01 bioRxivpreprint · 10 September 2026_
Gene conversion facilitates rapid evolution of inversions across avian immunoglobulin loci
- Structural variation
- Population genetics
IG loci annotated in 139 bird genomes show recurrent inverted duplications at IGH, found in no other vertebrate lineage, which the authors link to a pseudogene donor pool for somatic gene conversion.
Summary
Immunoglobulin loci were annotated in 122 Vertebrate Genomes Project and 17 California Conservation Genomics Project bird assemblies. Avian IGH loci carry recurrent inverted duplications that produce direct and inverted copies of the same repeat unit. Inversion density varies between species, and population data show the inversions change quickly. The proposed model: inversions keep producing near-identical pseudogenes that serve as donors for gene conversion, the mechanism chickens use instead of a large V(D)J repertoire. As predicted, IGH has few functional genes and many pseudogenes, IGL the reverse. Paired WGS and Iso-seq from one Red-winged Blackbird show a single IGLV gene diversified by conversion from neighbouring pseudogenes, and conversion at IGH.
What to take from it
- Inverted duplications at IGH are a bird-specific architecture
- Gene conversion from pseudogenes, known in chicken, looks general across birds
- IGH and IGL differ in functional-to-pseudogene ratio as the model predicts
- IG loci are a poor fit for a single linear reference in any bird
Review
The cross-species scale is the strength: 139 assemblies is enough to show the IGH architecture is general, and the IGH-versus-IGL prediction is a real test, not a post-hoc fit. The weak points are assembly and annotation. IG loci are long, repetitive and duplicated, exactly where assemblies collapse or misjoin, and inverted duplications are a known artefact type. The preprint needs to show the inversions are supported by spanning long reads or Hi-C per species, not only present in the assembly. 'Evolve rapidly' rests on population data whose sample sizes and genotyping method the abstract does not give. The expression test is one wild bird, which shows conversion happens but not how much of the repertoire it explains. Before using this, for example in chicken or duck disease studies, check per-locus read support and whether annotation used a consistent pipeline across assemblies of different quality.
02 medRxivpreprint · 7 September 2026_
Modeling pathway overlap increases accuracy of GWAS gene set enrichment
- Methods
- Human genetics
Gene Swap Randomization keeps pathway size and multi-pathway gene membership in the null, and shows shared genes alone can produce pathway enrichment in polygenic traits.
Summary
Pathway databases share many genes between pathways, and this overlap grows as databases expand. Under polygenicity, overlap can make a pathway look enriched without pathway-specific biology. Gene Swap Randomization (GSR) builds an empirical null that preserves pathway size and overlap structure. Applied to MSigDB across 12 traits and four methods (MAGMA, PascalX, GSA-MiXeR, PRSet), GSR improved agreement with external benchmarks (Open Targets, Malacards, DoRothEA, GTEx) in 60.8% of comparisons and 79.3% of disease-association benchmarks. Some pathway-specific enrichment remained after adjusting for overlap.
What to take from it
- Standard enrichment tests do not model gene sharing between pathways
- Overlap alone can create enrichment for polygenic traits
- GSR plugs into existing tools as a different null
- Gains are moderate: about 61% of benchmark comparisons improved
Review
The problem is real and widely ignored, and testing across four enrichment tools makes the result more than a quirk of one method. The evidence for improvement is weaker than the framing. 60.8% of comparisons improving means about 39% did not, and benchmarks like Open Targets are themselves partly built from GWAS, so concordance is not independent validation. A randomization null that preserves overlap also lowers power for small pathways nested in large ones; the preprint should report what is lost, not only what is gained. Want to see calibration under simulated null traits with realistic LD and polygenicity. For livestock work the issue applies directly, since pathway sets are usually mapped from human orthologs, which adds overlap. Use GSR as a sensitivity check alongside the standard test, not as a replacement, until independent benchmarks exist.
03 Molecular Biology and Evolution · 1 October 2026_
Identifying adaptive variation in spatially structured populations using low-coverage whole-genome sequencing data
- Population genetics
- Methods
A genotype-environment association model that carries genotype uncertainty from low-coverage data into a mechanistic model of local adaptation, migration, mutation and drift.
Summary
Most genotype-environment association methods either drop low-confidence loci or call the most likely genotype from low-coverage data, and fit environment with logistic regression. Structured migration between different climates can bend allele-frequency clines away from the smooth S-shape such models assume. The new method propagates genotype likelihoods into an evolutionary model of local adaptation, structured migration, mutation and drift. In simulations, both features reduced false negatives. It was applied to 3.7 million SNPs from North American rosy-finches.
