Alternative Splicing in Plant Spaceflight Transcriptomes

Overview

Characterizing post-transcriptional regulation under spaceflight-relevant stresses.

While transcriptional changes in spaceflight have been widely mapped, alternative splicing remains a major frontier. This pipeline leverages rMATS-turbo to quantify alternative splicing events, classifies potential nonsense-mediated decay (NMD) sensitivity with a frameshift heuristic, performs machine learning (UMAP, nested cross-validation, LOSOCV, and SHAP importance), and conducts multi-ontology enrichment analysis (GO, MapMan) to uncover the post-transcriptional footprint of space stress.

51–72%

Retained Introns

Retained Intron (RI) events are the largest class of significant events in 6 of 7 studies (51–72% of significant events). The exception is OSD-37, where only 10 events were significant (SE 4, RI 3).

79–86%

Heuristic NMD class

Share of significant events classified as potentially NMD-sensitive by a frameshift heuristic that counts every RI event as NMD-sensitive. It mostly reflects the RI share; NMD itself was not measured.

93 genes

Recurrent Targets

93 genes are significant in 2 or more of the 7 studies (85 in two, 7 in three, 1 in four).

01The Spaceflight Splicing Landscape

Retained introns are the most frequent class of significant splicing change.

Across spaceflight (ISS), altered gravity, simulated galactic cosmic ray (GCR) radiation, and growth in Apollo lunar regolith, rMATS detected significant splicing changes in every study (10 to 306 events at FDR < 0.05 and |ΔPSI| > 0.1). Retained introns are the largest class of significant events in 6 of the 7 studies. Whether these intron-retention events introduce premature termination codons and trigger NMD was not tested here: the NMD classification is a frameshift heuristic that counts every retained intron as NMD-sensitive.

Distribution of alternative splicing classes across studies — Distribution of significant alternative splicing events across 7 Arabidopsis studies. Retained introns (RI) are the largest event class in every study except OSD-37 (SE 4, RI 3).

Pairwise gene overlap and recurrence across studies — Pairwise overlap of differentially spliced genes between studies, highlighting core stress-responsive genes sharing splicing changes.

02Recurrent Splicing Targets: AT4G17310

Recurrence analysis finds genes significant in several studies across gravity, radiation, and lunar substrates.

Comparing significant genes across the 7 Arabidopsis studies identified 93 genes significant in 2 or more studies. The most recurrent is AT4G17310 (an uncharacterized protein in Araport11), significant in 4 of the 7 studies: fractional gravity (OSD-251), altered gravity (OSD-314), lunar regolith (OSD-476) and simulated GCR radiation (OSD-658). Seven further genes are significant in three studies, including SOS4 (AT5G37850, pyridoxal kinase) and CRK28 (AT4G21400, cysteine-rich receptor-like kinase 28).

AT4G17310 Splicing: The change is not uniform. The event types are retained intron in OSD-314 (ΔPSI −0.171) and OSD-658 (ΔPSI +0.416), and skipped exon in OSD-251 (ΔPSI −0.155) and OSD-476 (ΔPSI −0.132). The gene is not significant in the three ISS flight-versus-ground datasets (OSD-120, OSD-37, OSD-678).

03Machine Learning: Batch Effects & Generalization

UMAP and tree-based classification reveal that study-specific batch effects dominate biological signals.

Machine learning classification of the 199-sample × 9,875-gene Percent Spliced In (PSI) matrix achieves a pooled nested 5-fold CV AUC of 0.598 for predicting treatment (stress) versus control. However, 7-study Leave-One-Study-Out Cross-Validation (LOSOCV) performance drops to near or below random (mean AUC across held-out studies: RF 0.49, GBM 0.42, XGBoost 0.41), showing poor cross-study generalization. Unsupervised UMAP embedding reveals that samples cluster almost perfectly by study identity (ARI = 0.995) rather than by treatment status (ARI = -0.002). This represents a critical methodological finding: batch effects overwhelmingly dominate the alternative splicing landscape in public space omics.

UMAP clustering of splicing profiles — UMAP projection of splicing profiles across 199 samples. Near-perfect clustering by study identity (ARI = 0.995) demonstrates dominant batch effects.

7-study leave-one-study-out classification performance — LOSOCV performance across multiple tree-based models (RF, GBM, XGBoost), demonstrating the challenges of cross-study generalizability.

Nested k-fold cross-validation performance — Per-fold AUC of nested 5-fold cross-validation (Random Forest) on samples pooled across the 7 studies, showing moderate within-cohort predictive signal (mean AUC = 0.598 ± 0.084).

SHAP feature importance top genes — Top splicing biomarkers identified by SHAP importance (including AT1G56220, AT5G01770, AT1G48360) for classification models.

04Ontology & Functional Convergence

Circadian-related GO terms recur in four flight datasets, while other stressors show distinct enrichments.

GO over-representation of the genes with significant splicing changes (terms with FDR < 0.25 unless stated):

Circadian terms: These are enriched in four of the five ISS flight datasets: OSD-314 (FDR 0.032, 2 genes), OSD-120 (FDR 0.097, 2 genes), OSD-37 (FDR 0.091, 1 gene, RVE1) and OSD-678 (FDR 0.008, 4 genes). They are not enriched in OSD-251. OSD-120 (root tips) and OSD-678 (leaves) come from the same CARA flight, so they are not independent experiments.
Lunar Regolith (OSD-476): This study has the most significant events (306). Its top terms include regulation of photosynthesis light reactions, regulation of the mitotic cell cycle and telomeric chromosome region, each with 2 genes at FDR 0.138.
Fractional Gravity (OSD-251): The strongest enrichment is response to superoxide (2 genes, FDR = 0.001, 488x fold enrichment).
GCR Radiation (OSD-658): Top terms include response to herbicide (FDR 0.009), phospholipid metabolic process (FDR 0.013) and regulation of cytoplasmic translational initiation in response to stress (FDR 0.016).

05Splice Locus Visualizations

Exploring reconstructed splice graphs and junction-level splicing metrics.

Splice graph models were reconstructed for the top differentially spliced loci. Use the interactive explorer below to select a study, browse key differentially spliced genes, inspect splicing patterns in vector (SVG) or raster (PNG) formats, and zoom/pan to view specific exon junctions: