Widespread introgression across a phylogeny of 155 Drosophila genomes
Curation statements for this article:-
Curated by eLife
Evaluation Summary:
The authors present an impressive view of introgression across the Drosophila clade. There is strong support for signals of introgression along numerous branches of the phylogeny. However, the placement of these introgression events on the phylogeny and their impact on genome-wide patterns of relatedness are less clear.
(This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. Reviewer #2 agreed to share their name with the authors.)
This article has been Reviewed by the following groups
Listed in
- Evaluated articles (eLife)
- Evolutionary Biology (eLife)
Abstract
Article activity feed
-
Evaluation Summary:
The authors present an impressive view of introgression across the Drosophila clade. There is strong support for signals of introgression along numerous branches of the phylogeny. However, the placement of these introgression events on the phylogeny and their impact on genome-wide patterns of relatedness are less clear.
(This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. Reviewer #2 agreed to share their name with the authors.)
-
Reviewer #1 (Public Review):
A) The authors state that there is widespread introgression. While there are quite a number of deep introgressions in the phylogeny, it is unclear what proportion of the genome is involved. Presumably, to be significant in these tests it's not a tiny fraction of the genome in any one introgression. However, without that information, the paper doesn't really live up to the claim of pervasive introgression. Obviously how we use the word pervasive is somewhat subjective, but getting rough estimates of the proportion of the genome affected would be helpful to readers. It seems like the authors might be able to turn the asymmetry in the discordant trees into an estimate of the proportion.
B) The red arrows depicting the upper bounds on the timing of the introgression seem potentially quite confusing to the …
Reviewer #1 (Public Review):
A) The authors state that there is widespread introgression. While there are quite a number of deep introgressions in the phylogeny, it is unclear what proportion of the genome is involved. Presumably, to be significant in these tests it's not a tiny fraction of the genome in any one introgression. However, without that information, the paper doesn't really live up to the claim of pervasive introgression. Obviously how we use the word pervasive is somewhat subjective, but getting rough estimates of the proportion of the genome affected would be helpful to readers. It seems like the authors might be able to turn the asymmetry in the discordant trees into an estimate of the proportion.
B) The red arrows depicting the upper bounds on the timing of the introgression seem potentially quite confusing to the reader. Nearly all of them fall between sister clades, which is not consistent with the tests performed. The authors discuss this in the text and provide a different view in the supplement. I understand the desire to try and make an upper bound of the number of introgression events. However, some of the arrows on the phylogenies suggest histories and timings of introgression quite different from that seen in the pairwise matrices. For example, D eugracilis shared gene flow with the ancestor of the clade that included D biarmipes (if I'm reading clade 4 correctly in B) but the arrow on the phylogeny shows the introgression into the ancestor of the entire large clade that included D eugracilis and D melanogaster. That interpretation does not seem supported by the data and leads the reader to a quite different conclusion than the data suggests.
Given the focus of the paper on time tree and introgression, it seems a shame that there are no numbers put on how diverged these hybridized lineages were. Could lower bounds of the divergence be constructed using the minimum branch length that separates the two (confidence intervals could also be constructed for these numbers). In part, these numbers are interesting because of the relatively rich information on pre and postzygotic barriers and Coyne and Orr style analyses for Drosophila. Obviously, there are caveats that would have to accompany such an analysis/figure, but it would do a lot to enhance the biological interpretation of the paper. It may also serve to assure the reader that the introgression is between lineages of low enough divergence that hybridization is not implausible. Obviously, this analysis may be difficult for introgression events that are ard to place on the tree, so perhaps restricting the analysis to well-defined events may still be informative.
