Hypomorphic and hypermorphic mouse models of Fsip2 indicate its dosage-dependent roles in sperm tail and acrosome formation

Male infertility is a major issue in human reproductive health, affecting more than 20 million men worldwide and preventing natural conception (Niederberger, 2020; Visser et al., 2020). Asthenozoospermia, a symptom frequently observed in infertile men with less progressive motile sperm (Cooper et al., 2010), is caused by morphological and functional defects of the sperm flagellum. A severe asthenozoospermia with near-zero progressive sperm, namely, multiple morphological abnormalities of the sperm flagella (MMAF), is characterized by sperm with absent, short, irregular and/or coiled flagella (Ben Khelifa et al., 2014; Yang et al., 2015). Thus far, defects in several flagellum-associated genes, including DNAH1, CFAP43, CFAP44, SPEF2, QRICH2, ARMC2, CFAP58 and AKAP4 (Ben Khelifa et al., 2014; Baccetti et al., 2005; Tang et al., 2017; Coutton et al., 2018; Coutton et al., 2019; Liu et al., 2019b; Shen et al., 2019; He et al., 2020), have been identified in patients with MMAF.

As an evolutionarily conserved organelle, the sperm tail or flagellum is an axoneme-based structure that exerts a significant impact on sperm motility and fertilization (Eddy et al., 2003; Lehti and Sironen, 2017). In mammals, the sperm tail is structurally divided into three major parts: the midpiece (MP), the principal piece and the end piece (Fawcett, 1975). The MP comprises the outer dense fibers (ODFs), the mitochondrial sheath and the microtubule-derived axoneme (Lehti and Sironen, 2017; Piasecka et al., 2001). The principal piece contains the fibrous sheath in place of the mitochondrial sheath (Lehti and Sironen, 2017; Brown et al., 2003). The end piece is composed of the axoneme surrounded by the plasma membrane, without any accessory structures.

The fibrous sheath is a unique cytoskeletal structure in the principal piece of the sperm flagellum, comprising two longitudinal columns connected by circumferential ribs and surrounded by outer dense fibers (Eddy et al., 2003; Eddy, 2007). The fibrous sheath provides the sperm tail with mechanical support that modulates flagellar bending and defines the shape of the flagellar beat (Ricci and Breed, 2005). In addition, the fibrous sheath also serves as a scaffold for glycolytic enzymes and signaling molecules involved in regulating sperm maturation, capacitation, motility, hyperactivation and/or the acrosome reaction (Eddy et al., 2003; Eddy, 2007).

The acrosome is a unique membrane-bound organelle that covers the anterior portion of the head of the mature spermatozoon (Hirohashi and Yanagimachi, 2018). It derives from the Golgi apparatus and contains digestive enzymes, which are indispensable for the acrosome reaction and successful fertilization (Khawar et al., 2019). ACRV1 is an acrosomal matrix protein that is evolutionarily conserved among mammals (Cruz et al., 2020). It is predominantly located in both round and elongated spermatids in the testes; a polyclonal antibody of Acrv1 has been shown to be useful for staging the cycle of seminiferous epithelium in the mouse (Tang et al., 2012; Osuru et al., 2014). The lateral flow immunochromatographic assay of ACRV1 was recently implemented for the evaluation of male infertility (Bieniek et al., 2016).

Fibrous sheath-interacting protein 2 (FSIP2) is one of the largest proteins (6907 amino acids) in the human genome and a major component of the fibrous sheath (Brown et al., 2003). Six homozygous loss-of-function mutations of FSIP2 were recently reported, which could cause the typical human MMAF phenotypes (Martinez et al., 2018; Liu et al., 2019a). A completely disorganized fibrous sheath associated with axonemal defects was observed in all patients carrying FSIP2 mutations, with abnormalities in fibrous sheath, microtubules and dynein arm-associated proteins (Martinez et al., 2018; Liu et al., 2019a). FSIP2 is also reported to interact directly with AKAP4, the most abundant protein in the sperm fibrous sheath (Brown et al., 2003). However, the molecular mechanisms underlying the MMAF phenotypes caused by the mutations in FSIP2 remain undiscovered.

In this study, we identified a novel homozygous truncating mutation of FSIP2 in a patient with MMAF through whole-exome sequencing (WES). Based on this, we generated two transgenic mouse models: an Fsip2-knock-in (KI) mouse that mimics the truncation mutation, which displayed typical MMAF phenotypes; and an Fsip2-overexpression (OE) mouse, which exhibited longer sperm tails compared with wild-type (WT) mice. Truncation and overexpression of Fsip2 reciprocally affected proteins associated with the fibrous sheath, mitochondrial sheath and acrosomal vesicle. Intriguingly, we showed that two C-terminal motifs of Fsip2 were only conserved in species with internal fertilization. Thus, our results provide new insights into the roles of Fsip2 in the sperm tail and acrosome formation.

