Evidence for an FMR1 mRNA Gain-Of-Function Toxicity Mechanism in the Pathogenesis of Fragile-X Associated Premature Ovarian Insufficiency


 Fragile X-associated premature ovarian insufficiency (FXPOI) is caused by expansion of a CGG repeat sequence located in the 5’ untranslated region of the FMR1 gene. Women with FXPOI have a depleted ovarian reserve, resulting in amenorrhea, hypoestrogenism, and loss of fertility before the age of 40. FXPOI is caused by CGG sequence expansions to lengths between 55 and 200 repeats, known as a FMRI premutation, however the mechanism by which the premutation drives disease pathogenesis remains unclear. Two main hypotheses exist, which describe an mRNA toxic gain-of-function mechanism or that repeat-associated non-AUG (RAN) translation results in the production of an abnormal protein, called FMRpolyG. We have developed an in vitro granulosa cell model of the FMR1 premutation by ectopically expressing CGG-repeat RNA and FMRpolyG protein. We show that expanded CGG-repeat RNA accumulated in intranuclear RNA structures, and these aggregates were able to cause significant granulosa cell death independent of FMRpolyG expression. Furthermore, using an innovative RNA pulldown, mass spectrometry-based approach we have identified proteins that bind CGG-repeat RNA in granulosa cells in vitro, and thus may be deregulated as consequence of this interaction. Collectively, these data provide evidence for the contribution of an mRNA gain-of-function mechanism to FXPOI disease biology.


Introduction
The fragile X mental retardation 1 (FMR1) gene is located on the X chromosome and contains a CGG trinucleotide repeat sequence within its 5' untranslated region (5'UTR), expansions of which can result in both neurological and reproductive disorders. The polymorphic length of the repeat sequence is categorised into four different size ranges. Individuals who carry less than 44 CGG repeats have a normal repeat length that is usually transmitted in a stable manner from mother to offspring 1 , while having between 45 and 54 repeats is classi ed as intermediate, or grey zone. Although grey zone repeat lengths are not directly associated with any disease phenotypes, some CGG repeat instability has been reported, which results in variable repeat expansion during transmission 2 . Expansion of this CGG repeat sequence to more than 200 repeats is categorised as a full mutation, and underlies the severe neurodevelopmental condition fragile X syndrome 3 , which is the most common cause of inherited intellectual disability and autism in males, with patients suffering from a wide range of clinical, cognitive and behavioural dysfunctions. The range of 55-200 CGG repeats is considered a premutation, and some females with this develop what is now known as fragile X-associated premature ovarian insu ciency (FXPOI) 4,5 .
Premutations also result in the more recently described fragile X-associated tremor/ataxia syndrome (FXTAS), a multisystem neurological disorder with tremor and ataxia as its principal features, which was initially recognised in aging carriers but with clinical features potentially also present in children 6 .
Premature ovarian insu ciency (POI) is de ned by the depletion of the ovarian follicle population, resulting in amenorrhea, hypoestrogenism, and loss of fertility before the age of 40 years 7 . In addition to the direct impact on fertility, secondary consequences arising from estrogen de ciency compromise bone, cardiovascular and neurological health of affected individuals (comprehensively reviewed in 8 ). Despite advances in genomic technologies and the strides taken to unravel the genetic determinants of POI, abnormalities in the FMR1 gene are the only monogenic cause currently tested for in routine clinical practice. Approximately 20-30% of female premutation carriers develop FXPOI 9,10 , with these women having midrange CGG tract sizes between 70 and 100 repeats 9,11 . However, an additional ~ 20% of females with the premutation present with irregular periods and a further ~ 13% report di culty conceiving 12,13 . Even premutation carriers without signs of ovarian dysfunction have a menopause that is on average 5 years earlier than women in the general population 11 , thus although the premutation doesn't necessarily result in POI, it is clear that its presence impairs normal FMR1 gene function in the ovary with a range of clinical consequences.
