BGPT: Paper Review: Anthropocene genetic diversity loss in the marine tropics

Explore by Goal

Quick Explanation Copied

The paper reports that two tropical nearshore ponyfish species from the Philippines (museum 1908 vs modern 2018 samples) show genome-wide loss of diversity (π and heterozygosity) alongside strong recent bottlenecks and order-of-magnitude drops in effective population size (Ne), with minimal evidence of changing population structure over time (low temporal FST).

Long Explanation

Paper review (skeptical, evidence-based)

“Anthropocene genetic diversity loss in the marine tropics” (museum–modern temporal genomics)

DOI: 10.1101/2025.09.11.675623

Visualize first: core quantitative claims

Figure A — diversity metrics: % change (historic 1908 → modern 2018)

Values extracted from the manuscript text: average declines are reported as π −4.3%, HO −4.3%, HE −5.9% across both species (with species-specific ranges stated).

Figure B — temporal genetic differentiation: temporal FST

The manuscript reports temporal FST as very small values: 0.0047 (G. minuta) and 0.0065 (E. laterofenestra), consistent with limited changes in genomic composition through time.

Figure C — temporal Ne estimates (two estimators)

The manuscript reports two temporal Ne estimators: allele-frequency-change method (values in low thousands) and heterozygosity-loss method (values around ~650). Note these are on different statistical foundations and are not interchangeable; the manuscript discusses sensitivity of harmonic-mean temporal Ne to bottlenecks and possible age/spatial structure effects.

What the paper claims (and how it’s supported)

1) Temporal genomic signal is population-local (structure largely stable)

PCA and ADMIXTURE are reported to show limited changes through time, with temporal clustering and K=1 best supported; temporal FST values are very low (Figure B).
Skeptical lens: low FST does not automatically imply unchanged Ne or diversity; diversity erosion can occur with stable allele composition if drift/bottlenecks reduce within-population diversity while keeping mean allele frequencies similar. The manuscript’s approach explicitly tests diversity metrics and demographic history rather than relying on FST alone.

2) Diversity loss is reported as genome-wide erosion

The manuscript reports average declines of π, HO, and HE (Figure A), plus temporal shifts in the two-dimensional site-frequency spectrum and a Tajima’s D shift consistent with contraction in E. laterofenestra.
The paper states these declines can exceed null expectations in at least one species (E. laterofenestra) using permuted-population null models.

3) Demographic modeling supports recent bottlenecks, not just diversity noise

The authors use momi2 to compare multiple Ne-change models, selecting by AIC and using bootstrapped SFS to estimate uncertainty. The text reports best-fitting models that include deeper-time expansion and a more recent dramatic bottleneck (timing assessed on the order of the last century for both species, with an additional possible earlier start when including historical data).
Selection scans (BayeScan) reportedly detect only one temporal outlier SNP across both species, which the authors interpret as limited evidence that diversity loss was driven by strong, detectable genome-wide selective sweeps (noting reduced-representation design limitations).
Foundational caution (conceptual): momi2 and SFS-based Ne inference assume that the chosen model class can explain observed SFS shifts. Many distinct demographic histories can map to similar SFS patterns, especially with reduced genomic coverage and sample-size constraints—so “best AIC” is not the same as “unique truth.” (General SFS-based demographic modeling caveat aligned with the stated simulation-based evaluation rationale for momi2-like approaches.)

Skeptical critique: where the evidence is strong vs fragile

Strengths

Temporal sampling design: the work compares historic (1908) vs modern (2018) specimens from the same locality, explicitly targeting a temporal baseline.
Damage-aware handling: the manuscript quantifies historical DNA damage with mapDamage2 and recalibrates base quality, which is relevant when comparing historic vs modern short-read data.
Multiple complementary diversity readouts: π, HO, HE, 2D-SFS changes, Tajima’s D, and demographic modeling are all used, reducing dependence on any single statistic.
Data/code transparency claims: the manuscript reports deposition of sequence data in NCBI SRA and scripts/results on GitHub and Zenodo.

Fragilities / blind spots (what could change the conclusion)

Reduced-representation sequencing limits selection inference: the authors explicitly acknowledge that reduced genome coverage hinders detection of selection signatures.
Sample size per time point is small (11–92 individuals/time point after filtering; 184 total). Small N and missingness can inflate uncertainty or induce bias in SFS and diversity estimators (especially when downsampling for 2D-SFS).
Cryptic species/contamination filtering decisions: the manuscript removes putative cryptic species and removes individuals with anomalously high heterozygosity; such filters can affect allele frequency spectra and thus inferred Ne.
Ne estimators are sensitive to structure/age: the manuscript discusses that temporal Ne (especially harmonic means) can be influenced by age and spatial structure and may reflect local vs metapopulation Ne.
Geographic generality: conclusions are drawn from a single location in the Philippines. Even if the bottleneck is real there, extrapolation to “tropical marine fishes” requires caution.

Evidence connections to prior work (context-only)

The paper situates its results within broader claims about global genetic diversity loss and the skew of empirical evidence toward temperate regions, citing an Essential Biodiversity Variable framing and prior meta/landscape analyses.

Essential Biodiversity Variable framing for genetic diversity maintenance is discussed via Hoban (2022).
Anthropocene genetic diversity loss estimates across wild populations are discussed via Leigh et al. (2019) and Exposito-Alonso et al. (2022).
Temporal genomics as a method for measuring evolutionary responses is anchored in Clark (2025).

Reproducibility checklist (based only on the provided text)

Item	What the manuscript says	Risk if missing
Sequence data	Deposited in NCBI SRA (PRJNA998814, PRJNA998057, PRJNA998845, PRJNA999299) and linked via GEOME-DB arks.	Cannot re-run mapping/genotyping and re-check filtering & diversity estimates.
Scripts	Scripts for results available on GitHub and archived on Zenodo (DOI 10.5281/zenodo.15085162).	Hard to verify exact processing steps (e.g., downsampling/filters).
Bioinformatics methods detail	Mapping/quality filters (MAPQ≥30, discard secondary/supplementary reads), mapDamage2 recalibration, FreeBayes calling, biallelic SNP retention; PCA/ADMIXTURE settings and permutation/null framework are described.	Small differences in filters or damage handling can shift SFS and Ne.

Data availability and methods are explicitly stated.

Author review links (bespoke deep dives)

Tap to generate an independent author-focused review on BGPT.

Note: BGPT author-button links are created only for authors whose full names are explicitly extractable from the provided manuscript header text.

Feedback:

Updated: March 20, 2026