An infertile male patient who presented with primary infertility and normal sexual function was enrolled in our study. The analysis of three semen samples over 3 months showed normal semen volume, pH and concentration, but progressive motility was remarkably impaired (Table 1). Scanning electron microscopy (SEM) identified the MMAF phenotypes of the spermatozoa: short tails, thin principal pieces and abnormal MPs (Fig. 1A). Aberrant sperm ultrastructures were further confirmed by transmission electron microscopy (TEM): the mitochondrial sheath was dysplastic and disorganized, and the fibrous sheath was absent (Fig. 1B). In order to uncover the genetic cause of the MMAF phenotypes, we performed WES and found a homozygous 1 base pair (bp) insertion in FSIP2. A premature stop codon downstream of the 2790th residue was introduced by the c.8368_8369insC mutation, which was further confirmed by Sanger sequencing (Fig. 1C). This mutation was not found in the following sequence databases: dbSNP (https://www.ncbi.nlm.nih.gov/snp/); 1000 Genomes Project (https://www.internationalgenome.org/); ExAC (exac.broadinstitute.org/); or gnomAD (https://gnomad.broadinstitute.org/).

To mimic the mutation (c.8368_8369insC, p.2790fs) in FSIP2 found in the patient with MMAF and to evaluate the effects of Fsip2 overexpression, we generated the Fsip2-KI and Fsip2-OE mouse models. Sanger sequencing confirmed the insertion mutation (c.8137insC) located in the 16th exon of Fsip2 in the Fsip2-KI mice (Fig. 2A). We detected substantially reduced and increased mRNA expression levels of Fsip2 in KI and OE mice, respectively (Fig. 2B). The Fsip2 copy numbers of OE mice were at least 4.5-fold higher than those of WT mice and 70% of OE mice showed commensurately increased Fsip2 mRNA levels (Fig. 2C). Female homozygous KI mice were fertile, whereas male homozygous KI mice were sterile despite showing normal mating behavior with successful ejaculations and vaginal plug formation. By contrast, both the female and male OE mice were fertile and there was no significant difference in the testes weights of WT, KI or OE mice (Table S1). Sperm from WT, KI and OE mice were then extracted and analyzed with a computer-assisted sperm analysis (CASA) system. Compared with WT mice, the viability and motility of sperm from KI mice were significantly attenuated (Table 2). However, we observed no obvious differences in sperm viability or motility between WT and OE mice (Table 2).

The immunostaining results confirmed the absence of Fsip2 in the sperm from KI mice (Fig. 2D; Fig. S1). It was also noteworthy that Fsip2 was not only localized to the principal piece, but also found in the MP and the head of the sperm in WT and OE mice (Fig. 2D). This observation suggested that the existence and functionality of Fsip2 are not only restricted to the fibrous sheath.

More spermatozoa could be observed in the seminiferous tubules of OE mice (Fig. 2E). The sperm flagella of KI mice were short, curved and frayed (Fig. 2F), consistent with the morphology of the sperm from the patient with MMAF. SEM and TEM further revealed the MMAF phenotypes and corresponding ultrastructural defects of KI-mice sperm: the flagella were short, the fibrous sheathes were absent and the axonemes were exposed (Fig. 2G,H). We did not observe any structural alterations in the flagella when we compared OE and WT sperm.

Diff-Quick staining revealed some longer spermatozoa from OE mice (Fig. 3A). To determine the lengths of the sperm tails, MitoTrack signal (red) and Akap4 immunostaining (green) were used to indicate the MPs and principal pieces, respectively (Fig. 3B). Our results revealed that both principal pieces and MPs of sperm from OE mice were significantly longer than those from WT mice (Fig. 3C; P<0.01). The sum of the average lengths of the principal pieces and MPs of WT-mice sperm was 115.21±3.09 μm. If the sum of the lengths of the principal piece and MPs of an OE-mouse sperm was greater than 124.48 μm [115.21+(3×3.09)], the sperm was defined as a ‘super-long’ sperm. According to this definition, there were 7.4 times more ‘super-long’ sperm in OE mice compared with WT mice.

To explore the alterations in the gene expression profile of the testis when induced by the mutation or overexpression of Fsip2, the testes of WT, KI and OE mice were processed and analyzed in a pipeline of 10x Genomics single-cell RNA sequencing (RNA-seq). We captured 39,502 cells and partitioned them into 14 clusters, which were then classified into six cell types (spermatogonia, spermatocytes, round spermatids, elongating spermatids, Sertoli cells and Leydig cells) according to previously defined marker genes (Green et al., 2018) (Fig. 4A). Fsip2 was expressed in all cell types and significantly enriched in round spermatids (Fig. 4B).

The round spermatid cluster was then further subdivided into four subclusters (Round spermatid 1, 2, 3 and 4; Fig. 4C), corresponding to the four stages of round spermatids (RS2, RS4, RS6 and RS8, respectively) proposed recently (Chen et al., 2018). We observed no obvious differences between the cell proportions of the four stages of round spermatids from WT and KI mice; however, the proportion of round spermatid 2 in OE mice was substantially higher than in WT and KI mice (Fig. 4D). The marker genes of round spermatid 2 included Acrv1, Spaca1 and Spaca3, which were indispensable for acrosome formation.

