REVIEW 3 major objections 4 minor
Widespread remote introgression in the grass genomes
T0 review · 3 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash
Pith's one-line read The paper reports 622 remote introgression events between the two deep grass lineages, detected by a new pipeline called RIFinder.
desk verdict The two case studies and the dataset are solid, but the 622-event count rests on a branch-length filter that has never been shown to remove ILS. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The central object is RIFinder, a phylogeny-based remote-introgression detection pipeline. It works by clustering proteins into homology groups, inferring gene trees, collapsing monophyletic single-clade branches, splitting multi-copy trees into ortholog-like subtrees, then scoring each leaf's topological incongruence against a reference two-clade species tree (PACMAD vs. BOP); candidate events must survive a modified branch-length test that compares donor-to-acceptor distances with donor-to-sister distances, on the logic that introgression makes the donor-acceptor pair look younger than incomplete lineage sorting would. The threshold for a significant RI signal uses Cayley's formula for the number of possible unrooted topologies on a given set of tips.
What would settle it
Simulate the grass PACMAD/BOP species tree under a realistic deep coalescent model with zero transfer, run RIFinder on the simulated gene trees, and count how many RI events it reports; if pure incomplete lineage sorting routinely triggers its branch-length criterion, the 622-event claim collapses. A second check is direct: remap long-read assemblies of Cleistogenes songorica and Achnatherum splendens and confirm the 30-kb segment's boundaries and absence from Cson-B; if the segment is present across broad Chloridoideae diversity or shows genealogies that intermix with multiple outgroup species, the recent-transfer interpretation is undermined.
Extended reading notes
Core claim
On the paper's own terms, the discovery is that DNA transfer between subfamilies of grasses that diverged more than eighty million years ago is widespread: 622 candidate remote introgression events, traced to 543 distinct homologous genes, with the Pooideae receiving the most introgressed genes and the Bambusoideae the fewest. The introgressed copies show signatures of post-transfer local adaptation, with significant enrichment in stress-response protein domains. Two cases are worked out in detail: a roughly 30-kilobase Triticeae-derived segment in Cleistogenes songorica that is absent from its other subgenome and from other Chloridoideae, and whose genes respond to drought and heat stress; and the gramine biosynthetic gene cluster, whose AMIS and NMT components show discordant phylogenies placing Panicoideae and Chloridoideae copies inside the Oryzoideae lineage, implying that remote introgression contributed to the cluster's origin and diversification. The paper also reports that RIFinder recovers previously documented transfers, such as the Bx benzoxazinoid genes in Triticeae and the momilactone cluster in rice, and that simulated benchmarks yield high precision and recall.
Load-bearing premise
The count assumes that ordinary sorting of ancestral genetic variation cannot by itself make a donor and acceptor look closer in time than the sister lineage; if it can, many of the 622 'transfers' could be artifacts of that sorting.
Editorial extensions
If this is right
- Gene trees in grass phylogenomic data sets will need to be treated as potentially composite: a gene's history can include a foreign branch from the opposite deep lineage even when the species tree is uncontroversial.
- Functional screens for stress-tolerance genes in grasses should check for discordant phylogenetic placement, because the enriched stress-response signal means candidate adaptive genes may be introgressed rather than native.
- The asymmetry between donor and acceptor counts points to life history, geographic range, and reproductive timing as predictors of how much inter-lineage DNA a plant family acquires, which is a testable comparative hypothesis.
- The documented assembly of gramine biosynthetic gene clusters by remote introgression means the presence of a metabolic cluster in a species does not by itself establish vertical ancestry with other cluster-bearing species.
- If the Cleistogenes segment's drought stress response is causal, then multigenic stress modules can be acquired ready-made from sympatric distant relatives, which is an evolutionary shortcut unavailable under a purely vertical-gene model.
Reading between the lines
- The reported 622 events are a lower bound by the paper's own conservative filters, so denser sampling and k-mer or pangenome-based detection should raise the count; re-running the pipeline on the additional published grass genomes would be a direct test.
- If the branch-length assumption survives deep-coalescent benchmarking, the same workflow could be applied to other large plant families or animal radiations, potentially making remote introgression a general feature of eukaryotic genome evolution rather than a grass quirk.
- The sympatry of Achnatherum splendens and Cleistogenes songorica suggests an ecological predictor: deeply divergent species with overlapping ranges should show elevated RI enrichment; a systematic comparison of sympatric versus allopatric pairs across the 122 genomes could test this.
