Genomic repeatability and predictability of climate (mal)adaptation in a reef-building coral

Key results

1. 01

   ShownAdaptation to mean temperature is polygenic and mostly population-specific. Only a small set of 10-kb genomic windows was repeatedly associated with temperature across the five S. pistillata populations.

   Weighted-Z analysis over 31,828 10-kb windows flagged 814 to 1,359 candidate windows per population (empirical p 0.05). PicMin found eleven windows repeated at FDR q<0.1 (one in three, three in four, seven in all five populations), against a permutation null mean of zero. 188 windows passed at q<0.5. · Genomic architecture of thermal local adaptation…; Genomic signatures of temperature adaptation…; Repeatability of thermal local adaptation across…

2. 02

   ShownRepeatability at the functional level was also limited. GO enrichment of top candidate windows differed among populations, although it included temperature response and protein refolding. The repeated windows overlap genes linked to metabolism, protein turnover, apoptosis and cell adhesion.

   Using the top 1% of windows (319 per population), only 7% (14/192) of enriched Biological Process GO terms were shared by two or three populations. The eleven repeated windows overlap 15 genes (13 protein-coding, 2 non-coding RNAs), including peroxidasins and sulfotransferase family proteins. · Functional annotations of candidate adaptive…; Discussion; Functional enrichment in candidate windows under…

3. 03

   ShownEnvironment explains a small but significant share of genome-wide variation, and 38% of the explainable signal overlaps with geography. The unique environmental component that remains is modest, and the thermal and shore-distance axis captures 45% of it.

   GBR-wide variance partitioning: environment 3.0% and geography 2.8% adjusted R². The partial RDA conditioned on geography kept 1.4%, with 38% of the full model's explainable variance confounded with geography. Env-PC1 (shore distance, thermal metrics) captured 45% of the partial model's explainable variation. All models were significant (permutation p<0.05). · Predictability of local adaptation across the…; Multivariate genotype-environment associations…

4. 04

   ShownGenotype-environment relationships learned across the GBR, with the test population left out, closely predicted genome-wide adaptive genotype scores in held-out populations. Models trained on any single reef population predicted poorly.

   Out-of-sample Euclidean distances between observed and predicted scores on the first two RDA axes were smaller for GBR-wide training than for single-population training. The geography-conditioned partial RDA was the most accurate of three GBR-wide models. Heron Reef-trained models were worst; Central Reefs-trained models were generally better. · Predictability of local adaptation across the…; Discussion; Multivariate genotype-environment associations…; Figure legends

5. 05

   SuggestedProjected genomic offsets vary within and among regions and are highest at northern and inshore reefs by mid-century. Late in the century they become broadly high under high emissions, partly in thermal conditions outside the historical training range.

   For 2040-2050 the three SSPs gave similar offset distributions. For 2090-2100, northern offsets rose under SSP2-4.5 vs SSP1-2.6 and rose range-wide under SSP5-8.5 (Tukey post-hoc p<0.05), where Far North temperatures reach +2.3°C above the historical maximum. The abstract cites up to 30% at severe risk by 2100 under SSP5-8.5. · Abstract; Genomic predictions of local (mal)adaptation…; Discussion; Figure legends