Uncoupling transcription and translation through miRNA-dependent poly(A) length control in haploid male germ cells

Once synthesized, transcripts undergo extensive post-transcriptional modifications in both the nucleus and cytoplasm (Keene, 2007). In the nucleus, precursor mRNAs are processed into mature mRNAs by removing introns through splicing and adding 5′ caps and 3′ polyadenylated [poly(A)] tails. The poly(A) tail is crucial for nuclear export, stability and translation of mRNAs (Goldstrohm and Wickens, 2008; Nicholson and Pasquinelli, 2019). In eukaryotic somatic cells, most poly(A) tails of cytoplasmic mRNAs are shortened over time through deadenylation (Elkon et al., 2013). The shortening of poly(A) tail leads to reduced translational efficiency and increased degradation. Interestingly, the poly(A) tails of mRNAs can also be lengthened through cytoplasmic polyadenylation in specific cell types, including oocytes, early embryos and neurons (Wormington, 1994; Richter, 1999; Ivshina et al., 2014). In mature oocytes, although mRNAs have shorter poly(A) tails (<20 nt), they are stable and stored in ribonucleoprotein (RNP) particles without being translated (Lim et al., 2016; Morgan et al., 2017). Soon after fertilization, these maternal transcripts are re-activated by cytoplasmic polyadenylation, which lengthens the poly(A) tails up to ∼80-150 nt, followed by efficient translation to produce proteins that are essential for survival and growth of the embryos, from fertilization to zygotic genome activation (two-cell embryo stage and four-cell embryo stage in mice and humans, respectively) (Gohin et al., 2014; Lim et al., 2016). The physiological significance of such a long delay in translation lies in the fact that postfertilization development before zygotic genome activation requires many proteins, which must be synthesized using pre-transcribed and stored maternal transcripts. In neurons, transcribed mRNAs tend to accumulate in the cell body, and these transcripts are sequestered into RNP granules, which travel a long distance and then start translation when reaching the axon terminals (Richter, 2001; Andreassi and Riccio, 2009; Norbury, 2013). Similarly, the translationally repressed mRNAs in neurons tend to have shorter poly(A) tails. Once they reach the synaptic junctions, these transcripts undergo cytoplasmic polyadenylation to lengthen their poly(A) tails, followed by an efficient translation (Curinha et al., 2014; Hilgers, 2015). These findings suggest that poly(A) length control represents an integral mechanism underlying uncoupled transcription and translation.

In addition to oocytes and neurons, haploid male germ cells (spermatids) also display uncoupled transcription and translation (Kashiwabara et al., 2008; Idler and Yan, 2012). As soon as round spermatids start to elongate, transcription is shut down owing to the onset of nuclear condensation. However, from the onset of spermatid elongation (step 9 in mice) to the completion of spermatid differentiation into spermatozoa (step 16 in mice), there are numerous steps through which structurally sound spermatozoa are assembled (Hermo et al., 2010; Idler and Yan, 2012). As transcription ceases upon elongation (step 9), all proteins needed for the remaining steps of sperm assembly (steps 9-16 in mice) have to be produced using transcripts synthesized before the transcriptional cessation, i.e., in round spermatids (steps 1-8) and even in late pachytene spermatocytes. For example, Spata6 mRNAs start to be expressed in late pachytene spermatocytes, and its mRNA expression persists through the entire haploid phase. However, its protein is not detected until the sperm connecting piece/neck starts to assemble in step 9 spermatids (Yuan et al., 2015). Therefore, spermiogenesis, the process through which round spermatids differentiate into elongated spermatids and eventually spermatozoa, represents an excellent in vivo model for studying uncoupling of transcription and translation (Braun, 1998; Bettegowda and Wilkinson, 2010).

Uncoupling of transcription and translation in spermiogenesis is achieved through physical sequestration of mRNAs subjected to translational delay into the RNP granules, which exist as the nuage (also called intramitochondrial cement) in spermatocytes and the chromatoid body in round spermatids (Kotaja and Sassone-Corsi, 2007; Hermo et al., 2010). When spermiogenesis progresses to elongation steps, these mRNAs are gradually released from RNP granules and loaded onto polysomes to translate into the proteins required for sperm assembly (Braun, 1998; Bettegowda and Wilkinson, 2010; Hermo et al., 2010; Idler and Yan, 2012). Recent findings have shed light on the underlying molecular mechanisms. Briefly, our earlier work has revealed that RNP enrichment of mRNAs is a dynamic process, through which the overall length of 3′ untranslated regions (UTRs) becomes increasingly shortened compared with that of polysome-enriched mRNAs when late pachytene spermatocytes develop into the round and elongated spermatids (Zhang et al., 2017). The global 3′ UTR shortening is achieved through continuous shuffling of longer 3′ UTR mRNAs out of RNP granules followed by UPF1-3-mediated selective degradation (Bao et al., 2016) and by targeting shorter 3′ UTR mRNAs into RNP granules (Zhang et al., 2017). In this way, the overall 3′ UTR length of the entire mRNA transcriptome in elongating spermatids becomes shorter and shorter (Zhang et al., 2017). We have also reported data showing that both microRNAs (miRNAs) and m6A modification on pre-mRNAs are involved in the global shortening of transcripts and delayed translation (Tang et al., 2018, 2020). Precisely, proper m6A levels control correct splicing and, consequently, the expected length distribution of transcripts (Tang et al., 2018). Moreover, miRNAs target transcripts with longer 3′ UTRs through binding the distal binding sites to polysomes for translation followed by degradation, whereas transcripts with shorter 3′ UTRs only possess proximal miRNA binding sites, which, once bound by miRNAs, are targeted into RNP granules for stability and translational repression (Zhang et al., 2017).

