Genomic offsets predict observed kelp declines and suggest benefits of assisted migration in the Northeast Pacific

Kelp forests are widely distributed along temperate and polar coastlines worldwide and are among the world’s most productive and diverse marine ecosystems. Yet, due in part to ocean warming, they are declining and even disappearing in many parts of the world. While genomic tools can identify local adaptation and predict species’ responses to global change, these predictions have rarely been validated in the field, hampering their widespread use in conservation practice . Here, we applied a seascape genomics approach to investigate environmental adaptation in the two main canopy-forming species of the Northeast Pacific, Macrocystis tenuifolia and Nereocystis luetkeana. We leveraged whole -genome sequences of 598 individuals across 94 sites along the British Columbia and Washington coasts, together with 37 environmental variables. Both species showed genomic signatures of local adaptation, with distinct environmental drivers shaping adaptation in each species despite their co -occurrence across much of the studied area. Using gradient forests, we modelled the genetic turnover across e nvironmental gradients and predicted populations’ vulnerability (genomic offset) under projected environmental conditions. Genomic offsets differed greatly among regions and were positively correlated with kelp declines observed to date, especially in Macrocystis, validating the link between genomic models and outcomes in the field and allowing us to translate genomic predictions into an ecologically meaningful metric: the risk of extir pation under global change. Our models predict that assisted migration could significantly attenuate kelp’s vulnerability to global change. Across environmentally heterogenous coastlines, short-distance migration can often substantially reduce future genomic -environmental mismatches, but in many cases, long -distance migration would be most beneficial. Our results highlight the potential of sea scape genomics to predict vulnerability of populations to global change. Importantly, the validated link between our genomic models and ecological outcomes allows quantification of climate-driven extirpation risk and can inform conservation strategies to improve the resilience and sustainable management of these vulnerable ecosystems.

Kelp forests – Seascape genomics – Environmental adaptation – Global change – Genomic offset – Risk of extirpation – Assisted migration – Conservation .CC-BY-NC 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 4, 2026. ; https://doi.org/10.64898/2026.04.01.715974doi: bioRxiv preprint

Kelp forests, formed by large brown macroalgae of the order Laminariales, are widely distributed along more than 1/3 of the world’s coastlines (Jayathilake & Costello 2021). They represent some of the most productive and diverse marine ecosystems, providing habitat for a wide range of species, including invertebrates, fishes, and other seaweeds of ecological and economic importance (Steneck et al., 2002, Teagle and Smale, 2018; Pessarrodona et al., 2022 ). However, these ecosystems are quickly disappearing in many parts of the world, mainly due to ocean warming and marine heatwaves (Vergés et al., 2016; Smale, 2020; Tait et al., 2021; Starko et al., 2022; Wernberg et al., 2024) , though overgrazing by herbivores is also a major threat (Filbee- Dexter and Scheibling, 2014; Ling et al., 2015) . The IPCC ranks kelp forests as the second most vulnerable coastal marine ecosystem to global change, only after coral reefs (Pörtner et al. 2019). Despite this general trend of decline, kelp ecosystems in some places remain stable or have even increased in abundance over the past several decades (Pfister et al., 2018; Smale, 2020; Mora-Soto et al., 2024a; Starko et al., 2024) . A clearer understanding of intraspecific and interspecific responses to environmental gradients is therefore critical to better predict responses to ongoing climate change and to inform conservation, management, and restoration strategies of kelp forest ecosystems. In the Northeast Pacific, there are two main canopy-forming kelp species: the perennial giant kelp Macrocystis tenuifolia (hereafter Macrocystis) – until recently (Lindstrom 2023) included in Macrocystis pyrifera sensu lato (Demes et al. 2009) – and the annual bull kelp Nereocystis luetkeana (hereafter Nereocystis). As a result of past glaciation, the Northeast Pacific coastline is complex, featuring numerous fjords, islands, and shorelines of varying wave exposure levels, which creates a mosaic of microcli mates with high variability in key ocean variables (e.g., temperature, salinity, nutrients) over relatively short distances (Mora-Soto et al., 2024a; Starko et al., 2024; Man et al., 2025) . A recent genetic study revealed strong genetic structure associated with geography in both species (Bemmels et al., 2025). However, the extent to which populations are locally adapted to these environmental gradients and the amount of intraspecific genetic diversity for adaptive loci remain unknown. Importantly, complex environmental heterogeneity in the Northeast Pacific decouples environmental gradients from simple geographic proxies such as latitude (Mora-Soto et al., 2024a; Starko et al., 2024; Gendall et al. 2025). In many systems, spatial .CC-BY-NC 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 4, 2026. ; https://doi.org/10.64898/2026.04.01.715974doi: bioRxiv preprint and environmental heterogeneity are often strongly confounded , making it challenging to tease apart isolation-by-distance from true signatures of local adaptation ( e.g., Vranken et al. 2021; Wood et al. 2021 ; Minne et al. , 2025). In contrast, localities at the same latitude can vary substantially in environmental conditions in the Northeast Pacific (Mora-Soto et al., 2024a; Starko et al., 2024 ; Gendall et al. 2025 ), potentially enabling more robust inferences about genotype - environment relationships. As both species have recently experienced strong declines in some parts of the Northeast Pacific (e.g., Mora-Soto et al., 2024b; Starko et al., 2024 ; Gendall et al. 2025 ), interest in kelp forest restoration is rapidly growing (Wood et al., 20 24a; Dykman et al., 2025) . To preserve genetic integrity of restored sites, restoration best -practices that involve moving individuals from one location to another often recommend using local source populations whenever possible (Bucharova et al., 2017) . When suitable local sources are not available, population genetic structure has also been used to choose donor sources. This has been done successfully in other kelp species (Wood et al., 2020) and experimentally in bull kelp in British Columbia (BC) (L. Dykman and J. Baum, personal communication). In the absence of detailed knowledge about local adaptation or population genetic structure, government agencies in Alaska (Gruenthal and Habicht, 2022), BC (McConnell et al. 2024), and Washington (Cui, 2023) have adopted informal but conservative recommendations that kelp be transferred no further than 50 km from its geographic provenance for restoration and aquaculture. However, restoration with local germplasm is unlikely to be successful if environmental conditions have already shifted away from the historical conditions to which local genotypes were adapted (Houde et al., 2015; Coleman et al., 2020). To address such concerns, methods known as genomic offsets (GOs) have recently been developed to predict population responses to environmental variation in space and time (Fitzpatrick and Keller, 2015; Capblancq et al., 2020) . Genomic offset statistics rely on genotype -environment association (GEA) analyses to model changes in allele frequencies along current environmental gradients. Then these models are used to predict allele frequencies from environmental features and evaluate differences in predicted allele frequencies at pairs of temporal or spatial points (Capblancq et al., 2020; Gougherty et al., 2021) . Higher genomic offsets indicate that a population’s genomic composition will be less suited to the given environment in a different location or different time period. This approach offers a powerful toolkit to inform decisions about .CC-BY-NC 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 4, 2026. ; https://doi.org/10.64898/2026.04.01.715974doi: bioRxiv preprint selection of the optimal genotypes for restoration and the identification of areas with higher or lower vulnerability under changing climates. Though little is known about local adaptation in most marine species in general (Layton et al. 2024), previous studies in other kelp and seaweeds have calculated genomic offsets and identified environmental gradients and candidate loci involved in adaptation (Vranken et al. 2021; Wood et al. 2021; Minne et al. 2025), highlighting how this approach has the potential to inform restoration strategies in kelp (Coleman et al., 2020) . GEA analysis has also been recently applied to Nereocystis and used to identify candidate source populations for genetic rescue (Abbott et al. 2025); however, GEA was applied on only a very small portion of this species’ range (Puget Sound, Washington) and genomic offsets were not explicitly calculated. Furthermore, as GEA analyses rely on statistical associations between genetic variants and environmental variables, their predictions are rarely experimentally tested (Luo et al., 2025) . In species such as kelp for which long-term ecological monitoring datasets are available (e.g., Sta rko et al., 2024) , comparing genomic offsets to ecological performance could provide a powerful approach to validating GEA models and translating genomic offsets into quantified predictions about ecological outcomes. Here, we investigate the genetic bases of environmental adaptation in the two main canopy - forming species from the coasts of British Columbia and Washington: Macrocystis and Nereocystis. Our aims are to 1) identify the most relevant environmental gradients for each species; 2) predict the vulnerability of kelp forests to global change under various migration scenarios; and 3) validate our predictions with contemporary kelp persistence records. Together, these approaches provide a framework for linking environmental adaptation to future vulnerability and informing conservation strategies in Northeast Pacific kelp forests.

Genetic variation and population structure We obtained initial pools of 10,800,722 and 16,527,403 high -quality SNPs from re -sequencing 191 and 404 individuals of Macrocystis and Nereocystis, respectively. These individuals were collected from 33 and 61 sites (each with at least three individuals; hereafter “populations”). These initial SNP pools were previously generated by Bemmels et al. (2025) and were filtered to exclude first-degree relatives. For each species, we further processed the initial pools to generate two .CC-BY-NC 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 4, 2026. ; https://doi.org/10.64898/2026.04.01.715974doi: bioRxiv preprint filtered SNP datasets: 1) the full dataset filtered to a minimum minor allele frequency (MAF) of 0.05, excluding only redundant fully linked loci (r 2 < 0.99), which gave 375,110 and 1,228,456 SNPs for Macrocystis and Nereocystis, respectively; and 2) a low linkage disequilibrium (LD) dataset also filtered to a minimum MAF of 0.05 but including only SNPs with low LD (r 2 < 0.2 using sliding windows of 50 kb and 10 kb steps), which gave 43,933 and 116,707 SNPs for Macrocystis and Nereocystis, respectively. Usi ng the SNPs generated in these datasets, we calculated the allele frequency for each SNP and population and used the population -level allele frequency data for all subsequent analyses. The full datasets were used for genotype-environment associations, while the low LD datasets were used for principal component analysis. Environmental characterization of kelp forest habitats To characterize the environmental variation experienced by kelp, we retrieved 37 environmental variables from the Northeastern Pacific Canadian Ocean Ecosystem Model (NEP36 -CanOE) Climate Projections database (Holdsworth et al. 2021). These variables included mean, minimum, and maximum values for sea surface temperature, salinity, nutrient availability, and primary productivity (Table S1). For each variable, we obtained historical values (1986 -2005) as well as simulations for the peri od 2046-2065 under two global warming scenarios: moderate mitigation (RCP4.5) and no mitigation (RCP8.5). We also included fetch (a proxy for wave exposure) as wave exposure is known to impact kelp habitat suitability (e.g., Mora -Soto et al. 2024a, 2024b; Graham et al. 1997). Fetch was defined as the total distance from a focal ocean point to land (up to a maximum of 200 km) summed across 72 compass bearings separated by 5º each (Gregr et al. 2018) and was calculated using custom R scripts as the mean fetch across 100 evenly spaced points per pixel of a 1/36º-resolution grid (excluding points that fell on land). The coastline map used in fetch calculations was a composite of source maps for BC and Washington (GeoBranch BC 2002) and Alaska (Alaska DNR 2021), with both source maps at 1:250,000 resolution. January and July sea surface values of environmental variables (except fetch, for which only a single value applies) were extracted for a grid dataset at 1/36º resolution covering the range of distribution of the two species, comprising 4,447 and 7,479 observations for Macrocystis and Nereocystis, respectively. January and July were selected as proxies for winter and summer environments, respectively. The spatial masking for each of the species was approximated based .CC-BY-NC 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 4, 2026. ; https://doi.org/10.64898/2026.04.01.715974doi: bioRxiv preprint on species distribution data from the BC Marine Conservation Atlas ( BC Marine Conservation Analysis 2011). Using only the historical values, we ran a pairwise correlation analysis to filter out environmental variables with a Spearman correlation coefficient (r) > |0.8|. In general, January and July values of each variable showed low correlation, only Nitrate concentration in January (NO3_01) and July (NO3_07) passed the threshold , and we retained NO3_07. For each month (January or July), the minimum, mean, and maximum values of each variable showed strong correlation, so we retained the mean values. Among environmental variables, we retained sea water practical salinity (salt_01 and salt_07 ) over total alkalinity concentration , sigma potential density, and dissolved inorganic concentr ation, and pH over air-sea CO 2 flux. The final set comprised 14 environmental variables (Table 1 and S1). Principal component analysis using the 14 environmental variables (zero centered) from observed and sampled sites was performed using the R function prcomp. Identification of genomic regions associated with local environmental adaptation We used latent factor mixed models (LFMM) with ridge penalty as implemented in the lfmm version 2.0 R package (Caye et al., 2019) to test for associations between genotypes (allele frequencies) and each of the 14 environmental variables while accounting for population structure. In LFMM, the genotype matrix is the response variable in a regression mixed model, the environmental variables are the explanatory variables introduced as fixed effects, and population structure is modeled using latent factors. Late nt factors are unobserved variables estimated from the genotype matrix itself that capture covariance patterns shared genome wide. Then, the effect of environmental variables is tested on the residual component, largely reducing the rate of false positives (Caye et al., 2019) . The number of latent factors is chosen based on estimates of population structure (in our case using principal component analysis) and represents the most likely number of ancestral genetic groups. We used three latent factors for both species, as this i s the most likely number of ancestral genetic groups (Fig. S1). P -values were calibrated by calculating the genomic-inflation factor, an estimate of variance of z -scores, using the lfmm_test function (Caye et al., 2019). Then, calibrated p-values were combined using a windows-based approach, Weighted-Z Analysis (WZA) (Booker et al., 2024) . This approach combines information from multiple linked sites .CC-BY-NC 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 4, 2026. ; https://doi.org/10.64898/2026.04.01.715974doi: bioRxiv preprint within a-priori defined windows to identify regions (instead of individual SNPs) associated with environmental variables. The approach uses expected heterozygosity (estimated from minimum allele frequencies) to construct weights for SNPs, as sites with higher expected he terozygosity carry more information about the demographic history of populations (Booker et al., 2024) . The window size was set to 10 kb , and only windows with more than five SNPs were considered. A strict Bonferroni correction (p < 0.05/number of windows) was used for setting the significance threshold. Although windows chosen by WZA have higher average Z scores, not all sites within a window are necessarily adaptive. Therefore, for all SNPs within significant windows, we individually calculated pairwise correlations between allele frequency and each of the environmental variables (environment), latitude a nd longitude (geography), and the first four principal components from the PCA (population structure) and retained only those SNPs whose correlation was stronger with any environmental variable than geography and population structure variables. Hereafter, we refer to these SNPs as ‘adaptive’. To validate our set of adaptive SNPs and ensure that patterns were likely due to environmental adaptation rather than statistical artifacts, we generated, for comparison, a set of neutral SNPs by randomly selecting a similar number of windows not associate d with any environmental variable (all p-values > 0.1). Mantel tests and gradient forest analyses were performed for neutral and adaptive SNPs. Mantel tests were used to test for isolation by distance (IBD) and isolation by environment (IBE) under the expectation that adaptive SNPs should have greater IBE. For Mantel tests, environmental distances were calculated as Euclidean distances among the 14 environmental variables, while geographic distances were calculated as the minimum ocean distance (i.e., excluding land) between populations and were previously generated by Bemmels et al. (2025). Gradient forest analyses were used to inspect the rate of allele turnover along environmental gradients for neutral and adaptive SNPs. Adaptive SNPs are expected to show stronger and faster turnover rates than neutral SNPs due to the combined forces of selection and demography. These analyses were performed using the R packages vegan version 2.6.4 (Oksanen et al. 2022) and gradientForest version 0.1.37 (Ellis et al. 2012). .CC-BY-NC 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 4, 2026. ; https://doi.org/10.64898/2026.04.01.715974doi: bioRxiv preprint Genomic offsets to predict vulnerability to climate change By using the set of adaptive SNPs and the 14 environmental variables, we investigated spatial and temporal patterns of maladaptation to future climate change for both species. We used gradient forests (GF) (Ellis et al. 2012) to model the genetic turnover along present environmental gradients and predict genomic offsets to projected future environments. We estimated three complementary genomic offset (GO) variables, each quantifying the degree of mismatch between current genetic composition and future environmental conditions. First, we modelled local genomic offset, which quantifies in situ maladaptation and does not allow migration of genotypes. Next, we modelled forward and reverse genomic offset (as defined by Gougherty et al., 2021), which allow migration of genotypes across space but using different approaches. To aid understanding, in this study , we refer to forward and reverse genomic offsets as Population GO and Location GO, respectively. Population GO predicts how the current population genotype wil l fare in the future, assuming it can move to the most optimal spot available across the predefined mapping space. A low Population GO indicates that the genotype has a suitable future home, while a high Population GO means that the genotype will be less a dapted in the future than at present, regardless of where it moves. Location GO asks whether any current populations have a genetic composition that is well adapted to the future environmental conditions of that location. A low Location GO score means that there are populations currently available with a genomic composition that is well adapted to the location's future climate. A high Location GO indicates that no current genotype will be adapted to the future climate at the location. Under this framework, local GO fixes genotypes in place, Population GO fixes the genetics of a population but allow s its location to change, while Location GO fixes location, but allows genotypes to disperse to this location from any current population. To explore how Population GO and Location GO would be impacted by different kelp management strategies, we considered two additional versions of the GOs. The first version limited migration to a set maximum distance of 50 km, in line with current guidelines for the maximum distance that kelp can be transplanted in Alaska, BC, and Washington (Gruenthal and Habicht 2022; Cui 2023; McConnell et al. 2024). In this version of the GO analysis, Population GO 50 km asks how well the genotype at a location would do in the future, assuming it could only migrate or be transferred up to 50 km; Location GO 50 km asks if there are suitable genotypes to colonize its .CC-BY-NC 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 