance greater than the hamming distance between the replicates of individuals were retained, while others were considered potential clones and were removed from analysis. Initial screening of the DArT SNP data identified all individuals from the Lalang-garram Marine Park reefs in the inshore Kimberley region (Jackson Island, Haywood Island, Augustus Island and Okenia Island) as outliers, probably representing a cryptic species (Tables S2 and S3, Figure S1). These samples were excluded from downstream analyses. After excluding the Lalang-Garram Marine Park sample data, the remaining dataset returned 38,456 single nucleotide polymorphisms (SNPs) (  (Thomas et al., 2020). Furthermore, secondary SNPs located within the same fragment were removed as these are likely to be linked. To generate a dataset of putatively neutral loci with F ST outliers removed, we ran the filtered SNP genotype data through BayeScan 2.0 (Foll & Gaggiotti, 2008) using 20 pilot runs of 5000 iterations, followed by 100,000 iterations for sampling (Thomas et al., 2020).

| Population genetic connectivity
The package poppr (Kamvar et al., 2014) was used to calculate genotypic diversity measures on the neutral loci dataset, and the package StAMPP (Winter, 2012) was applied to determine significance in pairwise F ST and genetic differentiation between reef systems, reefs and sites (Nei, 1973). Furthermore, hierarchical analysis of molecular variance (AMOVA) was conducted to link variation in genetic differentiation between reef systems, reefs, sites and samples (see spatial classification in Table S1). Spatial patterns of population connectivity were estimated using discriminant analysis of principle components (DAPC) in package adegenet (Jombart, 2008). To construct the DAPC using the neutral loci dataset, optimal K was identified using the function find.cluster, retaining 600 PCs to include the highest percentage cumulative variance and lowest BIC score. Furthermore, for DAPC construction, all discriminant analysis eigenvalues were included. Additionally, the spatial structure of genotypes was investigated using fastSTRUCTURE 1.0 model-based Bayesian clustering (Raj et al., 2014), running 100 replicates across K ranging from 1-10 (total number of reefs) on the Pawsey supercomputer facility.
The ChooseK function within the fastSTRUCTURE algorithm was applied to determine the optimal K value that best explained the structure on the neutral loci dataset. The package PopGenReport (Adamack & Gruber, 2014) was used to calculate allelic richness.

| Genetic offset to climate change
Genetic offset is a term used to describe the mismatch of geneenvironmental associations (GEAs) under future climate conditions (Bay, Harrigan, Underwood, et al., 2018;Fitzpatrick & Keller, 2015). This is usually characterized by the Euclidean distance between present and future biological space (Ellis et al., 2012). Under this framework, we used two model algorithms, gradient forest (GF) and generalized dissimilarity models (GDM) in the R packages gradientforest (Ellis et al., 2012) and gdm (Fitzpatrick et al., 2021), respectively, to describe patterns of observed genetic variation under specified climate conditions at the 26 sample sites in WA (excluding the Lalang-garram Marine Park sites). In contrast to GF which partitions the genotype data along the gradient of environmental data, GDMs are not based on machine learning techniques and integrate distance matrices to fit gene-environmental responses using Isplines, which inform on the magnitude and slope of variables when explaining genetic turnover (Fitzpatrick et al., 2013;Fitzpatrick & Keller, 2015;Gibson et al., 2017). Once gene-environmental responses were identified at sample site locations, the models were then used to estimate regional spatial similarities in genetic composition in site-neighbouring regions to predict future mismatches in GEAs under climate change conditions (genetic offset) across reef systems in Western Australia, following the approach described in Fitzpatrick and Keller (2015).
Before running gradientforest and gdm, we identified outlier loci with significant GEAs using BayeScEnv (excluding samples from the GBR due to high genetic dissimilarity to WA samples), which is an adapted Bayesian approach that combines F ST differentiation at loci level with the selective pressure on allele frequencies driven by environmental and geomorphological conditions (de Villemereuil & Gaggiotti, 2015;Stucki et al., 2017). Loci outside the 95% false discovery rate threshold were considered outliers possibly under directional selection, and these were included in the genetic offset analyses.
