The complement system supports normal postnatal development and gonadal function in both sexes

Male and female infertility are clinically managed and classified as distinct diseases, and relatively little is known about mechanisms of gonadal function common to both sexes. We used genome-wide genetic analysis on 74,896 women and men to find rare genetic variants that modulate gonadal function in both sexes. This uncovered an association with variants disrupting CSMD1, a complement regulatory protein located on 8p23, in a genomic region with an exceptional evolution. We found that Csmd1 knockout mice display a diverse array of gonadal defects in both sexes, and in females, impaired mammary gland development that leads to increased offspring mortality. The complement pathway is significantly disrupted in Csmd1 mice, and further disruption of the complement pathway from joint inactivation of C3 leads to more extreme reproductive defects. Our results can explain a novel human genetic association with infertility and implicate the complement system in the normal development of postnatal tissues.

To confirm the biological role of CSMD1 in male and/or female gonadal function, we perturbed 208 its ortholog in a model organism. We generated a colony of Csmd1 wildtype, heterozygous, and 209 knockout mice and observed the effect of genotype on gonadal function and fertility (Methods,210 Figure 3D). In males, gross testis weight at necropsy did not differ significantly among Severity ("none", "mild", and "profound") and onset (postnatal day 34 through day 300) of the 219 degeneration phenotype vary greatly between individuals. In fact, different foci within the same 220 testis of Csmd1 knockout mice often show different stages of degeneration ( Figure 3B). Our 221 histological study of over 50 knockout animals uncovered two types of germ cell pathology 222 whose connection to each other is unclear. The first is a sequence of active loss of germ cells 223 within each tubule ( Figure 3B). Spermatogenesis begins to become disorganized, especially at 224 the late stages of spermiogenesis, with failure of spermiation, fewer numbers of elongating 225 spermatids in the lumen, and mixing of spermatid steps in stages IX-XII. This is followed by the 226 sloughing of all types of germ cells into the lumen; remaining germ cells can be observed in 227 unusual tubules that appear to be missing one or more waves of spermatogenesis, and these 228 eventually resolve as Sertoli cell-only tubules. Sloughed germ cells can be seen downstream in 229 the epididymis, and, occasionally they obstruct the rete testis leading to dilation of the tubules 230 (data not shown). These defects are most likely to arise due to disruption of interactions between 231 Sertoli and germ cells. The second pathology was an apparent depletion of spermatogonial stem 232 cells in the atrophic tubules; even in tubules with ongoing spermatogenesis, some areas show no 233 spermatogonia. Significantly fewer germ cells express the male germ cell antigen TRA98+ 234 (Poisson regression; P < 2 x 10 -16 ; Figures 3D and S3D), in both atrophic and normal tubules, 235 suggesting that knockout testes suffer from expression perturbations in addition to, or perhaps 236 presaging, loss of spermatogonia and frank degeneration. Together, these observations indicate 237 that the Csmd1 knockout mutation (i) is not fully penetrant; and (ii) may be influenced by 238 environmental and/or stochastic events. However, even after accounting for age covariates, 239 Csmd1 genotype segregates significantly with testes derangement status (P = 7.69 x 10 -3 ; 240 MANOVA; Figure 3C). Finally, we performed serial backcrossing for 9 generations on a subset 241 of mice to validate the effect of the Csmd1 null allele on a roughly constant genetic background 242 (Methods). We recapitulated the degeneration phenotype in these backcrossed male knockouts 243 ( Figure S3E), indicating that Csmd1 genotype status-not genetic background-was driving this 244 signal of degeneration. 