- The rice pangenome shows the gramine cluster is nearly restricted to one cultivated haplotype group with selection signals, implying RI-derived defense clusters may be subject to fast presence/absence evolution; searching other pangenomes for recently introgressed metabolic clusters would test that.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper introduces RIFinder, a phylogeny-based pipeline for detecting remote introgression (RI) between deeply diverged lineages, and applies it to 122 haploid grass genomes. The authors report 622 RI events between the PACMAD and BOP clades, functional enrichment in stress-response genes, a Triticeae-derived drought-associated segment in Cleistogenes songorica, and RI-mediated assembly of gramine biosynthetic gene clusters. The manuscript also describes simulation benchmarks, detailed case studies, expression analyses, synteny, and ecological niche comparisons.
Significance. If the main claim holds, the paper establishes that inter-subfamily gene transfer is widespread and adaptively important in grasses, a major expansion beyond documented cases of HGT and close-range introgression. The study is valuable for its large, openly described dataset, the public availability of RIFinder code and results, and the multi-pronged case studies. The C. songorica segment and gramine cluster examples are supported by independent evidence beyond the genome scan, including raw-read coverage, synteny, phylogenetic topology tests, expression, and metabolomics. The central quantitative claim, however, currently rests on a detection method whose ability to exclude incomplete lineage sorting has not been demonstrated, so the 622-event count should be treated as provisional until a proper null benchmark is supplied.
major comments (3)
- [Methods, 'Modified Branch-Length testing' and 'Simulation'] The central claim of 622 true RI events depends on mBLT excluding ILS, but mBLT is never benchmarked on an ILS-only simulation. The Simulation section only injects RI at rates 0.1%–2% and reports precision/recall; it does not run pure-ILS gene trees through RIFinder, so the false-positive rate under ILS is unknown. In addition, the mBLT null is stated as 'coalescence times should not differ significantly between topologies' while the same paragraph notes that ILS 'typically leads to older coalescence time.' Under the multispecies coalescent, the discordant topology labeled as introgression has a deeper coalescence, so d(Donor, Accepted) is expected to be longer, not equal, than d(Donor, Sister). Since the implemented test is a two-tailed independent t-test, the authors do not specify whether only the one-sided direction supporting introgression is retained; if significant differences are kept regardless of direction, ILS itself could generate exactly the signal the test is meant to exclude. Please add a no-gene-flow null simulation, report the false-positive rate at the same taxonomic sampling, and clarify the directional decision rule.
- [Methods, 'RI scoring'] The RI score threshold is described as determined by Cayley's formula n(n-2), where n is the number of taxa. The text appears to intend n^(n-2), but Cayley's formula counts the number of labeled trees on n labeled vertices and is not a distribution of discordance scores expected under ILS; applying it as a significance threshold is uncalibrated and unexplained. Because this threshold is one of the filters that converts candidate signals into '622 RI events,' its derivation needs to be replaced by an explicit null model or by calibration on simulations that include ILS. As written, the threshold is an ad hoc parameter.
- [Results, 'Inferring remote introgression events using RIFinder'] The performance evaluation is partly circular. The observed RI rate (0.03%–0.3%) is produced by RIFinder, and the simulation study then selects transfer rates (0.04% and 0.1%) at which RIFinder reports high precision and recall. Such a loop does not independently validate the observed rate. Please provide external positive controls, such as the previously documented Bx, momilactone, and Panicum-derived segments, alongside a negative control with no gene flow, in addition to the current simulation.
minor comments (4)
- [Methods, 'Compressing and splitting gene trees'] The pseudocode and surrounding text contain a literal placeholder for the minimum taxonomic representation threshold, written as 'default: ?'; this must be replaced with the actual value or formula used in the analyses.
- [Methods, 'Modified Branch-Length testing'] The notation is inconsistent: the text says distances to 'all donor leaves d(Donor, Accepted) and all sister clade leaves d(Donor, Sister),' but the second distance should presumably be d(Acceptor, Sister) or d(Donor, Sister) depending on which pairwise comparison is intended; please clarify the definitions of the three clades and the measured distances.
- [Figure 1b and Methods, 'Simulation'] The x-axis label says 'Transfer proportion' in the figure but the Methods text specifies transfer rates per lineage-million-years; the units should be made consistent.
- [Methods, 'Gramine-related gene expression'] The text says 'Primes used for constructs' but should read 'Primers'; also, the internal control genes are listed without a corresponding reference for the Z. latifolia 18S primer.
Circularity Check
Simulation validation loop uses RIFinder's own estimated rate as the 'biologically relevant' benchmark, self-referentially confirming the method.