The mechanisms that underlie compromised ovarian follicular function preceding the full development of FXPOI are unclear, but it is proposed these insults could occur at various stages of follicular development.
Findings from knock-in mouse models [14][15][16] generally show consensus in their reproductive physiology and demonstrate that the FMR1 premutation allele does not interfere with the establishment of the primordial follicle pool. However, the population of growing follicles exhibited increased atresia, affecting all growing follicle stages 14,15 . This follicle decline was paralleled by a decrease in litter size 17 . Although how this results in premature depletion of the ovarian reserve (i.e. non-growing follicle pool) is unclear, there are clear interactions between the growing and non-growing pools that regulate the activation of follicle growth 18, 19 .
At a molecular level, FXPOI shares many common features with the other premutation associated disorder, FXTAS, and advances made in understanding this neurological condition have also been applied to the pathogenesis of FXPOI (reviewed in 20 ). In premutation carriers, the FMR1 locus is transcriptionally active and mRNA levels are elevated 21 . Thus, a key hypothesis is that FMR1 mRNA gain-of-function toxicity may underlie FXPOI, a concept that originated from the pathogenesis of another trinucleotide expansion disease myotonic dystrophy 22,23 . In this model, FMR1 transcription is augmented and expanded CGG-containing transcripts accumulate into nuclear RNA foci, which bind and sequestrate speci c RNA-binding proteins and thus potentially inhibit their normal functions, compromising cell functions and potentially causing cell death [24][25][26][27] . It is important to note that in this model, toxicity arises because of the expanded CGG repeat itself, and not of overexpression of FMR1 protein product, as overexpression of FMR1 mRNA without a CGG repeat expansion does not trigger neuronal death or produce behavioural de cits 28 . A second (and non-exclusive) model has been proposed recently, based on the observation that expanded repeat sequences can be translated in absence of any AUG canonical start codon, through a mechanism named repeat-associated non-AUG (RAN) translation 29,30 . In the case of FMR1, expanded CGG repeats are predominantly translated into a polyglycine-containing protein, named FMRpolyG, which forms ubiquitin-positive intranuclear inclusions and which expression is toxic in cell and animal models 31, 32 Sellier, 2017 #4270}. Intranuclear inclusions of FMRpolyG have been detected in the brains of FXTAS patients 31,33−35 , as well as in non-CNS tissues 36 , including the ovarian stroma of a woman with FXPOI 37 , and in mural granulosa cells from six FMR1 premutation carriers 38 . Furthermore, FMRpolyG protein has been detected in the ovarian stroma of mice expressing expanded CGG repeats 37,39 . Collectively these data suggest that RAN translation may be involved in FXPOI.
To study the relative contributions of mRNA gain-of-function toxicity and RAN translation in the pathogenesis of FXPOI, we established an in vitro human granulosa cell line model of the FMR1 premutation by ectopically expressing CGG-repeat RNA and FMRpolyG protein. FMRP is expressed in granulosa cells of mature follicles in adult ovaries 40 and increased FMR1 transcript levels have been reported in granulosa cells of premutation carriers 41 . We found that expanded CGG-repeat RNA accumulated in intranuclear structures, and using an innovative methodology that combines RNA pulldown with stable isotope labelling by amino acids in cell culture (SILAC) high-throughput mass spectrometry (RP-SMS), we identi ed proteins that are speci cally sequestered by CGG RNA aggregates in granulosa cells in vitro and colocalised these endogenous proteins with CGG RNA aggregates. CGGrepeat RNA caused signi cant levels of granulosa cell death, which was independent of the presence of FMRpolyG protein. These data thus provide evidence for the contribution of the mRNA gain-of-function mechanism to FXPOI disease, and provide protein targets whose dysregulation may contribute to this pathological condition.