Previous studies have shown that cytoplasmic poly(A) polymerases and poly(A) binding proteins are essential for spermiogenesis (Kleene et al., 1994; Kashiwabara et al., 2002, 2016; Yanagiya et al., 2010). However, it remains unknown how poly(A) length is regulated during global shortening of 3′ UTRs and dynamic translocation of mRNAs between RNP granules and polysomes during spermiogenesis, owing to technical difficulties in determining the full-length sequences of the poly(A) tails. Although both TAIL-seq and PAL-seq have been developed as the next generation sequencing (NGS)-based methods for determining poly(A) tail sequences (Chang et al., 2014; Subtelny et al., 2014), the short reads of NGS (<300 nt) do not allow for accurate determination of full-length poly(A) sequences, thus compromising the analyses on relationships among 3′ UTR length, poly(A) tail length, exon splicing patterns and translational status. To overcome this hurdle, we developed a sensitive method based on the third-generation PacBio sequencing, which we termed as poly(A)-PacBio sequencing (PAPA-seq), similar to FLAM-seq (Legnini et al., 2019). PAPA-seq can accurately measure poly(A) length with reads covering the entire 3′ ends and the full-length transcripts. Using this method, we determined the dynamic changes in poly(A) length in differentiating spermatids during spermatogenesis. Moreover, we found that miRNAs play an essential role in the regulation of poly(A) length and thus, delayed translation. For the first time, our data demonstrate a crucial role of miRNA-dependent poly(A) length control during spermiogenesis.

Given that the poly(A) tail is well-known to affect mRNA stability and translational efficiency (Goldstrohm and Wickens, 2008; Nicholson and Pasquinelli, 2019), we set out to measure the poly(A) length in spermatogenic cells using PAPA-seq (Fig. 1A,B), a sensitive method similar to FLAM-seq (Legnini et al., 2019). To construct libraries for PAPA-seq, the poly(U) polymerase was used to incorporate a limited number of G and I nucleotides to the 3′ end of poly(A)-selected RNA, and the reverse transcription was then performed to generate cDNAs containing the full-length poly(A) tails followed by sequencing using the PacBio system (Fig. 1B). Spike-in RNAs were sequenced to cross-validate the PAPA-seq data (Fig. S1). Using a modified STA-PUT method (Zhang et al., 2017), pachytene spermatocytes, round and elongating spermatids were purified from adult mouse testes with purities of 90%, 90% and 75%, respectively (Fig. 1A; Fig. S2).

Using PAPA-seq, we first examined the UTR length during spermatogenesis. Consistent with our previous report using short-reads sequencing (Zhang et al., 2017), PAPA-seq revealed that the 3′ UTRs of mRNAs were progressively shortened when pachytene spermatocytes developed into the round and elongating spermatids (Fig. S3A). A similar trend was also observed in the 5′ UTRs, but to a lesser extent (Fig. S3B). Supporting our previous finding that m6A-dependent splicing activities increase with the progression of spermiogenesis (Tang et al., 2018), further analyses of the PAPA-seq data also revealed increased splicing events (alternative exon, exon skipping etc.) when pachytene spermatocytes developed into the round and then elongating spermatids (Fig. S4A). The global shortening of 3′ UTRs might be related to enhancement of translational efficiency because shorter 3′ UTRs provide fewer binding sites for regulatory factors, including RNA-binding proteins and small non-coding RNAs (sncRNAs) (Jia et al., 2013; Bao et al., 2016; Zhang et al., 2017).

Although poly(A) length has long been known to influence mRNA stability and translational efficiency (Goldstrohm and Wickens, 2008; Nicholson and Pasquinelli, 2019), it remains unknown how the poly(A) length is regulated and whether the poly(A) length control is involved in uncoupling of transcription and translation during spermiogenesis. Despite the progressively shortened 3′ UTRs (Fig. S3) and decreased overall isoform transcript length (Fig. 1C), we found that the poly(A) tail length was dynamically regulated in a biphasic fashion from pachytene spermatocytes to round and elongating spermatids (Fig. 1D). The poly(A) tail length first increased from pachytene spermatocytes to round spermatids, which may contribute to the longer half-life of mRNAs that are pre-protected for delayed translation in late spermiogenesis (Phase I, Fig. 1D; Fig. S5). In contrast, from round to elongating spermatids, the poly(A) length gradually decreased (Phase II, Fig. 1D), suggesting that the shortening of the transcript poly(A) tails coincides with the delayed translation progressing. A good agreement was also found when comparing this result with those in the previous reports showing that shortening of the poly(A) tails correlates with translational activation in spermiogenesis (Kleene et al., 1984; Kleene, 1989). For example, transcript isoforms of Spata6 mRNAs start to be expressed in late pachytene spermatocytes, and more isoforms continue to be expressed through the entire haploid phase. However, its protein is only expressed in the developing connecting piece in elongating (steps 9-12) and elongated (steps 13-16) spermatids (Yuan et al., 2015). We observed that the poly(A) length of Spata6 drastically increased from pachytene spermatocytes to elongating spermatids, and this increase in poly(A) length occurred before the peak of SPATA6 protein expression in elongated spermatids (Fig. 1E). Taken together, our PAPA-seq analyses revealed that mRNAs in round spermatids owned the longest poly(A) tails compared with pachytene spermatocytes and elongating/elongated spermatids despite the global shortening trend in overall length of the isoform transcripts. The increased poly(A) length synchronized with the delayed translation in spermiogenesis.