We identified the differentially expressed genes between round spermatids of OE and WT mice. Round spermatid 1 consisted of 156 upregulated and 101 downregulated genes; round spermatid 2 consisted of 512 upregulated and 599 downregulated genes; round spermatid 3 consisted of 365 upregulated and 261 downregulated genes; and round spermatid 4 consisted of 622 upregulated and 440 downregulated genes. Fsip2 was upregulated in all subclusters of round spermatids. Gene Ontology (GO) enrichment analysis indicated that the upregulated genes of round spermatids 1, 2, 3 and 4 of OE mice were involved in sperm flagellum and motile cilium (Fig. 4E), in agreement with the longer principal pieces and MPs observed in the sperm from OE mice (Fig. 3C). In addition, the upregulated genes were also significantly enriched in the GO terms of acrosomal vesicle, spermatid development and spermatid differentiation (Fig. 4E).

By contrast, we found far fewer differentially expressed genes among the four subclusters of round spermatids of KI and WT mice, comprising 13 (0 upregulated and 13 downregulated), 40 (17 upregulated and 23 downregulated), five (two upregulated and three downregulated) and six (two upregulated and four downregulated) genes, respectively. The top three downregulated genes for the four subclusters of round spermatids are summarized in Table S2. Fsip2 was downregulated in all subclusters of round spermatids. In addition, the top downregulated genes in KI mice included Tcte3 (Dynlt2a1), Tuba3b and Slxl1, which were primarily expressed in spermatocytes and round spermatids (Fig. 4F). Three uncharacterized genes, 1700017L05Rik, 1700109H08Rik and Gm36368 (mainly expressed in round spermatids), were also downregulated (Fig. 4F).

As mentioned above, Fsip2 was expressed in the sperm head (Fig. 2E); single-cell RNA-seq analysis indicated that the upregulated genes in round spermatids of OE mice were also enriched in the acrosomal vesicle (Fig. 4E). Our immunostaining results showed that, in the acrosome, Fsip2 was co-expressed with Acrv1, a known acrosomal marker (Fig. 5A). The Duolink Proximity Ligation Assay (PLA) further confirmed the interaction between Fsip2 and Acrv1 in the acrosome, because a red PLA signal was observed (Fig. 5B,C). In addition, expression of Acrv1 was significantly increased in OE mice compared with WT mice, whereas diminished expression of Acrv1 was observed in KI mice (Fig. 5D,E).

To reveal the impacts of truncation and overexpression of Fsip2 on protein expression in the testis, we investigated protein expression levels in the testes of WT, KI and OE mice using a proteomics approach. We quantified a total of 6693 proteins from KI mice, of which 656 were identified to be expressed differentially compared with WT mice, including 107 downregulated proteins and 549 upregulated proteins (Fig. 6A). In terms of OE mice, a total of 7342 proteins were quantified, of which 87 were identified to be expressed differentially compared with WT mice, comprising 57 downregulated proteins and 30 upregulated proteins (Fig. 6A).

Twenty-six sperm fibrous sheath-associated proteins and 26 mitochondrial sheath-associated proteins have been summarized by previous research (Lehti and Sironen, 2017). Among these, 22 fibrous sheath-associated proteins (Akap4, Akap3, Gapdhs, Ldhc, Spa17, Fsip1, Fsip2, Ropn1, Ropn1l, Ldha, Pde4a, Tekt2, Fscb, Cabyr, Hk1, Aldoa, Slc25a31, Gapdh, Rhpn2, Gsk3b, Stat1 and Gstm5) and 15 mitochondrial sheath-associated proteins (Klc3, Spata19, Tekt5, Prkar2a, Srsf10, Selenop, Cfap157, Vdac2, Ppp1cc, Gpx4, Tssk2, Ak2, Tex22, Nectin2 and Gopc) were detected in both the KI and OE proteomic data. Gene Set Enrichment Analysis (GSEA) showed that the 22 proteins associated with the fibrous sheath were significantly enriched in the downregulated proteins of KI mice and the upregulated proteins of OE mice (Fig. 6B; q<0.05). In addition, the 15 mitochondrial sheath-associated proteins and the 73 acrosomal vesicle-associated proteins showed a similar trend (Fig. 6B), despite being nonsignificant in OE mice (q>0.05).

Twelve fibrous sheath-associated proteins, nine mitochondrial sheath-associated proteins and 32 acrosomal vesicle-associated proteins, which were downregulated in KI mice and upregulated in OE mice simultaneously when compared with WT mice, accounted for 46.2% (12/26), 34.6% (9/26) and 43.8% (32/73) of known fibrous sheath-, mitochondrial sheath- and acrosomal vesicle-associated proteins, respectively (Fig. 6C). To validate the conclusions of our proteomics analysis, we performed western blotting and sperm immunostaining on Akap4, Akap3, Cabyr, Gapdhs and Odf2. The results indicated upregulation and downregulation of Akap4, Akap3, Cabyr and Gapdhs in the testes of OE and KI mice, respectively (Fig. 6D). This trend was not typical of Odf2, for which we observed decreased expression in KI mice but no increased expression in OE mice. Furthermore, staining of Akap4, Akap3 and Cabyr was absent in KI sperm, whereas Gapdhs and Odf2 staining depicted shorter and disorganized sperm tails in KI sperm (Fig. 6E).