Plasmids
Plasmids expressing 60 CGG repeats (referred to as 60x CGG) or 100 CGG repeats within the human FMR1 sequence, fused to GFP (referred to as 5'UTR FMR1 (CGG)100x GFP) have been described previously 33,42 . For cell viability assays, a bidirectional plasmid was created by inserting GFP with a CMV promoter and terminator sequence (ampli ed from pEGFP) into a 5'UTR FMR1 (CGG)100x GFP plasmid with the ACG start codon deleted (referred to as Δ5'UTR FMR1 (CGG)100x GFP) 33 ) using BglII and EcoRI restriction sites. To ensure the stability of the expanded CGG repeats, all CGG plasmids were transformed into NEB® Stable Competent E. coli (New England Biolabs, UK) and grown at 30ºC according to manufacturer's instructions.
RNA pulldown SILAC high-throughput mass spectrometry (RP-SMS) and Western blotting RNA pulldown coupled to stable isotope labelling by amino acids in cell culture (SILAC) mass spectrometry was carried out as described previously 45 . Brie y, HGrC1 cells were cultured in SILAC media (DC Biosciences, Dundee, UK), 'heavy' or 'light', supplemented with dialysed calf serum (DC Biosciences) to incorporate cells with heavy or light isotopes. Cell extracts were prepared and incubated with CGG (30x) RNA (Dharmacon, Cambridge, UK) coupled to agarose beads. Following a series of washes to remove unbound protein, proteins were electrophoresed into an SDS-PAGE gel (Bio-Rad, Watford, UK), and submitted for LC-MS/MS analysis performed using an Orbitrap™ mass spectrometer (ThermoFisher Scienti c). Data was analysed using the MaxQuant software 46 to determine the ratio of heavy-labelled peptides to light-labelled peptides, and identify proteins speci cally bound to CGG RNA. Pulldown experiments followed by Western blotting were used to validate mass spectrometry data. Pulldown was carried out as described above, and proteins were separated on an SDS-PAGE gel (Bio-Rad). Proteins were transferred onto Immobilon FL membrane (Millipore, Dorset, UK), which was blocked using Intercept blocking buffer (LI-COR Biosciences, Cambridge, UK). Western blotting was undertaken with anti-FUS (AMAb90549), anti-PA2G4 (15348-1-AP) and anti-TRA2β (ab31353) antibodies, at a dilution of 1:1000, overnight at 4ºC. Alexa Fluor 680-and 800-conjugated secondary antibodies (Molecular Probes) were used for detection (at 1:10,000) and blots were imaged on a LI-COR FC Odyessy.

Flow Cytometry Cell Viability Assay
HGrC1 and COV434 cells were seeded at a density of 300,000 cells per well of a 6 well plate. and transfected with either an empty pEGFP plasmid, Δ5'UTR FMR1 (CGG)100x GFP or 5'UTR FMR1 (CGG)100x GFP as described above. At 48h post transfection, cells were trypsinised, neutralised, and 6 wells per condition were pooled in order to have enough cells for ow cytometry analysis. Just before analysis, cells were incubated with DAPI (1:1000) for 3 min as a cell viability marker. Flow cytometry was carried out on a BD LSRFortessa™ and data analysed using BD FACSDiva™ software (version 8.0). Single cells were analysed for GFP expression (to identify positively transfected cells) and DAPI staining, with DAPI positive cells indicative of a compromised cell membrane.

Statistical analysis
All data are shown as mean ± standard error of the mean and were analysed using GraphPad Prism 9 software (GraphPad Software, Inc., San Diego, CA). Mann-Whitney and Freidman test statistics were carried out as appropriate. A p value of < 0.05 was considered statistically signi cant.

Results
Expanded CGG repeats within the FMR1 5'UTR form intranuclear RNA aggregates and FMRpolyG protein aggregates in granulosa cell lines To investigate the consequences of the FMR1 premutation in granulosa cells, we transfected a plasmid expressing 100 CGG repeats embedded within the 5'UTR of the human FMR1 gene and fused to the GFP in the glycine frame into two granulosa cell lines HGrC1 and COV434, and tested the formation of CGG RNA foci and FMRpolyG expression using RNA uorescence in situ hybridisation (FISH) and uorescence microscopy, respectively. Expression of this plasmid generated numerous intranuclear CGG aggregates in both granulosa cell lines, which could be observed at 24h post transfection. The RNA composition of these aggregates was con rmed as they were sensitive to RNase A treatment (Fig. 1A). Both granulosa cell lines were also able to translate this plasmid into FMRpolyG protein, as GFP-tagged protein was observed at 24h post transfection (Fig. 1B).