To further explore the effects of poly(A) length on mRNA translational repression and activation, we fractionated the cytosol of the three types of spermatogenic cells into RNP, mono- and polysome fractions using sucrose gradient centrifugation followed by PAPA-seq. By measuring OD₂₅₄, three fractions were observed: RNPs (the nuage/intramitochondrial cement in pachytene spermatocytes and the chromatoid body in round spermatids), monoribosome and polyribosome fractions (Fig. 2A). The purity of the fractions collected was validated through analyzing marker mRNAs well known for their distributions in RNP/polysome fractions (Fig. S6). Transcripts in the polysome fractions actively undergo translation, whereas those in the RNP fractions are translationally suppressed (Braun, 1998; Iguchi et al., 2006; Bettegowda and Wilkinson, 2010; Idler and Yan, 2012). Interestingly, through PAPA-seq, we found that the average poly(A) length of RNP-enriched mRNAs was only 1/40 of those enriched in the polysome fractions (Fig. 2B), suggesting a strong association between deadenylation of mRNAs and their localization to RNPs. This observation raises a crucial question: how are the mRNAs selected for deadenylation followed by sequestration in RNPs? Previous studies have shown that sncRNAs are enriched in RNPs (Zhang et al., 2017) and that argonaute proteins [e.g. AGO2, MIWI (PIWIL1) and MIWI2 (PIWIL4)] are abundant in the chromatoid body (Kotaja et al., 2006), suggesting a relationship between sncRNAs and deadenylation events of their target mRNAs. Following these clues, we hypothesized that the enhanced deadenylation of transcripts that are subjected to translational suppression in the RNP fractions during spermiogenesis might involve sncRNAs. To substantiate this hypothesis, we analyzed the miRNA-mRNA target relationship in RNPs and polysomes. We found that the miRNA targeting sites in the RNP-enriched transcripts appeared to be much more concentrated toward the 3′ ends compared with those in polysome-enriched transcripts in pachytene spermatocytes (Fig. 2C). When spermatogenesis progresses to round and elongating spermatid stages, the miRNA targeting sites at 3′ UTRs of the target transcripts gradually decreased and eventually disappeared, coinciding with the release of mRNAs from the chromatoid body/RNP in elongated spermatids (Fig. S7). The correlations among dynamic changes in miRNA targeting sites, mRNA sequestration and deadenylation suggest the involvement of miRNAs in these processes. This hypothesis is in agreement with previous reports, showing that miRNAs drive mRNAs into RNPs (Zhang et al., 2017) for deadenylation (Chen et al., 2009; Eulalio et al., 2009; Fabian et al., 2011; Eichhorn et al., 2016).

In somatic cells, miRNA-based deadenylation generally triggers the targeted mRNA decay (Chen et al., 2009). In contrast, the nuage and chromatoid body, which represent the RNP in spermatogenic cells functions to store transcripts (Peruquetti, 2015) and the fate of the deadenylated transcripts stored in RNP is not limited to decay. Further analyses of alternative polyadenylation site (APS) revealed that the vast majority of the RNP transcripts contain the APS motifs in the distal end of the 3′ UTRs, suggesting that these transcripts are not random degradation products (Fig. S8). To address the fate of deadenylated transcripts, we compared the poly(A) tail length distributions between RNP and polysome fractions in the three types of spermatogenic cells. Surprisingly, increased polyadenylation could be detected in both RNP and polysome fractions in round spermatids (Fig. S9). The presence of longer polyadenylated tails in polysome fractions could mean extended stability during translation, but how can we explain the longer polyadenylated transcripts in RNP fractions? These RNP-bound transcripts with longer poly(A) tails may function to move transcripts out of RNPs and reach the polysomes, thus switching from translational suppression to active translation. To test this hypothesis, we selected all of the transcripts with longer poly(A) tails in RNPs [>50 nt poly(A)] and examined their expression during spermiogenesis (Fig. 2D; Fig. S10). Indeed, levels of these transcripts decreased in RNP fractions but increased in polysome fractions from pachytene spermatocytes to round and elongating spermatids (Fig. 2D, framed by black dashed lines). In contrast, the transcripts with shorter poly(A) tails in RNPs (<5 bp) remained in the RNP fractions translationally repressed (Fig. 2E). A similar phenomenon was also observed in polyadenylated transcripts in round spermatid RNP fractions, in which the transcripts with longer poly(A) tails (>50 nt) moved out of the RNP fractions into polysome fractions, whereas those with shorter poly(A) tails (<5 nt) stayed in RNPs (Fig. S11). We also found that part of polyadenylated transcripts in RNPs quickly degraded in both RNP and polysome fractions (Fig. 2D, framed by blue dashed lines). These data may suggest that polyadenylated transcripts in the RNPs are either loaded onto the polysomes for active translation or subjected to degradation after translation (Fig. 2D; Fig. S11). However, it is difficult to provide the direct evidence to show the transcripts shifting between RNP and polysome. GO term enrichment analyses revealed that those polyadenylated transcripts encode proteins crucial for sperm assembly and sperm function, e.g., flagella development, acrosome assembly, motility and sperm-egg recognition, etc. (Fig. 2F). For example, Spata6 is expressed as multiple isoforms during spermiogenesis and functions to assemble the sperm connecting piece/neck (Yuan et al., 2015). Once Spata6 transcripts in the RNP phase gained the new long poly(A) tails [indicating the transcripts with re-polyadenylation or elongated poly(A) compared with the previous development stage) their levels dramatically decreased (Fig. 2G, left panel), whereas Spata6 levels in polysome fractions were upregulated with a gradual increase in poly(A) tail length (Fig. 2G, right panel). Another classic example of delayed translation is protamine mRNAs (Prm1 and Prm2) (Fig. S12). However, the poly(A) tails were cropped (50 bp) after releasing from RNP in the most abundant Prm isoform, suggesting that the suitable poly(A) length was required for delayed translation (Kleene, 1989). More interestingly, examination of the terminal sequences of all of the RNP-enriched transcripts revealed that the transcripts without new poly(A) tails (indicating the transcripts without polyadenylation compared with the previous development stage) displayed ∼10 times higher uridine content than the transcripts with new poly(A) tails [transcripts with re-polyadenylation or elongated poly(A) compared with the previous development stage] (Fig. S13D), suggesting a potential function of the uridine-rich motif in polyadenylation. Consistently, our ATAC-seq data (Tang et al., 2018) also show that most of the loci containing those re-polyadenylated mRNAs are already condensed in the pachytene and round spermatids (Fig. 2H; Fig. S14). Taken together, our data suggested that sncRNAs, especially miRNAs, likely mediate deadenylation of transcripts to be sequestered in RNPs, and the mRNAs with shorter poly(A) tails can be sequestered in RNPs.

