A Tiered Approach to Exome Sequencing Analysis in Early-Onset Primary Ovarian Insufficiency

A total of 199 women with EO-POI or familial POI attending UCLH met the inclusion criteria. Of these, 23 had syndromic POI (eg, Perrault syndrome, galactosemia) and were excluded. Of the remaining 176 women, a total of 149 women with POI were recruited to this study as part of the UCLH Reproductive Life Course Project (RLCP, Fig. 2 ) with informed consent. Of the entire cohort recruited (n = 149), 79.2% (n = 118) had sporadic, nonfamilial POI. The remainder of women, 20.8% (n = 31 individuals), had familial POI (ie, an affected sibling), coming from a total of 17 different kindreds of which 44.4% (n = 8) were consanguineous ( Fig. 2 ). Of the total cohort, 96.0% (n = 143) had EO-POI as defined by onset of POI before the age of 25 years (the other 6 being siblings in familial POI kindred) and 81.2% (n = 121) presented with primary amenorrhea. Of those with primary amenorrhea, 76.0% (n = 92) presented with complete absence of puberty or premenarchal pubertal arrest; the remainder presented with early-onset secondary amenorrhea (POI presenting with absent menstrual cycles <25 years after a short period of normal cycling) or familial POI ( Fig. 2 ). Unaffected family members were also recruited where possible (n = 37), with a focus on recruiting unaffected family members of women with familial POI (ie, 2 or more affected first- or second-degree family members with POI in the 1 kindred) (Table S3 ( 28 )). However, parental DNA was not available in many cases. Flow diagram of POI exome study recruitment. A total of 199 with POI met the inclusion criteria, of which 23 were subsequently excluded due to known syndromic causes of POI (eg, galactosemia, Perrault syndrome). Of the remaining 176 women, 149 with POI were recruited. A further 37 unaffected relatives were recruited, meaning 186 individuals were sequenced in total. Participants were considered in 2 main groups: familial POI and sporadic EO-POI (POI <25 years). EO-POI, early-onset primary ovarian insufficiency. Of the 118 women with sporadic EO-POI, 84.8% (n = 100) had primary amenorrhea and 62.7% (n = 74) had pubertal delay/arrest ( Fig. 2 ). A small proportion (15.3%, n = 18) had transient menstrual cycles as young girls (mean age of menarche 11.5 years [IQR = 10.5-13.0]; mean age of last reproductive cycle 13 years [IQR = 15.5-20.5]). For those with familial POI, inclusion criteria were not restricted to EO-POI. However, most women within this group presented with primary amenorrhea (67.4%, n = 21) and for those secondary amenorrhea (32.3%, n = 10), mean age of menarche was 13.1 years (IQR = 12.1-14.9) and mean age of last menstrual period was 22.7 years (IQR = 16.0-27.0). Only 6 women with familial POI had non-EO-POI. Notably, across the entire cohort of 149 women, none had a positive FMR1 premutation (Fragile X) screen and only 1 woman had positive ovarian and adrenal antibodies, despite all participants being tested (to note, as positive antibodies did not rule out a participant also having a genetic mechanism for POI identified, women with positive antibodies was still included for genetic analysis). Other clinical characteristics are outlined in Table 1 . Variant filtering followed processes outlined in “Materials and Methods” and Fig. 1 selecting only rare (<0.01% gnomAD 4.0)/novel variants that are predicted pathogenic in multiple in silico models. As described in “Materials and Methods,” variants were segregated into Category 1 (variant in PanelApp POI gene, n = 69); Category 2 (variant in gene associated with POI in the literature but not included on PanelApp, totaling 355 genes as listed in Table S2 ( 28 )) (or, PanelApp variant following an unexpected inheritance pattern); and Category 3 (forward screen for homozygous variants in novel candidate POI gene). Of 17 kindred with familial POI, 11 kindred (64.7%), including 20 affected women, had a genetic variant identified in a Category 1 or 2 gene ( Table 2 and Fig. 3A and 3B ; Table S4 ( 28 )). Of these 17 kindred, 6 kindred (35.3%) had variants in a Category 1 gene: homozygous variants in STAG3 , MCM9 , PSMC3IP , YTHDC2 , and ZSWIM7 ; and a heterozygous variant in POLR2C ( 32 ) (note: the YTHDC2 and ZSWIM7 variants have been published separately ( 30 , 31 )) ( Table 2 ). A further 5 families (29.4%) had Category 2 variant(s), including 3 kindred with single heterozygous variants in NLRP11 ( 33 ), IGSF10 ( 4 ), and PRKD1 ( 4 ); another kindred with compound heterozygous variants in PLEC ( 34 ); and a further family with oligogenic variants in PDE3A ( 9 ), POLR3H ( 35 ), MSH6 ( 36 ), and CLPP ( 37 ) (Table S4 ( 28 )). Notably, the participant with the Category 1 POLR2C variant also had heterozygous variants within Category 2 genes. The remaining 6 families (35.3%) had no genetic finding identified. Only 1 of the 8 consanguineous pedigrees did not have a genetic finding identified. Variants identified in POI cohort. (A) Variant by category across the cohort. Proportion of individuals/kindred (women from the same kindred considered as 1 entity for the purpose of this figure) across the sporadic and familial POI cohort with Category 1, Category 2, and Category 3 variants or combinations thereof. (B) Variant by probable mode of inheritance across the cohort. Proportion of individuals/kindred across the sporadic and familial POI cohort with single rare heterozygous variants, homozygous variants, likely compound heterozygous variants (note: parental DNA was often not available to confirm variants in trans), and variants in 2 or more different genes (likely polygenic). Only Category 1 or 2 variants are shown here. Those individuals without any variant identified or those with Category 3 variants (candidate homozygous variants) are not included in this figure. POI, primary ovarian insufficiency. Variants detected in the familial primary ovarian insufficiency cohort FPOI12 , FPOI13 , FPOI14 , FPOI15 , FPOI16 , FPOI17 no finding identified. Probable compound heterozygote; parental DNA not available for testing. Population frequency: gnomAD 4.0 allele frequency. Abbreviations: FPOI, familial POI; AR, homozygous variant, autosomal recessive inheritance; A D, heterozygous variant, autosomal dominant inheritance; CH, likely compound heterozygous inheritance. a Consanguineous family. b Category 2 gene. c Previously published in McGlacken-Byrne et al ( JCEM , 2022) ( 30 ). d Previously published in McGlacken-Byrne et al ( JCI Insight , 2022) ( 31 ). Of women with sporadic POI, 21.2% (n = 25/118) had a Category 1 variant(s) (n = 8 homozygous; n = 3 with 2 different variants in same gene (possible compound heterozygous); n = 14 single heterozygous variants) ( Table 3 , Table S5 ( 28 )). Genes included 12 green PanelApp genes; 4 orange PanelApp genes; and 4 red PanelApp genes. These 25 women were then screened for Category 2 variants, and 9 had 1 or more Category 2 variants in addition to the Category 1 variant(s), inferring possible oligogenic inheritance in these women ( Fig. 3A ; Table S5 ( 28 )). The remaining cohort of women with sporadic POI were then screened for Category 2 variants, which were found in a further 42.4% (n = 50/118 further women; 16 digenic/polygenic; 34 heterozygous; none biallelic). When both Category 1 and 2 variants were considered, 75 women (63.6%) had at least 1 Category 1 or 2 variant. In total, Category 1 or 2 variants in 2 or more genes were identified in 26/118 women with sporadic POI (22.0%), suggesting possible oligogenic or polygenic inheritance in this proportion. Variants detected in the sporadic primary ovarian insufficiency cohort Population frequency: gnomAD 4.0 allele frequency. SPOI78-118 had no finding identified. Probable compound heterozygote; parental DNA not available for testing. Abbreviations: AR, homozygous variant, autosomal recessive inheritance; AD, heterozygous variant, autosomal dominant inheritance; CH, likely compound heterozygous inheritance; SPOI, sporadic primary ovarian insufficiency. a Category 2 gene. b Category 3 gene. c Previously published in McGlacken-Byrne et al (JCI Insight, 2022 ( 31 )). d SPOI27: POI and situs inversus; SPOI77: POI and rod–cone retinal dystrophy; SPOI78: POI and chronic fatigue. e Consanguineous family. To identify novel genetic causes of POI, the cohort was then screened for Category 3 homozygous variants. Category 3 variants possibly associated with the POI phenotype in existing model systems/animal models were identified in 7 women ( Table 4 ; Table S5 ( 28 )). Of the 75 women with Category 1 or 2 variants, 5 women had 1 or more Category 3 variants (in PCIF1 , MMS22L , DND1 , MEF2A , RXFP3 ). Of the remainder without Category 1 or 2 variants, a further 2 women had 1 or more Category 3 variants potentially explaining their phenotype in 2 POI candidates (C4ORF33 and ARRB1 ). Homozygous variants in PDK1L1 