As CGG RNA foci are observed in some cell lines (e.g. COS7), but not in other (e.g. HEK293, HeLa, A172, U-937 etc., see 42 ), we con rmed these results using a second plasmid, which expresses 60 CGG repeats in isolation under the control of a CMV promoter. These repeats are deleted of their natural FMR1-hosting sequence and thus cannot express the FMRpolyG protein, where initiation occurs at near-cognate codons located upstream of the repeats within the 5'UTR of FMR1. RNA foci dynamics were analysed by RNA FISH at 24, 48 and 72h after transfection. HGrC1 and COV434 cells were transfected with either the 5'UTR FMR1 (CGG)100x GFP plasmid or the (CGG)60x construct. the expressed 60 CGG repeats formed intranuclear RNA aggregates that increased in size and number over time ( Fig. 2A and B), as reported in COS7 cells 42 . In contrast, the intranuclear CGG RNA aggregates that formed as a result of the expression of the 5'UTR FMR1 (CGG)100x GFP plasmid were more stable and did not evolve in size or number over time ( Fig. 2A and B). That CGG expanded repeats embedded in the natural FMR1 sequence robustly formed RNA foci in ovarian cell lines was unexpected, given that it has been shown that this mRNA should be exported into the cytoplasm for translation into FMRpolyG protein 33 . Therefore, we quanti ed CGG RNA foci and FMRpolyG expression in HGrC1 and COV434 cells by transfecting cells with the 5'UTR FMR1 (CGG)100x GFP plasmid and using RNA FISH followed by GFP immunocytochemistry to detect CGG RNA foci and FMRpolyG expression in the same cells at 48h after transfection (representative image of this analysis is shown in Fig. 3A). In HGrC1 cells, most transfected cells were positive for RNA foci only (73.5% ± 5.3%), with a smaller proportion (22.0% ± 2.5%; p = 0.0046) of cells expressing both CGG RNA foci and FMRpolyG protein and very few expressing FMRpolyG protein only (Fig. 3B). This differed markedly in COV434 cells, where most cells were positive for FMRpolyG protein only (81.1% ± 2.8%). However, CGG RNA foci were still observed in a signi cant proportion of cells, with or without accompanying FMRpolyG (18.9% ± 2.8%) (p = 0.0046 vs. FMRpolyG alone) (Fig. 3C). Taken together, the expression of premutation length CGG repeats in granulosa cells and the variable translation of these repeats into FMRpolyG protein, is consistent with a CGG RNA gain-of-function model contributing to FXPOI pathogenesis.

CGG-repeat RNA and FMRpolyG protein affect granulosa cell viability equally
We developed a ow cytometry-based assay to explore the individual effects of CGG-repeat RNA and FMRpolyG protein on the viability of HGrC1 and COV434 cells (Fig. 4A). To do this, we transfected cells with either a construct expressing the GFP alone as a negative control, the 5'UTR FMR1 (CGG)100x GFP plasmid, which is transcribed and translated into FMRpolyG-GFP, or a Δ5'UTR FMR1 (CGG)100x construct, which only produces CGG-repeat RNA given the deletion of the near cognate codon that facilitates translation of FMRpolyG, co-transfected with a GFP-expressing plasmid to allow selection of the transfected cells by FACS. At 48h post transfection, both CGG-repeat RNA and FMRpolyG protein caused signi cant cell death in both cell lines compared to cells expressing GFP only (p = 0.028)( Fig. 4B and C). However, there was no signi cant difference in the proportion of dead cells expressing CGGrepeat RNA only vs. FMRpolyG (33.3% ± 5.9% vs 29.1% ± 1.8% in HGrC1 cells, and 21.0% ± 0.9% vs 30.4% ± 4.3% in COV434 cells). Given that CGG-repeat RNA resulted in similar levels of cell death with or without FMRpolyG, this would suggest that accumulation CGG-repeat RNA alone can cause granulosa cell loss and drive FXPOI disease biology, supporting an mRNA gain-of-function toxicity model, while not excluding a contribution of FMRpolyG.
