2.1 Animal ethics

The germ cells and corresponding somatic cells scRNA-seq datasets used in this study were extracted from our published atlas (publicly available in the Gene Expression Omnibus (GEO) at NCBI under accession number ‘GSE148032’) spanning from E11.5 to PND7 (embryonic day 11.5 to postnatal day 7).¹¹ As previously described, male Blimp1-mVenus and Stella-ECFP (BVSC) transgenic mice were obtained from Mitinori Saitou laboratory (Kyoto, Japan). Female C57BL/6 mice were purchased from the Guangdong Medical Laboratory Animal Centre (Guangzhou, China). Female C57BL/6 mice in natural estrus were mated with male BVSC mice. The germ cells and somatic cells were collected using mVenus and ECFP signals ranging from E11.5 to E18.5. After birth, testicular cells were randomly selected without the assistance of FACS.

Mice were housed according to the ethical guidelines of South Medical University ethics committee (L2016149). Mouse experiments were performed in accordance with the U.S. Public Health Service Policy on Use of Laboratory Animals, and were approved by the Ethics Committee on Use and Care of Animals of Southern Medical University. Mice were kept in a standard 12 h light–dark cycle in the specific-pathogen free conditions, and were permitted for free access to water and food. The ambient temperature was 20–25°C and the humidity was 40–70%. All the mice we used were healthy and immune-normal.

2.2 Immunofluorescence staining

The mouse male gonads were fixed with 4% paraformaldehyde at 4°C overnight, and washed extensively with phosphate-buffered saline (PBS). Then, the tissues were dehydrated and embedded in paraffin, and then cut into sections of 5-μm (genital ridges and testis)/10-μm (embryos) thickness, respectively. Immunostaining was performed after deparaffinization and rehydration as previously described. The primary antibodies used were mouse anti-DDX4 antibody (1:500, Abcam #ab27591), mouse anti-PLZF (ZBTB16) antibody (1:100, Santa Cruz Biotechnology #sc-28,319), rabbit anti-ETV5 antibody (1:100, Novus Biologicals #NBP2-14950), rabbit anti-USF1 antibody (1:100, Abcam # ab180717), rabbit anti-ACY3 antibody (1:100, Abcam # ab197799), rabbit anti-PHGDH antibody (1:100, Abcam # ab211365), rabbit anti-ACY3 antibody (1:200, Abcam # ab197799), rabbit anti-PHGDH antibody (1:400, Abcam # ab211365), rabbit anti-TAGLN2 antibody (1:250, Abcam # ab121146), rabbit anti-PHGDH antibody (1:100, Abcam # ab211365). The secondary antibodies used were Goat Alexa Fluor 488 anti-rabbit IgG (1:500, Jackson ImmunoResearch #111–545-003), Goat Alexa Fluor 594 anti-rabbit IgG (1:500, Jackson ImmunoResearch #111–585-003), Goat Alexa Fluor 488 anti-mouse IgG (1:500, Jackson ImmunoResearch #115–545-003), Goat Alexa Fluor 594 anti-mouse IgG (1:500, Jackson ImmunoResearch #115–585-003). Images were captured with a ZEISS LSM880 confocal microscope.

2.3 Quantification of immunofluorescence

Intensity profiles in each channel along highlighted yellow dotted lines in the merged images are shown (Figure 6D).

2.4 Mouse spermatogonial stem cell culture

Mouse spermatogonial stem cell (mSSC) lines were derived from the P5.5 testes with the genetic background of B6D2F1 (C57BL/6 × DBA). mSSCs were cultured as previously described: StemPro-34 SFM (Invitrogen) supplemented with StemPro supplement (Invitrogen), 1% FBS, N2 (100×), 6 mg/mL D-(1)-glucose, 5 mg/mL BSA, 0.1 mM NEAA, 1 mM sodium pyruvate, 0.1 mM 2-mercaptoethanol, 2 mM L-glutamine, 10⁻⁴ M ascorbic acid, 10 mg/mL biotin, 30 ng/mL β-estradiol, 60 ng/mL progesterone (Sigma), 20 ng/mL mouse EGF, 10 ng/mL human bFGF, 10 ng/mL recombinant rat GDNF (R&D Systems) and 10³ U/mL LIF (Merck Millipore). Cells were passaged every 6 days by dissociating with 0.05% trypsin and changed the medium every 2 days.