and ABCA4 were also identified in 2 women (the woman with the DND1 variant and the C4orf33 variant, respectively), but in these cases these variants likely explained other aspects of the participant's clinical presentation rather than POI as both variants have been definitively associated with these other phenotypes (situs inversus and rod–cone retinal dystrophy, respectively). Proposed function and supportive evidence for identified variants Abbreviation: SPOI, sporadic primary ovarian insufficiency. Taking Category 1 and 2 variants together across both familial and sporadic cohorts, a total of 128 variants in 74 different genes previously associated with POI were identified in 63.1% (n = 95) women in the cohort studied (11 kindreds composed of 20 recruited individuals with familial POI, and 75 individuals with sporadic POI, as described above) ( Figs. 3A and 4 ; Table S6 ( 28 )). Considering women from the same kindred as 1 entity, a total of 86 individual women/kindred across the combined familial/sporadic cohort had Category 1 or 2 variants identified: likely oligogenic/polygenic variants in 32.2% (n = 28/86); homozygous variants in 11.5% (n = 10/86); presumed compound heterozygous variants in 3.4% (n = 3/86; parents not available for testing for these individuals, so variants could not be proven in trans); and single rare heterozygous variants in 52.9% (n = 46/86) ( Fig. 3B ). Notably, of these 46 heterozygous variants, although individually rare (<0.01% minor allele frequency [MAF]), 42 (91.3%) are also found in the heterozygote state in normal population controls (gnomAD 4.0) (Tables S4 and S5 ( 28 )). In addition to Category 1 and/or 2 variants, a total of 11 women carried homozygous variants in 13 candidate genes not previously associated with POI, including 6 with variants also identified from Category 1 or 2 genes ( Fig. 3B ; Tables S4 and 5 ( 28 )). Category 1 and 2 variants across the cohort. The 74 genes in which Category 1 or 2 variants (n = 128) were identified in this study cohort are shown (x-axis). The number of different variants in each identified POI gene are shown (y-axis). Individual variants are represented once in the graph, even if present in 2 or more individuals in the cohort. Compound heterozygous variants are included as 2 different variants (n = 2). Homozygous variants are included once (n = 1). Results were then compared between the familial and sporadic POI groups. More homozygous variants were found in familial POI than in sporadic (35.3% [n = 6/17] compared with 4.2% [n = 5/118]) (Fisher exact test, P < .001) and a nonstatistically significant trend towards more heterozygous variants found in sporadic POI compared with familial (36.4% [n = 43] compared with 17.6% [n = 3]) ( Fig. 3A ). Additionally, more Category 1 variants were found in the familial cohort than in the sporadic cohort (29.4% [5/17] compared with 13.6% [16/118]) ( Fig. 3B ). Proposed biological functions of identified genes as they relate to ovarian function (if known), as well as selected studies in which other variants in these genes have been described in POI, are listed in Table S7 ( 28 ). These processes span an entire reproductive life course and included ovary differentiation and development; primordial germ cell migration and growth, oogonia proliferation, meiosis (particularly prophase I in fetal life); folliculogenesis and maturation; maintenance of the resting follicle pool; and, ultimately, ovulation to produce an oocyte capable of fertilization ( Fig. 5 ). Four or more different variants were described in STAG3 , SETX , PATL2 , MLH1 , MCM9 , GDF9 , FSHR , BRCA2 , BMPR2 , and BMP15 ( Fig. 4 ). For most genes, only 1 variant within it was described. Most variants described were missense variants (90.7%, n = 117/128) with a small proportion (9.3%, n = 12/128) frameshift or stop gain loss of function variants. Of the 11 women carrying homozygous Category 3 variants in new POI candidate genes, 3 were loss of function variants (stop gain). Mapping POI-associated genes to key ovarian biological processes over a reproductive life course. POI-associated genes in which variants were identified in this cohort are mapped to key individual biological and cellular processes in fetal and adult ovarian development. PGC, primordial germ cell; dsDNA, double-stranded DNA. This figure was created in part on BioRender.com . Transformative figure; from McGlacken-Byrne et al JCEM ( 30 ). © The authors.

