The authors fully resolved my earlier comments and did an amazing job at improving the structure and presentation of their manuscript. I think their result are of value to a broad audience interested in adaptation, divergence and population genomics. I only have a few small further remarks.

Box 1 is a very helpful way of presenting the expected patterns! I think, perhaps, it would be helpful to clarify that the identification of the patterns may be dependent on the (early) timing of the sampling relative to the time of divergence. For example, I would expect the increased Fst relative to background in scenario 2 to show up only in a certain time window of divergence. If populations have been separated too long, this signal would get lost. In scenario 3, I believe Dxy should only be reduced if selection happened before the divergence event.

Figure 2b is a brilliant addition.

As before, it is great to see the presentation of how the relationship between these statistics changes along the divergence gradient in Figure 4 and how these changes in relationships match with theoretical predictions.

Supplementary Note 3 regarding the fd analysis does not seem to be referenced or mentioned in the main text.

In this manuscript, Shang and colleague study a radiation of Populus trees, to decipher the most likely scenario explaining speciation in this group. Specifically, by sequencing individuals from 7 species, they test 4 different speciation scenarios (divergence with gene flow, allopatric speciation, recurrent selection and balancing selection). They conclude that, in Populus, allopatric speciation fits most of the genetic differentiation observed in the genome and is the most likely mode of speciation in the genus. Authors also detected two introgression events in the radiation.

The study system is interesting and the authors use state of the art methodologies to answer their questions.

The authors seem to have considered reviewers comments on a previous draft but it is difficult to tell given that they did not provide a response letter of any kind. 

It seems that the manuscript has been extensively rewritten and is now clearer. I however still have a couple of concerns with the current version of the manuscript.

First, the distinction between model 3 and 4 involves more quantitative testing than qualitative testing, as the signs of the correlation between statistics are similar but just of a different magnitude. I therefore did not understand how the authors manage to distinguish  between these two scenarios looking strictly at genome wide correlations? 

Second, the authors use the term 'genomic landscapes' to describe the variation of many statistics along the genome. I find it a source of confusion personally.

Third, the authors recurrently mention that these landscapes are heterogeneous, but never illustrate it or test it. I think a test could be done to verify if the windows have higher, lower, or similar levels of divergence (or any other statistics) than the rest of the genome and see if it a different distribution than what can be expected by chance (using simulation data with an allopatric model possibly). I feel it is quite important given the aim of the paper is to understand why there might be heterogeneity in these landscapes.

Fourth, the authors mention repeatedly that these genomic landscapes are conserved across the speciation continuum based on correlations of statistics across the genome. However, the correlation coefficients that the authors find are significant but low for most statistics, implying that these landscape are somewhat similar but also somewhat different. I think the authors could do a better job at quantifying the amount of overlap across species. Also, some correlation coefficients decrease with time, indicating that they are actually not conserved.

Finally, I do not grasp the novelty of the current study. The message I get is 'Heterogeneous landscapes emerge because of different evolutionary process'. I feel that a deeper comparison of the authors' results with past literature in the Discussion section would clarify this point. 

I personally felt that the manuscript lacks clarity on the points I mentioned above and therefore do not recommand publication of the manuscript in its current form. 

Specific comments:

lines 54-55: Too vague. Strictly speaking we understand this now: Mutation, selection, gene flow and drift.

line 62: What are 'hotspots' of elevated genetic differentiation?

lines 72-74: what does 'highly heterogeneous genomic landscapes' mean?

lines 225-227: What else does shape it then?

line 240: I would argue that the genomic patterns are 'somewhat' conserved here.

Huiying Shang and colleagues present a detailed study on the genomic divergence landscapes between pairs of Populus tree species. They use more than two hundred whole-genome resequenced samples from 8 species to study genetic divergence along a continuum of speciation. By correlating measures of nucleotide diversity, relative divergence, absolute divergence, and recombination rate they set out a number of hypothesis that each well define how these relationships are expected to change as divergence increases between species. Doing this, they identify the common drivers of heterogeneity in genomic divergence to be background selection, with contributions of ancestral polymorphisms, gene flow and potentially selection.

I think the authors did an excellent job at introducing these complex concepts and outlining their hypothesis that can be neatly dissected with their data. The number of possible expected relationships between population genomic statistics is high and I was truly impressed how well the authors did at explaining their expectations and being able to follow how their results fit to them. I think the analysis and results are solid, but please allow me to give a few remarks that I hope could further increase the accessibility of this manuscript to a broad public.

First, I very much like Figure 1. Panel A shows expected correlations within or between species, whereas panel B shows how these relationships may change over time. In the results, however, results relating to panel A are presented at the end and it would have been more logical to me to have presented those results first (regarding that section, I would also suggest including the figures in the main text, rather than the supplement, because a quantification of each hypothesis is really unique! I think doing this could also improve understanding how much each of the different factors contributed to the divergence landscape.). Next, In Fig4 and Fig6 I would suggest improving the link with figure 1, by showing the results in the same order as in Fig1 and clearly pointing the reader to how they match Fig1 and what it tells about the drivers of genomic divergence.

Second, I think there’s a missed opportunity to better quantify the extent of gene flow between species, compared to using IBD and TreeMix. Would it be possible to calculate ABBA-BABA type of statistics (e.g. fd https://github.com/simonhmartin/genomics_general/blob/master/ABBABABAwindows.py) to understand what fraction of the differentiation heterogeneity results from admixture? Another set of analyses could include running SweepFinder in each population to identify common and unique selective sweeps.