The authors state: "It is also possible that some patterns we observe reflect scenarios where introgressed segments have persisted along some lineages but been purged along with others. This phenomenon could also cause older gene flow between sister lineages, which should generally be undetectable according to the BLT and DCT methods, to instead appear as introgression between non-sister lineages that our methods can detect. "
--Is this a likely explanation? If a reasonably large number of trees support the introgression genome-wide then by chance (either drift, hitchhiking, or selection of a random allele) it is unlikely that a signal would be lost in one lineage. Now selection could remove a particular ancestry along a lineage, but here specific we would need the introgression to occur into an ancestral population persist for a (perhaps long) time and then suddenly be selected out in only one of the daughter populations. That's not completely implausible but it doesn't seem a very general concern. -
Reviewer #2 (Public Review):
Suvorov and colleagues present a well-supported genome-scale phylogeny for 149 Drosophila species based on thousands of single-copy-orthologs. They then use several approaches to estimate the extent of introgression across the phylogeny, and report that it is common both recently and deeper in the past.
The main strength of this paper is that it uses a scale of sequencing that allows an assessment of genus-wide trends with reasonably good power. It also presents two new analysis approaches, but these represent fairly minor modifications of existing techniques to suit multiple gene alignments, and unfortunately their reliability is not evaluated in this paper. Nevertheless, the main finding that introgression is common appears to be well supported. This finding echoes those of similar recent studies on taxa …
Reviewer #2 (Public Review):
Suvorov and colleagues present a well-supported genome-scale phylogeny for 149 Drosophila species based on thousands of single-copy-orthologs. They then use several approaches to estimate the extent of introgression across the phylogeny, and report that it is common both recently and deeper in the past.
The main strength of this paper is that it uses a scale of sequencing that allows an assessment of genus-wide trends with reasonably good power. It also presents two new analysis approaches, but these represent fairly minor modifications of existing techniques to suit multiple gene alignments, and unfortunately their reliability is not evaluated in this paper. Nevertheless, the main finding that introgression is common appears to be well supported. This finding echoes those of similar recent studies on taxa such as cichlid fishes and Heliconius butterflies. The different approaches used, and different levels of sampling in these different studies do not allow for quantitative comparisons, leaving us with the somewhat vague conclusion that introgression is 'common' in all of these taxa. Perhaps most critically, the present paper does not delve any deeper into the evolutionary impacts of introgression, nor the factors at the species or genomic level that might determine its frequency. Below I describe some areas of concern in more detail.
1. Extent of introgression
Perhaps equally as interesting as the frequency of introgression per species across the phylogeny is the proportion of the genome of each species that is affected. Without such estimates, the full extent of introgression is difficult to assess.
2. Sampling effects
Since this paper is attempting to make an (admittedly crude) estimate of the extent of introgression in the entire genus, some discussion is needed to address the possible consequences of the fact that only around 10% of species in the genus are represented. For example, if sampling is very even, perhaps most ancient events would be detectable, but more recent events may tend to be missed simply because the species involved are not sampled.
3. Ancestral structure
The reasoning provided for dismissing the possible effect of ancestral population structure is unconvincing. First, the authors argue that it "seems less likely" that non-sister taxa would have bred more frequently in the ancestral population. However, this is the entire basis of the problem: it might be unlikely, but it can happen. Eriksson and Manica (2012 https://doi.org/10.1073/pnas.1200567109) provided a very reasonable scenario in which colonisation of a new region can lead to this pattern.
Second, the authors argue that QuIBL "should not be impacted by ancestral structure because this method searches for evidence of a mixture of coalescence times: one older time consistent with ILS and one time that is more recent than the split in the true species tree and that therefore cannot be explained by ancestral structure." This argument needs clarification. My understanding is that the split in the "true species tree" would also be inflated if there was ancestral structure.
My view is that ancestral structure leading to discordance between gene trees and species trees is itself an interesting phenomenon. In some ways, it is not conceptually distinct from introgression occurring soon "after" speciation if we consider ancestral structure as the beginning of a continuous speciation process, so I don't think it would weaken the paper to accept this as a possible contributing process.
4. Discordant count test
The statistical analysis in the DCT accounts for multiple testing of many triplets for introgression, but there is no mention of the fact that these triplets are non-independent. It is not clear to me whether this makes the correction used more or less conservative.
If there are any cases where the internal branch is long and the number of ILS gene trees is very small or zero, use of a chi-squared test may not be appropriate.
