Tissue identity is the dominant determinant of cross-species transferability of a porcine developmental programme

Liu, Tianyuan ORCID: https://orcid.org/0000-0002-8561-6239 2026. Tissue identity is the dominant determinant of cross-species transferability of a porcine developmental programme. MPhil Thesis, Cardiff University. Item availability restricted.

Preview	PDF - Accepted Post-Print Version Download (3MB) | Preview
	PDF (Cardiff University Electronic Publication Form) - Supplemental Material Restricted to Repository staff only Download (264kB)

Abstract

Reproducibility in pig research is hampered by the lack of a standardised, molecular definition of developmental stage. The core problem is that while the pig is an excellent preclinical model for human conditions, the molecular tempo of its development is vastly different from that of humans. Synchronising the maturation rates of the two species is essential to move preclinical research from a descriptive exercise to a predictive science. Current staging relies on chronological age or subjective morphological criteria, which fail to capture inter-individual variability and the asynchronous maturation of different organs. This thesis presents a transcriptomic atlas that systematically maps porcine development across five major tissues (brain, liver, muscle, lung, and blood) from 1,924 RNA-seq profiles in the PigGTEx resource, anchoring developmental stages to porcine physiological milestones (weaning, puberty, and reproductive maturity) and classifying each tissue at its native ordinal resolution by dropping, rather than merging, under-sampled stages. Developmental stage was modelled with a single partial-least-squares (PLS) regressor, used uniformly for within-pig staging, cross-species transfer, and biomarker extraction. The central result is that tissue identity is the dominant determinant of how a pig trained developmental score transfers to other species, with phylogenetic distance a weaker secondary effect. Projected onto the seven-species Cardoso-Moreira atlas, transfer was strongest for brain (ρ = 0.92) and heart (ρ = 0.89) and lowest on average for liver and ovary (ρ = 0.49 and 0.42), with heterogeneous values across species. Brain and heart transferred uniformly across all seven species and remained high even in chicken (ρ ≥ 0.95 across ∼320 million years of amniote divergence). A crossed design over six organs × seven species confirmed organ identity as the dominant determinant (Type-II ANOVA organ F(5,33) = 9.0, p = 2 × 10−5, random-factor reference p < 0.001; phylogeny significant only at the species level, Spearman −0.76, p =0.049, n = 7). The score independently confirmed lung, muscle, and adipose on the design-matched human developmental GTEx (dGTEx) resource (ρ = 0.59–0.67), each significant against a shared-symbol-scramble null and target-label permutation. Because the score is fit on pig and projected onto foreign-batch atlases, positive transfer rules out a pig-side technical-batch explanation but not every target-side or biological confound. Within-pig staging itself is reported with separability and deployment kept strictly separate. Stratified cross-validation (a batch-inclusive measure, not a deployment metric) shows developmental stage is separable in every tissue. The reproducible cross-study estimate is for muscle (grouped balanced accuracy 0.45 ± 0.04, ordinal ρ = 0.66, chance 0.20). Liver’s 0.39 estimate is a restricted four-stage sensitivity analysis (excluding neonatal and adult), not validation across its full six-stage scheme. For the remaining tissues developmental stage is heavily or perfectly confounded with the contributing study, so the cross-study quantity is not identifiable—a limit of the resource’s design, not evidence of absent biology. The PLS-VIP biomarkers are led by the muscle programme (accelerators MSS51/ZMYND17, ASB14, HOXC8, NEDD4; decelerators CPLX1, FGFRL1, DLK1), with the other tissues’ biomarkers reported under explicit study- and sex-confound caveats. This thesis makes four key contributions: (1) a physiologically anchored transcriptomic atlas for porcine postnatal development; (2) a single, transparent ordinal-PLS frame work that doubles as the engine for cross-species transfer and biomarker extraction; (3) the central finding that the porcine developmental programme transfers to other species in a tissue-dependent manner, with tissue identity the dominant determinant and phylogenetic distance a weaker secondary effect; and (4) an honest characterisa tion of within-pig staging that separates in-distribution separability from cross-study deployment, which is currently established for muscle and not identifiable for the study-confounded tissues; liver is retained only as a restricted four-stage sensitivity analysis. An interactive web application (https://pigdevstage.streamlit.app) makes the staging tool publicly available.

Item Type:	Thesis (MPhil)
Date Type:	Completion
Status:	Unpublished
Schools:	Schools > Engineering
Uncontrolled Keywords:	1. Porcine (pig) development 2. Developmental staging 3. Transcriptomics and RNA-seq 4. Cross-species transferability 5. Machine learning; partial least squares regression 6. Gene expression biomarkers
Date of First Compliant Deposit:	15 July 2026
Last Modified:	20 Jul 2026 08:59
URI:	https://orca.cardiff.ac.uk/id/eprint/188080

Actions (repository staff only)

Edit Item