To validate a role of miRNAs in poly(A) length control, we employed the Drosha conditional knockout (cKO) mouse model. Drosha encodes a nuclear RNase III enzyme essential for pre-miRNA cleavage (Han et al., 2006), and our previous study has shown that inactivation of Drosha exclusively in the spermatogenic cell linage through a cKO approach can abolish miRNA production in all developing male germ cells (Wu et al., 2012). Although Drosha cKO testes contain fewer spermatogenic cells due to germ cell depletion, there are still some pachytene spermatocytes, and round spermatids remained in the seminiferous epithelium (Wu et al., 2012), which we purified and pooled for PAPA-seq analyses.

To test whether miRNAs deadenylate and sequester mRNAs into RNPs, we examined the effects of miRNA deficiency on the poly(A) length of their target mRNAs in Drosha-null spermatogenic cells. RNA contents in the RNP fractions of the Drosha-null pachytene spermatocytes and round spermatids were significantly lower than those in the wild-type controls (Figs 2A and 3A), suggesting that these transcripts might fail to accumulate in the RNP granules in the Drosha cKO male germ cells. Moreover, the ratio of gene numbers in RNP versus polysome fractions in Drosha-null spermatogenic cells reduced by >50% when compared with that in wild-type spermatogenic cells (Fig. 3B), further suggesting that in Drosha-null male germ cells, the transcripts failed to be compartmentalized to RNP granules in the absence of miRNAs. The fact that the mRNA levels in RNP fractions of Drosha cKO cells were drastically reduced compared with those in RNP fractions of wild-type cells (Fig. 3C) also supports this possibility, because the transcripts that failed to be localized to RNPs undergo massive degradation, leading to drastically reduced RNA levels. The results agree with the phenotype observed in the Drosha cKO mice (Wu et al., 2012), suggesting that ablation of miRNAs disturbs the delayed translation in spermiogenesis. Further analyses of the poly(A) length in different cell types and cytosolic fractions uncovered that the average poly(A) length of RNP-enriched transcripts in Drosha-null male germ cells was significantly longer than those in wild-type cells (Fig. 3D; Fig. S15), suggesting the relationship between miRNA and poly(A) reduction. The increased average poly(A) length in RNP-enriched transcripts in the Drosha-null germ cells most likely resulted from relative enrichment of longer poly(A) transcripts due to failed compartmentalization of the transcripts with trimmed poly(A) tails into the RNP granules in the absence of miRNAs. To further support this notion, we examined the dynamic poly(A) change between RNPs and polysomes in the wild-type and the Drosha-null round spermatids (Fig. 3E). We found that these Drosha-null transcripts failed to be sequestered into RNPs, and appeared to be stuck in the polysome fractions with very long poly(A) tails (Fig. 3E). It is also noteworthy that partial RNP-enriched transcripts in the wild-type round spermatids possessed much shorter poly(A) tails, and the same set of transcripts were mostly stuck in Drosha-null RNPs with much longer poly(A) tails (Fig. 3E, framed by black dashed lines). The miRNA influence on transcript deadenylation and translation status is possible. For example, in wild-type round spermatids, the multiple isoforms of Spata6 were characterized by longer poly(A) in polysome and shorter poly(A) in RNP fractions (Fig. 3F). In sharp contrast, most of these transcripts were degraded in Drosha cKO cells, likely owing to mRNA degradation in the absence of miRNAs (Fig. 3F). As expected, a few transcripts with much longer poly(A) tails remained in the polysome fractions in Drosha cKO cells (Fig. 3F). Taken together, our data suggest that miRNA depletion might cause their targets to degrade and fail to compartmentalize RNAs into RNPs with poly(A) tail deduction.

While the overall changes in poly(A) length in Drosha-null male germ cells potentially support our hypothesis, it would be more convincing to demonstrate similar effects of depletion of specific miRNAs on the poly(A) length of their target mRNAs. Using the miR-506 knockout (KO) mouse line that we generated previously (Wang et al., 2020b), the effects of ablation of 18 miRNAs on their target mRNA, Fmr1, were investigated with an emphasis on poly(A) length and translational status. The miR-506 family contains 21 miRNAs transcribed from five large miRNA clusters encompassing a ∼62 kb region and a ∼22 kb region near Slitrk2 and Fmr1, respectively, on the X chromosome, most of which are preferentially expressed in the testis (Wang et al., 2020b). The KO line used in this study lacks 18 out of 21 miRNAs of the miR-506 family (Wang et al., 2020b). Fmr1 has been validated using both western blots and luciferase assays (Ramaiah et al., 2019; Wang et al., 2020b) (Fig. 4A). Fmr1 and the miRNA-506 family were both expressed in RNP/polysome fractions of spermatogenic cells (Fig. S16). We first analyzed Fmr1 mRNA levels using a semi-quantitative PCR and found no changes in Fmr1 levels in KO testes (Fig. 4B; Fig. S17), indicating that miR-506 did not degrade its target Fmr1. As shown in our previous report (Wang et al., 2020b), FMRP levels were reduced by ∼50% in the miR-506 family KO testes compared with the wild-type controls (Fig. 4B, middle panel). To unveil the differential patterns between mRNA and protein, we further examined the poly(A) length in the miR-506 family KO and wild-type testes using a poly(A) length PCR assay, as described previously (Kusov et al., 2001). Interestingly, the average poly(A) length of the Fmr1 mRNAs appeared to be doubled in the miR-506 KO testes (Fig. 4D, left panel), suggesting that the poly(A) length indeed affects protein translation efficiency. The poly(A) pattern of the miR-506 family KO mice was consistent with that of the Drosha cKO mice, supporting the suggestion that miRNAs in the testis may function to trim poly(A) tails to achieve RNP compartmentalization and delayed translation. Based on our proposed model, the Fmr1 mRNA would not be able to be deadenylated and sequestered into RNP granules in the absence of miR-506 family miRNAs. Indeed, our data showed that Fmr1 mRNAs levels decreased by ∼10% in RNP granules (Fig. 4C). Moreover, the poly(A) tails of Fmr1 mRNAs in RNPs are much longer in the miR-506 family KO testes than those in the wild-type testes (Fig. 4D, right panel), and no significant difference were observed in the polysome fractions (Fig. 4D). These findings are generally consistent with other genes (e.g. Skp2 and Snai2) found in the miR-506 KO testes (Fig. 4D). Overall, these results further support our hypothesis that miRNAs shorten the poly(A) length of their target mRNAs through deadenylation to sequester them in RNP granules for delayed translation.