5. Branch length test
The authors acknowledge that the BLT is "conceptually similar" to that of Hahn and Hibbins 2019 https://doi.org/10.1093/molbev/msz178, but to me it seems that the only material difference is the statistical procedure for testing for an significant difference between branch lengths.
An important consideration that appears to have been ignored is whether selection can impact the distribution of branch lengths, especially since many of the the BUSCO genes used here will be under strong selective constraint.
6. Intra-locus recombination
The paper needs to address the possible impact of intra-locus recombination on all of the introgression tests. For the DCT, I imagine that counts would be biased toward the species tree topology if the inferred trees span multiple distinct genealogies (see for example simulations by Martin and Van Belleghem 2017 https://doi.org/10.1534/genetics.116.194720 Figure S7). This might reduce test sensitivity.
Similarly, for the BLT, I would expect that true introgression would be more difficult to detect in the presence of recombination. It is possible that the block jackknife procedure of Hahn and Hibbins (2019, https://doi.org/10.1093/molbev/msz178) may be more suitable than the comparison of distributions of point estimates for genes used here.
-
Reviewer #3 (Public Review):
The authors compiled a collection of published and newly sequenced genomes to assemble the largest collection of Drosophila genomes to date. Using this dataset they extracted a set of single copy orthologs to use for phylogenomic analyses, with a focus on estimating a time-calibrated phylogeny and introgression.
This new dataset is a valuable resource that will serve the broader community of Drosophila researchers opening many new avenues for future phylogenomics research. The workflow of focusing on BUSCO genes for all comparative analyses is simple in a good way -- it is easy to understand how the data were collected and it should be easily reproducible -- which makes it easy to read past the genomics details and focus on the analyses of these data.
However, I feel this is an important aspect of the paper …
Reviewer #3 (Public Review):
The authors compiled a collection of published and newly sequenced genomes to assemble the largest collection of Drosophila genomes to date. Using this dataset they extracted a set of single copy orthologs to use for phylogenomic analyses, with a focus on estimating a time-calibrated phylogeny and introgression.
This new dataset is a valuable resource that will serve the broader community of Drosophila researchers opening many new avenues for future phylogenomics research. The workflow of focusing on BUSCO genes for all comparative analyses is simple in a good way -- it is easy to understand how the data were collected and it should be easily reproducible -- which makes it easy to read past the genomics details and focus on the analyses of these data.
However, I feel this is an important aspect of the paper that should receive more details, perhaps in the supplement. I may have missed it, but I could not find statistics about this ortholog data set. On average, how long is each locus, how many variable sites are there, how many taxa are missing data for any given locus due to paralogy? Do the BUSCO genes include both introns and exons? It is also unclear from the description exactly how the BUSCO genes were extracted from genomes. Are they extracted from the final assembled genomes, or do you perform variant calling after identifying them to call heterozygous site? If heterozygosity is excluded, how might this impact metrics such as the branch length tests, especially among close relatives? It likely impacts node age estimates as well?
The authors use this dataset to infer phylogenetic relationships among taxa using both ML concatenation (IQtree) and a two-step MSC approach (Astral) which yielded quite similar topologies, and they examined the impact of filtering loci with treeshrink, which had minimal impact. This new topology represents a substantial step forward for understanding the relationships among major Drosophila clades.
One of the main results of this study is a new set of node age estimates on the tree. For this they estimated branch lengths in mcmctree from a concatenated matrix of 1000 loci in the presence of fossil calibrations. The fossil calibration scheme selected as the best option includes three fossils, one dating the divergence at the split from mosquitos (uniform 195-230Ma) and two ingroup calibrations (U(43,64) and U(15,43)). To me, the credible intervals on node ages seem incredibly narrow. The authors mention this as an improvement compared to earlier studies, but they also mention later that the total amount of sequence data does not greatly impact node dating. So I'm a bit confused why the node ages are expected to be more accurate here. It seems to me that time calibrations should be most accurate when the greatest number of fossils are available, and when very appropriate Bayesian priors on set on the analysis. The effect of sequence variation is then relatively small. But here there are very few fossils, one of which is hugely distant, and so I would not expect highly precise age estimates. So I guess my question to the authors is, what do you think is going on here? Perhaps further description in the supplement of how the mcmctree method implemented here differs from traditional node dating done in a program like BEAST would help to clarify.