Here, we describe rare genetic variants in POI-associated genes in a high proportion of women with EO-POI and/or familial POI. The real-life cohort presented is mixed ancestry and uniquely enriched for early-onset, nonsyndromic POI; to our knowledge, it is the largest cohort of EO-POI described to date ( 19 ). Additionally, the very high recruitment rate reduces the risk of negative selection bias that might have confounded results. Our “detection rate” is overall similar to those reported by other exome sequencing studies over the last 5 years, many of which also identified variants in POI-associated genes in over half of the POI cohort examined ( 5 , 6 , 9 , 13 ). This adds weight to the emerging concept that a greater proportion of POI has a genetic basis than previously thought. We specifically advance our understanding of the genetic landscape of EO-POI. Importantly too, our judicious, evidence-based approach to exome sequencing analysis acknowledges that association does not necessarily imply causality. Key insights and reasons for caution are outlined below. Firstly, our data provide insight into the contribution of heterozygous variants to the POI phenotype. In both familial and sporadic POI, we found a high proportion of both rare/novel, predicted deleterious single heterozygous and digenic/oligogenic variants in genes from a carefully curated gene list based on previously proposed POI genes. Our variant identification rate is similar to or higher than previous studies sequencing mixed POI cohorts ( 5 , 6 , 9 , 13 ). Although our study focused on EO-POI, 1 large, published cohort described an EO-POI subset, and, similar to our “Category 1” detection rate in our data, found that 25.8% of women within the EO-POI group had a pathogenic or likely pathogenic variant in a POI-associated gene. Taken together, these studies suggest that at least a proportion of POI is oligogenic in origin. At face value, the high proportion of women with POI carrying rare heterozygote variants in POI-associated genes also suggests an inheritance model whereby haploinsufficiency of key POI-associated genes gives rise to clinical POI. However, this enrichment of rare heterozygous variants in POI-associated genes within POI cohorts needs to be considered alongside the presence or absence of these variants in population controls; the exponential rise in the number of genes associated with POI in recent years; and the fact that the strength of evidence underlying these individual gene–disease relationships significantly varies in the first instance. Some gene–disease relationships in POI are well-established, having been replicated in several cohorts and/or supported by functional evidence and convincing animal models. Many of these well-established genes are included on the “Green” Genomics England PanelApp (eg, MCM9 and STAG3 ) ( 38 , 39 ). Other gene–disease relationships in POI are less well established, having been described in only single cases or small cohorts with limited or no supporting functional evidence linking proposed loss of function variants within these genes to POI. Furthermore, in recent years, variants in several POI genes that classically follow recessive modes of inheritance (eg, MCM8 , BRCA2 , MCM9 ) have been reported in POI cohorts in heterozygous states, sometimes without convincing evidence for the contribution of haploinsufficiency to the POI phenotype ( 23 ). With these issues in mind, we used a hierarchical and categorical approach to variant filtering in this study which clearly highlighted the different tiers of evidence supporting gene–disease relationships in POI and allowed results to be divided into higher evidence (Category 1) and lower evidence (Category 2 and 3). The step of searching for individual heterozygous variants in population controls also revealed that most single heterozygous variants identified, although individually rare (MAF < 0.01%), were also found in presumed healthy gnomAD 4.0 controls. This echoes findings from a recent paper which evaluated the penetrance of heterozygous variants in POI-associated genes in both POI and control cohorts and found limited evidence for the pathogenicity of these variants in a heterozygote state ( 40 ). This does not mean that these identified variants have no possible relationship to POI but does call into question whether these single heterozygous variants can give rise to a POI phenotype in isolation. It may be that carrying multiple heterozygous variants in different genes contributes to a POI phenotype in a complex trait rather than Mendelian inheritance pattern (and that our stringent filtering approach missed other potential variants of significance in our cohort). The high rate of polygenicity in POI-associated genes suggested by this study would support this concept. Secondly, we propose that a clearly defined subgroup of women have autosomal recessively inherited—truly “monogenic”—POI. We identified a significant proportion of women—approximately 10%—carrying homozygous variants in Category 1 genes in our cohort, many of which have established