Environmental variables were selected based on their importance in delineating coral growth, settlement and survival ( Table 2 c Chlorophyll a in case 2 waters which represent coastal waters where inorganic particles concentration is higher than phytoplankton concentration. surface temperature (SST), SST anomalies (Zinke et al., 2018), water column optical parameters, geomorphological variables, and physical water column parameters. All variables were downscaled to the 250 m bathymetry resolution of Australia (Whiteway, 2009) using the nearest neighbour resampling approach (Gogina & Zettler, 2010) after smoothing and completing missing environmental data using kriging interpolation (Assis et al., 2018). Once downscaled, all variables were clipped to the 0-40 m bathymetry mask, representing the zone that most photic hard corals occupy (Veron & Marsh, 1988 pilot runs and 5000 burnin length). Posterior error probability incorporating the environmental factor (PEP g) < 0.05 was applied as recommended threshold to identify potential outlier loci or putative adaptive loci.
Putatively adaptive loci were selected for the GF analysis if they were polymorphic in more than 20% of sampled populations (Fitzpatrick & Keller, 2015) while all adaptive loci, identified using BayeScEnv, were used for GDM analysis. The gradientforest algorithm was based on 2000 regression trees per SNP and constructed with a depth of conditional permutation adjusted to the number of variables (Fitzpatrick & Keller, 2015) and a variable correlation threshold of 0.8. GF model performance was calculated. Variable importance was visualized using cumulative importance plots across individual and overall SNPs with positive R 2 values. For the GDM, the default model setting of three I-splines was used. GDM performance was assessed based on % deviance explained and the relative variable importance was represented by the sum of I-spline coefficients (Fitzpatrick et al., 2013).
To identify regional variation in GEA patterns and assess the future genetic offset of A. digitifera populations in WA, the study area of the 26 sites in WA was extended with a radius of 50 km (very few larvae disperse farther than 50 km [Graham et al., 2011;Jones et al., 2009;Underwood, 2009] Figure S3). On the largest geographical scale, neighbour-joining tree analyses on the neutral dataset revealed three broad groups (Figure 2a, labels correspond to the sites which can be found in Table S1); one cluster contained the offshore reefs When the GBR reef system (Pelorus Island reef) was excluded, the pairwise F ST between WA reef systems ranged between 0.02-0.081 (

| Genetic offset to climate change
Thirteen environmental variables were initially considered in our analyses ( Table 2). Nine variables remained after removing the most highly correlated (≥|0.80|) (  (Figures 3, Figures S4 and S5) and these variables were used for predicting regional, spatial and temporal patterns in GEAs. In the final GDM (% deviance explained = 89.9%), tidal height and SSTmax were the only environmental variables considered significantly driving genetic variation patterns across NWA sites ( Figure S6). Based on the GF and GDM analyses, we identified three distinct clusters in our dataset (Figure 3a [GF] and Figure S7 [GDM]): (1) Ningaloo  Table S12).   Figure S7). These patterns remained relatively consistent across the different climate scenarios, but were most pronounced in the extreme case. Secondly, we identified differences in the variance around the mean genetic offset between reef systems under the different climate change scenarios. For example, the level of variability in genetic offset in the Rowley Shoals, predicted in GF, was low compared to the other reef systems (Figures 3c and 4) while highest variability was predicted in GF and GDM in the Ningaloo Coast World Heritage Area (Figures 3c, Figure S7 and Table S13).

| DISCUSS ION
Using a panel of genomic-wide SNPs to explore the genetic diversity, population structure and mismatch in future gene-environment associations of Acropora digitifera across 12 degrees of latitude, we identified strong population differentiation among geographically separated reef systems, indicating restricted connectivity and limited gene flow between inshore and offshore reef systems in Western Australia (WA). Loci showing strong associations with temperature revealed varying genetic offsets among different reef systems. Based on the model results presented in this study, corals living closest to their thermal stress limit in low latitude regions, such as the inshore Kimberley reef system, were predicted to require a lower adaptive shift to be able to cope with future increases in temperature, compared to mid-latitude reefs. For example, populations in the Ningaloo Coast World Heritage Area were predicted to have pronounced gene-environment mismatches under future climate scenarios, highlighting their vulnerability to forecasted temperature changes and the need for large and rapid adaptive shifts to keep pace with climate change. This study shows that the potential of coral populations in WA to maintain gene-environmental associations under climate change is quite variable, complex and highly correlated with the relative regional temperature shifts, projected under climate change conditions. While the predictions of genetic offset in this study are based on future sea surface temperature conditions only, the importance of shifts in other factors such as fine scale future temperature anomalies cannot be ignored.