To reveal the sequence characteristics of Fsip2, we investigated the functional domains of human FSIP2 with a protein-domain prediction tool. Three FSIP2 domains were recognized (Fig. 7A), whereas the functions of the other regions of FSIP2 remained unelucidated. In light of this, we explored conserved motifs of Fsip2 using the Multiple EM for Motif Elicitation (MEME) online tool. Eight species (Homo sapiens, Mus musculus, Pogona vitticeps, Latimeria chalumnae, Acanthaster planci, Bombyx mandarina, Mizuhopecten yessoensis and Stylophora pistillata) were enrolled. A total of 30 motifs were identified in the eight species (Fig. 7B), 29 of which were included in H. sapiens and M. musculus (except for the 7th motif). Fsip2 of S.pistillata, M.yessoensis, B. mandarina, A.planci, L.chalumnae and P.vitticeps, consisted of two, four, three, six, seven and 21 motifs, respectively, showing an increasing tendency commensurate with organismal evolution.

In H.sapiens, M.musculus and P.vitticeps, tandem repetition of the first and third motifs was observed in the proximal C-terminal regions (Fig. 7B), corresponding to the three FSIP2 domains (Fig. 7A). Notably, the nonrepetitive sixth and fifth motifs [the 6518-6648 amino acid (aa)-segment in the human FSIP2 protein, Fig. 7C], which did not appear in A.planci, B.mandarina, M. yessoensis or S.pistillata, were only found in the C-terminal region of more complex life forms. H.sapiens, M. musculus, P. vitticeps and L. chalumnae show internal fertilization, whereas A.planci, B.mandarina, M. yessoensis and S.pistillata have external fertilization. This observation suggested that the sixth and fifth motifs in the C-terminal region are associated with internal fertilization.

FSIP2 is one of the largest proteins in humans, and its mutations can cause MMAF phenotypes (Martinez et al., 2018; Liu et al., 2019a). In this work, we detected a homozygous truncating mutation of FSIP2 in a patient with MMAF, which prompted us to investigate the underlying molecular mechanism by establishing mouse models for Fsip2. The Fsip2-KI mouse model mimicked the mutation and, as expected, exhibited MMAF phenotypes typical of those observed in the sperm of the patient. Additionally, reduced mRNA expression levels of Fsip2 were detected in KI mice, in agreement with a previous observation suggesting nonsense-mediated mRNA decay of FSIP2 (Liu et al., 2019a). By contrast, longer principal pieces and MPs were observed in the sperm of the Fsip2-OE mouse model, manifesting as the aforementioned ‘super-long’ sperm. To the best of our knowledge, this is the first report of a transgenic mouse model with elongated sperm. The reciprocal morphology of sperm from KI and OE mice provides preliminary evidence of this vital role of Fsip2 in the development of the sperm flagellum.

The transcriptome profiles of mouse testes were revealed recently, which provided a high-resolution cellular atlas, discriminative cell markers and previously unknown stage regulators of male germ cell development (Green et al., 2018; Chen et al., 2018; Lukassen et al., 2018; Jung et al., 2019). In the present study, we used single-cell RNA-seq to explore the alterations in gene expression profiles induced by the mutation and overexpression of Fsip2. The results showed that Fsip2 was widely expressed in all six cell types, with highest expression in round spermatids, in accordance with a previous study that suggested Fsip2 as a marker of round spermatids (Green et al., 2018). Interestingly, the genes upregulated among the four subclusters of round spermatids of OE and WT mice were involved in sperm flagellum development, consistent with the elongated tails observed in OE-mouse sperm. In addition, the upregulated genes were also significantly enriched in the GO terms concerning spermatid development. This result agreed with our Odf2 immunofluorescence observations of the advanced development of spermatids at stage XI-XII in the seminiferous tubules of OE mice relative to WT and KI mice (Fig. S2), suggesting that overexpression of Fsip2 leads to an acceleration of spermatid development. By contrast, far fewer differentially expressed genes were found among the round spermatids 1-4 of KI and WT mice, indicating that the mutation in Fsip2 exhibited a minor influence on the testicular expression profile. One of the top downregulated genes, Tcte3, is reported to encode a dynein light chain and disruption of Tcte3 in a mouse model resulted in asthenozoospermia (Rashid et al., 2010).

The proteomes of testes or sperm from Qrich2-, Akap4- and Akap3-deficient mice have been explored, from which the molecular basis of MMAF was determined (Shen et al., 2019; Fang et al., 2019; Xu et al., 2020). Herein, we also used a proteomics approach to reveal the impact of truncation and overexpression of Fsip2 on testis protein expression. GSEA analysis indicated that the 22 proteins associated with the fibrous sheath and 15 proteins associated with the mitochondrial sheath were simultaneously enriched in the downregulated proteins of KI mice and the upregulated proteins of OE mice. The 11 proteins associated with ODF also showed a similar tendency, although the enrichments were not significant (Fig. S3; q>0.05). According to the literature, five of these proteins (Akap4, Akap3, Cabyr, Gapdhs and Odf2) are associated with the sperm tail, and their complete disruption significantly impairs the development of sperm flagella or sperm motility in mouse models (Xu et al., 2020; Miki et al., 2002, 2004; Donkor et al., 2004; Young et al., 2016). Our western blot analysis confirmed the upregulation and downregulation of Akap4, Akap3, Cabyr and Gapdhs in the testes of OE and KI mice, whereas Odf2 did not exhibit significantly increased expression in OE mice.