4, 2026. ; https://doi.org/10.64898/2026.04.01.715974doi: bioRxiv preprint future climate within 50 km. The second additional version of the GO analysis is similar but instead of 50 km it considers migration within pre-defined geographic regions for each species (Population GOREGION and Location GO REGION) (Fig. 1A). These regions were largely delineated based on neutral population genetic structure (Bemmels et al. 2025) , with minor adjustments to reflect geographic boundaries and areas of conservation importance. Validation of genomic offsets To validate our genomic offset estimates, we used kelp abundance data from Starko et al. (2024) for each species and asked whether sites with extirpated populations show higher genomic offsets than sites with stable populations. We registered presence -absence at two timepoints: T1 (1997 - 2007) and T2 (2018-2021). The timepoints were separated by 14 to 27 years (median = 14 years for both species). Sites where the species was present at T1 were classified as ‘stable’ if the species was present at both timepoints or 'loss' if it was absent at T2. Sites where the species was absent at T1 were not included in the analyses. Then, for each species, we calculated local genomic offsets for 2020 as above. Values of environmental variables for 2020 were estimated as the average between historical (1986-2005) and projected values for the period 2046-2065 under no mitigation scenario (RCP8.5). Finally, for each site with presence-absence data (stable or loss), we extracted the genomic offset of the closest point in the grid (mean distance = 1.07 and 1.04 km for Macrocystis and Nereocystis, respectively), which gave us 295 and 447 sites with stable or loss data from 83 and 97 points with genomic offset estimates for Macrocystis and Nereocystis, respectively. Weighted logistic regressions were used to model the probability (hereafter risk) of extirpation as a function of the estimated genomic offset for 2020. Finally, for Macrocystis, we estimated the risk of extirpation by 2040 under the moderate mitigation scenario (RCP.4.5) and evaluated the potential impact of assisted migration on this risk. We evaluated four scenarios of assisted migration: 1) no migration; 2) unrestricted migration; 3) migration within regions; and 4) migration within 50 km. For these analyses, we chose Macrocystis over Nereocystis because model fit was substantially better for Macrocystis, and the moderate mitigation scenario because most predicted genomic offsets of the no m itigation scenario (RCP.8.5) fall beyond the range that the model can predict without requiring extrapolation. .CC-BY-NC 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 4, 2026. ; https://doi.org/10.64898/2026.04.01.715974doi: bioRxiv preprint

Environmental characterization of kelp forest habitats Using 14 environmental variables, we explored the environmental variation experienced by kelp forests in the region. In the environmental PCA, the first two axes explained 51% of variation (Fig. 1B). The environmental distribution of both species largely o verlaps (Fig. 1C), and our sampled sites span most of the environmental variation found in the region (Fig. 1B, C). Although our sampling spanned over seven degrees latitude, temperature did not follow latitude (Fig. 1D). Some of the northern regions, for example, inner Haida Gwaii, had higher summer temperature s than many southern areas. Also, most locations in the Strait of Georgia had higher summer temperatures than locations from similar latitudes in the Outer South Coast. Identification of genomic variants associated with local environmental adaptation We identified 41 and 77 10 kb windows significantly associated with environmental variation in Macrocystis and Nereocystis, respectively. Associated windows were widely distributed across the genome, containing 1,914 (5-120 per window) and 2,910 (9-102 per window) SNPs in Macrocystis and Nereocystis, respectively (Fig. 2A). Most environmental variables, all but sea water practical salinity in January and July (salt_01 and salt_07, respectively), were significantly associated with at least one genomic region (Fig. 2B). For Macrocystis, aragonite saturation state in January (omega_01), followed by dissolved oxygen concentration in January (O2_01) and nitrate concentration in July (NO3_07), showed the most associations while for Nereocystis, fetch (a proxy for wave exposure) , followed by mean sea surface temperature in January (temp_01) and July (temp_07), showed the most associations . Further selecting only SNPs that had a stronger association with environment than population structure or geography further restricted the datasets to 944 and 1,590 SNPs, hereafter ‘adaptive’ SNPs. For comparison, we used 1,000 and 1,600 SNPs from 41 and 77 randomly chosen windows not associated with any environmental variable (all p- values > 0.1) for Macrocystis and Nereocystis, respectively. Hereafter, we refer to this set of SNPs as ‘neutral’. We used two complementary approaches to compare adaptive and neutral sets of SNPs. First, we performed Mantel tests to evaluate patterns of isolation by distance (IBD) and isolation by environment (IBE). IBD was substantially stronger for Macrocystis than for Nereocystis (r2 = 0.81 .CC-BY-NC 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 4, 2026. ; https://doi.org/10.64898/2026.04.01.715974doi: bioRxiv preprint and 0.36, respectively; Fig 2C) , and it was slightly stronger for neutral than adaptive SNPs (r 2 = 0.81 vs. 0.75 and 0.36 vs. 0.35 for Macrocystis and Nereocystis, respectively; Fig. 2C). In contrast, IBE was stronger for adaptive than neutral loci in both species (r2 = 0.81 vs. 0.73 and 0.53 vs. 0.3 for Macrocystis and Nereocystis, respectively; Fig. 2C), providing validating evidence for the selection of adaptive loci. When IBD and IBE were analyzed together using multiple regression, both IBD and IBE were significant in both species and sets of SNPs (r 2 = 0.70 and 0.72 and 0.16 and 0.30 for neutral and adaptive SNPs in Macrocystis and Nereocystis, respectively). Finally, we used a machine learning approach (gradient forest, GF), to show that the allele frequency turnover was stronger and more rapid for adaptive than neutral SNPs in both species (Fig. 3). Together, these complementary analyses indi cate that our set of adaptive loci are more strongly structured by environmental variation than by geographic d istance alone and therefore likely reflect meaningful signals of local adaptation. Genomic offsets to predict vulnerability to climate change First, we used GF analysis to determine the relative importance of environmental variables shaping the distribution of adaptive genetic variation in each of the species. For Macrocystis, temp_07, O2_01, and temp_01 were the most relevant environmental variables (Fig. 3), with all three showing a strong and rapid allele frequency turnover (Fig. 3). For Nereocystis, fetch was the most relevant environmental variable, followed by total primary production in July (TPP_07), and temp_01 (Fig. 3C; 3D). Genomic offsets (GOs) were estimated as the genetic distance between predicted genotypes for current and future conditions for individual locations (Local GO). We also estimated the genetic distance between the predicted genotypes for future conditions and the best -matched current predicted genotype at any location (Location GO), as well as the genetic distance between the current genotype at a location and the best-matched predicted genotype at any location (Population GO). Genomic offsets (GOs) were large r for the warming scenario with no mitigation (RCP8.5) than the warming scenario with moderate mitigation (RCP4.5), though both estimates were strongly correlated (r = 0.97 and 0.91 for Macrocystis and Nereocystis, respectively), so we focus below on results of the moderate mitigation scenario (RCP4.5) to be conservative. In general, warmer sites during summer showed larger GOs than colder ones in both species (correlation .CC-BY-NC 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 4, 2026. ; https://doi.org/10.64898/2026.04.01.715974doi: bioRxiv preprint between local GO and temp_07 (r) = 0.57 and 0.31 for Macrocystis and Nereocystis, respectively). Among regions, the northernmost ones (the inner coast of Haida Gwaii and the North Coast; see Fig. 1A for definitions of regions) and Barkley Sound showed the highest values of Local GOs, indicating that these regions are the most vulnerable to global change (Fig. 4A). In addition, both Population GOs and Location GOs were significantly lower than Local GOs (Fig. 4B), indicating that migration can significantly attenuate maladaptation to future environments. This was also true when either 50 km (Population GO 50km and Location GO 50km) or within -region (Population GOREGION and Location GOREGION) migration restriction was employed, although attenuation was stronger when migration was allowed within regions (Fig. 4B). As migration is predicted to attenuate maladaptation to projected environments, for each species, we mapped the geographic distance between each site and its predicted optimal source population under future environments (Fig. 5). For Macrocystis, the Strait of Juan De Fuca, Haida Gwaii and the North Coast showed the lowest distances, indicating that local or nearby genotypes are the best option under future environments, regardless of whether estimated GOs are overall low (Strait of Juan De Fuca) or high (Haida Gwaii and North Coast). For all the other regions (i.e., North Vancouver Island, Outer South Coast, Central Coast, and Barkley Sound), long -distance migrations are required to maintain optimal genotype -environment combinations under future environmental conditions. For Nereocystis, the Strait of Juan de Fuca and North Vancouver Island showed the lowest distance values, while the remaining regions showed overall long distances between sites and their predicted optimal source population under future environments (Fig. 5). Genomic offsets predict local kelp declines To validate our genomic offset statistics, we used kelp distribution data from natural populations monitored at two timepoints that spanned a period of significant ocean warming: T1 (1997-2007) and T2 (2018 -2021) and model led the probability of extirpation as a function of the estimated genomic offset for 2020. If genomic offset represents an index of maladaptation, sites with extirpated populations should present larger genomic offsets than sites with stable populations. For both species, we found a significant association, in the predicted direction, between genomic offset for 2020 and the risk of extirpation (Fig. 6A), though model fit was substantially better for Macrocystis (p < 2e -16; pseudo-McFadden r2 = 0.58) than for Nereocystis (p = 0.029; pseudo - .CC-BY-NC 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 4, 2026. ; https://doi.org/10.64898/2026.04.01.715974doi: bioRxiv preprint McFadden r2 = 0.01). Then, for Macrocystis, we found that without migration, an overall high risk of extirpation across the region is predicted (median = 61%; Fig. 6B and 6C), indicating high vulnerability to climate change. All migration scenarios are predicted to attenuate maladaptation by significantly reducing the risk of extirpation (median = 8%, 28%, and 30% for unrestricted, regional, and 50 km migration scenarios, respectively; Fig. 6B and 6C). However, for the most vulnerable regions (e.g., Barkley Sound and Haida Gwaii), only unrestricted migration (scenario 2) could significantly attenuate maladaptation (Fig. 6B).