| Regional genetic structure across tropical North West Australia
We identified strong regional genetic differentiation in Acropora digitifera among, but not within, reef systems in WA with substantial exchange of beneficial alleles within systems. Consistent with the expectations of metapopulation structure, populations from the Great Barrier Reef showed strong genetic divergence from WA samples. Within WA, distinct genetic differences were identified between populations from the offshore reef systems, the inshore macrotidal Kimberley region and Ningaloo Coast World Heritage Area reefs. The spatial patterns of restricted exchange of genetic material between reef systems are similar to that observed in other brooding and spawning coral species in northwest Australia (Rosser et al., 2020;Underwood, 2009;Underwood et al., 2018). Our data, and other studies, indicate that contemporary larval exchange between offshore reefs (Rowley Shoals and Ashmore Reef) and the inshore Kimberley reefs (Adele Island and Beagle Reef) is restricted. To sustain local populations, the reef systems examined here are reliant on self-seeding and local recruitment to recover after disturbances and maintain population health. Hence, this study adds to the growing body of evidence highlighting the importance of local recruitment in maintaining healthy coral populations Thomas et al., 2017;Underwood, 2009;Underwood et al., 2020Underwood et al., , 2009. At a metapopulation scale, this dataset also highlights unexpected evolutionary linkages between the offshore NW shelf reefs

| Genetic offset across Western Australian reef systems
The genetic offset results indicate that the responses of Western Australian coral populations to climate change conditions are likely to be variable and spatially complex. The sensitivity and reactiveness of coral populations to changing environmental conditions have been described in the literature as fundamentally different in marine and terrestrial organisms (Burrows et al., 2011;Pinsky et al., 2019Pinsky et al., , 2013. More specifically, marine organisms have a broad and variable dispersal capacity (Kinlan & Gaines, 2003) and live close to their environmental limits. Hence, marine species are more responsive and sensitive to fluctuating environmental conditions, such as temperature anomalies, than terrestrial organisms (Pinsky et al., 2019(Pinsky et al., , 2013, which in turn could affect the magnitude of future genetic offset predicted in these populations. Our results indicate that there is variability in gene-environmental association mismatches under a range of climate change conditions. As expected, the largest gene-environmental mismatch was predicted under the extreme climate conditions (RCP 8.5 in 2090(RCP 8.5 in -2100 and revealed different degrees of genetic offset across the reef systems in WA. For example, A. digitifera populations at the inshore Kimberley region were predicted to experience the lowest mismatch in genetic variation under climate change conditions compared to other reef systems in WA, which supports the high resilience and adaptive potential predicted for this region in other studies (Richards et al., 2015;Underwood et al., 2020). In contrast, GEA mismatches were predicted to be highest in Ningaloo Coast World Heritage Area, indicating mismatches in local adaptive potential of these populations to increasing temperatures, especially under RCP 8.5 conditions in 2090-2100. Ningaloo has been predicted to serve as future stronghold of coral biodiversity under RCP 8.5 climate conditions in 2090-2100(Adam et al., 2021. However, coral reefs within the Ningaloo Coast World Heritage Area have been impacted over the last decade (Gilmour et al., 2019) with parts of the reef system been damaged in recent years by mass bleaching and cyclones Gilmour et al., 2019;Moore et al., 2012;Speed et al., 2013), with some reefs showing limited recovery (Babcock et al., 2021). Therefore, the level of GEA mismatches identified in this study may offset the potential for this region to function as future coral refugia.
Two hypotheses can be presented that could explain the pattern of genetic offset found across the study area. The first hypothesis is that the magnitude in genetic offset is strongly linked to the specific regional environmental conditions and the level of local adaptive potential to temperature conditions. More specifically, the extent that A. digitifera populations are adapted to their local unique environmental conditions (specifically thermal variability) is inversely related to the predicted genetic offset. This means that strong adaptation to local temperature conditions could result in lower future mismatches in gene-environmental associations and potentially increased resilience potential. Tidal height, SSTrange and SSTmax were identified as the strongest drivers of local adaptation and could be considered key environmental variables across all reef systems in tropical WA, although their influence diminishes from low to mid latitude reef systems (see GF, GDM and Samßada results).
