B (3:32)

Sure. Thank you, Anna. What I'd like to do is I want to talk about the reproductive components, the pathogenesis of them first, and start with the GNRH neurons in the hypothalamus. Now, GnRH is secreted in a pulsatile manner so that there's usually in typical cases without pcos, there's about one pulse per hour, and in pcos the pulse generator is more rapid. And you could wonder how you could differentiate the production of LH from fsh, since the gonadotrope makes both hormones and as it turns out, the gonadotrope makes an alpha subunit and a beta subunit for LH and an alpha subunit and beta subunit for fsh. And depending on the pulse frequency, there's preferential transcription of the beta subunit, so that the pulse frequency of GNRH is very important in modulating the release of LA ginfsh. Even more complicated, or more interesting perhaps, is that GNRH itself is regulated from above, from the arcuate nucleus and the rostral paraventricular area by neurons called the candy neurons. K N D Y stands for kisspeptin, dynorphin and neurokinin B, and those three neurotransmitters are responsible for GnRH secretion. Now, GnRH then stimulates the pituitary to make LHREvSH preferentially LH if it's a rapid pulse frequency, and that in turn causes sex steroids to be produced, there's a negative feedback by estrogen and testosterone at the levels of the pituitary and hypothalamus. But estrogen acts in a positive manner. And so it's quite complex, but diagrammatically it's relatively simple to understand. Clinically it's complicated in some ways but very easy in others. You mentioned the criteria that we use, and there are two sets of standard criteria, the so called nih criteria from 1990, 1992 and the Rotterdam criteria. And the major difference between the two is that Rotterdam introduced the ovarian ultrasound as an ancillary or an important additional test. I will talk about the first two components, the NIH definition, and then I'll talk about Rotterdam. NIH requires both ovulatory dysfunction, manifested as less than eight menstrual cycles per year, and hyperandrogenism, meaning clinical features of androgen excess such as hirsutism, acne, hair thin and hair loss and or hyperandrogenemia. And hyperandrogenemia usually comes in the form of testosterone or free testosterone, depending on the assay used and how reliable they are. Most often today it should be done, in fact, it really should be done by dual tandem mass spec, and those results are pretty reliable. Now the reason for the ultrasound is that there was interest in trying to subdivide the patient population with PCOS to determine if in fact they have polycystic ovaries. And I'll tell you the pluses and minus of doing so. The plus side is that when done properly and carefully and read by an expert radiologist, the test can be very useful, especially in indeterminate cases. But typically the ultrasound is read relatively casually and it's read to show multiple cysts. In fact, they're not cysts, they're antral follicles. And they're usually but not always lined up around the periphery of the ovary, making a sort of pearl necklace kind of appearance. But it can be over interpreted or under interpreted. And at least 20% of women without PCOS can be found to have radiographic criteria for polycystic ovaries. And about 20% of women with PCOS will not have polycystic ovaries. The volume of the ovary is important and the number of follicles. There are strict defining criteria for pco, morphology of the ovary. And so clinically, I think the approach that I've taken, and I think many have as well, is to see if the patient meets NIH criteria, I.e. ovulatory dysfunction and hyperandrogenism. And of course this is after excluding all other conditions that could mimic PCOS, such as 21 hydroxylase deficiency and hyperprolactinemia. Hypothyroidism and so on. But in trying to make a distinction as to whether someone has polycystic ovary syndrome, the ultrasound, if correctly done and properly read, can be very useful. But in most circumstances it's not absolutely necessary unless because there isn't a target intervention to change the number of follicles. So we rely mainly on these clinical markers.

C (9:10)

So the application of Rotterdam criteria have resulted in four phenotypes which have been designated phenotype A which is hyperandrogenism plus ovulatory dysfunction plus polycystic ovarian morphology, phenotype B, which is hyperandrogenism and ovulatory dysfunction, phenotype C, which is hyperandrogenism and polycystic ovarian morphology and phenotype D, which is ovulatory dysfunction and polycystic ovarian morphology. However, these diagnostic criteria have not been shown to identify biologically distinct phenotypes. In contrast, there seem to be three subtypes in cases with the NIH phenotype of ovulatory dysfunction and hyperandrogenism that are associated with unique genome wide significant loci. So there's the reproductive subtype where individuals seem to have a high LH level and high SHBG with low BMI and low insulin levels. There's a metabolic subtype where individuals have increased glucose and insulin levels with lower SHBG and LH levels. And then there's a background subtype where doesn't seem to be distinguishable patterns. David, do these subtypes come up clinically in the office in terms of how you counsel your patients?