Considering that this paper aims to infer the new best time calibrated tree for the Drosophila community, I think that the current description of fossil calibration schemes, which primarily refers to other publication names in the supplement, is insufficient. Which fossils are used in those studies, are you using those fossils as calibrations here, or are you implementing secondary calibrations based on their phylogenetic results? The reader should not have to read every one of those papers to understand the basis of the calibrations in this paper.
Fig.1 shows nodal age posterior probabilities. Are these 95% confidence intervals? The taxon labels are too small in this figure, both on the large tree and especially in the inset figure. The legend refers to fossil taxon names used for calibrations, but because it is still unclear to me where the fossils are placed on the tree. Are the calibrations indicated somewhere in the figure?
The authors demonstrate evidence of introgression by showing mostly overlapping evidence from two different types of tests. Together, these tests show that most major clades contain significant imbalanced discordance in gene tree counts or branch lengths. The taxon labels in Figure 2 are unfortunately quite unreadable, especially the matrix labels, which makes it difficult to interpret.
I do not see a reason for presenting new names and acronyms for the introgression tests used in this study. The "DCT" is described as being similar to a suite of existing tests which are also based on comparison of rooted-triplet gene tree frequencies. These methods have been presented in many frameworks (BUCKy, D-stat, f4, etc.) and the only difference here seems to be the precise method used to determine significance. Similarly "the BLT is conceptually similar to the D3 test" could be replaced by just saying we implemented the D3 test which we refer to here as a 'branch length test (BLT)' to clarify that you have not in fact created a new test (e.g., you say "The first method we developed was the discordant-count test...")
I am not very satisfied with the estimates of the "upper bounds" of introgression used here. It seems that there could possibly be many ways in which admixture edges could be drawn on the tree to explain the matrix of significant test results, and it is better to let formal network inference methods (e.g., SNAQ, Phylonet) infer these edges rather than guess at their placement. The current approach of "placing introgression events between pairs of branches for which most descendant extant taxa show evidence of introgression" leaves significant room for subjectivity.
The authors did implement phylonet, but not very exhaustively. Why only fit a single edge on the tree instead of multiple? The authors state "networks with more reticulation events would most likely exhibit a better fit to observed patterns of introgression but the biological interpretation of complex networks with multiple reticulations is more challenging". I don't think this type of result is any more complicated to understand than the current approach used by the authors of drawing edges manually. And it is much less subjective. The authors say that it is computationally intractable, and this may be true for clades above ~15 tips, but testing on smaller trees by subsampling 10-12 tips seems feasible. From my experience network inference using pseudo-likelihood methods in SNAQ or phylonet takes a few minutes to fit 1 edge, and a few hours to fit 2-3 edges.
Currently the two major results of the paper seem disjointed. The authors infer a time-calibrated tree, and they infer introgression events, but there is not much connection between the two. I applaud the authors on one hand for being cautious in interpreting their "upper bounds" of introgression to say too much about when they think introgression has occurred in the context of the time-calibrated tree. I think there is insufficient confidence in the introgression timing estimates to do that. But, what about the inverse relationships? Does this extent of introgression across the tree impact your confidence in the estimated timing of divergence events? One expectation would be that it is biasing all of the divergence times to appear younger. See my suggestions for addressing this.
Overall, this study presents an impressive new dataset and important new results that greatly impact our understanding of the evolutionary history of Drosophila. Although the estimates of node ages and introgression events may be imperfect, they are clearly a step forward. It is clear from these results that introgression has occurred throughout the history of Drosophila, and this study paves the way for further investigation of these patterns, as the authors propose in their conclusions.
-
-
-
-