2.5 Derivation of Tagln2 knockdown mSSCs

Two shRNAs against Tagln2 were designed using pLKO.1-TRC shRNA system. For lentiviral vector production, HEK293T cells cultured in 10-cm dishes and transfect at 70–80% confluency using Lipofectamine® LTX and PLUS™ Reagent (Invitrogen) in Opti-MEM (Thermo Fisher). After incubation for 48 h in DMEM (containing 10% FBS), the supernatant was filtered and concentrated by ultracentrifugation. Oligonucleotides used in this study were listed in the Data S9. For lentiviral transduction, mSSCs were dissociated into single cells and 2.5 × 10⁴ cells per well were cultured in a 12-well plate. Lentivirus was added after 24 h and the medium was changed after 48 h. Puromycin was used to select transduced mSSCs at a final concentration of 0.4 μg/mL.

2.6 Cell proliferation, cell cycle and cell death analysis

For the cell counting assay, mSSCs were initially plated at 2.5 × 10⁴ cells per well in 12-well plates, with passaging every 6 days, and the total cell numbers in each well were counted from passage 1 to passage 3.

The EdU proliferation assay was performed using the EdU proliferation kit (EdU Flow Cytometry Assay Kit, Yeasen) according to the manufacturer's instructions. The cells were incubated with EdU under normal cell growth conditions and fixed with 4% formaldehyde. After permeabilization, EdU-stained cells were analysed using flow cytometer (CytoFlex, Beckman).

For analysis of cell death in knockdown mSSCs, freshly collected cells were stained with annexin V and PI (Annexin V-FITC/PI Apoptosis Detection Kit, Keygen) and analysed using flow cytometer (CytoFlex, Beckman).

2.7 Quantitative PCR

Total RNA was extracted using TRIzol (TIANGEN, Y1809) Reverse transcription was performed using HiScript III Reverse Transcriptase (Vazyme, R302-01). Quantitative PCR (q-PCR) was conducted with the AceQ q-PCR SYBR Green Master Mix (Vazyme, Q121-02) on a LightCycler+96 Real-Time PCR system (Roche). All data were normalized to the expression of housekeeping gene Gapdh and calculated by ΔCq or ΔΔCq. All q-PCR primer pairs were listed in Data S9.

2.8 Single-cell RNA-seq data processing

Single-cell RNA sequencing raw reads were processed to remove template switch oligo (TSO) sequence and polyA tail sequence. And then the reads with adapter and low-quality bases were removed to obtain clean reads, The clean reads were aligned to the mm10 mouse reference using STAR¹⁹ (version 2.7.1a) with options ‘–outFilterMultimapNmax 3’ and ‘–outFilterMismatchNmax 4’. To count uniquely aligned reads with genes, genomic features were then added to reads in BAM file using featureCounts of subread (version 1.6.4). ‘count’ tool in UMI-tools was used to remove PCR duplicates based on barcode and UMI information. Gene expression level was presented with log(TPM/10 + 1) unless specifically mentioned.

2.9 Estimation of dropout events

For each overlapped gene in germ cells between this dataset and the GSE184708 dataset, the dropout event of each gene was calculated based on the ratio of the number of cells whose gene count was zero in total number of germ cells.

2.10 Dimension reduction and cell-type annotation

Seurat²⁰ package (version 4.4.0) was conducted to perform dimension reduction and clustering analysis. For normal scRNA-seq data, first, raw UMI count matrix was used to create a Seurat object. We normalized the gene expression measurements for each cell with the parameter ‘scale.factor = 100,000’. Subsequently, 4000 highly variable genes were selected to perform PCA on the scaled data. Based on ‘JackStraw’ results, top 15 significant PCs were chosen to construct a KNN graph and run non-linear dimensional reduction (UMAP). ‘FindClusters’ function was used to cluster the cells with the parameter ‘resolution = 0.3’.