Previous findings using sperm immunofluorescence staining corroborate the absence of FSIP2 in the principal piece of sperm from a patient with MMAF (Liu et al., 2019a). Herein, also using immunofluorescence staining, we showed that Fsip2 was expressed in the principal piece, MP and sperm head. Duolink PLA further confirmed the interaction between Fsip2 and Acrv1 in the acrosome. Additionally, expression of Acrv1 was significantly increased and decreased in OE and KI mice, respectively. These results, together with the findings that upregulated genes of round spermatids in OE mice were enriched in the acrosomal vesicle and that the acrosomal vesicle-associated proteins were reciprocally affected by the dosage alterations of Fsip2, collectively suggest a fascinating role of Fsip2 in the acrosome.

A full understanding of the sequence characteristics of human FSIP2 has proved elusive. The use of a conventional protein domain prediction tool was unable to provide sufficient information because only three of the FSIP2 domains were recognized (Martinez et al., 2018). Here, using a motif discovery tool, we observed conserved motifs of Fsip2 in eight species. Intriguingly, among the identified motifs, the sixth and fifth motifs were only conserved in the species with internal fertilization, suggesting a potential connection between these two motifs and internal fertilization. Coincidentally, a recent study found that the accessory structure (fibrous sheath) of the flagellum enhanced the progressive motion of mammalian sperm in high-viscosity media (Gadêlha and Gaffney, 2019), representative of the internal fertilization environment. Conversely, the migration of sea urchin sperm lacking a fibrous sheath was significantly compromised in the same environment (Gadêlha and Gaffney, 2019). These results indicated that the fibrous sheath is essential for internal fertilization. In addition, a patient with MMAF with a 5463-aa truncation in FSIP2 exhibited an MMAF phenotype similar to those of three other patients in whom FSIP2 was truncated to sequences as short as 304 aa, 536 aa and 761 aa (Martinez et al., 2018). This implies that the region of FSIP2 required for fibrous sheath formation is beyond 5463 aa and located in the C-terminal region. Given that fibrous sheaths are unique to sperm of internally fertilized species, it could be inferred that the C-terminal region is indispensable for internal fertilization, in accordance with our conclusions drawn from our motif analysis.

Interactions among Cabyr, Ropn1 and Akap3, and between Akap3 and Akap4 have been validated (Brown et al., 2003; Li et al., 2011). Fsip2 was also found to bind Akap4, probably with the 1040-1705-aa segment (Brown et al., 2003). However, three patients in whom FSIP2 was longer than 2500 aa still presented with typical MMAF phenotypes (Martinez et al., 2018; Liu et al., 2019a). In addition, AKAP4 immunostaining was absent in sperm that carried a truncated FSIP2 of 5463 aa from the patient with MMAF (Martinez et al., 2018). Therefore, we speculate that there are two binding sites for AKAP4 on FSIP2, the first located in the 1040-1705-aa segment and the second residing on the C-terminal region, i.e., the sixth and fifth motifs, respectively, with the second binding site being indispensable for the formation of the fibrous sheath.

We hereby propose a molecular mechanism for MMAF induced by FSIP2 truncated mutations. Under physiological conditions, ROPN1 and CABYR bind AKAP3, which also interacts with AKAP4. AKAP4 binds the two binding sites of FSIP2 simultaneously to form the normal fibrous sheath and sperm flagellum. However, when FSIP2 is truncated because of loss-of-function mutations, the second binding site is lost, which significantly impairs the interaction with AKAP4, leading to the MMAF phenotypes.

In summary, our experiments and analyses based on mouse models with hypermorphic and hypomorphic alleles indicated dual crucial roles for Fsip2 in sperm tail and acrosome formation. The molecular mechanism underlying the MMAF phenotypes caused by mutations in FSIP2 was also elucidated. Our findings, together with previously reported recurring amplifications of FSIP2 in testicular germ cell tumors (Litchfield et al., 2015), suggest a complex role for Fsip2 in sperm-egg recognition, spermatogenesis and tumorigenesis.

An infertile 28-year-old man, found to have asthenozoospermia at the Center of Reproductive Medicine, Guangzhou Women and Children's Medical Center, was enrolled in this study. Semen samples were collected by masturbation after 2-7 days of sexual abstinence and were examined after liquefaction for 30 min at 37°C. The analysis was repeated three times over 3 months. The study was approved by the ethics committee of Guangzhou Women and Children's Medical Center. Informed consent was obtained from the patient.

Genomic DNA (gDNA) was extracted from the peripheral blood of the patient. WES and data analysis were performed as described previously (Sun et al., 2019). The insertion mutation found by WES was further confirmed by PCR using Q5 High-Fidelity Polymerase (5×, NEB), with the specific primers targeting the selected candidate variant (forward: 5′-AAGGTGAGCCCTAAGGACAA-3′; reverse: 5′-TTTGGCTTACCCGATGAAAT-3′). The PCR products were then validated by Sanger Sequencing.

