BGPT: Paper Review: Unique fingerprint of marine ectotherm body size change during hyperthermal crises

Explore by Goal

Quick Explanation Copied

Quick take: The authors compile ~6k–9k published body-size change records and show a consistent Lilliput/dwarfing during crises with significantly greater magnitude and volatility during hyperthermal (warming) events; warming magnitude modestly predicts dwarfing (slope ≈ -0.018 lnRR per °C). Key strengths are a large open dataset and reproducible code; major caveats are uneven taxonomic/temporal sampling, heterogenous measurement types, potential non-independence, and low explanatory power of the warming–lnRR regression (R2=0.018). Data & code: Zenodo (DOI) available.

Long Explanation

Visual summary — Key signals in the paper

Visuals first, brief interpretation after. Figures reproduce paper summary metrics (values taken directly from the manuscript / SI summary provided).

Interpretation, strengths, and detailed critique (evidence-cited)

Main claim reproduced: The manuscript compiles ~5,900–9,000 lnRR records (paper text: 5912 main changes; compiled dataset ~9000 records overall from fossil, historical, modern studies) and reports that median body-size change during crisis intervals is significantly negative (consistent Lilliput/dwarfing), with recovery intervals showing positive shifts—this is supported by the authors' analyses and the Zenodo dataset .
Hyperthermal fingerprint: The authors find that hyperthermal crises (those with substantial warming) show roughly double the median magnitude of lnRR (|lnRR| = 0.144 vs 0.065) and >2× volatility (SD = 0.274 vs 0.104) compared to non-hyperthermal events (Wilcoxon p<0.0004). This is central to their novelty claim that warming-driven crises have a distinct 'fingerprint' of stronger and more volatile species-level dwarfing .
Warming magnitude correlation is weak but significant: Regression reported slope = -0.018 lnRR per °C, R2 = 0.018 (p < 0.0001): statistically significant across events but explaining only ~1.8% of variance—implies warming matters, but much variance remains because of interacting stressors and sampling heterogeneity .
Context with modern experimental and observational work: The result that warming links to smaller body sizes aligns with modern temperature–size literature (TSR and field studies), e.g., broad fish time-series showing many species shrink with warming, but responses vary among taxa and contexts (some species grow) . The preprint's cross-timescale synthesis is therefore consistent with but not identical to modern patterns: warming is an important driver but not a sole determinant.
Major strengths
- Scale & openness: large, cross-taxonomic dataset spanning fossil → modern records and open archival of data/code at Zenodo (DOI), enabling reproducibility and reanalysis ).
- Clear statistical design: lnRR standardizes disparate size measures; non-parametric tests mitigate unit/normality issues; jackknife used to test influence of individual studies.
- Important conceptual distinction: separating hyperthermal vs non-hyperthermal events and splitting species-level vs above-species effects helps identify different proximate mechanisms (physiological/ecophenotypic vs selective extinction/turnover).
Primary weaknesses and blind spots (how results can mislead)
- Sampling heterogeneity and bias: The dataset pools measurements of different size metrics (length, area, centroid size, log area) across taxa, epochs, and methods; lnRR helps but cannot fully remove biases introduced by variable measurement practices, preservation biases in fossils, and differing sample sizes/measurement error across studies — authors acknowledge this (Methods / Limitations) ).
- Independence and pseudoreplication: Multiple time-series derived from the same specimens at different taxonomic/temporal resolutions risk non-independence. The authors used an 'independence indicator' and randomized selection in some analyses, but residual dependence can bias p-values and effect sizes, particularly when many records cluster around a few well-studied events (end-Permian, PETM) — jackknife partially addresses this but cannot replace hierarchical modeling that explicitly models nested structure.
- Weak explanatory power of warming regression: Slope significant but R2 = 0.018 indicates warming magnitude explains only ~1.8% of variance in lnRR across events; the remainder likely reflects interacting drivers (oxygen stress, productivity collapse, acidification, sampling variance, taxon-specific responses). The claim that warming is a 'major driver' is defensible qualitatively but quantitatively limited by low R2 — the paper carefully states 'major' in an interpretive sense but readers must note statistical modesty (small fraction of variance explained).
- Event classification ambiguity: Assigning 'hyperthermal' vs 'non-hyperthermal' per event relies on literature-derived temperature estimates that vary in quality and spatial representativeness; per-event maxima may not reflect regional or habitat-specific thermal exposure for the measured taxa.
- Scale-dependence & taxonomic breadth: The effect magnitudes increase with taxonomic and temporal scale (authors report scaling effects). That means assemblage-level signals may partly reflect shifts in community composition rather than within-species physiological change. The paper acknowledges mechanism differences by scale, but explicit hierarchical partitioning (within vs between species contributions) could be strengthened.
- Possible publication bias: Studies reporting significant size changes may be preferentially published; although the dataset is broad, it may over-represent positive/interesting findings; authors include Dryad baseline data to partially address background sampling but residual bias is possible.
Methodological improvements I would request before strong causal claims
1. Run multilevel (mixed-effect) models with random effects for source publication, taxon (nested species→genus→class), region, and event to explicitly partition variance and account for non-independence; contrast results to their current non-parametric tests.
2. Quantify and explicitly model measurement heterogeneity (e.g., grouping by metric type: length vs area) or use subset analyses restricted to a single metric where possible.
3. Use sensitivity analyses reclassifying events (vary warming thresholds) and evaluating effect on the hyperthermal vs non-hyperthermal contrasts.
4. Estimate effect sizes and uncertainties at taxon-level (random-slope meta-analytic framework) to measure how consistent the warming–lnRR response is across clades.
What would falsify the paper's main conclusions?
- Reanalysis using hierarchical models with publication/taxon random effects shows no significant median negative lnRR during crises globally (i.e., medians not different from zero).
- No significant difference in magnitude/volatility between hyperthermal and non-hyperthermal crises after accounting for sampling and nested effects.
- Warming magnitude no longer correlates with lnRR when improved temperature reconstructions or alternative event metrics are used.
Confidence in the central empirical claims: The dataset and statistics convincingly show a recurring negative median shift during crises (Lilliput-like), and the hyperthermal vs non-hyperthermal contrast is statistically robust in their analyses; however, causal attribution to warming as the primary driver is supported but quantitatively weak (low R2) and should be framed as one of multiple interacting drivers. For modern comparisons, heterogeneity of species responses in observational datasets (e.g., many fish species do not uniformly shrink) cautions against over-generalization .