RNA pulldown-SILAC mass spectrometry (RP-SMS) identi es proteins that bind CGG aggregates As CGG-repeat RNA caused signi cant granulosa cell death, we sought to identify proteins that are associated with CGG repeats, which could potentially be sequestered by RNA aggregates in granulosa cells and become dysregulated causing cell death. We used a novel methodology that combines RNA pulldown with SILAC high-throughput mass spectrometry (RP-SMS 45 ) to reveal that 111 cellular proteins were enriched at least two-fold (H/L ratio ≥ 2) in the CGG pulldown compared to the beads-only control ( Fig. 5A and B, Supplementary le 1). Among those identi ed were many known RNA splicing and RNA binding proteins, including hnRNPH proteins and MBNL1, that have previously been reported to associate with CGG RNA in mouse brain and COS7 cells 42 . Among the proteins identi ed, two were of particular interest: FUS, which is an RNA binding protein involved in amyotrophic lateral sclerosis and DNA damage repair 47 and PA2G4 (also named ErbB3-binding protein 1 (EBP1) that binds ribosomal RNA involved in cell proliferation 48 . Disruption of these two cellular functions could lead to granulosa cell death and subsequent follicle loss as observed in FXPOI. As a positive control, we also tested TRA2β as this RNA splicing factor was shown to co-localise in COS7 cells expressing premutation length CGG RNA and has been hypothesised to be involved in FXTAS disease progression 42 .
Repeat RNA pulldown experiments followed by Western blotting showed that the binding of FUS, PA2G4 and TRA2β was speci c to CGG RNA, as no protein bands were detected in the beads-only control (Fig. 5C). Next, we tested for co-localisation of these candidates with RNA aggregates in HGrC1 cells cotransfected with 60 CGG repeats and GFP-or HA-tagged proteins, combining FISH with immunocytochemistry at 24h post transfection. We observed that the presence of CGG RNA changed the distribution of FUS, PA2G4 and TRA2β expression inside the cell with areas of co-localisation (Fig. 6). We also screened other in vitro-identi ed protein candidates (CNBP, DDX5, DHX9, G3BP2 and TAF15), however they showed no co-localisation with CGG RNA aggregates (data not shown). To further con rm the binding of FUS, PA2G4 and TRA2β with CGG mRNA repeats, we quanti ed the co-localisation of CGG RNA and endogenously expressed FUS, PA2G4 and TRA2β in HGrC1 and COV434 cells using antibodies speci c to each protein (Fig. 7A). Results from a 2D analysis of 40 individual cells from over three repeated experiments, showed on average 13.0% and 21.5% co-localisation of CGG-repeat RNA with FUS in HGrC1 and COV434 cells respectively, 29.3% and 8.7% co-localisation of CGG-repeat RNA with PA2G4 and 30.7% and 45.5% co-localisation of CGG-repeat RNA with TRA2β, in HGrC1 and COV434 cells, respectively (Fig. 7B). In addition to the marked differences of co-localisation between the two granulosa cell lines, this co-localisation tended to be highly variable, which may be a consequence of the poor cell viability measured at this timepoint (Fig. 4).