Cell types were assigned based on the expression of canonical marker genes and the differential expressed genes (DEGs) of each cluster.

2.11 Cell–cell communication analysis

The R package CellChat²¹ (version 1.5.0) and NicheNet²² (version 2.1.5) were used to characterize cell-to-cell interactions as previously reports.^{23, 24} For CellChat analysis, we divided the seurat object into subsets based on sampling different time-points and computed the communication for each time-point. Following the established workflow, normalized counts were uploaded into CellChat, with standard preprocessing conducted using the identifyOverExpressedGenes, identifyOverExpressedInteractions and projectData functions with specific parameters. Potential ligand-receptor interactions were then calculated using the computeCommunProb, computeCommunProbPathway and aggregateNet functions with standard parameters. For NicheNet analysis, Germ cells were categorised into PGCs, ProSPG and SPG. We defined germ cell as receiver cells, and the somatic cells of the corresponding time-points as sender cells. Nichenet_seuratobj_aggregate function was used to connect the results of Seurat with the analysis of NicheNet.

2.12 Single-cell regulatory network inference and clustering analysis

The single-cell regulatory network inference and clustering (SCENIC)²⁵ pipeline (as implemented in pySCENIC) was performed to infer TFs, gene regulatory networks (GRN) of germ cell development as previously described.^{26, 27} The co-expression network was calculated by runGenie3 and the regulons were identified by RcisTarget. The regulon activity for each cell was scored by AUCell.

2.13 Weighted gene co-expression network analysis

The R package high-dimensional weighted gene co-expression network analysis (hdWGCNA)²⁸ was used to identify pivotal module genes involved in germ cell. First, we employed the setup for WGCNA function to construct an expression matrix from the Seurat object. Subsequently, using the MetacellsByGroups function, we built averaged ‘metacells’ based on cell types (with 20 nearest neighbour cells and a maximum shared cell count of three between two metacells). Following this, we normalized and standardized the metacell matrix through various means. Next, we used the SetDatExpr function to set the expression matrix and the TestSoftPowers function to calculate the topological indicators of the network, assisting in the selection of the optimal soft threshold. Ultimately, the ConstructNetwork function was utilized to establish a co-expression network, and the ModuleEigengenes function was used to compute the ‘eigenvectors’ of the modules to obtain module feature genes.

2.14 Self-organizing map analysis

The kohonen package in R was used to train a self-organizing map (SOM) for 3838 metabolism related genes (downloaded from https://www.gsea-msigdb.org/gsea/msigdb). UMI count was first normalized to TPM and median value was calculated for each type of germ cells to generate a gene-cell type matrix, which is used as the input data for SOM training. The total number of map units was set to 100. Visualization was performed with custom R code.

2.15 RNA-seq library construction and sequencing

Total RNA was extracted from sh-NC and sh-Tagln2 mSSCs using TRIzol Reagent (Ambion). 5 ng total RNA per sample was purified through poly(A) selection and subsequently utilized for library construction using the STRT-seq method with 3′ terminal cDNAs enrichment. Sequencing was performed on NovaSeq-PE150.

2.16 Statistical analysis

The experimental data was statistically analysed using unpaired two-tailed t test to compare differences between different groups with GraphPad Prism.