B (10:25)

They do in the sense that the reproductive component and the metabolic component. In fact, what I believe is that the syndrome is misnamed, it's not properly named, and it should be called the reproductive metabolic disorder. I'm not the only one that feels that way, but it better reflects the constellation of metabolic and reproductive disturbances. And I talked about the reproductive pathogenesis, but the metabolic components are usually reflected by body mass index, that's very high, and obesity and hyperinsulinemia and sometimes hyperlipidemia as a consequence, and sometimes other effects that insulin has, such as on testosterone production by the ovary. So yes, I do think it's important to distinguish the phenotypes because women will want to know whether they're at risk for diabetes, whether their metabolic syndrome is present, and in some cases it is, in some cases it isn't. Now, I should point out that as you may have noticed, insulin fasting, insulin, for example, is not a diagnostic criterion for pcos, although many women have it measured. And we don't know what to do with that because BMI will directly correlate with insulin levels and it's not a dichotomous variable and it's a continuous variable. So the higher the bmi, the higher the insulin. It gets a little tricky. So I would say that it's useful, very useful, to look at these genetic pathways that are correlated with the reproductive and or metabolic features of the syndrome.

C (12:13)

So the purpose of this study was to investigate one whether these subtypes were present in more broadly ascertained PCOS cases using the Rotterdam criteria, 2 whether the subtypes had differences in additional PCOS related clinical variables not used for clustering, and 3 whether the differences aligned with distinct biologic pathways. And with that, I'll bring it back to Chase.

A (12:34)

Anna, thank you. And David's already given us a nice preview of some advantages of if we were able to understand better how we should be splitting PCOS up into different groups, we'll be thinking about how the authors try to make that identification and confirm that that is a good way of doing that. And then we'll come back to that in the discussion. We do want to spend a while here in the Methods thinking about how the authors approach this analysis and they use a relatively sophisticated analysis. Where we'll start with, though is just thinking about the basics of their study design before we get into the particulars of their statistical methodology. So the methodology here is a relatively straightforward one from a study design perspective, and that is just that they use a cross sectional study. We've looked at cross sectional studies plenty of times here in this podcast. Just as a reminder of what that is, a cross sectional study is taking a group of individuals and looking at all of their information at a single point in time. So while a study may go on for many, many years, and that is the case for this study, all of the data that's collected for each individual subject is viewed as having been collected at a single point in time. That has a lot of strengths to doing that because there isn't any longitudinal following of these individuals. You don't lose people over time. You can also be fairly extensive in what data you're collecting. One of the challenges, though, that goes with that is that it's not always easy which is the cause and which is the effect. As you analyze this study, you have to make some decisions and what you're going to label as a cause and what you're going to label as an effect. And in some evaluations that's easier to identify what is likely to be the case. But it is one of the downsides here. So we'll come to what the authors do with their stats here in a second. First of all, though, let's think about the patients. The authors report that their subjects were girls and women of European ancestry in their ages range 13 to 45. And where these individuals came from is they were all patients who attended clinic at the Reproductive Endocrinology and Fertility at the Erasmus University Medical center, Rotterdam between February 1993 and February 2021. And for how these individuals were diagnosed with PCOS that did evolve over time, we've heard a little bit already about different criteria that exist. So before 2003, the individuals had to have anovulation plus PCOS morphology and or hyperandrogenism. And then between 2003 and 2018 the Rotterdam criteria were used. And then after that, international guideline for pcos, though in general I would say that's the same as the Rotterdam. David already pointed out the importance of evaluating people for other conditions that can imitate pcos. And so these women had exactly that done. They were looked at for adrenal, pituitary and ovarian disorders, and those were ruled out. The subjects were excluded if they had used hormonal contraceptives within three months prior to the screening and or they were not fasting at the time of that screening. And all of the patients underwent a standardized assessment that included the collection of clinical variables, labs and imaging. Okay, so now for the statistics that were used here. The authors, as we already mentioned in the very title of this article, used a cluster analysis. And this is a sophisticated analysis. And the reason that you would do a cluster analysis is if you have a group of subjects and you would have reason to think that there would be distinct subgroups within that, but you don't have a reason to group them together already, or you don't have a good understanding of how they should be grouped together. What a cluster analysis allows you to do is that you can feed in a bunch of variables and that this analysis will help you identify if there are distinct subgroups based on how the different variables cluster together. So it's a good terminology that's used there. There are many different types of cluster analyses, many different decisions that have to be made in exactly how you're going to do that. We're not going to go into those details, but that's the purpose of doing a cluster analysis. And then if your cluster analysis has allowed you to find distinct Groups, you can then compare those groups once they have been identified through this analysis. The way that they started was using some age adjusted variables and they looked at BMI, testosterone, SH, DHEAs, LH, FSH, insulin and glucose. Those were the variables that they started with. And that analysis was initially performed on subjects who met the Rotterdam criteria and then in a subset who met the NIH criteria. And with that, they identified those three subtypes that Anna has already pointed out for us, and we'll mention them again because they're important ones to keep in mind. So the first one is that reproductive subtype, and those are women with higher LH and SHBG levels and lower BMI and insulin. The second subtype is the metabolic subtype, and those are women with higher glucose and insulin levels, but lower SHBG and lh. And then finally that background that we couldn't really identify a clear signal on them with any of those variables. Then after those subtypes were confirmed, these different subtypes were then compared using both different clinical variables. And we'll hold off on listing those. Anna's going to mention some of the important ones here in the results, so we'll let her review those for us. And then finally, once those three subtypes were identified and those different clinical variables were compared between those subtypes, they were compared with the distribution from the phenotypes in Rotterdam and NIH with those three subtypes. And again, Anna's going to go through those with us. So that's a high level overview of the stats that were used here. I'm now going to turn things back over to Anna and she's going to walk us through what the authors found with that cluster analysis and then how those different subtypes compare to one another.