pathogenicity in POI when inherited in an autosomal recessive manner (eg, MCM8 , MCM9 , STAG3 , and PSMC3IP ). Unsurprisingly, consanguineous kindred had a high incidence of autosomal recessive POI; all but 1 had a homozygous genetic variant identified. This figure of 10% is higher than previously reported studies, including those with EO-POI mixed within their cohorts ( 19 , 33 ). This suggests that women with EO-POI are more likely to carry recessive variants in POI associated genes than those with later-onset POI. Many of these genes have clearly defined functions in human ovarian development and function (eg, DNA repair, meiosis). It may be that recessively inherited, EO-POI represents 1 end of a genotype–phenotype spectrum, with POI as complex polygenic trait more commonly defining POI occurring at a later age. Potentially, carrying an increasing number of these polygenic pathogenic variants could translate to a more severe POI phenotype. For a further group of women, POI may represent a complex trait phenotype lying just outside the continuum of normal age at natural, nonpathological menopause reflecting a burden of common single-nucleotide polymorphisms associated with earlier menopausal age. It is possible too that certain common single-nucleotide polymorphisms may confer genetic vulnerability to gene–environment effects that increase an individual risk of developing POI or physiological menopause at an earlier than expected age. Ultimately, larger and POI-specific genome-wide association studies and exome-wide rare variant association studies will be required to definitively answer these questions. Thirdly, we advance our understanding of recessively mediated POI by proposing some convincing novel candidate genes for POI worthy of further exploration. Our gene-agnostic approach of searching for recessively inherited variants across the cohort resulted in the identification of promising novel candidate genes for POI. Indeed, the evidence for some of these candidate homozygous variants is possibly stronger than the evidence underpinning some of heterozygous variants described in Category 2 POI-associated genes in this study. We discuss these novel candidate genes below. A subset of candidate genes for POI identified in this study—including PCIF1 , DND1 , and MEF2A— have convincing corresponding animal models of gonadal insufficiency and therefore warrant further investigation in humans. A homozygous variant in PCIF1 , a methyltransferase of N6,2-O-dimethyladenosine ( 41 ), found in a woman with sporadic POI adds weight to the emerging view that complex epitranscriptomic regulatory networks are key for normal ovary development ( 42 ). Furthermore, Pcif1 mutant flies have a reduced fertility phenotype that is exaggerated in females ( 43 ). DND1 homozygous variants have been identified in males with nonobstructive azoospermia ( 44 ) and a novel stop–gain homozygous variant was identified here in a woman with sporadic POI and situs inversus from a consanguineous family. Mouse models have demonstrated that DND1-dependent mRNA destabilization is required for primordial germ cell survival in mice ( 45 , 46 ); we suggest a role for it in female ovarian function in humans. Notably, this woman also had a homozygous variant in PKD1L1 which likely independently explains the situs inversus phenotype ( 47 , 48 ). We also identified a recessive variant in MEF2A , a gene with established myogenic functions, suggesting a possible role for it in human female ovarian function. Mef2 has been shown to have a role in normal germ cell development in mouse follicular cells, with Mef2 females only partially fertile ( 49 ). Specifically, it has been proposed to regulate expression of the nuclear receptor Nr4a1 in mice ( 50 ). Pathogenic variants in genes involved in DNA repair (including BRCA2 , ZSWIM7 , MCM9) are recognized causes of POI, and we propose 2 further DNA repair candidates here. Firstly, a homozygous variant was found in MMS22L , which together with RAD51 is required for functioning homologous recombination. RAD51 has been previously implicated in the pathogenesis of POI ( 36 , 51 ); MMS22L represents a plausible new POI candidate gene. RXFP3 is another DNA repair gene known to regulate the DNA damage response pathway via its ligand, relaxin 3 ( 52 ). Several relaxin genes, including RXFP3 , are expressed in male reproductive organs and during folliculogenesis in mice ( 53 , 54 ); it may therefore have an unexplored role in human gonadal function. The remaining 2 homozygous Category 3 variants identified in this study, C4orf33 and ARRB1 , do not have a recognized role in gonadal development but warrant further exploration. The woman with the C4orf33 variant is from a consanguineous family and has both POI and rod–cone retinal dystrophy with 2 separate homozygous variants identified: a missense variant in C4orf33 and a frameshift variant in ABCA4. The latter is an established cause of rod–cone dystrophy and likely explains her ophthalmological phenotype ( 55 ). This highlights that, while some “nonsyndromic” genetically