We characterized genomic patterns of environmental adaptation in two canopy -forming kelp species and demonstrated that many kelp populations are at risk of extirpation in the coming decades, but this risk could be reduced through assisted migration of pre -adapted genotypes to match future climates. Despite largely co -occurring across BC and Washington, the species differed in the environmental gradients that showed the strongest signatures of adaptation. Geographic sites and regions differed substantially in their predicted vulnerability to future climates and in their geographic distance to the source population with the optimal genomic match to future climates (i.e., lowest genomic offset). Genomic offsets significantly predicted kelp loss observed to date, thus validating the link between genomic vulnerability and actual ecological outcomes and allowing us to more confidently use genomic offsets to predict the risk of extirpation in future decades. We also explored several possible assisted migration management scenarios and found that current conservative policies for kelp movement do allow scope for assisted migration to reduce extirpation risk, but that longer distance migration is needed to substantially reduce extirpation risk for many populations. Over all, our results confirm the ecological relevance of genomic offsets and make predictions about the effects of different management strategies and which populations are most vulnerable, demonstrating the utility of genomic offsets to guiding kelp conservation strategies in the face of global change. Kelp populations show signatures of local adaptation to environmental gradients Because predictions of genomic maladaptation rel y on historical genotype -environment associations, accurately identifying genomic regions underlying environmental adaptation is critical for predicting response to climate change. Most studies in kelp have used reduced .CC-BY-NC 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 4, 2026. ; https://doi.org/10.64898/2026.04.01.715974doi: bioRxiv preprint representation sequencing methods (e.g., Vranken et al. 2021; Wood et al. 2021 ; Minne et al. , 2025) or sampled across small geographic areas (Abbot et al., 2025), thus missing many adaptive regions due to low genetic and/or spatial resolution (Lowry et al., 2017; Layton et al., 2024). Our study addresses those main caveats by combining wh ole genome sequencing, a comprehensive sampling across the Northeast Pacific Coastline, and high -resolution climate data (both present and future). We identified the most relevant environmental drivers of local adaptation, and several genomic regions strongly associated with environmental gradients. Interestingly, even when species distributions largely overlap across most of the studied area, the most relevant environmental drivers of local adaptation were distinct. For Macrocystis, the mean temperature in July showed the most rapid and strongest genetic turnover, but it had little importance for Nereocystis, while the opposite was true for fetch, a proxy for wave exposure. The importance of fetch for Nereocystis but not Macrocystis adaptation may reflect the wider distribution of Nereocystis in BC (Druehl 1978). Nereocystis has a higher tolerance than Macrocystis to low salinity, allowing it to occupy both exposed outer coasts as well as inner seas and narrow fjords. Consequently, Nereocystis occupies a wider gradient in fetch, perhaps making it a more important axis of adaptation overall. It is more surprising that the mean temperature in July was important for adaptation in Macrocystis but not Nereocystis, given that extreme summer temperatures have been linked to die -off in both species (Rogers-Bennett and Catton 2019; Mora-Soto et al., 2024a; Starko et al., 2024). However, because Nereocystis has historically been capable of occupying some of the warmest waters in BC, including the north Salish Sea (Mora -Soto et al. 2024b), it is possible that summer temperatures have not historically been highly stressful for Nereocystis and were not a primary axis of adaptive population differentiation. Though fetch and summer temperature are known to be relevant for kelp distribution and persistence (Pfister et al., 2018; Mora-Soto et al., 2024a; Starko et al., 2024; Man et al., 2025) , to our knowledge, this is the first study showing different adaptive responses between co-occurring kelp species in the Northeast Pacific Coastline . This information is critical for designing management zones based on environmental factors and predicting species’ vulnerability to global change (Yu et al., 2022; Lachmuth et al., 2023; Jacquemart et al., 2025). .CC-BY-NC 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 4, 2026. ; https://doi.org/10.64898/2026.04.01.715974doi: bioRxiv preprint Genomic offsets and importance for conservation The identification of regions most vulnerable to global change is key to prioritizing conservation and restoration efforts (Theissinger et al., 2023; Timm et al., 2023; Wood et al., 2024 b). In our study, genomic vulnerability differed greatly among regions (though also within regions), with warmer sites showing greater risk than colder ones. Similar patterns were observed in other kelp species and regions, e.g., Ecklonia radiata in Australia (Vranken et al., 2021; Minne et al., 2025). However, unlike other regions where ocean temperature (and predicted genomic vulnerability) follows a latitudinal gradient, in the Northeast Pacific, the complex geography of the coastline creates a mosaic of cold and warm microclimates across the latitudi nal gradient , resulting in mosaics of differential predicted vulnerability to global change. Supporting this, large differences in kelp abundance are associated with thermal gradients within distances as short as 16 km (Starko et al., 2022). As historical temperatures at warm edges represent the extreme of species’ tolerance, relatively small temperature increases can make these environments unsuitable for the species (Rehm et al., 2015; Fredston-Hermann et al., 2020; Vranken et al., 2021; Minne et al., 2025). For this same reason, populations from the warmest sites are more susceptible to marine heat waves (Tait et al., 2021; Starko et al., 2022) . The effects of marine heat waves are not included in our models and will likely exacerbate populations’ vulnerability to global change (Rogers-Bennett and Catton 2019; Tait et al. 2021). We identified the northernmost region, which includes the inner coast of Haida Gwaii and the North Coast, as the most vulnerable to global change. Declines in kelp abundance have been observed in these regions (Starko et al. 2024), especially at the warmest sites (Gendall et al. 2025), suggesting a direct link between kelp abundance and ocean warming (Starko et al., 2024; Gendall et al., 2025). Though this pattern is counter to the current paradigm that low latitude populations are most at risk, ocean temperature – the best predictor of genomic vulnerability in our study – did not follow a latitudinal gradient in the Pacific Coastline, which is characterized by high microclimate variation. Besides the northernmost regions, other regions like Barkley Sound for Macrocystis and the Strait of Georgia for Nereocystis also showed high vulnerability, which is in line with previous local - scale studies showing declines in kelp abundance in these regions (Mora-Soto et al., 2024a; Starko .CC-BY-NC 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 4, 2026. ; https://doi.org/10.64898/2026.04.01.715974doi: bioRxiv preprint et al., 2024). In contrast, regions like North Vancouver Island, the Strait of Juan de Fuca, and the Central Coast of BC showed the lowest vulnerability, suggesting kelps are overall resilient in these areas, which is supported by local-scale studies showing that kelp abundance has been stable in these regions (Mora-Soto et al., 2024a; Man et al., 2025) . Therefore, our study identified vulnerable regions where management interventions should be prioritized and helps explain contrasting patterns of kelp vulnerability/resilience observed in local-scale studies. The qualitative agreement between regions predicted to be most vulnerable and those that have exhibited recent declines suggests that kelp populations in the most vulnerable areas may already be facing maladaptation, highlighting the need for flexible kelp management strategies that consider local adaptation. Genomic offset predictions are rarely validated (but see Fitzpatrick et al., 2021; Luo et al., 2025; Verrico et al., 2026) , and to our knowledge, no studies have validated genomic offset in marine species to date (Layton et al., 2024) , which hinders its implementation in marine conservation programs. Here, using kelp distribution data obtained mostly from aerial images, we found the expected positive association between genomic vulnerability and the risk of extirpation in both species, thus validating genomic predictions. Validation models performed substantially better for Macrocystis than for Nereocystis, suggesting that factors other than the environmental ones are relevant for Nereocystis persistence. For example, the presence and density of herbivores (mainly sea urchins) likely affect kelp persistence across the range (Filbee-Dexter and Scheibling, 2014; Ling et al., 2015; Starko et al., 2022) . Though s ea urchins consume both Macrocystis and Nereocystis, as they prefer young kelp recruits, the negative impact of overgrazing could be strongest for the annual species Nereocystis. In addition, the range of model prediction is narrow compared with the genomic offset estimates, as the most vulnerable sites (i.e., Local GOs > 0.06) are not represented in the abundance dataset. For the same reason, we might be overestimating the risk of extirpation for Macrocystis at sites