3.1 A single-cell transcriptome map during fetal PGCs to SPG transition

To investigate the dynamic transcriptome changes and molecular events characterizing the transition from fetal PGCs to SPG, we extracted the germ cells and corresponding somatic cells covering this period from our published atlas spanning from E11.5 to PND7 (embryonic day 11.5 to postnatal day 7).¹¹ Dimensional reduction and clustering analysis were performed to group cells according to the global gene expression signatures (Figure 1A). As expected, the distribution of germ cells aligns along the developmental trajectory, predominantly clustering into six groups, including mitotic PGCs, mitotic arrest PGCs, quiescent ProSPG (Q-ProSPG), transitional ProSPG (T-ProSPG), undifferentiated SPG (Undiff.ed. SPGs) and differentiating SPG ( Diff.ing SPG).^{7, 9, 17} Referring to the traditional naming system^{16, 17, 29} (Table S1), mitotic PGCs correspond to multiplying ProSPG (M-ProSPG); mitotic arrest PGCs and Q-ProSPG constitute primary transitional ProSPG (T1-ProSPG); T-ProSPG includes both intermediate ProSPG (I-ProSPG ) and secondary transitional ProSPG (T2-ProSPG). These germ cell groups exhibited distinct expression patterns of marker genes (Figure 1B), such as Prdm14³⁰ in mitotic PGCs and Nanos2³¹ in mitotic arrest PGCs. As showed in the heatmap, T-ProSPG and Undiff.ed. SPG had considerable overlap in marker gene expression, including Gfra1, alongside the newly identified genes like Hhex and Tagln2.³² In contrast, Diff.ing SPG had elevated expression of differentiation-associated genes, such as Dmrt1.³³ Meanwhile, four somatic cell types were identified as progenitor cells, peritubular myoid cells (PMC) and interstitial progenitors (IP), leydig cells and sertoli cells, respectively. The proportion of these cell types across development was also recorded (Figure S1A).

As demonstrated above, our single-cell transcriptome sequencing dataset provided comprehensive coverage of both time-points and cell types during fetal PGC to SPG. We then assessed the quality of our data and found that, compared to data generated using 10× genomics sequencing, our data from single-tube amplification-based libraries exhibited significant advantages in terms of gene detection rate and dropout rate^{34, 35} (Figure S1B,C), which would be conductive to the following in-depth data mining in this study.

3.2 The transition from PGCs to SPG is characterized by global hypertranscription

Utilizing our high-precision continuous dataset, we found that the number of genes detected in germ cells dynamically changes during development (Figure 1C). Notably, the highest gene expression is detected at the Q-ProSPG, regardless of whether the threshold is ‘detected (log(TPM/10 + 1) >0)’ or ‘highly expressed (log(TPM/10 + 1) >1)’ (Figure 1C). Additionally, the Q-ProSPG also has the highest number of genes detected per single cell (Figure 1D). Next, we analysed DEGs from mitotic PGC to T-ProSPG, and observed a substantial increase in the number of upregulated genes, especially during transition I (from mitotic PGCs to mitotic arrest PGCs) and transition II (from mitotic arrest PGCs to Q-ProSPG) (Figure 1E). GO term analysis revealed that these upregulated genes are highly related to reproductive processes, general developmental processes and biological processes and regulation (Data S1). We then validated our findings using published data from bulk RNA-seq of mouse male germ cells and found that the dynamic changes in transcriptional activity in germ cells were confirmed (Figure 1F and Figure S1D). Furthermore, based on the expression of housekeeping genes and germ cell-specific expressed genes, as well as the detection of genes in somatic cells (Figure S1E,F), we demonstrated that these results were not due to single-cell sequencing technology bias or sampling bias.

Furthermore, we examined the dynamic changes in gene sets associated with several classical biological processes (Figure S1G and Data S1). These distinct patterns demonstrated that our data can sensitively capture the characteristics of the cell types using a specific gene set, primarily based on the relative expression levels and the number of highly expressed genes. Importantly, consistent with the aforementioned findings, the ProSPG consistently captures a greater number of genes.

In conclusion, we observed a prominent change of global transcriptional activity in fetal and perinatal germ cells with day-by-day resolution for the first time. Previous report noted a global hyperactivity of the germline transcriptome at E13.5 and E15.5.³⁶ However, in our study, we found that the hypertranscription in E16.5-P7 germ cells is more pronounced and intense (Figure 1F and Figure S1D).

3.3 Hypertranscription is companied by global regulon dynamics in germ cells

Cell-fate transition is highly associated with the transcriptional state, which always arises from an underlying GRN.²⁵ Within this network, a set of canonical TFs and non-canonical regulators, such as cofactors, enzymes and other regulators, mutually regulate one another along with their downstream target genes. The regulators and their target genes constitute subnetworks, referring to ‘regulons’, are capable of defining cell-fate. Based on the transcriptome dynamics described above, we asked whether hypertranscription in germ cells were driven by TFs and other regulators. To find the candidate regulators governing the transition from fetal PGCs to ProSPG, then to SPG, we performed single-cell regulatory network inference and clustering (SCENIC) analysis to score the activity of regulons.