mediated POI is associated with other features arising from the same genetic variant (eg, hypopituitarism and POI resulting from pathogenic RNPC3 variants) ( 56 ), in some women—particularly from consanguineous families—there can be 2 separate recessive variants explaining complex phenotypes. The other variant identified in this woman, C4orf33 , is highly expressed in the testis and epididymis in males in the Human Protein Atlas but has no known function in the ovary. The remaining Category 3 gene identified in this study, ARRB1 (encoding B-arrestin), functions in mice as part of the Hippo signaling pathway, interacting with YAP to modulate downstream transcription of YAP responsive genes ( 57 ). YAP1 expression is required for normal human ovarian development, important for primordial follicle growth and activation ( 58 ). Additionally, the NF2–YAP pathway has been associated with POI previously, with variants in BNC1 resulting in an ovarian insufficiency phenotype ( 59 ). Taken together, our work advances our understanding of the genetic architecture of POI. We identify several avenues for further exploration as well as reasons for caution. Like others, we found genetic variants in genes previously associated with POI in a high proportion of this POI cohort. While this does support the concept that a significant fraction of POI has an underlying genetic contribution, we caution that a genetic finding does not equal a genetic diagnosis. Specifically, the strength of evidence supporting the pathogenicity of each identified variant, and the evidence for its proposed model of inheritance, needs to be considered. For those variants with limited functional evidence (here named Category 2), we suggest there is a need to revisit gene–disease relationships in the first instance. We also propose that a proportion of POI is polygenic in origin, possibly constituting the severe end of a complex trait phenotypic spectrum of age at natural menopause. We also conclude that a distinct subset of EO-POI is truly monogenic, particularly in the context of consanguinity, and arises frequently from homozygous variants in key, established POI-associated genes. These genes frequently have functions related to pivotal ovary developmental pathways such as meiosis and DNA repair; indeed, pathogenic variants within these genes were among the first identified in sequencing studies of consanguineous POI pedigrees in the early days of genetic testing for POI. The complexity and uncertainty around genetic testing in POI has clinical consequences: for instance, beyond a recommendation to perform a karyotype and Fragile X screen, there is no publicly available targeted gene panel for POI in the UK (ie, as part of the NHS National Genomic Test Directory). Clearly, there is a pressing need to elucidate the genetic architecture of POI more fully to guide which variants should be routinely returned to women with POI as a genetic diagnosis and which genes should (or should not) be included on panels available for routine clinical genetic testing in the future. At the same time, exploring in more detail the novel candidate genes identified in this study would accelerate our understanding of human ovarian function and, in time, may provide a convincing genetic diagnosis to a greater proportion of women with POI.

Women and girls with POI over the age of 16 years and their unaffected family members were invited to participate in the Reproductive Life Course Project (RLCP) at University College London Hospital (UCLH). All participants who joined the RLCP gave written, informed consent (if >18 years) and assent (if <18 years, along with cosigned parental consent) and continued to receive usual clinical care for POI at the Reproductive Medicine Unit of UCLH. The RLCP has ethical approval from the NRES Committee London-Chelsea (15/0877) and is sponsored by UCLH. Inclusion eligibility criteria included: (1) a confirmed POI diagnosis in accordance with the European Society of Human Reproduction and Embryology (ESHRE) guidelines (4 months, estrogen deficiency, and raised follicle-stimulating hormone >40 IU/L on 2 occasions at least 1 month apart) ( 20 ); (2) EO-POI (primary amenorrhea with or without absent puberty/pubertal arrest) and/or familial POI (2 or more first-degree or second-degree relatives within the same family with POI); (3) no underlying cause for POI identified (eg, iatrogenic POI); and (4) a 46,XX karyotype. All women underwent a Fragile X screen ( FMR1 analysis for CGG repeats). Women with a known genetic diagnosis of an established clinical syndrome definitively associated with POI were excluded (eg, confirmed Perrault syndrome). Otherwise, extra-ovarian clinical characteristics were recorded ( Table 1 ). Unaffected family members of patients meeting the criteria above were also recruited with informed consent, if appropriate. The current and historic medical records at UCLH of all patients recruited to the RLCP were examined in detail. Data collected included demographic details, family history, details of diagnosis, current clinical status, and