BGPT: Paper Review: Alternative 3′ UTR polyadenylation is disrupted in the rNLS8 mouse model of ALS/FTLD

Explore by Goal

Quick Answer Copied

Key takeaway:

Re-analysis suggests the rNLS8 (TDP-43 ΔNLS) model has widespread 3′UTR alternative polyadenylation (APA) changes that (i) track gene-expression/protein-abundance correlations and (ii) show partial overlap with independent TDP-43 perturbation models—supporting both TDP-43 nuclear loss-of-function and toxic cytoplasmic mechanisms as contributors, while also leaving causality and cell-type resolution as major uncertainties.

Long Answer

Paper Review (science-focused, skeptical, evidence-based)

Manuscript: Alternative 3′ UTR polyadenylation is disrupted in the rNLS8 mouse model of ALS/FTLD (10.1186/s13041-025-01174-1)
What they did (high level): computational APA re-analysis (APAlyzer + REPAC) of existing RNA-seq datasets from rNLS8 and multiple comparison models, plus pathway enrichment and cross-omics correlation checks, and cross-species conservation comparisons.

1) Visualizing the paper’s quantitative claims (from provided full text)

Below figures are constructed only from numbers explicitly stated in the provided paper text.

2) Evidence map: what is supported vs what is inferential

Supported by explicit computation: number of genes/sites with differential 3′UTR APA in rNLS8 neocortex after re-analysis (APAlyzer: 217; REPAC: 309 sites; 199 genes).
Supported (but correlation only): APA-linked 3′UTR length changes explain a small fraction of expression variance overall (R²≈0.02), but larger fraction when restricting to APA-significant genes (R²≈0.38).
Supported (and time-dependent): APA–protein correlation exists but is modest (e.g., R²≈0.14 early stage) and varies by proteomics timepoint, suggesting measurement-timing/causality uncertainty.
Supported as mechanistic signal, not proof: overlap of APA genes with independent TDP-43 perturbation datasets is used to argue for both loss-of-nuclear-function and toxic gain-of-function. This is inferential but grounded in stated overlaps.
Supported at face value: some conservation with human ALS cell models is claimed (~23 APA genes (~10% of APAlyzer APA genes) disrupted in at least one human in vitro model).

3) Cross-paper context: how this fits (and where it may differ)

Consistent theme: TDP-43 nuclear loss can produce widespread APA changes detectable by high-resolution 3′ end sequencing.

Relevance to rNLS8 paper: The rNLS8 re-analysis also emphasizes overlap signals consistent with TDP-43 nuclear depletion contributing to APA dysfunction, but it relies on bulk RNA-seq-based APA inference rather than 3′ end-seq resolution as described in the human study.

Cell-type interpretation risk: In rNLS8, microglial transcriptome profiling suggests stage-specific inflammatory programs, but bulk approaches can dilute or miss cell-type specific APA.

Relevance to rNLS8 paper: The APA paper finds fewer APA changes in isolated microglia than in whole brain, and interprets this partly as microglia lacking TDP-43 pathology; however, cell-state differences, sensitivity of APA callers, and absence of single-cell APA measurement can complicate causal attribution.

4) Skeptical critique: key blind spots and likely failure modes

A) Bulk RNA-seq inference for APA can be underpowered for cryptic/nearby polyA sites.

APA calling from standard RNA-seq can miss or mis-assign polyA usage—especially when read length/fragmentation and library prep bias mapping to 3′ ends. The authors themselves explicitly note bulk RNA-seq limitations for APA and cryptic APA site detection.

B) Correlation ≠ causation (APA may be a downstream marker).

Even when APA length correlates with gene expression and proteomics, the explanatory power is often modest overall and time-dependent, consistent with measurement alignment constraints—or with APA reflecting broader dysregulation rather than being the primary driver.

C) Cross-model overlap is susceptible to pipeline + annotation differences.

Shared genes across datasets depends on shared reference annotations, read lengths, and preprocessing. Without uniform re-processing (same APA library assumptions), overlaps could under- or over-estimate true biological conservation. The paper uses re-analysis but still must reconcile differing datasets (sex, age, read length, depths).

D) Human conservation is limited and context-dependent.

The paper reports small numbers of shared APA genes across certain human in vitro models (e.g., only 2 by APAlyzer between mouse rNLS8 and specific iPSC motor neuron datasets in one comparison). That suggests partial—not comprehensive—translation of mouse APA events to human cell models.

5) What would most improve causal confidence?

The strongest missing link is experimental validation of selected APA events (especially those with the most disease-relevant overlap signals). The paper itself frames the work as supportive and calls for additional validation.

3′ end sequencing validation for a small panel of high-confidence APA genes (to confirm true polyA site usage changes rather than readthrough/annotation artifacts). (Supported by the authors’ discussion of bulk RNA-seq limitations for cryptic APA; causality direction is not yet proven.)
Functional perturbation of individual polyA events (reporters or isoform-specific manipulations) to test whether APA changes drive protein-level outcomes rather than tracking them.
Cell-type resolved APA (single-cell or nuclei-based 3′ end strategies) to confirm whether the APA dysfunction is neuronal-autonomous or partly glial/non-neuronal via indirect coupling.

6) Buttons for deeper exploration (BGPT)

This agent can iteratively check overlaps/consistency and build additional validation-oriented figures, using any supplemental tables if provided.

Feedback:

Updated: April 18, 2026