Concrete, testable follow-ups (experiments & analyses)

Hierarchical reanalysis (priority): Fit a mixed-effects meta-analytic model of lnRR with random intercepts for study, taxon (species nested), region, and fixed effects for crisis type (hyperthermal vs not), measurement metric, temporal resolution; include per-measurement sampling variance where available to weight observations. This will yield variance components and test whether the hyperthermal signal survives when controlling for nested structure.
Within-species experimental test: For a representative modern marine ectotherm (e.g., small mollusk or crustacean with broad range), run multi-factor lab experiments combining warming, low-O2, and food-reduction to map size-at-age/size-at-maturity responses and test whether warming alone vs interacting stressors produce the species-level volatile dwarfing envisioned for hyperthermals.
Paleo-physiological proxy crosswalk: For well-dated fossil series (e.g., foraminifers, belemnites), jointly analyze size time-series and local proxies for temperature, δ13C, redox (trace metals, pyrite), and productivity proxies to disentangle correlated drivers at event scales.

Data & code access (reproducibility)

The authors state data and code are archived on Zenodo (https://doi.org/10.5281/zenodo.15096088) and Dryad sources supplement background records; this strongly supports reproducibility and further reanalysis .

Notes: All quantitative numbers in visuals (median |lnRR|, volatility, regression slope/R2) are taken from the manuscript's reported summaries. For detailed re-analysis we recommend the hierarchical modelling steps above using the archived Zenodo dataset.

Feedback:

Updated: March 19, 2026