C (18:12)

So in total, 2,510 girls and women with PCOS diagnosed using the Rotterdam criteria were included in the total cohort. Eight participants were excluded because of a glucose greater than 7 millimoles per liter or 126 milligrams per deciliter. Of the remaining 2,502 girls and women, 1,067 met NIH criteria. So investigation of the previously defined three NIH subtypes showed the following distributions. The metabolic subtype, high bmi, glucose and insulin levels with relatively low LH and SHBG levels was 41%, essentially 1,026 out of the 2,502 participants. And this was driven primarily by the BMI and insulin levels with the reproductive subtype, higher fsh, lh, shbg, with relatively low BMI and insulin levels was present in 18%, so 450 of the 2,502 participants. This was mostly driven by LH and SHBG, and the background subtype with no distinguishable pattern in the phenotypic trait distributions was 41%, or 1,026 out of the 2,502 patients. Age didn't differ between the three subtypes, and all other variables showed significant differences between all of the three subtypes except for DHEAS levels, which were significantly higher only in the metabolic subtype. The authors compared the clinical variables not used for clustering to determine whether the subtypes captured additional distinctive biologic features related to the reproductive or metabolic pathway. In the reproductive subtype, girls and women had higher AMH and total follicle counts compared to participants in the metabolic group, and participants in the metabolic group had higher triglyceride and LDL levels and lower HDL levels and higher blood pressures. David, is there anything surprising about this with the additional clinical variables?

B (20:08)

Well, yes and no. I mean I think that it in a sense confirms that there are subphenotypes within the PCOS realm. The overall diagnosis of pcos, which the cluster analysis initially separated reproductive and metabolic and the biomarkers that they used were consistent with those clusters. So it's not diagnostic and they didn't mean it to be diagnostic, but it's consistent with the reproductive versus the metabolic phenotype. And I think that the genetic studies which I think we'll talk about next, were very important in terms of trying to discover metabolic pathways that could be responsible for these findings.