where genomic offset exceeds the range of model prediction. Another factor likely affecting validation models is the time lapse between observations. The median time lapse is 14 years, for most observations starting in 2004 – 2007, thus missing sites where extirpation occurred before the beginning of the study period . Also, the time lapse was originally thought to capture the effects of a sustained heat event during 2014 – 2016 on kelp persistence (Starko et al., 2024), and according to our models, summer temperatures are more relevant for Macrocystis than for Nereocystis. Including more sites covering a broader .CC-BY-NC 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 4, 2026. ; https://doi.org/10.64898/2026.04.01.715974doi: bioRxiv preprint area, longer timeseries, the frequency and intensity of extreme weather events (e.g., heatwaves), and proxies for both biotic interactions and human pressure is expected to improve our models and to better inform kelp forest conservation and restoration. To this end, remote sensing technologies could represent a powerful, high-throughput, and cost and time-efficient method to further validate genomic predictions in kelp. Complementary offset statistics identify valuable germplasm for restoration Restoration efforts in kelp are becoming important in several regions of the globe (Eger et al., 2022; Wood et al., 2024 b) and promising approaches for large -scale restoration are under development (Fredriksen et al., 2020; Dykman et al., 2025). Yet, there are no clear guidelines for the optimal source population selection, which is an important factor determining restoration success (Houde et al., 2015). Here, we show that local genotypes are rarely the best adapted under projected future environmental conditions and long -distance migrations among regions are often required to minimize genotype -environment mismatch. In BC and Washington, the 50 km rule restricts the transfer of kelp between populations more than 50 km apart (Cui 2023; McConnell et al. 2024), narrowing down the options for the optimal source population selection. Our results show that moving genotypes either within 50 km or within geographic regions (largely based on neutral genetic structure) both offer significant and similar attenuation of extirpation risk, even though geographic regions as defined here often extend much further than 50 km. This finding likely reflects the high microclimatic variation along the coasts of BC and Washington, w here geographically proximate locations can experience dramatically different environmental conditions (but see Fig. 1D). One caveat to our approach is that we use all possible kelp habitats as a possible donor, but neither kelp species occupies all possible locations in its range. This means that we may predict an ideal donor population to exist at a location where there is no current kelp. From an operational standpoint, this issue could be partly mitigated by using a geographically broader area from which to select donor populations, to help improve the chances of identifying locations with locally adapted kelp present. For example, as most regions we defined extend over a much larger area than 50 km (Fig. 1A), the region-wide assisted migration policy may be more practical choice than sourcing within 50 km, even though the theoretical benefit under the assumption of total occupancy .CC-BY-NC 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 4, 2026. ; https://doi.org/10.64898/2026.04.01.715974doi: bioRxiv preprint of all sites is small (Fig. 6C). Another caveat is that our approach assumes local adaptation to microclimates. In other words, even if kelp is present in a given microclimate, that population may not necessarily be locally adapted and thus may not provide benefits as a donor population (Dykman et al., 2025). A potential solution to overcome this caveat is the characterization of the adaptive genetic variation of potential donor populations, i.e., testing if donors indeed represent novel genotypes predicted to be well adapted to an outplanting environment. Th ough whole - genome sequencing for many samples may be cost -prohibited, our study provides a low number of genomic regions (< 100 per species) strongly associated with environmental gradients, which can be target -sequenced to allow cost -effective genotyping. Combining target sequencing with pool-seq approaches would further reduce genotyping costs, facilitating its adoption in conservation programs. Genomic offsets and extirpation risk are lowest of all when there are no restrictions placed on migration distance. This suggests that extreme long -distance migration has the potential to substantially improve resilience in most regions. The potential adap tive benefits of long-distance assisted migration should be weighed against the possibility of outbreeding depression (Frankham et al. 2011) between groups that may have been genetically isolated for long periods of time (Bemmels et al., 2026) . In addition, our models do not consider the possibility that populations may be able to adaptively evolve (i.e., experience adaptive change in allele frequencies), reducing the need for long-distance migration, although recent declines in the Northeast P acific Coastline have been linked to ocean warming (Starko et al., 2022; Gendall et al., 2025) suggesting that kelp species are not adapting fast enough to keep pace with global change. Also, long -distance migrations may negatively alter biotic interaction s of kelp (e.g., with urchins or coralline algae), which are not included in our models but are potentially relevant for local adaptation (Twist et al., 2024). Finally, we also acknowledge that deliberately altering the genetic composition of natural populations raises bioethical issues that would require careful consideration from stakeholders before implementation (Coleman et al. 2020). Despite potential side -effects and uncertainties, long-distance migration may be the only viable option for restoring ar eas where kelp have long been extirpated, and nearby populations either do not exist or are declining. Regardless of the management policy pursued for the transport of individuals, we identified geographic sites that are the most likely to exhibit genomic mismatch with future climates (i.e., .CC-BY-NC 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 4, 2026. ; https://doi.org/10.64898/2026.04.01.715974doi: bioRxiv preprint more vulnerable to global change). These sites are mostly distributed in Haida Gwaii and the North Coast for both species, and additionally in Barkley Sound for Macrocystis. Some sites within these regions are still predicted to have a high risk of extirpation under most migration management scenarios, suggesting that even targeted interventions may be insufficient to protect some kelp populations in BC from extirpation. Importantly, n orthern regions harbour the highest genetic diversity within BC and Washington for both species (Bemmels et al. 2025), suggesting that they may be important reservoirs for genetic diversity in general, potentially including adaptive variation to factors other than climate. This combination of factors (high risk and high genetic diversity) highlights that Haida Gwaii and the North Coast should be a priority for conservation resources and research about the impacts of global change. In contrast, areas predicted to have a lower mismatch with future climates (such as North Vancouver Island and Juan de Fuca) could be ideal candidate locations for marine protected areas or other tools emphasizing in situ conservation.

Genomic vulnerability is becoming an increasingly important tool to predict maladaptation driven by global change. However, the lack of validation of such predictions has prevented the widespread implementation of GEA analyses in marine and terrestrial conservation programs. Here, using a seascape genomics approach in two kelp species, we identified areas along the Northeast Pacific Coastline of high genomic vulnerability that should be prioritized in conservation planning. We also link genomic vulnerability to extirpation risk, which allows us to quantify the predicted benefit of assisted migration in ecologically meaningful terms (i.e., risk of extirpation). The link between genomic offsets and extirpation observed to date further suggests that the most vulnerable populations may already be facing maladaptation to changing climates, highlighting the need for intervention to ensure their persistence in changing environments. Our

can inform the selection of source pop ulations for assisted migration or restoration in regions where local populations have been extirpated and can help policymakers calculate the costs and benefits of assisted migration policy. Nonetheless, further validation (e.g., of transplanted populations) and incorporation of non-environmental factors potentially affecting kelp persistence are needed to improve model predictions and to better inform conservation and restoration. .CC-BY-NC 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 4, 2026. ; https://doi.org/10.64898/2026.04.01.715974doi: bioRxiv preprint

We thank the Gitga'at, Gitxaała, Haida, Haisla, Heiltsuk, Kitasoo-Xai'xais, Kitselas, Kitsumkalum, K'ómoks, Mamalilikulla, Metlakatla, Tlowitsis, and Wei Wai Kum First Nations for permission to reanalyze DNA sequences from samples collected from their territories. For a Biocultural Notice of cultural rights and responsibilities regarding these samples see Bemmels et al. (2025). Funding was provided by the Natural Sciences and Engineering Research Council of Canada (RGPIN - 2021-02482 to GLO) and Genome British Columbia (GIRAFF Grant to GLO and LHR ). Postdoctoral support was provided by a Mitacs Accelerate Postdoctoral Fellowship to JBB and a Biodiversity Research Centre Fellowship to FH.