Based on the SCENIC results, 315 relatively highly active regulons were identified, including 135 canonical TFs and 180 non-TF regulators (Figure 2A and Data S2). To delineate the step-wised regulator regulation landscape, regulon activity dynamics of six types of germ cells were examined (Figure 2B). Eleven modules of regulons with distinct activity changing patterns were classified into two categories among these consecutive types of germ cells. Category I represented regulons were specifically active mainly in a single stage, such as TFAP2C regulon in mitotic PGCs and non-TF regulator TAGLN2 in T-ProSPG. While those were active in multiple inconsecutive stages were involved in category II, such as LHX1 regulon in both mitotic PGCs and Undiff.ed. SPG (Figure 2B). Among these, the expression of two uncharacterized factors, ETV5 and USF1, were tracked by immunostaining (Figure S2).

Transcription is generally driven by the TFs and their associated regulators, as well as the transcription machinery. Through SCENIC analysis, we found that there were the highest number of active regulons in ProSPG, consistent with the hypertranscription in germ cells. Next, we conducted enrichment analysis on these regulators to elucidate their roles (Figure 2C). The active regulators in mitotic arrest PGC, Q-ProSPG and T-ProSPG, which are PGC to SPG transition-related regulators, were more specifically involved in transcription processes compared to PGC-related and SPG-related regulators, including ‘generic transcription pathway’, ‘DNA-templated transcription’, ‘RNA polymerase II pre-transcription events’ and ‘epithelial cell differentiation’ (Figure 2C). While other regulators are involved in ‘germ cell development’, ‘mitotic cell cycle’ and ‘stem cell population maintenance’. These results indicated that hypertranscription in germ cells is driven by the stage-specific active TFs and non-TF regulators, orchestrating the global transcriptome reconfiguration in germ cells.

3.4 Metabolism profiling during the PGC to SPG transition

Cell-fate transition during gonadal germ cell and somatic cell development involves the participation of metabolism-related genes.^{37, 38} However, the expression dynamics of these genes in the context of the PGC to SPG development is hitherto unknown. To this end, metabolism-related genes were clustered by the SOM, and we found that each cell cluster enriched different sets of metabolism-related genes (Figures 3A and S3A and Data S3). Through single-cell transcriptomic profiling, we identified that 2586 out of 3795 metabolism-related genes were highly enriched in germ cells. These genes were subsequently classified into three categories based on co-expression patterns: PGC mitosis-related, PGC to SPG transition-related and SPG-related. Interestingly, the largest number of metabolism-related remains PGC to SPG transition-related, consistent with the aforementioned hypertranscription.

PGC mitosis-related genes were enriched in GO terms (Figure S3B) such as ‘cell cycle’, ‘organophosphate biosynthetic process’, ‘regulation of chromosome organization’, ‘DNA replication’ and ‘cellular responses to stress’, indicating their association with the active mitosis state of PGCs.^{37, 39} Metabolism-related genes involved in the PGC to SPG transition were enriched in GO terms such as ‘monatomic ion transmembrane transport’, ‘transport of small molecules’, ‘monocarboxylic acid metabolic process’, ‘metabolism of lipids’ and ‘biological oxidations’. These metabolic events, primarily occurring in ProSPG, facilitate the transition from PGC to SPG. In contrast, genes highly expressed in SPG were enriched in biological processes such as ‘purine metabolism’, ‘cell cycle, mitotic’, ‘amino acid metabolic process’, ‘glutathione and one-carbon metabolism’ and ‘l-serine biosynthetic process’. These pathways indicated the transition from a quiescent state to active cell cycling, relying on critical metabolic pathways involving purine, amino acids and glutathione.