Uncoupling of transcription and translation is prominent during spermiogenesis (round spermatid differentiation into spermatozoa) (Braun, 1998; Bettegowda and Wilkinson, 2010; Idler and Yan, 2012), oogenesis (maternal transcript production) (Bettegowda and Smith, 2007; Sha et al., 2019), preimplantation embryonic development (protein production before zygotic genome activation) (Cui and Kim, 2007; Gohin et al., 2014; Lim et al., 2016) and neuronal cell functions (mRNA synthesis in the cell body and translation in the axon) (Richter, 2001; Andreassi and Riccio, 2009; Norbury, 2013). Several potential mechanisms have been identified to achieve uncoupled transcription and translation, including physical sequestration of mRNAs and proteins in RNP granules (Braun, 1998; Bettegowda and Wilkinson, 2010; Zhang et al., 2017), 3′ UTR length control through alternative polyadenylation (Di Giammartino et al., 2011; Mayr, 2017) and 3′ UTR length-dependent selective decay of transcripts by UPF proteins (Boehm et al., 2014; Bao et al., 2016). The poly(A) length has long been known to regulate transcript stability and translational efficiency (Goldstrohm and Wickens, 2008; Nicholson and Pasquinelli, 2019). However, investigations into the role of poly(A) length control have just begun to emerge (Morgan et al., 2017; Legnini et al., 2019; Liu et al., 2019; Nicholson and Pasquinelli, 2019). This is primarily due to a lack of sensitive methodologies that allow for accurate determination of the full-length poly(A) tail sequences. The third-generation deep sequencing technologies, e.g., PacBio and Nanopore sequencing, allowed us to develop a sensitive method, which we termed PAPA-seq, to determine the full-length sequences of not only poly(A) tails but also the rest of the entire transcripts. There are several design differences between FLAM-seq (Legnini et al., 2019) and PAPA-seq, although the overall concept is similar. First, FLAM-seq was tailed by G and I bases using the USB poly(A) length assay kit (764551KT, Thermo Fisher Scientific), which is proprietary. In our protocol, we demonstrated that the GI tail could be effectively added by poly(U) polymerase (M0337, New England Biolabs) in the precise GI mixing concentration. Second, we tested many different oligos and identified the one with the highest efficacy in batch testing. These oligos could generate fewer dimers in the PCR. Lastly, the oligo modification is different. Specifically, FLAM-seq used iCiGiCAAGCAGTGGTATCAACGCAGAGTACATrGrGrG. We used the LNA modification in the 3′ end to increase the template switching activity. Using PAPA-seq, we discovered that the average length of the poly(A) tails is ∼100 nt. However, it can be as long as 1000 nt in some transcripts in the three spermatogenic cell types (pachytene spermatocytes, and the round and elongating spermatids). The poly(A) tails are much longer in round and elongating spermatids than those in pachytene spermatocytes. This pattern aligns well with the highest number of transcripts subjected to delayed translation in round and elongating spermatids, compared with pachytene spermatocytes. As longer poly(A) tails tend to have enhanced stability and translational efficiency, the peak of poly(A) lengthening in round and elongating spermatids may reflect the peak of delayed translation of those pre-synthesized, RNP-enriched transcripts.

Previous reports have shown that poly(A) binding proteins (PABPs) interact with the deadenylation complex (CCR4-NOT-Tob and PAN2-PAN3) to cause increased mRNA decay and repressed translation by shortening their poly(A) tails (Flamand et al., 2016; Yi et al., 2018). However, this mechanism targets all transcripts, causing massive degradation, whereas in developing spermatids with uncoupled transcription and translation, deadenylation occurs in the transcripts to be sequestered into the RNP granules. The selective deadenylation must be mediated by a factor with sequence specificity, e.g., miRNAs. Indeed, our data strongly support such a role of miRNAs. Therefore, miRNAs appear to play an important role by regulating the poly(A) length through binding their target mRNAs. In spermatogenic cells, miRNAs count for ∼10% of total sncRNAs (Zhang et al., 2017). Other sncRNAs, for example, tsRNA and piRNA, may also be able to sequester mRNAs into RNPs (Vourekas et al., 2012; Kim, 2019). Moreover, the molar ratio of sncRNAs versus large RNAs in RNPs is over 100:1, implying that the excessive amount of sncRNAs may target other regions of mRNAs to translocate mRNAs (Zhang et al., 2017). Therefore, in the absence of miRNAs, the mRNAs with pure poly(A) tails cannot be deadenylated and thus, failed to be compartmentalized into the RNP granules. However, a few transcripts remain associated with RNPs, which may be due to other targeting sncRNAs and/or poly(A) patterns. It is noteworthy that cytosine-enriched poly(A) tails still can manage to be phase-separated into RNP granules because the cytosine-enriched poly(A) tails have much reduced PABP binding affinity when compared with the pure poly(A) tails (Lunde et al., 2007) and are therefore more capable of changing their internal structures to increase hydrophobicity, driving RNP phase separation independent of miRNA-mediated deadenylation. Even if the miRNA depletion caused the massive RNA degradation, some RNAs (Spata6) could preserve the longer poly(A) to maintain the longer stability in order to support the essential protein expression. Overall, the miRNA binding is crucial, but not the only factor to sequester the transcripts into RNPs. Other factors, such as RNA binding proteins, non-A nucleotide distribution in poly(A) tails and other sncRNAs could also affect the mRNA storage in RNPs.