Abbott, E., C. Alves-Lima, J. Toft, and F. Alberto. (2025). Evidence of local adaptation in Puget Sound’s threatened bull kelp ( Nereocystis luetkeana ) populations. bioRxiv Preprint 2025.10.20.683549. Alaska DNR. 2021. Alaska coastline 1:250,000. North Slope Science Initiative, Dataset 851. https://catalog.northslopescience.org/dataset/851 Attridge, C. M., K. D. Cox, B. Maher, S. Gross, E. G. Lim, K. R. Kattler, and I. M. Côté. 2024. Studying Kelp Forests of Today to Forecast Ecosystems of the Future. Fisheries 49: 181– 187. Bemmels, J. B., S. Starko, B. L. Weigel, K. Hirabayashi, A. Pinch, C. Elphinstone, M. N. Dethier, et al. 2025. Population genomics reveals strong impacts of genetic drift without purging and guides conservation of bull and giant kelp. Current Biology 35: 688-698.e8. Bemmels, J. B., K. J. Kroeker, S. R. Palumbi, R. A. Bay, K. M. Gruenthal, S. C. Lindstrom, M. M. Osmond, G. L. Owens . 2026. Spatial inference of ancestor locations suggests northern refugia for canopy -forming kelps in the Northeast Pacific . bioRxiv Preprint 2026.01.09.698529. Booker, T. R., S. Yeaman, J. R. Whiting, and M. C. Whitlock. 2024. The WZA: A window-based

for characterizing genotype–environment associations. Molecular Ecology Resources 24. British Columbia Marine Conservation Analysis Project Team. 2011. Marine Atlas of Pacific Canada: A Product of the British Columbia Marine Conservation Analysis. Available from www.bcmca.ca. Bucharova, A., W. Durka, N. Hölzel, J. Kollmann, S. Michalski, and O. Bossdorf. 2017. Are local plants the best for ecosystem restoration? It depends on how you analyze the data. Ecology and Evolution 7: 10683-10689. Capblancq, T., M. C. Fitzpatrick, R. A. Bay, M. Exposito-alonso, and S. R. Keller. 2020. Genomic Prediction of (Mal)Adaptation Across Current and Future Climatic Landscapes. Annual Review of Ecology, Evolution, and Systematics 51: 245–269. .CC-BY-NC 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 4, 2026. ; https://doi.org/10.64898/2026.04.01.715974doi: bioRxiv preprint Caye, K., B. Jumentier, J. Lepeule, and O. François. 2019. LFMM 2: Fast and Accurate Inference of Gene -Environment Associations in Genome -Wide Studies. Molecular biology and evolution 36: 852–860. Coleman, M. A., G. Wood, K. Filbee -Dexter, A. J. P. Minne, H. D. Goold, A. Vergés, E. M. Marzinelli, et al. 2020. Restore or Redefine: Future Trajectories for Restoration. Frontiers in Marine Science 7. Cui, T. 2023. State Environmental Policy Act (SEPA) Mitigated Determination of Nonsignificance (MDNS): Vashon Kelp Forest (No. Permit SHOR22 -0017). King County Department of Local Services, Renton, WA, USA. Demes, K. W., M. H. Graham, and T.S. Suskiewicz. 2009. Phenotypic plasticity reconciles incongruous molecular and morphological taxonomics: the giant kelp, Macrocystis (Laminariales, Phaeophyceae), is a monospecific genus. Journal of Phycology 45: 1266 – 1269. Diggon, S., J. Bones, C. J. Short, J. L. Smith, M. Dickinson, K. Wozniak, K. Topelko, and K. A. Pawluk. 2022. The Marine Plan Partnership for the North Pacific Coast – MaPP: A collaborative and co-led marine planning process in British Columbia. Marine Policy 142. Druehl, L. D. 1978. The distribution of Macrocystis integrifolia in British Columbia as related to environmental parameters. Canadian Journal of Botany 56: 69-79. Dykman, L. N., S. C. Steell, J. B. Bemmels, G. K. Melchers, B. D. Timmer, G. L. Owens, C. J. Neufeld, and J. K. Baum. 2025. Testing the roles of local adaptation and genetic diversity to improve Giant kelp (Macrocystis pyrifera) restoration. Restoration Ecology. Eger, A. M., E. M. Marzinelli, H. Christie, C. W. Fagerli, D. Fujita, A. P. Gonzalez, S. W. Hong, et al. 2022. Global kelp forest restoration: past lessons, present status, and future directions. Biological Reviews 97: 1449–1475. Ellis, N., S. J. Smith, and C. R. Pitcher. 2012. Gradient Forests: calculating importance gradients on physical predictors. Ecology 93: 156-168. Erlandson, J. M., M. H. Graham, B. J. Bourque, D. Corbett, J. A. Estes, and R. S. Steneck. 2007. The kelp highway hypothesis: Marine ecology, the coastal migration theory, and the peopling of the Americas. Journal of Island and Coastal Archaeology 2: 161–174. Fitzpatrick, M.C., and S. R. Keller. 2015. Ecological genomics meets community-level modelling of biodiversity: mapping the genomic landscape of current and future environmental adaptation. Ecology Letters 18: 1-16 Fitzpatrick, M.C., V. E. Chhatre, R. Y. Soolanayakanahally, and S. R. Keller. 2021. Experimental support for genomic prediction of climate maladaptation using the machine learning approach Gradient Forests. Molecular Ecology Resources 21: 2749-2765 Frankham R, Ballou JD, Eldridge MDB, Lacy RC, Ralls K, Dudash MR, Fenster CB. Predicting the probability of outbreeding depression. Conservation Biology 25: 465–475. Fredriksen, S., K. Filbee-Dexter, K. M. Norderhaug, H. Steen, T. Bodvin, M. A. Coleman, F. Moy, and T. Wernberg. 2020. Green gravel: a novel restoration tool to combat kelp forest decline. Scientific Reports 10. Fredston-Hermann, A., R. Selden, M. Pinsky, S. D. Gaines, and B. S. Halpern. 2020. Cold range edges of marine fishes track climate change better than warm edges. Global Change Biology 26: 2908–2922. Gendall, L., M. Hessing -Lewis, A. Wachmann, S. Schroeder, L. Reshitnyk, S. Crawford, L. C. Lee, et al. 2025. From archives to satellites: uncovering loss and resilience in the kelp forests of Haida Gwaii. Frontiers in Marine Science 12. .CC-BY-NC 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 4, 2026. ; https://doi.org/10.64898/2026.04.01.715974doi: bioRxiv preprint GeoBC Branch. 2002. NTS BC coastline polygons 1:250,000. B.C.’s Map Hub, ID 5ebbc1fd1d7e456684b8d1a3f10186e4. http://governmentofbc.maps.arcgis.com. Gougherty, A. V., S. R. Keller, and M. C. Fitzpatrick. 2021. Maladaptation, migration and extirpation fuel climate change risk in a forest tree species. Nature Climate Change 11: 166– 171. Graham, M. H., C. Harrold, S. Lisin, K. Light, J. M. Watanabe, and M. S. Foster. 1997. Population dynamics of giant kelp Macrocystis pyrifera along a wave exposure gradient. Marine Ecology Progress Series 148: 269-279. Gregr, E., M. Peterman, and J. Lessard. 2018. Coastline fetch estimates for Pacific Canada. Marine Spatial Ecology Section, Fisheries and Oceans Canada, Nanaimo, BC, Canada. https://open.canada.ca/data/en/dataset/412431c4-7363-410e-86a4-76feb9a6dcde. Gruenthal, K.M., and C. Habicht. 2022. Literature review for implementation of the 50-50 rule for cultivation of seaweeds and other aquatic plants in Alaska. Alaska Department of Fish and Game, Regional Information Report No. 2A22-01, Juneau. Holdsworth, A. M., L. Zhai, Y. Lu, and J. R. Christian. 2021. Future changes in oceanography and biogeochemistry along the Canadian Pacific continental margin. Frontiers in Marine Science 8: 602991. Houde, A. L. S., S. R. Garner, and B. D. Neff. 2015. Restoring species through reintroductions: Strategies for source population selection. Restoration Ecology 23: 746–753. Jacquemart, A. S., A. Tigano, M. K. Gale, T. Weir, H. G. M. Ward, C. M. Wong, E. J. Eliason, et al. 2025. Application of Genomic Offsets to Inform Freshwater Fisheries Management Under Climate Change. Evolutionary Applications 18. Lachmuth, S., T. Capblancq, S. R. Keller, and M. C. Fitzpatrick. 