These results reflect the variety of unique environmental conditions documented in the studied reef systems (Gilmour et al., 2019;Richards et al., 2009Richards et al., , 2018Richards et al., , 2014Richards et al., , 2015Speed et al., 2013;Thomas et al., 2020;Zinke et al., 2018). For example, the inshore Kimberley is known for its specialized coral communities that are able to survive harsh and variable environmental conditions (Richards et al., , 2014(Richards et al., , 2015Underwood et al., 2020). These coral populations are probably adapted to the high turbidity, extreme tides (>11 m) and high temperatures that are typical for the region (Richards et al., , 2013Underwood et al., 2017Underwood et al., , 2020. In contrast, offshore reef systems such as the Rowley Shoals and Ashmore Reef are more isolated, surrounded by oligotrophic clear oceanic waters with a smaller tidal range and have experienced variable levels of heat stress over the last decade, impacting coral communities in these regions (Gilmour et al., 2019;Thomas et al., 2020;Zinke et al., 2018).

Conversely, fringing reefs at the Ningaloo Coast World Heritage
Area, characterized by high total suspended matter conditions, experience variable ranges of sea surface temperature conditions and frequent cyclone activity (Zinke et al., 2018).
Our results also showed that the variability in regional environmental conditions between reef systems is correlated to the spatial scale of these systems ( Figure S8) as well as the spatiotemporal resolution of the variable data integrated in the models. For example, as site locations are more spread out across the Ningaloo Coast World Heritage Area and inshore Kimberley reef system, more environmental variation could be integrated into the GF and GDM models compared to smaller areas such as Ashmore Reef and the Rowley Shoals.
To interpret the genetic offset results and understand the environmental processes at play, it is important to understand that the fine scale spatial and temporal microhabitat temperature type variation that we see for example at the Rowley Shoals, such as daily fluctuations in temperature, are not resolved in the GF and GDM models. Due to habitat variation (e.g., lagoon vs. outer reefs), the Rowley Shoals experience variable fine scale environmental conditions (Gilmour et al., 2019(Gilmour et al., , 2022. Such fine scale spatial and temporal variation within environmental variables can have subtle yet profound impacts on the resilience potential of coral populations (Thomas et al., 2020).
The second hypothesis is that the magnitude of temperature shifts across latitude drives the regional genetic offset predictions across WA. This could explain why mid latitude reefs were predicted to experience higher genetic mismatches to cope with future climate conditions compared to those in low latitude regions. When comparing  in 2090-2100 between Ningaloo Coast World Heritage Area and inshore Kimberley reef systems, we observed a dramatic shift in the magnitude of change in future temperature conditions. In particular, SSTmax within the Ningaloo Coast World Heritage Area increases from ~27-28°C to 31-32°C ( Figure S8), which has also been predicted in other studies (Saha et al., 2018). In comparison, a smaller change in SSTmax was predicted (from ~31-32°C to 33.5-34°C) within the inshore Kimberley reef system ( Figure S8). This shows that when coral populations are locally adapted to temperature conditions, drastic temperature shifts could result in higher predicted distances between present-day and future genetic composition and therefore an increased genetic offset. These regional differences in future temperature conditions show that many Ningaloo reefs would need to adapt to a larger increasing temperature change than the inshore Kimberley populations. The hypothesis that inshore Kimberley coral populations are highly adapted to extreme temperature conditions which could benefit their resilience potential to future climate conditions, has been suggested previously (Richards et al., 2015).
However, whether these populations have already reached their adaptive limit and therefore are restricted in their ability to persist under future temperature conditions is unknown.
Also, the GF and GDM models that were used to assess the genetic offset are associated with certain assumptions and, in some cases, these provide limits to interpretation. For example, the outcomes presented here are based on future changes across certain temperature variables (SSTmax and SSTrange), assuming no future changes in migration, reproductive success, brood stock, mutation rate and local adaptation potential, or shifts in anomalous conditions or population dynamics. All of these factors are considered to potentially influence coral reef resilience under climate change conditions but are difficult to project over time. Many studies have discussed the impact of extensive heat stress (Zinke et al., 2018(Zinke et al., , 2015, driving the large scale degradation of coral reefs and the erosion of population structure Gilmour et al., 2019;Hughes et al., 2017;McManus et al., 2021. Underwood et al., 2007. Hence, the recovery capacity of coral populations in WA reef systems is highly dependent on the extent and frequency of anomalous heat stress events, which are predicted to intensify towards midhigh latitude regions along the WA coastline over the next decades (van Hooidonk et al., 2016(van Hooidonk et al., , 2014(van Hooidonk et al., , 2013. However, future thermal stress metrics have not been integrated in the GF and GDM models to estimate future genetic offset due to high collinearity with other temperature related variables, even though thermal stress has impacted all coral reef systems investigated in this study to some extent (Gilmour et al., 2019) and is likely to have structured the local adaptive capacity of coral populations. Furthermore, an increasing body of evidence is highlighting how rising sea level (projected to increase up to 1.4 m in Fremantle, Southwest of Australia [Carson et al., 2016]) not only impacts the distribution of coral populations but also affects accretion levels of coral reefs (Cornwall et al., 2021), thereby compromising the structural integrity of these habitats.
