PCOS Causes: What Mendelian Randomization Can—and Can’t—Show

PCOS has long looked like a metabolic disorder tangled with reproductive hormones, inflammation, body weight, and possibly the gut. A 2024 genetic analysis strengthens the case that some of those factors sit upstream of PCOS rather than merely traveling alongside it. The finding matters, but it does not hand anyone a new supplement protocol or prove that a single gut bacterium causes ovarian symptoms.

The study used Mendelian randomization (MR), a method that uses genetic variants associated with an exposure—such as fasting insulin or a microbial taxon—as proxies for lifelong differences in that exposure. Because people inherit these variants before PCOS develops, MR can reduce two familiar problems in observational research: reverse causation and confounding. It can sharpen causal hypotheses. It cannot make weak genetic instruments strong, erase genetic pleiotropy, or turn a risk association into treatment advice.

A broad genetic screen found metabolic, inflammatory, microbiome, and iron signals

In a 2024 two-sample MR study by Zhao et al., researchers screened more than 400 potentially modifiable exposures across eight domains against a PCOS genome-wide association study. The domains included lifestyle, socioeconomic factors, body measurements, metabolic traits, hormones, iron metabolism, inflammatory markers, and 196 gut-microbiota taxonomic groups.

The metabolic results were the least surprising: insulin resistance, higher BMI, and related adiposity measures emerged as putative causal risk factors. That fits clinical experience. Insulin resistance and excess central adiposity commonly cluster with PCOS, and a 2024 metabolomic MR analysis also supports a causal role for metabolic pathway disruption in female reproductive endocrine disorders that include PCOS.

The important restraint is that MR has trouble separating traits that travel together. Fasting insulin, BMI, waist-to-hip ratio, and HbA1c are tightly correlated. A univariable MR result for insulin can partly capture the effect of adiposity, and a result for adiposity can partly capture insulin-related biology. A review of PCOS MR research by Wang et al. explains that multivariable MR is needed to estimate their independent contributions. Existing multivariable analyses have produced inconsistent results. Metabolic dysfunction is central to PCOS; the evidence does not yet identify insulin resistance or central adiposity as the one primary cause independent of the other.

The gut signal is plausible, but the genus-level claims are fragile

Zhao et al. identified several microbial findings, including Oscillibacter and Alistipes as putative risk factors and Bifidobacteriaceae as putatively protective. An independent 2024 bidirectional MR study by Mao et al. also examined whether gut microbiota influence PCOS or whether PCOS changes the microbiota. Its direction-of-effect findings support the possibility that microbiota contribute to PCOS risk, rather than dysbiosis being entirely a downstream consequence of the condition.

There is a credible biological route. A 2024 review of gut microbes in PCOS describes lower microbial diversity, altered Bacteroidetes-to-Firmicutes patterns, and impaired short-chain-fatty-acid production in people with PCOS. Those changes could influence insulin sensitivity and systemic inflammation, both relevant to ovarian androgen regulation. Plausible is not the same as proven: the review does not establish that Oscillibacter, Alistipes, or any other MR-identified genus directly alters ovarian steroid production or luteinizing-hormone signaling.

Microbiome MR is especially prone to weak-instrument bias. Genetic variants usually explain very little variation in bacterial abundance; Wang et al. note that microbiome instruments can have F-statistics near or below the conventional threshold of 10 and explain less than 0.5% of the variance in abundance. When the genetic proxy is weak, a genus-level MR estimate can point in the wrong direction. The two MR studies increase confidence that the gut microbiome deserves serious study in PCOS. They do not validate a commercial plan to "lower Oscillibacter" or selectively raise one named taxon.

Iron may explain an observational paradox, not justify supplements

The iron result is more counterintuitive. Zhao et al. reported that genetically predicted higher serum iron and ferritin appeared protective against PCOS in their MR framework. That conflicts with a familiar observational pattern: in a 2021 cross-sectional study, ferritin correlated positively with insulin resistance and visceral adiposity index among women with PCOS.

The two designs ask different questions. The cross-sectional study measured ferritin after participants had PCOS and metabolic dysfunction, so elevated ferritin could reflect the disease state or its metabolic consequences. MR instead uses inherited differences related to baseline iron status, which is less vulnerable to reverse causation. A separate 2024 MR study of mineral metabolism and PCOS provides relevant methodological corroboration for studying iron markers this way.