2023. Assessing uncertainty in genomic offset forecasts from landscape genomic models (and implications for restoration and assisted migration). Frontiers in Ecology and Evolution 11. Layton, K. K. S., M. S. O. Brieuc, R. Castilho, N. Diaz -Arce, D. Estévez-Barcia, V. G. Fonseca, A. P. Fuentes -Pardo, et al. 2024. Predicting the future of our oceans —evaluating genomic forecasing approaches in marine species. Global Change Biology 30: e17236. Lindstrom, S. C. 2023. A reinstated species name for north -eastern Pacific Macrocystis (Laminariaceae, Phaeophyceae). Notulae Algarum 290: 1–2. Ling S. D., R. E. Scheibling, A. Rassweiler, C. R. Johnson, N. Shears, S. D. Connell, A. K. Salomon, et al. 2015. Global regime shift dynamics of catastrophic sea urchin overgrazing . Philosophical Transactions of the Royal Society B 370: 20130269. Lowry, D. B., S. Hoban, J. L. Kelley, K. E. Lotterhos, L. K. Reed, M. F. Antolin, and A. Storfer. 2017. Breaking RAD: an evaluation of the utility of restriction site -associated DNA sequencing for genome scans of adaptation. Molecular Ecology Resources 17: 142-152. Luo, Y. C. M. Lors , E. H. Lawrence -Paul, and J. R. Lasky. 2025. Experimental validation of genome-environment associations in Arabidopsis. Molecular Ecology 34: e70129. Man, L., R. V. Barbosa, L. Y. Reshitnyk, L. Gendall, A. Wachmann, N. Dedeluk, U. Kim, et al. 2025. Canopy -forming kelp forests persist in the dynamic subregion of the Broughton Archipelago, British Columbia, Canada. Frontiers in Marine Science 12. McConnell, A., C. Neufeld, L. Romero, and K. Truman. 2024. Seaweed Industry Development Plan: Regional District of Mount Waddington. LGL Limited Environmental Research Associates, Sidney, BC, Canada. Minne, A. J. P., S. Vranken, D. Wheeler, G. Wood, J. Batley, T. Wernberg, and M. A. Coleman. 2025. Strong Environmental and Genome -Wide Population Differentiation Underpins .CC-BY-NC 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 4, 2026. ; https://doi.org/10.64898/2026.04.01.715974doi: bioRxiv preprint Adaptation and High Genomic Vulnerability in the Dominant Australian Kelp (Ecklonia radiata). Ecology and Evolution 15. Mora-Soto, A., S. Schroeder, L. Gendall, A. Wachmann, G. R. Narayan, S. Read, I. Pearsall, et al. 2024. Kelp dynamics and environmental drivers in the southern Salish Sea, British Columbia, Canada. Frontiers in Marine Science 11. Oksanen, J., G. L. Simpson, F. G. Blanchet, R. Kindt, P. Legendre, P. R. Minchin, R. B. O’Hara, et al. 2022. Vegan: community ecology package. R package, https://cran.r- project.org/web/packages/vegan/index.html. Pessarrodona, A., J. Assis, K. Filbee-Dexter, M. T. Burrows, J. Gattuso, C. M. Duarte, D. Krause- et al. 2022. Global seaweed productivity. Science Advances 8: eabn2465. Pfister, C. A., H. D. Berry, and T. Mumford. 2018. The dynamics of Kelp Forests in the Northeast Pacific Ocean and the relationship with environmental drivers. Journal of Ecology 106: 1520– 1533. Pörtner, H.O., Roberts, D.C., Masson-Delmotte, V., Zhai, P., et al. 2019. IPCC special report on the ocean and cryosphere in a changing climate. Intergovernmental Panel on Climate Change, Geneva. Rehm, E. M., P. Olivas, J. Stroud, and K. J. Feeley. 2015. Losing your edge: Climate change and the conservation value of range-edge populations. Ecology and Evolution 5: 4315–4326. Rogers-Bennett, L., and C. A. Catton. 2019. Marine heat wave and multiple stressors tip bull kelp forest to sea urchin barrens. Scientific Reports 9: 15050. Smale, D. A. 2020. Impacts of ocean warming on kelp forest ecosystems. New Phytologist 225: 1447–1454. Starko, S., C. J. Neufeld, L. Gendall, B. Timmer, L. Campbell, J. Yakimishyn, L. Druehl, and J. K. Baum. 2022. Microclimate predicts kelp forest extinction in the face of direct and indirect marine heatwave effects. Ecological Applications 32. Starko, S., B. Timmer, L. Reshitnyk, M. Csordas, J. McHenry, S. Schroeder, M. Hessing -Lewis, et al. 2024. Local and regional variation in kelp loss and stability across coastal British Columbia. Marine Ecology Progress Series 733: 1-26. Steneck, R. S., M. H. Graham, B. J. Bourque, D. Corbett, J. M. Erlandson, J. A. Estes, M. J. Tegner. 2002. Kelp forest ecosystems: biodiversity, stability, resilience and future. Environmental Conservation, 29: 436-459.Tait, L. W., F. Thoral, M. H. Pinkerton, M. S. Thomsen, and D. R. Schiel. 2021. Loss of Giant Kelp, Macrocystis pyrifera, Driven by Marine Heatwaves and Exacerbated by Poor Water Clarity in New Zealand. Frontiers in Marine Science 8. Teagle H. and D. A. Smale. 2018. Climate - driven substitution of habitat- forming species leads to reduced biodiversity within a temperate marine community . Diversity and Distributions, 24: 1367-1380. Theissinger, K., C. Fernandes, G. Formenti, I. Bista, P. R. Berg, C. Bleidorn, A. Bombarely, et al. 2023. How genomics can help biodiversity conservation. Trends in Genetics 39: 545–559. Timm, L. E., N. Tucker, A. Rix, S. LaBua, J. A. López, K. M. Boswell, and J. R. Glass. 2023. The untapped potential of seascape genomics in the North Pacific. Frontiers in Conservation Science 4. Twist, B. A., F. Mazel, S. Zaklan Duff, M. A. Lemay, C. M. Pearce, and P. T. Martone. (2024). Kelp and sea urchin settlement mediated by biotic interactions with benthic coralline algal species. Journal of Phycology, 60: 363–379. Vergés, A., C. Doropoulos, H. A. Malcolm, M. Skye, M. Garcia -Pizá, E. M. Marzinelli, A. H. Campbell, et al. 2016. Long -term empirical evidence of ocean warming leading to .CC-BY-NC 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 4, 2026. ; https://doi.org/10.64898/2026.04.01.715974doi: bioRxiv preprint tropicalization of fish communities, increased herbivory, and loss of kelp. Proceedings of the National Academy of Sciences of the United States of America 113: 13791–13796. Verrico, B. M., T. Capblancq, M. C. Fitzpatrick, and S. R. Keller. 2026. Reciprocal evaluation of genomic offset predictions of climate maladaptation with independent empirical datasets. The American Naturalist 207: 415-432. Vranken, S., T. Wernberg, A. Scheben, A. A. Severn-Ellis, J. Batley, P. E. Bayer, D. Edwards, et al. 2021. Genotype–Environment mismatch of kelp forests under climate change. Molecular Ecology 30: 3730–3746. Waters, M. R. 2019. Late Pleistocene exploration and settlement of the Americas by modern humans. Science 365. Wernberg, T., K. Krumhansl, K. Filbee-Dexter, and M. F. Pedersen. 2019. Status and Trends for the World’s Kelp Forests, in: C. Sheppard, ed. World Seas: An Environmental Evaluation, pp. 57-58. Elsevier. Wernberg, T., M. S. Thomsen, J. K. Baum, M. J. Bishop, J. F. Bruno, M. A. Coleman, K. Filbee - Dexter, et al. 2024. Impacts of Climate Change on Marine Foundation Species. Annual Review of Marine Science 16: 247–282. Wood, G. V., E. M. Marzinelli, A. Vergés, A. H. Campbell, P. D. Steinberg, and M. A. Coleman. 2020. Using genomics to design and evaluate the performance of underwater forest restoration. Journal of Applied Ecology 57: 1988–1998. Wood, G. V., E. M. Marzinelli, A. H. Campbell, P. D. Steinberg, A. Vergés, and M. A. Coleman. 