The animal experiments were approved by the Experimental Animal Ethics Committee of Guangzhou Medical University; all animal experiments were performed in accordance with the guidelines and regulations of the Committee. To mimic the FSIP2 mutation (c.8368_8369insC, p.2790fs) found in the patient with MMAF, an Fsip2-KI mouse line (c.8137insC) was generated by homology-dependent DNA repair using the genome-editing method based on CRISPR-Cas9 technology. All mice used in this study were of a C57BL/6J genetic background. Guide RNA used was 5′-ATCAAGAACAAGTTATCTGCTGG-3′ and donor DNA was 5′-AGCAGTACTAAGACCAAAATCAAGAACAAGTTAagcGCTGGAGAGAAAAcCTCCAAGAGAGAGCAGACCAAAACCGCCCTTGGGCTGCCACAAACTCCAC-3′. The mixture of Cas9 mRNA, gRNA and donor DNA was injected into fertilized eggs with recognized pronuclei from superovulated 6- to 8-week-old female C57BL/6J mice in M2 medium. The injected zygotes were cultured in an environmentally controlled incubator with 5% CO₂ at 37°C, and the blastocysts obtained were then transferred into the uteri of pseudo-pregnant C57BL/6J females. The inserted mutation was confirmed using PCR on the offspring. The genotyping primers used were as follows: forward, 5′-AACTCAGCCCAAAGAACAGCCC-3′; reverse, 5′-TCCGTAGGATAACCTGCACCCA-3′.

Transgenic mice in a C57BL/6J background were generated by using BAC clone ID RP23-221K3 (CloneDB, https://www.ncbi.nlm.nih.gov/clone/669242/), which included Fsip2 and its flanking sequences. The BAC DNA (1 ng/μl in circular form) was injected into fertilized oocytes and integrated into the genome randomly. The integration of the BAC clone into the mouse genome was confirmed using PCR primers at each end of the BAC sequence: 5′ reaction, 290 bp (forward: 5′-CAGCAAGGAAACAATGGTTACAC-3′; reverse: 5′-CCAGAAGTGCAGTCGTAAAAGTC-3′); 3′ reaction, 545 bp (forward: 5′-TCCGCACCCGACATAGATAATAAC-3′; reverse: 5′-CATGACTGCTCTGACAACACAC-3′). The copy number of Fsip2 was determined as previously described (Habib et al., 2018). First, the mouse genomic DNA was extracted using DNeasy Blood and Tissue Kit (Qiagen), according to the manufacturer's instructions. Second, to estimate the copy number of Fsip2, the qPCR reaction mixture was prepared by adding 10 ng of the genomic DNA into PowerUp SYBR Green Master Mix (A25742, Applied Biosystems) containing 10 mM of the following primers: Fsip2 247 bp (forward: 5′-AAATGAGCAAAAGCCAGGGG-3′; reverse: 5′-CTGTTCGGGTGTTTTCTGCA-3′) and Dicer1, 260 bp (forward: 5′-CTGGTGGCTTGAGGACAAGAC-3′; reverse: 5′-AGTGTAGCCTTAGCCATTTGC-3′). The reaction mixture was subjected to real-time qPCR for 40 cycles on the Applied Biosystems System Quant 6 flex.

The ΔΔCT method was used for data analysis. Each assay was performed in triplicate for each sample. Dicer1 was used as the copy number reference (two genomic copies). The copy number for Dicer1 was constant in both WT and OE mice (Habib et al., 2018). The copy number of Fsip2 was two for the WT littermates.

Total RNA was extracted from the testes using TRIzol kit (15596026, Life Technologies), followed by RNA purification using Direct-zol RNA mini prep (R2050, Zymo Research). cDNA was synthesized using the Thermoscript RT-PCR system (1713181, Invitrogen). Real-time qPCR reactions were carried out using two sets of primers, one close to the 5′ end of Fsip2 (forward: 5′-TCACACGATTCCAAAACTGG-3′; reverse: 5′-AAGCGTTGTTCCTCTGCTGT-3′), and the second close to the 3′ end of Fsip2 (forward: 5′-TGATGAGGAGGAGGTTGTCC-3′; reverse: 5′-TTTCAGGTTGCTTGTGCTTG-3′). Gapdh served as the reference (forward: 5′-ATCCAGAATACATGGTTTAC-3′; reverse: 5′-GTTGATCTCAAGGTTGTC-3′).

The fertility of each genotype was tested using adult male mice (8-12 weeks old). Briefly, a male mouse was caged with two WT female mice. Vaginal plugs were checked for each morning; once a vaginal plug was identified (day 1 post coitus), the female with the vaginal plug was placed in a separate cage and the male was allowed to rest for 2 days. Another female was then placed in the cage for another round of mating. The female was deemed not pregnant if it had not generated any pups by day 22 post coitus, and it was euthanized to confirm the lack of pregnancy. At least 20 mice were tested for each gender and genotype.

Cauda epididymal sperm were collected immediately from adult male mice after euthanization and washed with PBS. The sperm were then released by making an incision in the epididymal tissue, placed in Sperm Rinse medium (Vitrolife) and allowed to swim up for 30 min at 37°C. Sperm viability and motility was examined using the Automatic Sperm Cytomorphology Analyzer (CASA) system with phase-contrast microscopy (Microptic S.L.). For morphology assessments, Papanicolaou (BASO) staining of mouse sperm was performed. Briefly, the smears prepared from the epididymal cauda sperms isolated above were fixed in 95% ethanol for 15 min, rehydrated in an ethanol gradient and stained with Papanicolaou staining according to the manufacturer's instructions (DA0191, Leagene Biotechnology).

