PCOS and Mendelian Randomization: Too Soon? Free

Polycystic ovary syndrome (PCOS) is a highly prevalent disorder affecting 5% to 15% of women worldwide, and is the single most common cause of hirsutism, a devastating problem for many, the most common cause of subfertility in Western countries, and, after family history and obesity, the most common risk factor for type 2 diabetes mellitus in premenopausal women. It is also associated with a number of obstetrical, cardiometabolic, vascular, mental, and malignant morbidities, with an economic burden conservatively estimated to be USD8 billion (2020) yearly in the United States alone (1).

Zhu and Goodarzi have examined the causes and consequences (ie, as the outcome or as the exposure, respectively) of PCOS by reviewing published Mendelian randomization (MR) studies (2). MR allows us to test whether an observed association between a marker or risk factor and a disease is actually causal or simply reflects “reverse causation” (ie, the effect of the disease on the marker) or confounding by other factors (eg, environmental conditions) (3,4). In some cases, causality could be better tested through improved control of cofounders in study design, improved measurement of and statistical adjustment for potential confounders, or through replication in different populations. However, in most situations testing for causality will require long-term randomized clinical trials, which often are neither practical nor ethical.

MR gets around these limitations by studying the relationship of a third factor (the genetic determinant of the marker) to the disease, since gene alleles are assumed to follow Mendel’s second law—the law of random assortment—and are allocated at random at conception. Hypothetically such an association would not be subject to either confounding or reverse causation. In this setting “gene” generally means gene variants (eg, polymorphisms, often detectable by single nucleotide polymorphisms, etc.).

It is important to note that the validity of MR studies will be negatively impacted if there are direct effects of the gene variant on the disease which are not mediated through the risk factor, or if there are other gene–environment or gene–gene interactions, or if population stratification is present (ie, if cases and controls are somehow genetically different, for example, due to clustering of genetically similar individuals within a population) (4).

Zhu and Goodarzi reported that MR studies so far suggest that obesity, testosterone, fasting insulin, sex hormone–binding globulin, menopause timing, male pattern balding, and depression may play a causal role in PCOS (2). Alternatively, they observed that PCOS may increase the risk of estrogen receptor positive breast cancer, decrease the risk of endometrioid ovarian cancer, and have no direct causal effect on type 2 diabetes, coronary heart disease, or stroke. Notwithstanding their interesting and provocative conclusions, the study by Zhu and Goodarzi (2) most importantly highlights how much we do not know.

Firstly, the study focuses on PCOS which is not a single “disease.” In fact, the term “syndrome” is still used for PCOS precisely because we are not yet sure whether it is 1 or many disorders. For example, there are now 4 phenotypes recognized in PCOS, which we have unimaginatively named Phenotypes A to D (5). While we are more or less certain that Phenotypes A and B (a.k.a. “classic PCOS”) seem to be the same disorder based on observational studies noting similar levels of morbidity, the same cannot be said of Phenotype D (a.k.a. “nonhyperandrogenic PCOS”).

The proportion of the different PCOS phenotypes differs according to whether PCOS is detected in the clinical setting or in an unselected (medically unbiased) population (6). Subjects detected in the clinical (referral) setting tend to be more obese, more hyperandrogenemic, and have a greater proportion of Phenotype A and lesser of Phenotypes B and C, although similar in the prevalence of Phenotype D (~20%). Many of the studies reviewed by Zhu and Goodarzi use data from previous genome-wide association studies (GWASs), which often recruited their subjects in the clinical setting or are self-diagnosed. Consequently, we can estimate that at least 60% of subjects included in PCOS GWASs are of a single phenotype: Phenotype A.

Secondly, MR depends on the use of genetic variants in genes of known function that reliably relate to a modifiable risk factor. MR will be most efficient when the connection between the genetic variables (instrument variables) and the risk factor is strong and the variants have little pleiotropy (ie, the potential to have more than 1 specific phenotypic effect). However, if the variants have multiple effects, at least 1 of which has a causal effect on the disease through some other pathway not involving the risk factor, or if the association between the variants and the disease is confounded by population stratification or by other variants it is in linkage disequilibrium with, then the estimated MR association between the markers and the disease will be distorted. Additionally, when investigating effects of functional polymorphisms, it is possible that these effects will be also distorted by environmental exposure.

Unfortunately, we still know very little regarding the action of susceptible variants, particularly as they relate to PCOS. And even less about their pleiotropy, or whether the gene in question acts directly on the disorder itself or are modified by other genes or the environment. Additionally, GWASs identify loci, not genes, with loci so far identified optimistically accounting for less than 20% of PCOS heritability. All which should lead us to question the validity of MR in PCOS.

Elucidating these questions will take investment and effort. Unfortunately, funding for PCOS research in the United States, despite its high prevalence, morbidity, and impact on quality of life, is much less than that of other reproductive or disorders of similar morbidity and, further, has been declining (7). In contrast to other disorders of similar morbidity, National Institutes of Health (NIH) funding for PCOS research is provided by a single Institute, the National Institute of Child Health and Development, although that Institute does not even mention the disorder in its current strategic plan. Moreover, PCOS is not currently listed in the NIH Research, Condition, and Disease Categories reporting (8). The limited scope, funding, and prioritization of PCOS research have left significant gaps in research—gaps that need to be urgently addressed.

Abbreviations

Abbreviations
 
• GWAS

  genome wide association study

 
• MR

  Mendelian randomization

 
• PCOS

  polycystic ovary syndrome

Additional Information

Disclosures: R.A. is consultant for Spruce Biosciences and Fortress Biotech.

Data Availability

Data sharing is not applicable to this article as no datasets were generated or analyzed during the current study.

References