The lack of germ cell depletion in miR-506 KO testes makes this KO mouse line advantageous due to minimal secondary effects. Fmr1 protein FMRP levels decrease by half, whereas its mRNA levels remain unchanged despite their poly(A) tails doubling their length in the absence of 18 Fmr1-targeting miRNAs in the KO testes in vivo. This finding is surprising because it is against the common belief that miRNAs degrade the target mRNAs in somatic cells (Shyu et al., 2008). However, this result does support our proposed model in spermiogenesis. Specifically, in the absence of targeting miRNAs, the mRNAs fail to be deadenylated. Although Fmr1 mRNAs managed to get into RNPs, their translation in elongating and elongated spermatids does not occur normally, leading to a decreased protein level, because of the dysregulated poly(A) tails. This is in sharp contrast to the wild-type situation where Fmr1 with shorter poly(A) tails are stored in RNPs, supporting our notion that miRNAs trigger the deadenylation of their targeted mRNAs, and deadenylation of mRNAs effectively controls the delayed protein translation in spermiogenesis.

The function of the chromatoid body is still a source of debate, but the general consensus is that the chromatoid body determines the transcript fate: degradation or storage for delayed translation (Matsui et al., 2000; Keeling et al., 2006; Kotaja and Sassone-Corsi, 2007; Moser and Fritzler, 2010, 2013; Meikar et al., 2011; Ostareck-Lederer and Ostareck, 2012; Piccolo et al., 2014; Hofweber and Dormann, 2019). Disturbing the chromatid body components caused severe defects in spermiogenesis (Vasileva et al., 2009). However, a definitive link between poly(A) length and the dynamic changes of transcripts between RNP and polysome subcytoplasmic compartments has been challenging to prove. Here, we report a clear association between the poly(A) length and translational status, as reflected by relative enrichment in the RNP and polysome fractions. Our inference is that the RNP-stored transcripts are re-adenylated and shifted to polysomes for delayed translation in later developmental stages, which represents a model for the delayed translation in spermiogenesis. However, the present study is limited by difficulties in distinguishing the early stored transcripts from the nascent RNAs in the re-polyadenylated RNA population, considering transcription remains active in round spermatids. Therefore, it is also possible that the transcripts in RNPs were degraded directly without re-adenylation, and the same mRNAs on the polysomes represent nascent mRNAs. However, the fact that the re-polyadenylated mRNAs in round spermatids display a similar shift from RNP to polysomes in the elongating spermatids, in which transcription has long ceased, suggest that re-polyadenylation occurs to the pre-existing rather than nascent mRNAs. Nevertheless, further experimental evidence showing depolyadenylation and re-polyadenylation on the same transcript is warranted. Of course, the possibility that the polyadenylated mRNAs in RNPs are mainly nascent mRNAs does exist. In summary, here, we have reported evidence supporting miRNA-dependent poly(A) length control in male germ cells, which represents a mechanism underlying uncoupled transcription and translation in the haploid phase of spermatogenesis.

All wild-type and KO mice used in this study were on the C57BL/6J background and housed under specific pathogen-free conditions in a temperature- and humidity-controlled animal facility at the Nantong University (China) and the University of Nevada, Reno (USA). The animal protocols were approved by the Animal Care and Use Committee of Nantong University or University of Nevada, Reno. Male germ cell-specific Drosha conditional KO mice (Drosha^loxp/loxp mice were bred with Stra8-iCre mice) and global KO of X-linked miR-506 family were generated and genotyped at the University of Nevada, Reno as previously described (Wu et al., 2012; Zhang et al., 2017; Wang et al., 2020a,b) (Fig. S18).

Mouse tail or ear snips were lysed in 120 μl of lysis buffer (40 mM NaOH, 0.2 mM EDTA, pH 12) for 30 min∼1 h at 95°C, followed by neutralization with 120 μl of neutralizing buffer containing 40 mM Tris-HCl (pH 5) as previously described (Wang et al., 2020a). PCR reactions were conducted using the GoTaq Green master mix (M7123, Promega). The PCR conditions were as follows: initial denaturation at 95°C for 2 min, followed by 35 cycles of amplification (denaturation at 95°C for 30 s, annealing at 55°C for 30 s, and elongation 72°C for 1 min) with a final elongation at 72°C for 5 min. The primers used for genotyping are listed in Table S1.

Pachytene spermatocytes, round and elongating/elongated spermatids were purified from adult mouse testes using the STA-PUT method (Zhang et al., 2017). The bovine serum albumin (BSA) gradients (0.5-4%) were prepared in the EKRB buffer (K-4002, Sigma-Aldrich), supplemented with sodium bicarbonate (1.26 g/l), L-glutamine (0.29228 g/l), Penicillin and Streptomycin mix (Thermo Fisher Scientific, 10,000 U/l), MEM non-essential amino acids (Thermo Fisher Scientific, 1 ml 100× per l), MEM amino acids (20 ml 50× per l) and cycloheximide (100 ng/ml) (pH 7.2-7.3). Eight testes were pooled each time for cell purification. After being collected and decapsulated, testes were placed into 10 ml of the EKRB buffer containing 5 mg collagenase (Sigma-Aldrich) for 12-min digestion at 32°C to disperse the testicular cells. Once dispersed, the testicular cells were washed three times using the EKRB buffer followed by trypsin digestion by incubation in 10 ml EKR buffer containing trypsin (Sigma-Aldrich, 0.25 mg/ml) and DNase I (Sigma-Aldrich, 20 μg/ml) at 37°C for 12 min with occasional pipetting to facilitate cell dispersion. Thoroughly dispersed testicular cells were washed three times, followed by centrifugation (at 800 g for 10 min) and re-suspension in 10 ml of 0.5% BSA. The cell suspension was passed through 50 μm filters and the filtrate was saved for loading onto the STA-PUT apparatus for sedimentation. After 3 h sedimentation at 4°C, fractions were collected from the bottom of the sedimentation chamber. A total of 30 fractions of 15 ml each were collected. After centrifugation, the supernatants were removed and the cells in each fraction were re-suspended. The cell purity was determined by microscopy examination based on cell morphology, as described previously (Morgan et al., 2019). Fractions containing the same cell types were pooled followed by centrifugation (at 800 g for 10 min) to collect purified pachytene spermatocytes, round spermatids and elongating/elongated spermatids.