-
other
[Results, 'Inferring remote introgression events...' and Methods, 'Simulation' (Fig. 1b)]
"InPoaceae,RIeventswerequantifiedtorangefrom0.03%to0.3%ofallgenes,respectively(See next; Supplementary Fig. 4).Atthisbiologicallyrelevanttransferfrequency(0.04%and0.1%),RIFinderdemonstratedrobustperformance,achievinghighaccuracy(95.9%to98.2%,93.5%to96.3%)andhighrecall(85.0%to87.6%,83.5%to84.0%)acrossthethreescenarios(Fig. 1b).Theseresultsindicatedthatthesimulationframeworkreflectedbiologicallyrealisticintrogressiondynamics,therebyreinforcingconfidenceintheapplicabilityofRIFindertoreal-worldgenomes."
The paper first uses RIFinder to quantify the real-world RI rate (0.03%–0.3% of genes), then selects simulation rates (0.04% and 0.1%) as 'biologically relevant' based on that estimate, and finally shows RIFinder performs well at these simulation rates. The simulation therefore benchmarks the method at values that are outputs of the same method, so the good performance cannot independently validate the method's rate estimate or the confidence placed in the real-data count. This self-referential loop does not by itself force the 622-event count, but it makes the claimed simulation-based reinforcement of RIFinder's applicability circular rather than an external test.
full rationale
The load-bearing claim of 622 remote introgression events rests on RIFinder's phylogenetic discordance scoring, the modified Branch-Length Test, and topology filters, none of which is defined in terms of the final event count. The mBLT's null hypothesis ('coalescence times should not differ significantly between topologies') is inconsistent with the paper's own statement that ILS 'typically leads to older coalescence time,' and no ILS-only simulation is run; this is a serious correctness risk but not a circular derivation. The only identifiable circular step is the simulation-rate calibration loop, where the observed RI rate is used to select simulation conditions and then invoked to confirm RIFinder's reliability. This is self-referential but not fully reductive: the simulation does not generate the 622 events, and the case studies (Cleistogenes segment, gramine cluster) are supported by independent synteny, expression, and topology tests. Prior self-citations (e.g., Wu et al. 2022a/b) are used as recall benchmarks but are not load-bearing for the central claim. Overall, the paper has a moderate circularity in its validation logic, but the central derivation retains independent content.
Assumptions & free parameters
free parameters (4)
- RI score threshold (Cayley formula) =
n^(n-2), with default '?' in manuscript
- Bootstrap support cutoff =
70
- Minimum taxonomic representation threshold =
? (default undefined in Methods)
- Scoring weight in RI pseudocode =
undefined
assumptions (3)
- domain assumption The reference species phylogeny and the assignment of species to BOP and PACMAD clades are correct.
- ad hoc to paper Under incomplete lineage sorting, branch-length distances between donor and acceptor do not differ significantly from donor and sister (the mBLT null).
- domain assumption Phylogenetically ambiguous signals (low bootstrap, long branches) can be removed without biasing the RI catalog.
Cite this review
Pith. "Pith review of Widespread remote introgression in the grass genomes." pith.science (2026). https://pith.science/paper/F2RMHKUN
@misc{pith2026250707761,
author = {Pith},
title = {Pith review of: Widespread remote introgression in the grass genomes},
year = {2026},
howpublished = {\url{https://pith.science/paper/F2RMHKUN}},
note = {Machine review of arXiv:2507.07761}
}
read the original abstract
Genetic transfers are pervasive across both prokaryotes and eukaryotes, encompassing canonical genomic introgression between species or genera and horizontal gene transfer (HGT) across kingdoms. However, DNA transfer between phylogenetically distant species, here defined as remote introgression (RI), has remained poorly explored in evolutionary genomics. In this study, we present RIFinder, a novel phylogeny-based method for RI event detection, and apply it to a comprehensive dataset of 122 grass genomes. Our analysis identifies 622 RI events originating from 543 distinct homologous genes, revealing distinct characteristics among grass subfamilies. Specifically, the subfamily Pooideae exhibits the highest number of introgressed genes while Bambusoideae contains the lowest. Comparisons among accepted genes, their donor copies and native homologs demonstrate that introgressed genes undergo post-transfer localized adaptation, with significant functional enrichment in stress-response pathways. Notably, we identify a large Triticeae-derived segment in a Chloridoideae species Cleistogenes songorica, which is potentially associated with its exceptional drought tolerance. Furthermore, we provide compelling evidence that RI has contributed to the origin and diversification of biosynthetic gene clusters of gramine, a defensive alkaloid chemical, across grass species. Collectively, our study establishes a robust method for RI detection and highlights its critical role in adaptive evolution.
Reviewed August 6, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.