Third, do reads of more distant species map equally well to the reference genome. It might be good to verify this, so that no correlation biases are created because more distant species have less reads mapping, resulting into reduced diversity values.

Fourth, regarding the title “Conserved genomic landscapes of differentiation”, the point of conservation is a bit lost to me (after the abstract) as most of the later figures visualize how relationships between statistics change as species diverge. I might thus suggest presenting a more direct quantification of the conservation of the differentiation landscape. I do agree that the drivers of genomic differentiation may be highly conserved.

Small comments:

Be careful in the use of ‘linked selection’ and ‘background selection’. They are not the same and I think they are sometimes used as interchangeable (e.g. in the abstract L44 and L46).

L87-92: Also observed in Heliconius (and including introgression) (Martin et al. 2019 Plos Biology; Edelman et al. 2019 Science; Van Belleghem et al. 2021 Evolution)

L124: “different scenarios can be hypothesized regarding the extent of local gene flow…”. I found this sentence confusing because the four scenarios do not just focus on gene flow, but include balancing selection, background selection and selective sweeps. This becomes clear further in the manuscript, but I would argue to modify the sentence here.

L145: “…polygenetic adaptation from standing genetic variation”. Does this include patterns arising from linked selection in the common ancestor?

L153: “recombination (r)”

L154: I would suggest “across the speciation continuum”

L154: “…negative correlation of Fst and r become increased with advancing differentiation…”. Is this assuming no gene flow? I imagine that with gene flow, this correlation can also become increased.

L158: “The positive correlations between p and r should remain highly correlated with each other under background selection”. Can selective sweeps also contribute here to that relationship?

L178: Populus trichocarpa

L179: P. trichocarpa

L226: Fig. 2c instead of Fig. 1c?

L229: Can shared haplotypes in the IBD analysis also result from incomplete lineage sorting?

Fig3a: Why do introgression edges go from tree edges to tree tips and not tree edge to tree edge? (Maybe I am just not familiar with the method)

Fig4e-f: Would it be possible to visualize and add the genome-wide correlation of p and r?

L285: Regarding the low p but high r in P. grandidentata, I noticed that this species has undergone consistent population decline. Could that have affected p independently from r?

L318: “correlation coefficients for…”. Correlation with what in this case? I assume with divergence level, but this took me some time to understand.

L341: Is “respectively” needed here?

L342: Again, I would suggest “across the speciation continuum”

L352: “…expect negative correlations between Fst and p  or r and the relationships become stronger as da increases”. Would there not be a limit to this, e.g., when Fst reaches 1?

L353-354: This is a difficult sentence to me. Should it read: “Besides, the positive correlations between Dxy and p should be highly correlated with da …”?

L359-361: “…p  and Dxy showed significantly positive correlations while the trend became weaker as divergence increased”. Is this not expected as the proportional contribution of ancestral p to Dxy and the current p will decrease over time? Hence, the correlation will decrease?

Fig6: I would suggest fixing the y-axis so that all plots are readily comparable.

L387: There seems something wrong with this sentence.

L405-406: It is unclear to me how support was found for reproductive barriers.

In this manuscript, Shang and colleague study a radiation of Populus trees, to decipher the most likely scenario explaining speciation in this group. Specifically, by sequencing individuals from 7 species, they test 4 different speciation scenarios (divergence with gene flow, allopatric speciation, background selection and balancing selection). They conclude that, in Populus, allopatric speciation fits most of genetic differentiation observed in the genome and is the most likely mode of speciation in the genus. Authors also detected two introgression events in the radiation.

The study system is interesting and the authors use state of the art methodologies to answer their questions.

However, the authors explain very succinctly the two last models they test for and the associations of statistics they expect to emerge if these models are true. Specifically I did not understand model 4 (balancing selection) and how this model differed from model 1. Also, the distinction between model 3 and 4 involves more a quantitative testing than a qualitative testing, as the signs of the correlations between genetic statistics are similar but just of a different magnitude. I therefore did not understand how the author did manage to distinguish these two scenarios?

I personally felt that the manuscript is poorly written in parts, making some sections unintelligible. Indeed, I did not understand a couple of result sections (Conserved genomic landscapes across the continuum of divergence, Correlated patterns of genome-wide variation across speciation continuum, and Scenarios of genomic patterns of differentiation) because of both the writing, and the many reasoning shortcuts used in these sections.

I therefore do not recommand publication of the manuscript in its current form. 

I think the above mentioned sections of the manuscript need to be rewritten. This is more of a personal preference, but the flow of the manuscript could be changed to present the main results of the manuscript (i.e. the hypothesis testing) earlier on. In that line of thinking I would put the Between species variability in demographic trajectories section in the SI, and bring from the SI Figure S11 which I feel is central to the results of this paper.

Specific comments:

lines 143-145: expand reasoning.

line 148: why is this expected?

Figure 2c: There is more colour than cluster number for K=5. Why is there green colour in pade and pdav individuals?

line 287: gene density appears from nowhere. Integrate it in the introduction.

lines 289-291: Why? expand reasoning.

lines 305-306: How where selected the 'representative' species pair?

lines 316-317: reformulate. Fst along the genome. What does 'independant' refers to for species pairs?

lines 317-318: I do not understand this sentence.

lines 318-322: I do not understand this sentence.

lines 328-330: expend reasoning.

line 342: the speciation continuum?

lines 345-346: I don't understand this sentence.