2021. Global vulnerability of a dominant seaweed points to future -proofing pathways for Australia’s underwater forests. Global Change Biology 27: 2200-2212. Wood, G. V., K. Filbee-Dexter, M. A. Coleman, J. Valckenaere, J. D. Aguirre, P. M. Bentley, P. Carnell, et al. 2024a. Upscaling marine forest restoration: challenges, solutions and recommendations from the Green Gravel Action Group. Frontiers in Marine Science 11: 1364263. Wood, G. V., K. J. Griffin, M. van der Mheen, M. F. Breed, J. M. Edgeloe, C. Grimaldi, A. J. P. Minne, et al. 2024 b. Reef Adapt: A tool to inform climate -smart marine restoration and management decisions. Communications Biology 7. Yu, Y., S. N. Aitken, L. H. Rieseberg, and T. Wang. 2022. Using landscape genomics to delineate seed and breeding zones for lodgepole pine. New Phytologist 235: 1653–1664. .CC-BY-NC 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 4, 2026. ; https://doi.org/10.64898/2026.04.01.715974doi: bioRxiv preprint Figure 1. Environmental variation and sampled sites . A: Pre-defined geographic regions used in this study. Nereocystis is present in all regions, while Macrocystis is present in all regions but Puget Sound and Strait of Georgia. B: Principal component analysis using 14 environmental variables with low correlation (r < |0.8|). Background gray and green dots correspond to sites where both species and only Nereocystis are potentially present (Puget Sound and Strait of Georgia), respectively, based on their bro ad-scale biogeographic distributions. Brown and dark green main dots represent sampled populations used in this study for Macrocystis and Nereocystis, respectively. C: Density plots showing the environmental distribution of occurrence (gray) and sampled (brown or dark green) sites for Macrocystis (top panels) and Nereocystis (bottom panels). D: Mean sea surface temperature in July showing the complex geographic pattern of environmental variation along the BC coastline. Circles corresponded to sampled sites, which may be plotted rounded to the nearest 0.5º to protect information about culturally sensitive sites (see Biocultural Notice in Bemmels et al., 2025). Figure 2. Genetic architecture of environmental adaptation. A: Manhattan plots showing the association between 10 kb windows and environmental variables; only the top association across variables is shown for each window. B: Number of significant associations found for each environmental variable. C: Isolation by distance (IBD) and isolation by environment (IBE) for neutral and adaptive loci for Macrocystis (top panels) and Nereocystis (bottom panels). Figure 3. Genetic turnover along environmental gradients. Relative importance of environmental variables in the gradient forest models and genomic turnover for adaptive (red line) and neutral (black line) SNPs across environmental gradients. For each species, the top three environmental variables are shown. Figure 4. Genomic vulnerability to climate change. A: Geographic distribution of genomic offsets for Macrocystis (top panels) and Nereocystis (bottom panels). Open circles represent genotyped populations used to build the models. B: Genomic offsets estimated for grid data. Figure 5. Geographic distance between each recipient population and its optimal source under future climates for Macrocystis (left) and Nereocystis (right). Figure 6. Genomic offsets predict local kelp declines. A: Weighted logistic model between genomic offsets for 2020 and the risk of extirpation. B: Map of the risk of extirpation by 2040 for Macrocystis. C: Median risk of extirpation under four scenarios of migration in Macrocystis. .CC-BY-NC 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 4, 2026. ; https://doi.org/10.64898/2026.04.01.715974doi: bioRxiv preprint Table 1. Final set of environmental variables used in this study. Mean, minimum, and maximum values of sampled populations are shown. * Following Gregr et al. (2018). Variable Description Unit Source Mean Minimum Maximum Fetch log10 mean Sum Fetch log10 (m) This study* 5.8 4.6 6.8 NO3_07 Nitrate concentration in July mmol m-3 NEP36-CanOE 3.4 5.4E-07 13.7 O2_01 Dissolved oxygen concentration in January mmol m-3 NEP36-CanOE 305 276 346 O2_07 Dissolved oxygen concentration in July mmol m-3 NEP36-CanOE 284 256 309 omega_a_01 Aragonite saturation state in January NEP36-CanOE 1.7 1.1 1.9 omega_a_07 Aragonite saturation state in July NEP36-CanOE 2.1 1.2 2.5 PH_01 pH in January NEP36-CanOE 8.1 8.1 8.2 PH_07 pH in July NEP36-CanOE 8.1 8.0 8.2 salt_01 Sea water practical salinity in January PSU NEP36-CanOE 28.3 18.3 31.1 salt_07 Sea water practical salinity in July PSU NEP36-CanOE 28.2 14.6 31.8 temp_01 Mean sea surface temperature in January Deg C NEP36-CanOE 7.2 3.3 9.4 temp_07 Mean sea surface temperature in July Deg C NEP36-CanOE 13.6 10.2 17.6 TPP_01 Total primary production in January mol m-3 s-1 NEP36-CanOE 1.0E-08 2.2E-09 2.2E-08 TPP_07 Total primary production in July mol m-3 s-1 NEP36-CanOE 6.7E-08 8.0E-11 1.3E-07 .CC-BY-NC 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 4, 2026. ; https://doi.org/10.64898/2026.04.01.715974doi: bioRxiv preprint Supplementary materials Figure S1. Genetic structure for Macrocystis (left) and Nereocystis (right) using 43,933 and 116,707 SNPs in low linkage disequilibrium. Each dot represents a population (site) with at least three individuals. Putative ancestral genetic clusters are shown in different colours. .CC-BY-NC 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 4, 2026. ; https://doi.org/10.64898/2026.04.01.715974doi: bioRxiv preprint Table S1. Environmental variables used in this study. The final set of variables retained after pairwise correlation < |0.8| is highlighted in bold. Variable Description Units Fetch log10 mean Sum Fetch Alkalini_01 Total alkalinity concentration in January mmol m-3 Alkalini_07 Total alkalinity concentration in July mmol m-3 Cflx_01 Air-sea CO2 flux in January mol m-2 s-1 Cflx_07 Air-sea CO2 flux in July mol m-2 s-1 DIC_01 Dissolved inorganic concentration in January mmol m-3 DIC_07 Dissolved inorganic concentration in July mmol m-3 maxO2_01 Maximum oxygen concentration in January mmol m-3 maxO2_07 Maximum oxygen concentration in July mmol m-3 maxPH_01 Maximum pH in January maxPH_07 Maximum pH in July maxtemp_01 Maximum temperature in January Deg C maxtemp_07 Maximum temperature in July Deg C minO2_01 Minimum oxygen concentration in January mmol m-3 minO2_07 Minimum oxygen concentration in July mmol m-3 minPH_01 Minimum pH in January minPH_07 Minimum pH in July mintemp_01 Minimum temperature in January Deg C mintemp_07 Minimum temperature in July Deg C NO3_01 Nitrate concentration in January mmol m-3 NO3_07 Nitrate concentration in July mmol m-3 O2_01 Dissolved oxygen concentration in January mmol m-3 O2_07 Dissolved oxygen concentration in July mmol m-3 omega_a_01 Aragonite saturation state in January omega_a_07 Aragonite saturation state in July PH_01 pH in January PH_07 pH in July salt_01 Sea water practical salinity in January PSU salt_07 Sea water practical salinity in July PSU sigma_01 Sigma potential density in January kg m-3 sigma_07 Sigma potential density in July kg m-3 TCHL_01 Total chlorophyll (diatoms + nano chlorophyl concentration) in January mg m-3 TCHL_07 Total chlorophyll (diatoms + nano chlorophyl concentration) in July mg m-3 temp_01 Mean sea surface temperature in January Deg C temp_07 Mean sea surface temperature in July Deg C TPP_01 Total primary production (diatoms + nano) in January mol m-3 s-1 TPP_07 Total primary production (diatoms + nano) in July mol m-3 s-1 .CC-BY-NC 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 4, 2026. ; https://doi.org/10.64898/2026.04.01.715974doi: bioRxiv preprint .CC-BY-NC 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 4, 2026. ; https://doi.org/10.64898/2026.04.01.715974doi: bioRxiv preprint A B C D .CC-BY-NC 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 4, 2026. ; https://doi.org/10.64898/2026.04.01.715974doi: bioRxiv preprint r = 0.81 r = 0.36 r = 0.75 r = 0.35 r = 0.73 r = 0.3 r = 0.81 r = 0.53 Isolation by distance Isolation by environment A B C .CC-BY-NC 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 4, 2026. ; https://doi.org/10.64898/2026.04.01.715974doi: bioRxiv preprint .CC-BY-NC 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 4, 2026. ; https://doi.org/10.64898/2026.04.01.715974doi: bioRxiv preprint A B .CC-BY-NC 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 4, 2026. ; https://doi.org/10.64898/2026.04.01.715974doi: bioRxiv preprint .CC-BY-NC 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 4, 2026. ; https://doi.org/10.64898/2026.04.01.715974doi: bioRxiv preprint Stable Loss n = 83 p < 2e-16 pseudo r 2 = 0.53 n = 97 p = 0.0289 pseudo r 2 = 0.01 A B C Stable Loss .CC-BY-NC 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 4, 2026. ; https://doi.org/10.64898/2026.04.01.715974doi: bioRxiv preprint