245 In females, we observed severe inflammatory changes associated with foam cell infiltration, and, 246 rarely, ovarian cysts in a subset of Csmd1 knockouts (Figures 4A and 4B). Foam cells are 247 multinucleated phagocytic macrophages which have become engorged with lipid, and are 248 associated with ovarian aging. We performed Oil Red O staining which showed highly elevated 249 lipid signal in the ovarian stroma of knockouts compared to age-matched controls, indicating a 250 phenotype of premature ovarian aging in knockout animals ( Figure 4A). Csmd1-deficient 251 females had significantly smaller ovaries by mass when controlling for age, (p = 8.1 x 10 -3 ; 252 Figure 3D; Figure 4C). Furthermore, knockout females showed significantly more atretic 253 follicles and fewer normal pre-ovulatory follicles at necropsy (p=3.5 x 10 -3 ; Hotelling t-test; 254 Figure 4D). To evaluate whether these biometric and histologic changes were also associated 255 with reproductive performance, we estimated female time to pregnancy based on retrospective 256 husbandry records. We generated a null distribution of time to conception which demonstrates 257 distinct periodicity corresponding to the mouse female estrous cycle lasting 4-5 days ( Figure   258 4E). Next, we stratified our population by maternal genotype. For Csmd1 wildtype mothers, the 259 bulk of conceptions occurred within the first estrous cycle as expected (Foldi et al., 2011), 260 whereas most Csmd1 knockout mothers became pregnant after two or more cycles (β GT = 10.4; P 261 = 0.012). A small minority of knockout females required many cycles to achieve pregnancy (> 262 60 days). Circulating gonadotropin levels did not differ between wildtype and knockouts after 263 controlling for estrous stage, suggesting that this reduction in mating success was not secondary 264 to impaired hormonal input along the HPG axis (Methods, Figure S4). Instead, if Csmd1 265 knockout females bear a reduced ovarian reserve, there may be a reduced probability of 266 conception per cycle due to a smaller oocyte target for male sperm. Interestingly, while knockout 267 females achieved fewer pregnancies per estrous cycle, the average number of offspring born per 268 pregnancy did not differ significantly between wildtype and knockout mothers (x ̅ wt = 6.6 (95% 269 ); x ̅ ko = 6.9 (95% CI [5.7-8.1]); Figure 3D). However, pups borne of Csmd1 270 knockout mothers suffered from significantly higher mortality rates during the neonatal period (1 271 -10 days) when compared to wildtype/heterozygous mothers (% mortality WT/het = 10.5% (95% 272 CI [3.6% -17.5%]); % mortality KO = 50.0% (95% CI [30.0% -70.0%]); Poisson regression P = 273 7.93 x 10 -7 ; Figure 5A). We performed necropsy on expired offspring which revealed an absence 274 of milk spots, suggesting death by starvation. Because neonatal mortality segregated with 275 maternal genotype but not offspring genotype or paternal genotype, we hypothesized that this 276 increase in mortality could be explained by a nursing defect in Csmd1-deficient mothers. 277 Therefore, we performed IF to confirm that CSMD1 is expressed in the normal mammary gland 278 through the adult life cycle of wildtype animals ( Figure 5B). CSMD1 is observed on both 279 luminal epithelial cells and myoepithelial cells of the mammary ducts, and on numerous stromal 280 cells (Figures 5B and 5C). Ductal cell expression of CSMD1 appears to be regulated throughout 281 the life cycle, with lowest expression seen in virgins, increasing in mid-pregnancy and lactation, 282 with maximal expression during involution. Mammary glands from knockout females showed 283 reduced density of the epithelial branching network during mid-pregnancy and post-nursing, 284 likely explaining the lack of milk available to nursing pups ( Figure 5D). Visual comparison of 285 duct morphology in nulliparous wild type and knockout animals suggested that the main 286 structural defect was a highly reduced incidence of lateral branches prior to pregnancy ( Figure   287 5E), a conclusion that was statistically supported by quantitative image analysis ( Figure 5F).