For SEM, spermatozoa samples were isolated, washed in PBS and fixed in 2.5% glutaraldehyde and 2% paraformaldehyde in 0.15 M sodium phosphate buffer overnight at 4°C. The sperm were then washed in the buffer and collected on Nucleopore filters or glass coverslips, subjected to critical point drying and coated with gold/palladium. Samples were examined in a HITACHI S-3000N scanning electron microscope at 20 kV.

For TEM, spermatozoa samples were fixed under the same conditions as described for SEM, postfixed in 2% osmium tetroxide in cacodylate buffer and embedded in Lowicryl resin. Sections were stained with uranyl acetate and lead citrate, and examined in a Hitachi H-7500 transmission electron microscope at 80 kV.

The cauda epididymis sperm were isolated as described above, fixed with 4% paraformaldehyde (PFA) or 95% ethanol, permeabilized with freshly prepared 0.3% Triton X-100 in PBS, blocked with 10% goat serum for 1 h at 37°C and then incubated with primary antibodies (Table S3) with optimal dilution overnight at 4°C. The slides were allowed to stand at room temperature (RT) for 1 h and were then washed and incubated with secondary antibody Alexa Fluor 488 goat anti-rabbit IgG (H+L) (Table S1) for 1 h at RT. Finally, the slides were washed and mounted using anti-fade mounting media with DAPI. The images were captured at high magnification using a fluorescence microscope with a phase-contrast channel (Leica Microsystems).

To estimate the length of MPs and principal pieces of sperm of OE and WT mice, we used the immunostaining method described above in combination with phase-contrast microscopy. The MP and principal piece were stained with MitoTrack (red) and Akap4 antibody (green), respectively. The measurements were performed using Leica microscopy image processing and quantification tools. At least 100 sperm from each sample were used to estimate the length of the MP and principal piece.

Testes were surgically removed and fixed in modified Davidson's Fluid (mDF) (ServiceBio), dehydrated in ascending grades of ethanol and embedded in paraffin. Then, 3- to 4-μm-thick sections were prepared. For the histological assessments, sections were stained with Hematoxylin and Eosin according to the standard protocol. PAS staining was also conducted on sections from the testis using periodic acid solution and Schiff's reagents (395B, Sigma-Aldrich). The morphology was observed and captured using PANNORAMIC slide scanners (3DHITECH); the images were processed and exported using CaseViewer (version 2.4).

Paraffinized fixed testicular tissue sections were hydrated in descending grades of ethanol, followed by antigen heat activation using citrate buffer (pH 6) for 20 min. The slides were then permeabilized with 0.3% Triton X-100, blocked with 10% goat serum for 1 h at RT and incubated with primary antibodies (Table S3) overnight at 4°C. Next, the slides were allowed to stand for 1 h at RT, followed by two washes with PBS for 5 min, incubated with the secondary antibody (goat anti-rabbit Alexa Fluor 488 or goat anti-mouse Alexa Fluor 568; A11034 or A11004, Invitrogen; Table S3) and mounted with Fluoroshield with DAPI (ab104139, Abcam). Images were acquired using a fluorescence microscope (Leica).

The total protein was extracted from the mice testes. Approximately 50 mg of testes were homogenized on ice with prechilled RIPA lysis buffer containing a 10 mM cocktail protease inhibitor (Roche) and PMSF using a tissue homogenizer, allowed to set for 30 min with occasional shaking, centrifuged at 12,000 RPM (16,000 g) for 15 min, followed by denaturation at 95°C for 10 min with 4× loading buffer (Bio-Rad) containing 2-mercaptoethanol (Sigma-Aldrich). The denatured proteins were separated on SDS-polyacrylamide gels and transferred to a PVDF membrane for immunoblotting analyses. β-actin was used as the loading control.

The protein-protein interaction between Fsip2 and Acrv1 was detected using the Duolink PLA kit (Duolink In Situ Red Starter Kit Mouse/Rabbit, DUO92101, Sigma-Aldrich). The protein-protein interaction was indicated when a red signal with DuoLink PLA was generated (i.e. when the physical distance between the two proteins was less than 20 nm). According to the manufacturer's instructions, the sperm smear slides were fixed and permeabilized, blocked with Duolink Blocking Solution at 37°C in a preheated humidity chamber for 60 min and then incubated overnight at 4°C with antibodies raised from two different species [i.e. rabbit anti-Fsip2 antibody (bs-16187R, Bioss) and mouse anti-Acrv1 antibody (sc-398536, Santa Cruz Biotechnologies)]. Next, the smears were washed and incubated with Duolink PLA Probe PLUS and MINUS diluted in antibody diluent buffer for 1 h at 37°C. Unbound PLA probes were removed with washing buffer A, followed by ligation steps for 30 min at 37°C. The PLA signal was amplified by incubating the slides with amplification solution for 100 min at 37°C. Finally, the slides were washed with buffer B, mounted using Duolink In Situ Mounting Medium with DAPI and captured using a confocal microscope (Leica TCS SP8 Microsystems).