We fractionated the purified spermatogenic cells into RNP, monoribosome and polyribosome fractions using a continuous sucrose gradient ultracentrifugation method, as previously described (Zhang et al., 2017). In brief, a continuous sucrose gradient (15%-50%) was prepared by carefully overlaying 15% sucrose onto 50% sucrose followed by diffusing for 3 h at 4°C. The 15% and 50% sucrose solutions were prepared in a lysis buffer [containing 150 mM potassium acetate, 5 mM magnesium acetate, 2 mM DTT, protease inhibitor cocktail (Sigma-Aldrich, 1×), RNase inhibitor cocktail (Sigma-Aldrich, 1×), cycloheximide (100 ng/ml) and 50 mM HEPES, pH 7.5]. Freshly purified pachytene spermatocytes, round spermatids and elongating/elongated spermatids were homogenized in the lysis buffer freshly supplemented with 0.5% Triton X-100 and 0.25 M sucrose. The homogenates were centrifuged at ∼500 g for 15 min at 4°C to remove tissue debris, unbroken cells and nuclei. The supernatant was loaded onto the continuous 15-50% sucrose gradient followed by centrifugation at 150,000 g (35,000 rpm) for 3 h at 4°C. A tiny hole was punched gently at the bottom of the tubes for fraction collection. Twenty-four 500-μl fractions were collected followed by UV spectrometer measurement for OD₂₅₄.

Both sides of epididymis were removed from adult C57/B6 mouse. Cauda were minced in 1 ml of pre-balanced TYH buffer (Easycheck, M2050). The mixture was incubated at 37°C, 5% CO₂ for 10 min. After the incubation, the mixture was transferred to a 15 ml falcon tube with 3 ml pre-warmed TYH buffer at 37°C, 5% CO₂ for 30 min. Swimming-up sperms were then collected by carefully aspirating 1 ml from the mixture followed by 10 min of centrifugation at 10,000 g. After washing twice with PBS, sperm were resuspended in 1 ml of Cellbanker 2 (Zenoaq) and were transferred to −80°C for storage.

Total RNA from all samples (cell sample RIN>8, polysome sample RIN>8, RNP not applicable, sperm sample RIN>3) was centrifuged to discard inhibitor pellet. mRNA was purified using Dynabeads^® mRNA Purification Kit (Life Technology) according to the manufacturer's instruction. mRNA was checked using a 2100 Bioanalyzer RNA picochip to ensure integrity. Qualified mRNA was denatured at 65°C, chilled on ice and then added to GTP-ITP mix (0.5 mM each; TriLink BioTechnologies) and 1× NEBuffer 2 (New England Biolabs), 2 U poly(U) polymerase and 40 U RNase inhibitor, followed by incubation at 37°C for 1 h. The mRNAs tailed with GTP and ITPs were then purified by 1.8× volumes RNA cleanup Ampure beads (A63987, Beckmann Coulter) and eluted in 10.5μl H₂O, followed by reverse transcription. Then 2μl 3' coding DNA sequence (CDS) primer (10 μM) (5′-AAGCAGTGGTATCAACGCAGAGTACNNNNNNCCCCCCCCCCCCTTT-3′) was added into the purified GI-tailed mRNA and incubated at 72°C for 3 min, followed by cooling down to 42°C at the 0.1°C/s speed. Then a master mix containing iso-template switch oligo (5′-AAGCAGTGGTATCAACGCAGAGTACATrGrG+G-3′, where +G indicates locked nucleotide), SSII superscript transcriptase, 1× first strand buffer, DTT, dNTP and RNase inhibitor was added. The reaction was incubated at 42°C for 90 min and stopped by incubating at 70°C for 10 min. The full-length mRNA library reaction was set up with the following reagents: 1× KAPA HiFi master mix, PCR primers (5′-AAGCAGTGGTATCAACGCAGAGT-3′) and 10 μl of synthesized cDNA. The PCR conditions were as follows: initial denaturation at 95°C for 2 min, followed by 15-18 cycles of amplification (denaturation at 98°C for 20 s, annealing at 65°C for 15 s, and elongation at 72°C for 4 min) with a final elongation at 72°C for 7 min. The amplified library was then purified with 1× and 0.4× Ampure XP DNA Beads separately and resuspended in 42 μl H₂O. We diluted 1 μl of the equal mass mixed library 5× and 1 μl was checked on high sensitivity DNA chip (Agilent Technologies).

The purified PCR libraries were submitted to the Genomics core facility of BGI Genomics for PacBio sequencing. Sequencing libraries were prepared using the PacBio Amplicon Template Preparation and Sequencing Protocol (PN 100-081-600) and the SMRTbell Template Prep Kit 1.0-SPv3 according to the manufacturer's guidelines. Sequencing on the Sequel was performed in Diffusion mode using the Sequel Binding and Internal Ctrl Kit 2.0. Every library was sequenced on one SMRT Cells 1M v.2.1 with a 1×1200 min movie. Circular consensus sequence (CCS) reads were generated within the SMRT Link browser 5.0 (minimum full pass of three and minimum predicted accuracy of 90).