288
The complement pathway is dysregulated in Csmd1 knockout mice 289 The primary protein sequence of CSMD1 shares homology with complement-interacting proteins 290 (Kraus et al., 2006). Complement acts as an inflammatory/phagocytic signal in the innate immune 13 system (Liszewski et al., 1996), and recent work has shown that classical complement components 292 C1q and C3 are also responsible for microglia-mediated phagocytosis of excess neuronal cells in 293 a normal developmental process known as synaptic pruning (Schafer et al., 2012 Group of the Psychiatric Genomics, 2014) and complement C4 (Sekar et al., 2016) have also 296 been associated with schizophrenia in independent, well-powered human association studies. 297 Furthermore, some of the most significantly associated variants previously associated with 298 azoospermia encompass the greater MHC locus, which include complements C2, C4 and factor 299 B (Ni et al., 2015;Zhao et al., 2012). Csmd1 is also known to inhibit the classical complement 300 pathway in vitro (Escudero-Esparza et al., 2013;Kraus et al., 2006). Thus, to consolidate the putative 301 roles of complement with Csmd1-mediated pathology, we investigated the activity of 302 macrophages and complement component C3 in wildtype and Csmd1-null gonads. C3 mRNA is 303 detectable in whole testes and ovaries, and in testicular germ cells at multiple stages of 304 spermatogenesis ( Figure 6A). C3 and Csmd1 mRNA expression are anticorrelated throughout 305 spermatogenesis. Macrophages, the immune cells most commonly associated with complement-306 mediated phagocytosis, are found in the interstitial space between seminiferous tubules (Figure In wildtype ovaries, we observed a localization of C3 and macrophages that support the 315 hypothesis that complement-mediated phagocytosis and cellular remodeling are processes that 316 regulate normal gonadal function. Interestingly, C3 is localized to the oocyte surface in normal 317 developing follicles, colocalized with CSMD1, and then observed to be diffused in large 318 amounts throughout the corpus luteum, which is devoid of CSMD1 (Figures 6D and 6E). 319 Macrophages are a prominent cell type in the ovary and associated with, but excluded from 320 entering, healthy follicles; they invade corpora lutea and degrading follicles ( Figure 6E). As 321 predicted in a model of C3-mediated phagocytosis by macrophages, C3 colocalizes with 322 macrophages in the corpus luteum as well as in atretic follicles. Interestingly C3 is abundant 323 within the early follicular antrum (probably in follicular fluid), suggesting that C3 may be 324 important for remodeling the connections between granulosa cells during antrum formation 325 ( Figure 6E). It has previously been reported that activated C3 is present in human follicular fluid 326 at levels comparable to sera, but its physiological role in folliculogenesis, ovulation or 327 fertilization is unknown (Perricone et al., 1992). 328 329 Finally, we observed a pattern of C3 and CSMD1 expression in wildtype mammaries that also 330 supports the notion that CSMD1-complement interactions are dysregulated in the pathologies 331 observed in CSMD1 knockouts ( Figure 6F). As early as puberty, C3 can be seen in high levels 332 within the mammary duct lumen of virgin animals. We speculate that C3 may be involved in the 333 process of lumen formation, as lower levels of C3 are observed in lumens that are just beginning 334 to open and contain dissociated cells. C3 is also expressed within vesicles of specific subsets of 335 CSMD1-positive stromal cells, likely macrophages or eosinophils.
Based on previous findings that CSMD1 is a negative regulator of C3, we predicted that removal 337 of C3 would partially or completely alleviate the morphological degeneration and fertility defects 338 observed in Csmd1 knockout mice. To test this prediction, we generated a colony of C3/Csmd1 339 double knockout (DKO) mice. Surprisingly, we found no evidence of rescue in DKO males or 340 females ( Figure S6). Instead, we observed an unmasked phenotype of more severe histological 341 degeneration in all DKO females, characterized by even more invasion of foam cell 342 macrophages, extensive pyknosis, and deformed follicles. We also observed profound 343 inflammatory changes in the mucosal layer of the oviduct ( Figure S6B). We monitored the 344 fertility of 19 DKOs (10 males and 9 females), and of these, only 4 (21%) produced progeny 345 after at least 3-7 months of mating (3 males and 1 female; Table S3, Figure S6C). The average 346 litter size resulting from successful mating was small compared to wildtype (mean size 4.25 347 pups). These extreme phenotypes are not observed in Csmd1 nor C3 single knockouts, indicating 348 that the combined effect of Csmd1 and C3 on fertility is synergistic. We used a human genetic screening approach to identify genes that modulate male and female 353 gonadal function, and identified a strong candidate, the complement regulator CSMD1. The 354 human phenotypes that we studied were ascertained for having abnormal, early loss of germ cell 355 development, and we observed defects in gametogenesis in both male and female Csmd1  S4). We did not observe any stage-specific accumulation or depletion of germ cells in either sex, 371 nor, as mentioned above, any tell-tale signs of excess apoptosis that is usually seen in such cases 372 (Lipkin et al., 2002;Yatsenko et al., 2015). We observed no significant differences in PCNA marker 373 levels between Csmd1 wildtype and knockout testes of adult animals (data not shown).  In conclusion, we have used human genetics and animal models to identify a likely role for the 506 complement system in postnatal developmental processes across multiple tissues in the body.

