Metformin’s effect on ESR1 splicing in breast cancer cells depends on how densely the cells are grown
Metformin did not produce a stable, dose-responsive shift in ESR1 splicing across the full range tested in breast cancer cells grown at high seeding density. Instead, the preprint shows a treatment effect that changes with culture context, a reminder that splicing biology can look persuasive in one assay condition and disappear in another.
In MCF-7 cells and in the tamoxifen-resistant TAMR1 subline, investigators examined whether metformin altered the balance between full-length estrogen receptor alpha and the truncated ERα46 isoform generated by skipping exon 1. At high density, metformin from 0.25 to 5 mM failed to produce a consistent direction of change in the full-length share of ESR1 transcript. The reported percentages moved irregularly rather than along a clean pharmacologic gradient, which weakens any claim of a straightforward dose effect in that setting.
The density dependence is the central biological signal here. In low-density MCF-7 cultures, the earlier shift persisted over time, with the full-length transcript share rising at 48 hours as metformin concentration increased. In TAMR1 cells, however, the direction and magnitude of the response were not reproducible across the same kinds of conditions, and the high-density comparison that had originally suggested a metformin effect did not hold up when the dose range was expanded.
HMGA1a fell, but its direction of change does not fit the splice pattern seen in TAMR1 cells
The most direct mechanistic lead examined in this study was HMGA1a. Metformin lowered HMGA1a protein in MCF-7 cells and showed the same tendency in TAMR1 cells, regardless of seeding density. That result matters because HMGA1a had already been shown by this group to promote ERα46 generation by trapping U1 snRNP at an upstream pseudo-5′ splice site.
A reduction in HMGA1a should therefore be expected to reduce ERα46 formation. The observed TAMR1 behavior ran in the opposite direction under low-density conditions, which makes HMGA1a a poor explanation for the splicing phenotype in that model. The mechanistic mismatch is not a footnote; it is the main reason the authors move away from HMGA1a as the carrier of metformin’s effect on ESR1 processing.
This is the kind of result that is easy to miss if only the direction of a band shift is read off a gel. In splicing regulation, the identity of the relevant RNA-binding factor matters as much as the final isoform ratio, and the data here argue that a fall in one candidate regulator does not automatically explain the transcript outcome.
RBM3 is the more plausible splicing mediator in the metformin response
A second protein, RBM3, fits the biology more cleanly. Metformin lowered RBM3 protein in TAMR1 cells at 24 and 48 hours. RBM3 is already known as a protein through which metformin alters splicing in myotonic dystrophy, so the convergence on this factor gives the preprint a more credible mechanistic route than the HMGA1a pathway.
That does not mean RBM3 has been proven to mediate the ESR1 splice shift here. It does mean the candidate list has narrowed in a biologically sensible way. A drug effect that tracks with exposure time and appears only under certain culture densities is exactly the kind of pattern one would expect if metformin were acting through a stress- or RNA-processing program rather than a simple on-off receptor effect.
The exposure-time data reinforce that point. In low-density MCF-7 cultures, the metformin-associated shift in full-length ESR1 transcript remained visible at 48 hours. The temporal persistence suggests that the response is not a fleeting artifact of early handling, but neither is it robust enough to be treated as universal across growth states.
The xenograft readout does not resolve the treatment comparison cleanly
The authors also revisited RT-PCR panels from their earlier xenograft work. Those tumor samples detected full-length ESR1 transcript in the untreated and tamoxifen-plus-metformin groups, but not in the single-agent groups. Several lanes failed to amplify, leaving the figure too incomplete to separate treatment effects with confidence.
This matters because a noisy endpoint in tumor tissue can easily be overread as mechanistic validation when it may simply reflect amplification failure, loading variability, or inadequate discrimination between arms. The preprint does not provide a clean in vivo resolution of the splicing question, so the strongest evidence remains the cell-based density and exposure experiments.
A one-sentence verdict is unavoidable here: the xenograft panel does not rescue a uniform model of metformin-driven ESR1 splicing.
Why this preprint changes the way the metformin signal should be read
For breast cancer biology, the practical implication is narrow but important. Metformin can still be linked to changes in ER-related splicing machinery, yet the direction of the effect depends on cellular context rather than a universal dose rule. In a proliferative system such as MCF-7 or in a resistant derivative such as TAMR1, cell density changes the transcriptional and splicing environment enough to alter the apparent drug response.
That kind of context sensitivity has direct implications for experimental design. Studies that compare only one seeding density, or that rely on a single time point, may misclassify a conditional effect as a consistent one. In breast cancer work focused on endocrine resistance, that distinction affects how confidently one can connect metformin to ER isoform balance, and whether a splicing mechanism should be pursued as a therapeutic vulnerability or regarded as an assay-dependent observation.
The most defensible conclusion from the preprint is also the simplest: metformin lowers HMGA1a and RBM3 protein in these breast cancer models, but HMGA1a is unlikely to explain the ESR1 splicing pattern, and the reported splice shift does not survive a high-density dose-response test. Before anyone treats this as a mechanistic basis for endocrine modulation, the next step has to be tighter validation of RBM3-linked splicing changes under controlled culture conditions, with cleaner tumor readouts than the current xenograft RT-PCR panels provide.