RNA from testes [both wild type and the miR-506 family KO (XW)] or fractions (RNP and polysome) were extracted using mirVana™ miRNA Isolation Kit (AM1560, Thermo Fisher Scientific) according to the manufacturer's instructions. To obtain the full-length poly(A) tail, 100 ng of RNA from testes or fractions was used for GI tailing. GI tailing was performed using 2 U poly(U) polymerase (M0337S, New England Biolabs) supplied with a final concentration of 1× first strand buffer (from SMARTScribe™ Reverse Transcriptase, 639538, Takara), 2 U/μl RNase inhibitor (M0314L, New England Biolabs) and GTP (10106399001, Sigma-Aldrich) and ITP (I0879, Sigma-Aldrich) mix (0.5 mM each) in a 10 µl reaction at 37°C for 1 h. Then 1 µl of reverse transcription (RT) primer 1 (10 µM) was added into the GI-tailed mix after GI tailing, and incubated at 72°C for 3 min. When the temperature reached 42°C, 9 µl of RT mix containing a final concentration of 1× first strand buffer, 1 mM DTT, 1 mM dNTP, 2 U/μl RNase inhibitor (M0314L, New England Biolabs), and 2.5 U/μl SMARTScribe Reverse Transcriptase was added into the mix and incubated at 42°C for 60 min, followed by deactivation at 70°C for 10 min.

PCR reactions were conducted using the GoTaq Green master mix (M7123, Promega). The PCR conditions were as follows: initial denaturation at 95°C for 2 min, followed by 35 cycles of amplification (denaturation at 95°C for 30 s, annealing at 55°C for 30 s, and elongation at 72°C for 2 min) with a final elongation at 72°C for 5 min. PCR amplification products were generated using two primers to determine the poly(A) length. A gene-specific forward primer (Fmr1 3′ UTR F2) designed upstream of the polyadenylation site and universal reverse primer (PCR primer 1) was used to generate a product that includes the poly(A) tails of the gene-of-interest. Finally, the PCR products are separated on agarose gels. The poly(A) tail lengths of the gene of interest are the sizes of poly(A) PCR-amplified products minus the calculated length of the gene-specific forward primer to the putative polyadenylation start site. qPCR was performed using Fast SYBR™ Green Master Mix (4385612, Thermo Fisher Scientific). The qPCR conditions were as follows: initial denaturation at 95°C for 10 min, followed by 40 cycles of amplification (denaturation at 95°C for 15 s, annealing and elongation at 60°C for 1 min). The relative RNA expression levels were normalized to Gapdh or Actb. The primers used for PCR and qPCR are listed in Table S2.

Western blot was performed as previously described (Wang et al., 2020b). GAPDH was used as internal control. The antibodies used are as follows: anti-DDX4 antibody (1:800, ab27591, Abcam), anti-β-actin antibody (1:6000, ab27591, Affinity), anti-FMRP antibody (1:800, ab27455, Abcam), anti-GAPDH antibody (1:6000, G9545, Sigma-Aldrich).

First, we used NCBI BLAST (version 2.2.28+ with parameters ‘-outfmt 7-word_size 5’) to map 5′ and 3′ primers to CCS reads, then used in house Perl script to parse standard pairs of 5′ and 3′ primers to CCS reads as the full-length isoform. Next, we trimmed the primer sequence and reported the unique molecular identifier in each full-length isoform. Finally, each isoform was oriented from 5′ to 3′ end (Fig. S19).

We developed a special modified sliding window algorithmic approach to accurately and error-tolerantly detect poly(A) tails. For example, for the poly(A) tail TCGAAATCAAGAAAAACAAAAAA, we listed all the windows without overlap (from 3′ to 5′: AAAAAA, AAAAAC, AAG, AAATC, TCG) and obtained the percentage of A in each window (100%, 83.33%, 66.66%, 60%, 0%), which were defined as the parameter w1. Then using a sliding window starting from the 3′ end, we can get the percentage of A in the total window, defined as the parameter w2. Based on the parameters optimized and benchmarked against a set of manually annotated poly(A) tail estimated from UHRR datasets, the w1 and w2 parameters were set to w1≥50% and w2≥70%. As for the example above, we listed all the sliding regions (AAAAAA w1=100% w2=100%, AAAAACAAAAAA w1=83.33% w2=91.66%, AAGAAAAACAAAAAA w1=66.66% w2=80%, AAATCAAGAAAAACAAAAAA w1=60% w2=80%, TCGAAATCAAGAAAAACAAAAAA w1=0% w2=69.56%), so we can define the poly(A) as AAATCAAGAAAAACAAAAAA.

After poly(A) tails detection and trimming, the remaining fraction of each isoform was mapped to the mm10 genome using GMAP (version 2018-05-30) with parameters ‘-f samse -n 0 --min-intronlength 9 --max-intronlength-middle 500000 --max-intron length-ends 10000 --trim-end-exons 12’. Then, using cDNA_Cupcake (https://github.com/Magdoll/cDNA_Cupcake) python script collapse_isoforms_by_sam.py to collapse all samples isoforms, based on collapsed output, we used in house Perl script to get the isoform expression quantity in each sample. After collapse, nonredundant isoforms were detected using cuffcompare (version v2.0.2) assigned to Ensemble mm10 annotation gene models.

The CDS coding frame and UTR region were predicted using TransDecoder (Oakley et al., 2014) from the nonredundant isoforms. The predicted CDS were further confirmed using NR and Pfam databases.

Computational prediction of miRNA targets is a crucial initial step in identifying miRNA:mRNA target interactions for experimental validation. In order to find possible targets, miRanda and TargetScan were used. The intersection targets with appropriate filter conditions such as MFE scores were taken for further analysis. We used miRanda (John et al., 2004) (with parameters ‘-en -20 -strict’) and TargetScan (Agarwal et al., 2015) (with default parameter) to get the target genes of miRNA, extracted intersection or union of the target genes as a final prediction result.

Both unpaired one-tailed Student's t-test and Wilcoxon rank sum test (a non-parametric or distribution-free test) were used for statistical analyses. The majority of the data followed a lognormal distribution. The Student's t-test was also performed on the logarithm data.