To validate the metabolism predictions generated by scRNA-seq at the protein level, we selected markers previously unreported in reproductive systems. We examined the expression pattern of PHGDH (Figure 3B,D), which played a critical role in the early steps of L-serine synthesis. Consistent with the mRNA expression pattern, PHGDH was barely detectable in germ cells at E13.5 but highly expressed in ProSPG and SPG, suggesting its role in regulating the quiescent state and re-entry into mitosis. This may also relate to the switch in energy metabolism pathways, with germ cells relying less on aerobic respiration post-colonization into genital ridge.³⁹ Additionally, PHGDH might function as a sensor for 3-phosphoglycerate,⁴⁰ thereby regulating short-term cell-fate. Similarly, we observed the expression pattern of ACY3 in germ cells, which was involved in biological oxidations^{41, 42} (Figure 3C,E).

3.5 Cell–cell interaction analysis during fetal to postnatal stages

Germ cell development and differentiation are fundamental biological processes facilitated by intricate cell–cell interactions and communications within the microenvironment of the gonads. However, the signalling pathways that drive the transition of PGCs into SPG remain to be elucidated. Besides, we were interested in investigating whether intercellular communication contributes to hypertranscription in germ cells. Here, we have reconstructed a day-by-day single-cell molecular interaction map spanning from fetal to postnatal stages.

After colonizing the genital ridge, PGCs and fetal gonadal somatic cells are relatively interspersed.⁴³ After E12.5, male germ cells were surrounded by somatic cells to form cord (Figure 4A). Interestingly, upon observing the communication strength between germ cells and somatic cells, we found that interactions were most intense from E14.5 to P1 (Figure 4B). Integration of the temporal trends of intercellular communication intensity and the number of detected genes in germ cells suggested that intercellular communication signals from somatic cells to germ cells significantly contribute to hypertranscription in germ cells (Figure 4B).

This period from E14.5 to P1 coincides with the formation of cords and the maturation of the basement membrane, indicating that the differentiation of PGCs into SPGs was closely associated with the maturation of testicular structure.⁴⁴ To further investigate how somatic cells participated in germ cell differentiation, we analysed the intercellular interactions between germ cell and somatic cell. Notably, we found that the interaction intensity gradually changed alongside the developmental process, as seen from germ cells↔progenitors interactions to germ cells↔Sertoli cells and finally became germ cells↔PMC and IP interactions (Figure 4C). Next, the detailed day-by-day interaction strength and numbers of sub-pathways involved in ‘cell–cell contact’, ‘extracellular matrix(ECM)-receptor’ and ‘secreted signalling’ were arranged (Figures 4D and S4A and Data S4). Consistent with previous studies,^{11, 45-48} we observed dynamic changes in signalling pathways such as KIT, BMP, TGFβ and Notch during development. Then we summarized the vast and intricate signalling pathways involved in these interactions (Figure 4E). The key signalling pathways include: for Sertoli cells→germ cells, COLLAGEN,⁴⁹ AMH,^{50, 51} GAS (growth arrest-specific gene 6, GAS6),^{52, 53} WNT,^{54, 55} junctional adhesion molecule (JAM),^{56, 57} secreted phosphoprotein 1 (SPP1)⁵⁸ and hedgehog (HH).^{59, 60} For PMC and IP → germ cells, COLLAGEN,⁶¹ LAMININ,⁶² PROS, MK⁶³ and angiopoietin-like protein (ANGPTL). Additionally, cell–cell interaction networks were constructed in a daily resolution (Figure 4F and Figure S4B), and the detail data could be found in Data S4.

3.6 Potential association between gonadal somatic cells and hypertranscription in germ cells

Given that the ascending temporal trends in intercellular communication intensity (initially) and the number of detected genes in germ cells (subsequently) (Figure 1 and Figure 4B), it is reasonable that cell–cell interactions signalling from gonadal somatic cells might participate in the regulation of hypertranscription in germ cells. To test this hypothesis, ligand-target activity was predicted using NicheNet²² (Figure 5A and Data S5). The regulatory network connected the ligands in somatic cells and targets in germ cells at the corresponding time-points.