What to take from it
- Calling genotypes from low-coverage data biases GEA; this keeps the uncertainty
- Migration between climates can hide adaptive clines from logistic models
- Gain reported is fewer false negatives in simulation
- Relevant to low-pass sequencing of landraces and local breeds
Review
The two problems it targets are real: low-coverage data are now the default for many non-model and livestock population studies, and hard calls at 1-2x distort allele frequencies. Building migration into the model is a sound reason to expect better behaviour than a GLM. What is missing from the abstract is the false-positive side: reducing false negatives is easy if the threshold moves. The simulations are generated under a model close to the one being fitted, which favours it. Needed before use: false-positive rate under model misspecification (e.g. isolation by distance, demographic expansion), runtime at millions of SNPs, and comparison with ANGSD-based alternatives and LFMM on the same data. The rosy-finch application has no ground truth, so it shows feasibility, not accuracy.
04 medRxivpreprint · 23 September 2026_
Schizophrenia-associated polygenic liability and structural genomic risk demonstrate broadly distributed neuropsychiatric associations in the All of Us Research Program
- Human genetics
- Structural variation
In 71,992 All of Us participants, carriers of recurrent neurodevelopmental CNVs and people with high schizophrenia PRS both show raised risk across many psychiatric phenotypes, with no sign of interaction.
Summary
71,992 participants were analysed, 2,115 carrying SVs overlapping curated recurrent neurodevelopmental CNV loci. CNV carriers had higher odds of schizophrenia diagnosis (OR 1.49, 95% CI 1.09-2.03); each SD of schizophrenia PRS raised odds of a broad psychosis-spectrum phenotype (OR 1.07). Deletions had larger effects than duplications. Psychosis-spectrum prevalence rose from 5.4% to 6.7% between the lowest and highest PRS deciles, with carriers higher at every decile. PRS x CNV interaction estimates were near zero. Both forms of risk extended to neurodevelopmental, cognitive and internalizing phenotypes.
What to take from it
- Rare CNV and common-variant risk both act beyond schizophrenia itself
- Effects look additive: no PRS x CNV interaction detected
- Deletions carry more risk than duplications
- PRS effect is small: 1.3 percentage points across deciles
Review
Joint analysis of CNVs and PRS in one diverse cohort is useful, and the additive result agrees with earlier work in UK Biobank and clinical samples. The weaknesses are measurement and power. 'Structural variants overlapping' a CNV locus is a loose carrier definition: partial overlaps and calls from short-read SV pipelines may include carriers who do not have the pathogenic deletion or duplication. EHR diagnoses undercount psychiatric disease and differ by health-care access, which varies by ancestry in All of Us. With 2,115 carriers, a near-null interaction is weak evidence of absence; the confidence interval on the interaction term matters more than the point estimate. The PRS is described as 'reduced', and its portability across ancestries needs to be shown. Before relying on this, ask for carrier calls validated against array or read-depth data and interaction CIs by ancestry.
05 bioRxivpreprint · 10 September 2026_
FANCM restrains structural genome evolution and defines a synthetic lethal dependency in BRCA1-deficient breast cancer
- Cancer genomics
- Structural variation
Timed FANCM depletion in BRCA1-deficient breast cancer cells increases tandem duplications, including megabase-scale ones, that form near transcribed regions.
Summary
BRCA1-FANCM synthetic lethality was shown in human BRCA1-deficient breast cancer cells. Controlled FANCM depletion allowed SVs to be tracked over successive divisions before the cells lost fitness. FANCM loss increased the short tandem duplications typical of BRCA1 deficiency and allowed larger, including megabase, duplications and other rearrangements. New duplications were enriched at Pol II-occupied regions, and FANCM depletion brought replication machinery closer to elongating RNA polymerase II. BRCA1-altered tumours with low FANCM expression showed similar features, and new SVs were linked to local expression changes.
What to take from it
- FANCM is required for survival of BRCA1-deficient cells
- Without FANCM, tandem duplications grow in number and size
- New duplications cluster at transcribed regions
- A mechanism linking transcription-replication conflict to SV formation
Review
The time-resolved design is a good way to watch SVs form rather than infer them from end-point tumours, and the Pol II association gives a testable mechanism. Limits: the work is in cell lines, and the abstract does not say how many lines, clones or replicates, or how SVs were called and at what size threshold, which decides whether 'megabase TDs emerge' means a few events or a shift in the distribution. Pol II regions are also open, early-replicating and easier to map, so enrichment needs a matched background. The tumour comparison uses low FANCM expression, a correlation that may track other features of those tumours. For readers outside cancer, the point is a defined process producing tandem duplications at genes; whether anything similar shapes germline SV is not addressed.