investigations performed to date. Clinical characteristics of the cohort Abbreviations: FSH, follicle-stimulating hormone; HRT, hormone replacement therapy; LH, luteinizing hormone; TPO, thyroid peroxidase antibodies. Participants provided EDTA blood samples. DNA was extracted from whole blood using the QIAamp DNA Blood Midi/Maxi kit (QIAGEN N.V.) as per protocol. An Exome CG enrichment panel (Nonacus, Birmingham, UK) was used for library preparation. Precapture library preparation and enrichment were performed following the Nonacus Library Prep v2 (HT) and Cell3 Target Enrichment protocol for Next Generation Sequencing protocols (Nonacus, Birmingham, UK). Processes were automated on the Hamilton StarLet robot (Hamilton, Reno. NV, USA). Library qualitative checks were undertaken using a Tapestation 4200 platform (Agilent, CA, USA). Libraries were sequenced on the NovaSeq 6000 (S4, 2 × 151 bp, Illumina) using paired-end sequencing and a S4 flowcell (UCL Genomics). Reads were aligned against the human reference genome sequence (NCBI, GRCh38) using the Burrows-Wheeler Aligner (BWA-MEM) ( 21 ). Platypus software (v0.8.1) was used for variant calling using standard parameters ( 22 ). Variant filtering was performed using QIAGEN Clinical Insight Interpret software (QIAGEN N.V. 2024). Variants with call quality ≥20 and a read depth ≥10 (median read depth 75; IQR 46-138) were kept. Unless the variant a well-established relationship with the pathogenesis of POI, filtering only retained variants that were (1) predicted pathogenic in silico on at least 2 of 3 pathogenicity prediction tools (Combined Annotation Dependent Depletion score >15 ( 23 ); PolyPhen2 ( 24 ); and SIFT ( 25 )); (2) rare/novel (global minor allele frequency in gnomAD v4 <0.01% for biallelic variants (compound heterozygous/autosomal recessive); <0.005% for polygenic and single heterozygous variants; adjusted P < .0001); (3) enriched in the POI cohort compared with controls (gnomAD v4; 2-tailed Fisher exact testing); and (4) classified as pathogenic or likely pathogenic as per the American College of Medical Genetics (ACMG) criteria ( 26 ). All synonymous changes were excluded, unless they were associated with splice site loss up to 7 bases into an intron or predicted to affect splicing using MaxEntScan ( 27 ). Downstream filtering then considered 3 categories of genes/variants. The entire cohort was initially screened for Category 1 variants. Category 1 variants included those within genes included on the Genomics England Primary Ovarian Insufficiency PanelApp (January 2024, v1.67; n = 69 genes; https://panelapp.genomicsengland.co.uk/panels/155/ ; Table S1 ( 28 )), including those in the “Green” (n = 31 genes), “Amber” (n = 22 genes), and “Red” (n = 16 genes) categories. This is a virtual gene panel which was used in the UK 100 000 Genomes Project (a project where 100 000 genomes from 85 000 individuals accessing health care on the National Health Service [NHS] were sequenced), which uses expert crowdsourcing to assign genes associated with presumed monogenic phenotypes into 3 tiers (“Green”: high evidence for established pathogenicity in POI; “Amber”: emerging POI gene for which there is currently moderate evidence; “Red”: emerging POI gene for which there is currently low evidence). The entire cohort was then screened for a manually curated list of Category 2 variants (including in individuals carrying a Category 1 variant). These variants were those within genes associated with POI in published research studies not included on the POI PanelApp (n = 355 genes; supportive studies listed in Table S2 ( 28 )) or variants in Category 1 genes included on the POI PanelApp but following different inheritance patterns to what is expected or established for that gene (eg, heterozygous variants in genes which usually follow recessive inheritance patterns, Table S1 ( 28 )). Given that biallelic variants in key ovary-related genes are established causes of POI, the cohort was then screened for Category 3 variants, which were defined as homozygous variants in any gene meeting the above-described filtering criteria not included in Category 1 or 2, with the aim of identifying new candidate genes for POI ( Fig. 1 ). Variant filtering approach: Categories 1-3 were derived from the virtual POI gene panel used in the UK 100 000 genomes project on the genomics England primary ovarian insufficiency PanelApp (Jan 2024, v1.67; n = 69 genes; https://panelapp.genomicsengland.co.uk/panels/155/ ; Table S1 ( 28 )); “Green” (high evidence for pathogenicity, n = 31 genes), “Amber” (emerging POI gene with moderate evidence, n = 22 genes), “Red” (emerging POI gene with low evidence, n = 16 genes) categories. POI, primary ovarian insufficiency; EO-POI, early-onset POI; MAF, minor allele frequency. Statistical analysis (Fisher exact testing, 2-tailed) was performed using GraphPad Prism v9.1.1 (GraphPad Software). A P value of less than .05 was considered statistically significant. The Benjamini–Hochberg approach was used to adjust for multiple testing with cutoff adjusted P values of .05 ( 29 ).