C (20:58)

So the authors next assess which phenotype features used for the diagnosis of Rotterdam PCOS were captured by the subtypes. So the metabolic subtype predominantly had phenotype A and as a reminder that's ovulatory dysfunction plus hyperanargism plus polycystic ovaries on ultrasound. So that was 72%, while 9.4% had phenotype B, which is ovulatory dysfunction, hyperandrogenism, 4.2% had phenotype C, hyperandrogenism and polycystical raison imaging and 13% had phenotype D, which is ovulatory dysfunction and polycystic ovaries. On imaging in the reproductive subtype, 52.9% had phenotype A, 2.4% phenotype B, 2.2% phenotype C, and 41% had phenotype D. The background subtype had predominantly phenotype D that was 63.6%, while 26% had phenotype A, 4.6% phenotype B and 2.3% phenotype C. Additionally, after performing the cluster analysis, they divided the total cohort into classic NIH and non NIH Rotterdam based on the diagnostic criteria. Comparison of the two subsets showed higher prevalence of the metabolic subtype in the classic NIH subset compared to the non NIH Rotterdam subset, so that was 61.7% versus 18.3% which was significant, whereas in the non NIH Rotterdam subset the background subtype was the most prominent subtype, so that was 62.4% versus 20.8% which was also significant.

A (22:29)

All right, Anna, thank you. So now we'll move into the discussion and see how the authors took all this different information and tried to help us think through that. So we'll start with where the authors think about this and how they conclude some things here. So they state first of all that pcos is a heterogeneous condition and that the current diagnostic criteria, which are based on expert opinion, don't well capture this heterogeneity and they point that out by saying that different phenotypes are actually similar genetically. However, in previous work that they have done and published, a data driven approach using the analysis of phenotypic traits that identified those three subtypes that reproducible reproductive, metabolic and background subtypes in a European ancestry cohort of cases from the US moved this along and the reason that they said that is that's because those were identified with unique genetic loci, suggesting that they actually captured distinct biological causal pathways. And in this study in particular, this replicated those subtypes in a Dutch European ancestry PCOS cohort of girls and women that fulfilled the Rotterdam diagnostic criteria. To summarize those main findings that Anna just walked us through, the first one is that that metabolic subtype was comprised mostly of what we might think of as a classic PCOS phenotype, that phenotype A is identified for us, and that the reproductive subtype was mostly comprised of phenotypes A and D, that D being ovulatory dysfunction and PCO morphology. And finally that background subtype, the one without clear identifiable patterns, was mostly the phenotype D. A few other important findings and implications that we could take from this. David has already alluded to those, but the authors were explicit about how we could be thinking about this. First of all, in the reproductive subgroup, the higher lh, shbg, AMH and total FOLLICLE count all suggest that alterations in folliculogenesis may be present. However, in the metabolic subgroup, the higher bmi, glucose and insulin levels and the unfavorable lipid profile suggest an increased cardiovascular risk. And then another important finding that the authors point out, as Ann has mentioned already, is that age did not differ between the clusters, suggesting that reproductive and metabolic features are present from an early age. Okay, David, you gave us a hint already about how this could be helping us, but elaborate on that a little bit more. How might these findings help change our thinking about pcos if they do?

B (25:04)

Sure. So you mentioned that these findings may exist at an early age. That's one way in which they're important because a lot of these girls, when they're tested, may already have impaired glucose tolerance, they may be overweight, they may have other components of the syndrome that usually don't get focused upon. In other words, it's not just the reproductive component that is treated. It may be that the metabolic aspects of the disorder are already present from an early age. And even if it doesn't mandate treatment at that age, the surveillance for these metabolic disturbances should probably be made more frequently or more intensively. So that's one component. The other is that by looking at data, there are certain signals that suggest that the pathways that involve follicular, as you mentioned, follicular genesis are targeted more in the reproductive group and that the metabolic group has defects more in the pathways that affect metabolism, for instance. So I think this blitting approach, the phenotypes A, B, C, D and E I think get very confusing. So I would say that it's best, I mean, for me it's easier to remember that one of them is the NIH criteria. That's just hyperandrogenism and ovulatory dysfunction. And the Rotterdam includes the ovarian morphology and that alone is potentially important to distinguish one from the other.

