Donor Background Overpowered PMOS Status in the iPSC Transcriptome
A genotyped study of 14 patient-derived induced pluripotent stem cell lines found no robust transcriptomic signature separating polyendocrine metabolic ovarian syndrome from control cells. Investigators worked with seven women meeting all three Rotterdam criteria for PMOS and seven controls matched by body mass index category, then asked a blunt question: does disease status still stand out once donor genetics, BMI, and line-to-line variation are taken seriously?
It did not.
Across 16,489 expressed genes in bulk RNA-seq, no gene met the false discovery rate threshold for differential expression between PMOS and control lines. The same null pattern held when expression was quantified with an alternative read strategy and when count-based models were used instead. In practical terms, the cell lines behaved more like a collection of individual donors than a shared disease model, even before differentiation or hormone stimulation entered the picture.
A Larger Panel Narrowed the Gap Left by Earlier Small Studies
Earlier iPSC comparisons in this disease space relied on three or four donors per group and reported hundreds to thousands of differentially expressed genes. This study was designed to test whether those signals would survive in a more disciplined framework: more donors, genotyping, BMI matching, and BMI-adjusted analyses. With seven donors per group, the design could detect 1.8-fold differences per gene, a much tighter window than the roughly 3- to 6-fold differences implied by the earlier small-panel work.
That shift in power matters because cell lines can easily produce convincing-looking false positives when donor effects are large and the sample is thin. Here, exact enumeration of all 3,432 possible group splits showed that the number of nominally associated genes in the true PMOS-versus-control comparison was no greater than random 7-versus-7 partitions of the same lines. The permutation result was unambiguous: p = 0.75.
The biological message is straightforward. In undifferentiated iPSCs, the background genome of the donor appears to dominate the transcriptome more than PMOS status itself.
Hormonal Differences Existed, But They Did Not Translate Into a Disease Expression Program
The clinical donors were not interchangeable. Women in the PMOS group had higher anti-Mullerian hormone, testosterone, and free androgen index than controls, which fits the endocrine phenotype the study set out to model. Yet those hormonal differences did not map onto a coherent gene-expression signature in the reprogrammed cells.
Neither diagnosis nor any measured hormonal or metabolic trait explained more expression variance than chance. PMOS-associated risk-locus genes and Hallmark pathways also failed to show enrichment. The mismatch between serum phenotype and transcriptomic readout is a warning against assuming that a clinically obvious endocrine disorder will reveal itself in an undifferentiated pluripotent state.
The likely reason is biological as much as technical. iPSCs are reset cells, and much of the hormonal circuitry relevant to ovarian steroidogenesis, follicle signaling, and metabolic crosstalk is not fully active in that state. A disease driven by endocrine responsiveness may only become visible after differentiation into the relevant lineage or after exposure to the right hormonal context.
XIST Loss in Control Lines Raised a Separate Quality Concern
Two control lines showed near-complete loss of XIST expression, and a third was borderline. XIST is central to X-chromosome inactivation, so reduced expression can signal erosion of that epigenetic state. The finding is not proof of X-chromosome inactivation failure, but it is strong enough to complicate interpretation of any iPSC transcriptome that depends on female cell lines.
This is the kind of technical issue that can quietly distort disease modeling. If one subset of lines drifts in X-inactivation status, the resulting expression noise can mimic or obscure disease-associated biology, especially in a disorder centered in women and in cells derived from female donors. The authors also note that one clone per donor leaves room for clonal variation, which cannot be separated fully from donor background in this design.
A single aberrant clone can shape the apparent biology of an entire donor.
Genetic Risk Was Higher in PMOS, Yet Overlap Remained at the iPSC Level
Within a cohort of 129 genotyped women, the PMOS polygenic risk score was higher in PMOS than in controls, with an odds ratio of 1.81 per standard deviation and an area under the curve of 0.65. That result supports a measurable genetic burden for the disorder at the population level. Even so, PMOS and control iPSC donors overlapped fully in polygenic risk, which means the transcriptomic null result cannot be dismissed as a simple failure to sample genetically distinct groups.
The study also examined PMOS risk-locus genes against a 202-line iPSC proteome. Endocrine PMOS-locus genes such as FSHR and AMH were not expressed in the undifferentiated lines, while DNA-repair and growth genes were. That pattern reinforces the central limitation of the model: the most biologically intuitive disease genes may be silent until the right differentiated state is reached.
For translational work, this is not a minor caveat. It means that a negative iPSC transcriptome in the undifferentiated state does not rule out disease biology; it only narrows where that biology is likely to emerge.
What This Means for PMOS Modeling Going Forward
The study makes a strong case for larger, genotyped, BMI-aware donor panels when modeling PMOS in iPSCs. It also argues for controlling XIST erosion and for selecting donors based on polygenic risk rather than diagnosis alone. Most of all, it suggests that future work should move beyond undifferentiated cells and into hormone-responsive derivatives where the endocrine phenotype has a chance to appear in the transcriptome.
The authors do not show that PMOS lacks a cell-based molecular signature. They show that 14 undifferentiated lines were not enough to reveal one, and that donor variation dominated the readout under the conditions tested. Disease-associated differences in differentiated cells remain possible, as do effects missed by a single clone per donor. For now, the cleanest interpretation is that donor background is the loudest signal in this model, and the disease signal, if present, will need a more appropriate cellular stage to surface.