To simplify analysis, we combined mitotic PGCs and mitotic arrest PGCs as PGCs, combined Q-ProSPG and T-ProSPG as ProSPG, and combined undiff.ed. SPG and diff.ing SPG as SPG. Then, comparisons of ligand-target activities between ProSPG versus PGCs and SPG were conducted, respectively (Figure 5A and Data S5). These two profiles consistently showed that ProSPG exhibited the strongest ligand-target activity. Meanwhile, targets regulated by somatic cell-derived ligands were notably abundant, and the average regulatory potential per target was higher in ProSPG. These findings suggested a significant role of gonadal somatic cells in driving hypertranscription in germ cells.

Next, we examined the regulatory status of 315 regulators uncovered above (Figure 5B). The prioritized ligand-target matrix highlighted the top 20 regulators relevant to the transition from PGC to SPG (Figure 5C). For instance, Tagln2 was found to be regulated by members of the TGF-beta superfamily (Tgfb1 and Inhba), the transmembrane receptor protein serine/threonine kinase signalling pathway (Nog), and the nerve growth factor family (Bdnf) from somatic cells (Figure 5C). Thus, the intercellular interaction between gonadal somatic cell and germ cell might also involve in the regulation of hypertranscription in germ cells.

3.7 TAGLN2 is involved in SSC maintenance and differentiation

As above mentioned, key non-TF regulators were also identified in SCENIC analysis, with an indirect or collaborative effect in the regulation networks. Among these 180 non-TF regulators, Tagln2 caught our attention for its high regulon activity and potential roles in ProSPG to SPG transition (Figure 2B). Then, WGCNA also confirmed that Tagln2 was the fifth hub regulator during ProSPG to SPG transition (Figures 6A,B and S5 and Data S6). Interestingly, Tagln2 was the only non-TF regulator in the top hub genes. Tagln2, encoding a actin-binding protein, is involved in actin filament, and actin cytoskeleton organization during cell invasion, cell proliferation, cell migration and epithelial cell differentiation.^64-67 As a marker of differentiated smooth muscle, we wondered whether Tagln2 played critical roles in germ cell development (Figure 6C,D). To this end, Tagln2-related biological processes were examined during cell development, and the results showed that epithelial cell differentiation and actin organization were highly active in ProSPG and SPG (Figure 6E).

To further explore whether Tagln2 participated in the regulation of survival, maintenance and differentiation of germ cells, we knocked down Tagln2 in mSSCs (Figure 7A,B and Figure S6A). Tagln2 knockdown impaired the mSSC clone formation and decreased cell proliferation (Figure 7C and Figure S6B,C). Meanwhile, knockdown of Tagln2 promoted the cell death of mSSCs (Figure 7D). Genome-wide transcriptome was then performed (Figure S6D). We found that Tagln2 knockdown resulted in low expression of marker genes related to SSC maintenance and pluripotency (Figure 7E), such as Etv4, Etv5, Ret, Gfra1, Lhx1, Neurog3, Bcl6b and Pou5f1.^68-74 Interestingly, Phgdh and Acy3 were also downregulated upon Tagln2 knockdown (Figure 7E), implying the roles related to SSC maintenance of these two metabolism markers. In contrast, genes related to SPG differentiation were aberrantly upregulated, consistent with the fact that mSSC could not be maintained after Tagln2 knockdown. Next, the genes abnormally upregulated after Tagln2 knockdown were enriched in spermatogenesis, meiotic chromosome segregation and extracellular matrix organization, while the abnormally downregulated genes were involved in glycoprotein metabolic processes (Figure 7F and Data S7). Then, signalling pathways related to cell–cell interactions and communications involved in germ cell development were analysed upon Tagln2 knockdown (Figure 4D and Figure 7G). Tagln2 knockdown significantly impaired cell–cell interaction networks, such as ‘VEGF signalling pathway’, ‘Neuroactive ligand-receptor interaction’, ‘Cell adhesion molecules’, ‘ECM-receptor interaction’, ‘ErbB signalling pathway’ and ‘Cytokine-cytokine receptor interaction’. Finally, Tagln2 regulation network was constructed in the context of ProSPG to SPG transition based on these results (Figure 7H and Data S8). Together, TAGLN2 was identified as a key regulator involved in male germ cell survival, maintenance and differentiation.