A (26:58)

So we'll think about that a little bit more here in a minute. But first of all, we'll go through the strengths and weaknesses as identified by the authors. First of all, their strengths. They point out that this analysis had a large and thoroughly phenotyped cohort. They point out that they were able to explore additional traits and that finally that they were able to investigate the distribution of those Rotterdam based phenotypes. A few limitations that they list. First of all, they state that there is a possibility of a referral bias. I'll make an aside to point out with a referral bias is what that is and why that's A concern here is that a referral bias exists if you have a center that has a focus or that's a specialty center, is that you may not be getting a good representation of the general population. If you are taking subjects purely from your center, you may have only the hardest cases or the unusual cases or something that led them to being referred to your group. And so you may have to be a bit cautious as you think about trying to apply this to the general population that it may not be well represented. Okay, so that was the first of the author's limitations to go back to their list. They did point out that this was retrospective and as a cross sectional study, that this was also not longitudinal. They point out that their clusters were not validated with uncorrelated biomarkers. And then finally, something that they spent some time talking about earlier in their paper is that this cohort included only individuals, only women of European ancestry. David, we wanted your help thinking about this a little bit more. So they have a fairly homogenous population. So how big of a limitation is this? Knowing that there is a big genetic predisposition to PCOS is your area of expertise. And the authors also spend a while pointing out several other studies that have been done and other populations. And so does that reassure us a little bit more that these results may be broadly applicable?

B (28:52)

You're right. It's a homogeneous population and you can't really extend the findings to other populations. But I think those data are forthcoming.

A (29:03)

We'll now move into the author's conclusions and as a summary of their findings, the authors state that the three PCOS subtypes were replicated in this study with the identification of additional traits that differed between these subtypes, suggesting different etiologies and clinical characteristics. And the authors finish with what I would describe as an implication, and they state that clustering enables data driven diagnosis of PCOS and the identification of these subtypes will allow for precision medicine approaches. We'll come back to that implication shortly. But before we do that, Anna, let's start with you how you would think about the quality of this report overall.

C (29:43)

I thought this was a really helpful study. We already discussed that it's a very prevalent condition in young women. And I see these young women come to my office and I tell them that PCOS stands for polycystic ovarian syndrome and not polycystic ovarian disease, because, yes, it is this constellation of signs and symptoms. And my patients find it very frustrating that they have just some of these symptoms or all of These symptoms and how that affects management. And the typical triad of the treatment options that I offer to patients are typically estrogen containing birth control pill, metformin, spironolactone, myo inositol. And now having evidence that really helps us to better define these subtypes is helpful in terms of saying, can there be better precision medicine? So, yeah, I thought it was a great paper.

A (30:35)

David, let's finish up with you. Give us your thoughts. And I'm really interested as we're thinking about this, you've alluded to the benefits of using this type of splitting is that it may actually help us identify women who have the reproductive component, that we may be able to tailor an approach to them. So based on this, do you think we're there? So are we in a position to where we should adjust our screening based on what subtype you seem to have or perhaps our therapy or prevention that we're doing? Or is it still a bit premature for that? And we need to be waiting for more data to really modify what we've been doing for many decades in PCOS care.

B (31:12)

I think it's a bit premature, but it's important to recognize that these subphenotypes, they need to be replicated, but they're probably real and that individuals who have the metabolic traits need to be potentially treated or monitored or have surveillance for their development in the future, because it depends on what age these patients present. And I agree with what Anna said. PCOS is the syndrome. And I'm sure you've had this experience where you've had somebody come to you with morphologic evidence of PCO and they think they have this syndrome where they don't have a regular period or they don't have hyperandrogenia. And it's important to make sure that they meet at least the NIH criteria and or the Rotterdam criteria before making that diagnosis. I think it does a disservice to over diagnose this and also to under diagnose it. So to answer your question, yes, I think it's useful. Do I think we're at the point where we're going to be clustering our patients? I don't know quite yet that we're going to, but it hints at that.

A (32:30)

And with that, I would like to thank Anna Goldman and David Ehrman for joining me for this month's edition of Endocrine Feedback Loop. I hope that you all learned as much as I did and look forward to you all joining us again next month. And now you're in the loop. This has been Endocrine Feedback Loop Endocrine Feedback Loop is brought to you by the Endocrine Society with production oversight by Brandy Brown and Andrew Harmon. If you want to like and subscribe, you can find us on Apple, Spotify, or wherever you get your podcasts. We'd love to hear your feedback on this episode or the podcast itself. Please email us@podcastron.org Endocrine Feedback Loop is a free service of the Endocrine Society. To learn more or to become a member, visit the society's website at www.endocrine.org.