Review report of JEB-2020-00186 titled “Using gradient forest to predict climate response and adaptation in cork oak”

Using thousands of SNP markers, the authors investigated population genetic structure and gene-environmental association in the cork oak. The authors argue that the present study expanded previous investigations with aims at (1) implementing a population structure analysis based on the nearest neighbor haplotype “coancestry” and at (2) combining population genomics and gene-environment associations to further study the local adaptation of these tree keystone species (Line 119-121). As this species is a keystone tree, the present study gives substantial insights into forestry and conservation biology. The approach seems sound and the manuscript is easy to read. In the context of general evolutionary biology, however, I have several concerns about the sample size limitation and novelty of this study.

Major comments:
1. The number of populations seems too small to investigate fine-scale gene-environmental associations, given that the 17 populations were sparsely sampled across different geographical area separated by sea. Specifically, the result that PCNMs contributed more to the variation might be overrepresented due to the sampling design. As the authors acknowledge in the abstract (Line 21), this study is still at a preliminary stage. More sampling across all the area or intensive sampling on important regions may be conducted in the future.

1. Although this study aims at the prediction of future adaptation, genetic mechanisms underlying the local adaptation are still unclear. The authors detected some SNPs with annotation of “mannosylglycerate hydrolase” and “quinolinate synthase” (Table 1), but also found unknown functions. The set of candidate genes could have been subject to gene-set enrichment analyses.

2. As the author recognized in the conclusion (Line 495-500), not only the clustering analysis but also model-based approach, such as coalescent simulation, would be required to infer evolutionary history. As the authors detected major genetic clusters, they could have estimated the divergence time and extent of gene flow among the clusters.

3. Together with the coalescent simulation, projection on the past climates would also be valuable to infer evolutionary process. In particular, I am interested in the SNP distribution during the last glacier maximum. These variables could be compiled through the WorldClim database.

4. The predictability of GF seems too low to conclude the adaptation to future climate change. The results showed “The GF models that explained better the variation was the adaptive SNPs model (mean R2 = 14.0%)” (Line 322), though the adaptive model was slightly better than the null model.

5. I am not convinced with what differentiates the present study from previous ones. There have already been related studies on the cork oak adaptation to future climate change (Modesto et al 2014 Tree Genetics & Genomics; Pina‐Martins et al. 2019. Global Change Biology). While the present study obtained the larger amount of genomic data, the study populations were quite similar to previous investigations and still limited to perform further modeling (Line 495-500). Complementary experiments such as common garden experiments using sapling were not incorporated as well, and the present study seems at a progressing stage.

6. I agree that gene-environmental association is suitable for studying local adaptation in long-lived trees; however, several studies on herbaceous plants have already been illustrated isolation-by-environment with some implications into adaptation to future climate change (Lee & Mitchell-Olds 2013 Molecular Ecology; Anderson et al. 2015 American Naturalist). It seems unclear how the present study addresses new questions in evolutionary biology.

Minor comments:
8. To my knowledge, there have also been genomic resources for the cork oak (Ramos et al. 2018 Scientific data 5: 180069). The authors could have leveraged the existing resource more for comparative genomics and selection scan.

1. Line 322: For machine learning approach like GF, area under the receiver operating characteristic curve (AUC) might have been a standard index of species distribution modelling.

2. Line 221: |z| (effect size) -> |r| (Pearson’s correlation coefficient)?

Review report of JEB-2020-00186.R1 titled “Using gradient forest to predict climate response and adaptation in cork oak”

The authors have carefully addressed the major concerns raised during the first review, which makes me convinced with the scope of this study. Now I agree that the sample size is reasonable in comparison with the other studies on tree genomics, and the present study has focused on the local adaptation to current or future climates. It would be great if the following suggestions are considered in the final version. I look forward to seeing future progress of this project.

1. Line 253 (“genetic offset” sensu Fitzpatrick & Keller, 2015): The citation of “genetic offset” is shown in the method section for the first time, but it seems better to place it at the first appearance of the “genetic offset” words in the introduction.

2. Line 223: I understand the definition of R2 in GF. Given that it is not equal to a standard R2 of linear regression, I suggest that the authors cite Ellis et al. (2012) at the first mention to the split importance.