507
When combined with existing observations from mammalian brain and other model organisms, 508 we predict that macrophage mediated complement activity on self cells is a normal and highly 509 controlled process in many developmental systems in metazoans. Our work highlights the need 510 for deeper investigation into the role of immune system components in reproductive tissues, and 511 the opportunities that such work can have to illuminate and connect common biological 512 processes that produce disease in more complex contexts across the body. and immunofluorescence experiments as with the original colony (described below; Figure S7). 562 We performed microsatellite genotyping of these littermates to estimate the C57BL/6 563 background after backcrossing (Washington University Rheumatic Disease Core   to a single mouse pathologist in a blinded fashion. All samples received a score of 0 (no 674 damage), 1 (mild damage), or 2 (profound damage) (see Figure S3A for examples). In order to 675 estimate the effect of genotype on score, we fit a linear analysis of variance model:

Materials and Methods
where y ijk is the damage score for individual k, µ average damage score across all animals, α i is 678 the effect of genotype i, β j is the effect of age j, and ε ijk is the random error associated with the 679 kth observation. We performed immunofluorescence as described above on a pair of 34 day old male littermates 683 (the same individuals as seen in Figure 3A) using anti-TRA98 antibody (Abcam ab82527). We 684 generated count data for total cells (filtering based on size and shape), and for TRA98-positive cells (filtering based on green fluorescence) using the ImageJ software package. In order to 686 estimate the effect of genotype on TRA98 cell count, we fit the following linear model: Where y i is the TRA98-positive count in image i, and X 1 is the genotype (Csmd1 wildtype versus 689 knockout), and X 2 is the total cell count. ε i is the nuisance variable for image i.
Where y i is the number of total oocytes in bilateral ovaries of individual i, and X 1i and X 2i are 710 genotype and age, respectively. ε i is the nuisance variable for individual i.
Where y i is the estimated time to conception for mating pair i, X 1i is maternal genotype 726 (wildtype, heterozygote, or knockout), X 2i is maternal age at conception, and X 3i is paternal age 727 at conception. ε i is the nuisance variable. We bred 44 females (8 wildtype, 27 heterozygote, and 9 homozygote) with 41 males (4 wildtype, 732 26 heterozygote, and 11 homozygote) over a period of 10 months to produce 99 litters, totaling 688 live births. All 9 parental genotype permutations [wt dam x wt sire , wt dam x het sire … hom dam x 734 hom sire ] were represented multiple times (excepting het dam x wt sire ). We counted deaths in during 735 the neonatal period (defined as 1-10 days by convention, although the vast majority of deaths 736 occurred within 24-48 hours) and subtracted from the live birth total to obtain the final number 737 of surviving pups (550 total). Next, we stratified each litter by maternal and paternal genotype 738 status (Csmd1 wildtype or heterozygous versus knockout) and fit the following linear model: Where y i is the number of surviving pups in litter i, and X 1i and X 2i are the maternal and paternal 741 genotypes, respectively. ε i is the nuisance variable for litter i.
Human genetic studies were carried out using existing datasets, 2/3 of which were generated by 805 large epidemiological studies; thus, we simply used the sample sizes of cases and controls that 806 were available to us. Based on empirical findings for diseases with similar genetic architecture 807 (e.g. autism and schizophrenia) we hypothesized that sample sizes of approximately 500-1000 808 cases and thousands of controls would be sufficient to detect rare, large effect variants such as 809 the 16p11 deletion that has a frequency of ~1% in cases of autism, which was originally detected  We declare no competing personal or financial interests.