Discussion
Much of the research into the molecular mechanisms underlying the pathology of FMR1 premutationassociated conditions has focussed on the neurological aspects. This has led to two main hypotheses being proposed, which describe an mRNA gain-of-function or a RAN translation-based mechanism to explain how the FMR1 premutation drives the pathogenesis of these disorders. In this study, we sought to investigate whether these hypotheses could contribute to the ovarian dysfunction observed in FXPOI. To do this, we expressed CGG-repeat RNA, with and without accompanying FMRpolyG (the RAN translation protein product) in two human granulosa cell lines HGrC1 and COV434 and explored the consequences of this. The HGrC1 cell line is derived from granulosa cells of antral follicles 43 , and to our knowledge, is the only human cell line that possesses characteristics of granulosa cells belonging to early stage follicles. Other granulosa cell lines have been derived from follicles after in vitro fertilisation, and are therefore luteinised, or have been established from granulosa cell tumours. This was thought to be the case for COV434, although recent ndings have questioned the true origin of this cell line 49 , thus the present data using HGrC1 cells may be more relevant to ovarian follicular function. Whilst we acknowledge this in vitro model does not truly recapitulate premutation granulosa cells found in vivo, the scarcity of FXPOI patient tissue with ovarian follicles available for study necessitates this compromise. Furthermore, this cellbased model enables the study of human-speci c disease mechanisms where expanded CGG-repeat expression is restricted to one cell type.
Following expression of 60x or CGG-repeat RNA, deprived of its natural FMR1 5'UTR hosting sequence, in these granulosa cell lines, we observed the formation of intranuclear RNA aggregates that increased in size and number over a 72h period. This nding has been described in other cell lines 42 , however expression of microsatellite repeats in isolation, without their natural hosting sequence in which they are normally embedded, can lead to a nuclear retention bias where the mRNA is not exported into the cytoplasm for translation and accumulates in the nucleus instead 33 . Therefore, we additionally expressed 100 CGG repeats embedded in its natural FMR1 5'UTR sequence, which resulted in large stable RNA aggregates. This observation is interesting as we anticipated this mRNA to be exported and RAN translated into FMRpolyG protein. Indeed, CGG RNA aggregates are rare occurrences in brain tissue from transgenic mice engineered to express 99 CGG repeats within the human 5'UTR FMRI gene 33 , and other knock-in mouse models expressing premutation-length CGG repeats 42 . Given that our data show that both cell lines had an accumulation of CGG RNA aggregates which then resulted in cell death, in particular HGrC1 cells which were unable to translate the CGG-repeat mRNA into protein e ciently, we suggest that an mRNA gain-of-function mechanism is pertinent to FXPOI pathogenesis.
In support of the mRNA gain-of-function hypothesis, numerous studies have made efforts to identify the various proteins that can be sequestered by CGG-repeat mRNA and subsequently potentially deregulated or displaced from their physiological RNA targets [25][26][27]42 . Furthermore, the CGG-repeat mRNA is known to adopt secondary structures such as intramolecular hairpins that may recruit speci c RNA binding proteins 50 . However, this work has mostly been undertaken in relation to FXTAS models of disease, and thus identi ed candidate proteins with relevance only to neuronal cells. Here we used a novel RNA pulldown method that enabled us to identify proteins that speci cally bind to CGG-repeat RNA in granulosa cells, making this the rst study to investigate this RNA-protein interaction in the ovary. Ideally, pulldown experiments to seek CGG RNA interactants should be undertaken using cellular material isolated from FXPOI patients, but given that premutation granulosa cells are usually only isolated after IVF treatment, it is di cult to ascertain how representative those cells are of early stage follicular cells. Instead, we chose to use HGrC1 granulosa cells that readily form CGG RNA foci upon transfection. Some of the candidate proteins identi ed here, such as TRA2β, MBNL1, and DDX5, were also identi ed in brainbased studies 25,42 . Of our candidate proteins, none have established connections to ovarian cell biology, thus we validated FUS and PA2G4 given their well-characterised roles in DNA damage repair 47 and cell proliferation 48 respectively, and TRA2β given its hypothesised involvement in FXTAS disease progression. We hypothesised that sequestration of these proteins and subsequent deregulation of their function could be potentially detrimental for granulosa cells, and this could underlie the follicle loss that characterises POI. We con rmed co-localisation of FUS, PA2G4 and TRA2β with CGG-repeat RNA, but given its high variability and the toxicity of CGG repeats to these granulosa cell lines, it was not possible to carry out extensive time course analyses which would have allowed us to de ne how quickly these proteins were recruited to the intranuclear aggregates. The impact of sequestration of FUS and PA2G4 on their normal cellular functions also remains to be determined as an observation of protein co-localisation within RNA foci is not a de nitive indication of their sequestration and loss of function. Indeed, MBNL1 has been shown to co-localise with CGG inclusions in FXTAS patients 25 , yet the splicing events coordinated by MBNL1 are not altered in CGG-expressing cells or in FXTAS patients 42 . Furthermore, the observed toxicity may result from partially compromised collective function of several proteins.