Overall, the genetic offset is sensitive to a wide array of future changes that are not easily incorporated into the models, hence reevaluating these with additional data is warranted.
A second potential limitation in this study, is that the selection of outlier loci was based on statistical analyses in BayeScEnv, rather than a priori knowledge of adaptive SNPs as seen in Fitzpatrick and Keller (2015). However, gathering this type of information requires a large scale controlled experimental setup (Bay, Harrigan, Buermann, et al., 2018). Hence, confounding effects of neutral loci could have influenced gene-environmental responses in the models and have led to over-or underinflation of future genetic offset predictions.
Other confounding factors that need to be considered include the correlation between geographic distance with differences in environmental conditions as some correlated variables appear to be identified as important variables in the GF and GDM models (Table S14 and Figure S9). This shows that more distant sites tend to be environmentally more distinct than neighbouring sites, which could potentially inflate the model predictions.
Whether the broadscale projects of gene-environmental mismatches that we described here for A. digitifera are transferable to other coral species with similar or different reproductive modes is unknown. In contrast to broadcast spawning corals that release gametes in the water column that can travel over large distances, brooding corals release larvae in close proximity to the parents, which makes the latter particularly more vulnerable to changing climate conditions. Based on our findings, we hypothesise that brooding coral populations, which are highly adapted to local conditions, could experience even higher mismatches in gene-environmental associations with the increasing rate of future temperature shifts.
Based on these projections, we can assume that coral populations at tropical reef systems in WA, which predominantly depend on local recruitment to replenish populations after disturbance events, will respond differently to climate change pressure. As the potential for populations to adapt to climate change conditions is strongly correlated with the magnitude in temporal temperature shifts, populations such as those in the inshore Kimberley showed to experience the lowest mismatch in genetic variation under future temperature shifts. While inshore Kimberley populations are predicted to experience the lowest genetic offset across reef systems in WA, it is uncertain whether these populations have the capacity to respond and adapt fast enough to keep up with increasing frequency and magnitude of temperature change. Therefore, the gene-environmental associations analyses in this study provide the building blocks for future research to investigate rates of adaption and whether the shifts in population genetics are likely to convey greater resistance of coral reef systems to future heat stress. Nevertheless, the increased pressure of climate change, variability in environmental responses as well as spatial and genetic isolation of coral populations in WA, calls for regionally tailored conservation and management strategies to monitor how the metapopulation responds to the increased intensity of climate disturbances in the future.

| CON CLUS ION
This study identified an increasing vulnerability of coral populations in Western Australia to rising global temperatures. It also supports the notion that reef systems in WA are highly adapted to local environmental conditions, reproductively isolated from neighbouring systems, and therefore self-reliant for population maintenance and genetic rescue. However, our data also revealed pronounced differences in genetic offset among our sampled reefs, offering a glimmer of hope that some reef systems, such as the inshore Kimberley, may fare better than others under climate change conditions. However, inferences about future adaptive potential for populations are based on the observed distribution of heat adapted alleles, which are strongly correlated with the background exposure to higher temperatures. Furthermore, our results show that the capacity of populations to maintain present-day adaptive potential under climate change conditions is highly dependent on the magnitude of regional temperature change predicted in the future. Nonetheless, the primary factor determining the impact of climate change on coral reefs is the frequency and severity of temperature increases, which typically overwhelm the latent adaptive capacity of many reefs and habitats. Variation in adaptive capacity will slow the degradation of some populations on some reefs; however, reducing rates of temperature increase generated through carbon emissions remains the most effective means maintaining coral reef ecosystems into the future. Given the prediction of recurrent mass mortality events in the future, broadly evaluating the metapopulation structure and the adaptive capacity of populations provides useful information for the prioritization of limited conservation resources.