Testes of WT, KI and OE mice were isolated and dissociated into single-cell suspensions as previously described (La Salle et al., 2009). Three mice were used for each genotype. Briefly, the testis was excised and the tunica albuginea was removed. Seminiferous tubules were digested with Collagenase IV (C5138, Sigma-Aldrich), trypsin from bovine pancreas (T9201, Sigma-Aldrich) and DNase I from bovine pancreas (14365, USB), lyophilized, filtered and resuspended. For each sample and prepared libraries, ∼5000 cells were loaded into one channel of the Chromium system according to the protocol of 10x Genomics. The sequencing process was conducted on an Illumina HiseqX machine in a paired-end 150 bp mode.

The Cell Ranger v3.0.0 mkfastq (10x Genomics) was used to demultiplex the cellular barcodes. Adapters, low-quality reads and low-quality bases were removed using Trimmomatic software (Bolger et al., 2014). Basic statistics of the quality of the clean data reads was performed with FastQC.

The clean reads were aligned to the murine reference genome (mm10) using the Cellranger count command. The R package Seurat (3.1.0) (Satija et al., 2015) was applied to convert the unique molecular identifier (UMI) count matrix to Seurat objects. The genes expressed in fewer than ten cells and cells with fewer than 200 expressed genes were removed. The expression data were normalized using the NormalizeData function, in which the UMI counts were divided by the total number of UMIs per cell, multiplied by 10,000 for normalization and then log-transformed. After normalization, 2000 highly variable genes were identified using the FindVariableGenes function. We then used the FindIntegrationAnchors and Integratedata functions to combine the nine samples from WT, KI and OE mice. Principal component analysis (PCA) was performed with the RunPCA function. The top 30 PCA components were used for the RunTSNE function and the FindClusters function. Fourteen clusters were identified for WT, KI and OE mice. The marker genes were found using the FindConservedMarkers function.

Marker genes for major cell types in the murine seminiferous tubule have been proposed previously (Green et al., 2018), which were used to identify the major cell types of clusters 1-14. The eighth, fifth, second and fourth clusters were recognized as elongating spermatids. The tenth, sixth, third and eleventh clusters were recognized as round spermatids. The ninth, twelfth and first clusters, and the seventh cluster, were recognized as spermatocyte and Sertoli cells, respectively. The thirteenth and fourteenth clusters were recognized as spermatogonia and Leydig cells, respectively. Twelve subclusters of WT, KI and OE mice that were recognized as round spermatids were extracted and analyzed. The differential expressed genes between the corresponding subclusters of OE and WT mice and between KI and WT mice were analyzed with the function FindMarkers. GO analysis was performed on the downregulated genes using the GO enrichment analysis tool of Omicshare (https://www.omicshare.com/tools/home/report/goenrich.html). GO terms with FDR-corrected P<0.05 were considered significant.

Tandem mass tags (TMT) proteomics for WT, KI and OE mice was performed in accordance with previously described protocols (Fang et al., 2019). GSEA was conducted using GSEA software v3.1.0 (www.broadinstitute.org/gsea/index.jsp) separately against four protein sets: (1) fibrous sheath-associated proteins (Lehti and Sironen, 2017) (Akap4, Akap3, Gapdhs, Ldhc, Spa17, Fsip1, Fsip2, Ropn1, Ropn1l, Ldha, Pde4a, Tekt2, Fscb, Cabyr, Hk1, Aldoa, Slc25a31, Gapdh, Rhpn2, Gsk3b, Stat1 and Gstm5); (2) mitochondrial sheath-associated proteins (Lehti and Sironen, 2017) (Klc3, Spata19, Tekt5, Prkar2a, Srsf10, Selenop, Cfap157, Vdac2, Ppp1cc, Gpx4, Tssk2, Ak2, Tex22, Nectin2 and Gopc); (3) ODF-associated proteins (Lehti and Sironen, 2017) (Tekt2, Tekt4, Cst8, Vdac2, Vdac3, Cdk5, Odf1, Odf2, Odf3, Uap1 and Ak1); and (4) acrosomal vesicle-associated proteins downloaded from the MSigDB (https://www.gsea-msigdb.org/gsea/msigdb/cards/GOCC_ACROSOMAL_VESICLE).

Protein domains of FSIP2 were predicted by the SMART server (Letunic and Bork, 2018) (http://smart.embl-heidelberg.de/). Conserved motifs were predicted by using the MEME motif discovery tool (Bailey et al., 2009) (http://meme-suite.org/tools/meme). The aa sequence of FSIP2 of H.sapiens (NP_775922.3) and Fsip2 of other seven species [M.musculus (A2ARZ3.3), P.vitticeps (XP_020670029.1), L.chalumnae (XP_014351833.1), A.planci (XP_022093257.1), B.mandarina (XP_028029952.1), M.yessoensis (XP_021368192.1) and S.pistillata (PFX30584.1)] were included. The minimum and maximum widths of each motif were set to be 60 aa and 200 aa, respectively. Each motif had to be recognized in at least two sequences. The visual representation of motifs was generated by TBtools (Chen et al., 2020).

The data from all experiments were expressed as the mean±s.d. Statistical analyses were performed with R package 3.5.1 and Microsoft Excel. P-values were calculated using an unpaired two-tailed t-test. Statistical significance was set at P<0.05.

We are very grateful to the patient who participated in this study. We acknowledge Linglong Huang for his kind help with electron microscopy experiments.

The peer review history is available online at https://journals.biologists.com/dev/article-lookup/doi/10.1242/dev.199216.