Together our data support the involvement of an mRNA gain-of-function hypothesis to the development of FXPOI. While RAN translation of the expanded CGG-repeat mRNA may also be a contributing factor, these data did not support it as a major cause of granulosa cell death in addition to the effect of the CGGrepeat RNA only. In murine models of FXTAS, the expression of FMRpolyG was pathogenic, with these mice exhibiting inclusion formation, motor phenotypes and reduced lifespan, while the sole expression of CGG-repeat RNA did not induce any of these features 33 ; however, no examination of ovarian tissue was carried out. In contrast, Shelly et al recently studied the fertility phenotype of two premutation mouse models, but in these the expression of coding Fmr1 was not altered and therefore there was no interference due to decreased FMRP expression 39 . Only expression of both CGG RNA and FMRpolyG led to a progressive loss of fertility with age 39 although expression of CGG-repeat RNA alone was su cient to impair key ovulatory processes in response to exogenous hormones. CGG RNA aggregates have not been reported in women with FXPOI, however FMRpolyG inclusions have been observed in the ovarian stroma of one woman with FXPOI 37 and the mural granulosa cells of six premutation carriers 38 . While investigations for CGG RNA aggregates in human FXPOI ovarian tissue have not been undertaken, it is possible that the cellular stress induced by CGG-repeat RNA and FMRpolyG expression causes atresia of ovarian follicles before RNA aggregates are visible by FISH or immunostaining. Derivation of ovarian somatic cells from FXPOI patient iPSCs may be informative as a model, as these cells will carry the human genetic landscape as well as the disease 39 .
In conclusion, the present data support the involvement of an mRNA gain-of-function mechanism in driving the pathogenesis of FXPOI through the accumulation of large stable nuclear aggregates formed from expanded CGG-repeat RNA, which can cause signi cant granulosa cell death independent of FMRpolyG expression. Furthermore, the identi cation of proteins that could potentially be deregulated in granulosa cells as a result of interactions with CGG aggregates also supports an mRNA gain-of-function toxicity model.   Expanded CGG-repeat mRNA is not e ciently translated into FMRpolyG protein in HGrC1 cells. HGrC1 and COV434 cells were transfected with a plasmid expressing 100 CGG repeats within the FMR1 5'UTR and RNA FISH followed by GFP immunocytochemistry were used 48h after transfection to identify the colocalisation of CGG RNA aggregates and FMRpolyG protein expression. A representative image is Page 17/19 shown in (A). Scale bars represent 500µM and 20µM, respectively. Quanti cation of cells expressing either CGG RNA only, CGG RNA and FMRpolyG or FMRpolyG only in HGrC1 (B) and COV434 (C) cells.
Data is presented as the mean ± SEM of four individual experiments. P=0.0046 for both cell line experiments.

Figure 4
HGrC1 and COV434 cell viability following expression of CGG-repeat RNA only or CGG-repeat RNA and   Co-localisation of exogenously expressed FUS, PA2G4 and TRA2β, and 60x CGG-repeat RNA. HGrC1 cells were co-transfected with a plasmid expressing 60x CGG repeats and a plasmid expressing a GFP-or HAtagged protein of interest, and RNA FISH followed by immunocytochemistry were used 24h after