The most economical interpretation is that high ferritin in clinical PCOS cohorts may often be a marker of metabolic dysfunction rather than an upstream driver of PCOS. But that remains a causal hypothesis, not an iron-treatment result. No randomized trial in this evidence base shows that raising iron or ferritin improves ovulation, hormones, insulin resistance, or PCOS symptoms. Iron can cause harm at excessive levels. Do not treat this genetic finding as a reason to take iron unless a clinician has identified iron deficiency and recommended replacement.

Why the study cannot settle causation on its own

The screen was ambitious, but breadth raises its own problem. Testing more than 400 exposures at a conventional 0.05 threshold would yield roughly 20 false positives by chance if no correction were applied. The available publication text does not allow independent verification of its false-discovery-rate approach, effect estimates, confidence intervals, p-values, or sensitivity-test outputs for every exposure.

Those missing details matter. MR depends on the exclusion restriction: the genetic variants should influence PCOS through the exposure being tested, rather than through another pathway. Metabolic and inflammatory variants may affect energy expenditure, body composition, glycemic control, and reproductive hormone signaling at the same time. MR-Egger, weighted-median, leave-one-out, and outlier analyses can probe some of those concerns, but they cannot guarantee a clean causal estimate—and the relevant outputs cannot be confirmed here for each headline result.

Population also matters. The exposure and outcome GWAS datasets were predominantly European ancestry. PCOS presentation, hormonal profiles, and microbiome composition differ across populations, so the estimates may not transfer cleanly to East Asian or South Asian groups. That limitation is especially relevant for microbiome findings, because microbial composition varies substantially with ancestry, geography, diet, and environment.

MR can improve the direction of the question—does this exposure plausibly contribute to PCOS?—without answering the clinical question of how much changing that exposure will help an individual patient.

What changes in practice?

For metabolic health, the practical priorities remain familiar because intervention trials, not MR alone, support them. Dietary patterns that improve energy balance and insulin sensitivity, resistance exercise, and approaches that reduce central adiposity can improve metabolic and endocrine outcomes in PCOS. An umbrella review of randomized-trial meta-analyses found evidence across nutritional interventions, although the strength and clinical importance varied by outcome and intervention.

Intermittent fasting has some recent trial-level synthesis behind it, but the evidence is still limited. A 2025 meta-analysis included five studies and reported a mean body-weight reduction of 4.25 kg, with a 95% confidence interval from 7 kg lower to 1.5 kg lower. That is a pooled anthropometric result, not proof that fasting prevents PCOS or that it suits every person. A ketogenic diet meta-analysis also reported improvements in weight, metabolic markers, and endocrine parameters, but diet choice should account for preferences, medical history, sustainability, and medications.

Microbiome-directed treatment is reasonable as an adjunct, not a precision-microbe prescription. A 2024 systematic review of randomized trials evaluated probiotics, prebiotics, and synbiotics for insulin resistance and hormonal imbalance in women with PCOS. A 2025 randomized trial tested Lactobacillus helveticus plus Bifidobacterium longum for eight weeks. These studies make microbiome modulation clinically testable; they do not establish that every probiotic works, that benefits persist, or that a product can reproduce a genetic association with Bifidobacteriaceae.

Polyphenols, including curcumin, also have intervention evidence for metabolic, inflammatory, and hormonal markers. A systematic review and meta-analysis of curcumin trials reported benefits for insulin resistance, lipid profiles, and hormonal markers. These are useful signals, but they are largely biomarker and metabolic outcomes rather than proof of improved fertility, pregnancy outcomes, or long-term disease prevention.

The verdict

This MR landscape supports a more causal view of PCOS as a metabolically addressable disorder and gives the gut microbiome a credible place in the research agenda. Insulin-related biology, adiposity, and inflammation have the strongest practical relevance, even though their independent causal weights remain unresolved. The microbiome findings merit replication with stronger genetic instruments. The iron signal is an explanation to investigate, not a supplement recommendation. For patients, the evidence still points to lifestyle and medical care built around insulin sensitivity, cardiometabolic risk, symptoms, and personal goals—not an assay-driven hunt for a single bacterium or a bottle of iron.