Efficient generation of marmoset primordial germ cell-like cells using induced pluripotent stem cells

Reconstitution of germ cell fate from pluripotent stem cells provides an opportunity to understand the molecular underpinnings of germ cell development. Here, we established robust methods for induced pluripotent stem cell (iPSC) culture in the common marmoset (Callithrix jacchus [cj]), allowing stable propagation in an undifferentiated state. Notably, iPSCs cultured on a feeder layer in the presence of a WNT signaling inhibitor upregulated genes related to ubiquitin-dependent protein catabolic processes and enter a permissive state that enables differentiation into primordial germ cell-like cells (PGCLCs) bearing immunophenotypic and transcriptomic similarities to pre-migratory cjPGCs in vivo. Induction of cjPGCLCs is accompanied by transient upregulation of mesodermal genes, culminating in the establishment of a primate-specific germline transcriptional network. Moreover, cjPGCLCs can be expanded in monolayer while retaining the germline state. Upon co-culture with mouse testicular somatic cells, these cells acquire an early prospermatogonia-like phenotype. Our findings provide a framework for understanding and reconstituting marmoset germ cell development in vitro, thus providing a comparative tool and foundation for a preclinical modeling of human in vitro gametogenesis.


INTRODUCTION
The germline, a lineage that ultimately form the gametes, is the fundamental component of the life cycle in metazoan species, ensuring perpetuation and diversification of the genome across generations. In addition, the germline is the foundation of totipotency, since combination of the gametes at fertilization gives rise to totipotent zygotes that establish all embryonic and extraembryonic lineages necessary for production of a new organism. The germline first arises during early embryonic development as primordial germ cells (PGCs), which subsequently migrate to the developing gonads and ultimately produce either spermatozoa or oocytes through complex and sex-specific developmental pathways 1 . Accordingly, aberrancies associated with PGC development can lead to infertility and a variety of genetic and epigenetic disorders in offspring. Therefore, a precise understanding of how PGCs develop bears significant implications not only for reproductive medicine but also towards a better understanding of a breadth of human diseases.
Although much has been learned from murine genetic studies regarding the cellular dynamics, signaling, genetic and epigenetic requirements accompanying PGC specification 1,2 , the scarcity of germ cells and complexity of their development and cellular interactions has limited deep understanding of transcriptional regulatory networks and epigenetic bases of germ cell development. The last decade, however, has witnessed remarkable progress towards establishing in vitro gametogenesis (IVG) technologies as an alternative approach to study germ cell development. Remarkably, through the stepwise recapitulation and validation of developmental milestones starting with pluripotent stem cells [embryonic stem cells (ESCs) or induced pluripotent stem cells (iPSCs)], the entirety of mouse germline development has been reconstituted in vitro, culminating in the successful generation of fertilization-competent oocytes and spermatozoa, and healthy offspring 3,4 . These landmark studies have been followed by successful development of human induced pluripotent stem cell (iPSC)-based germline reconstitution methods, in which pre-meiotic oogonia and prospermatogonia-like cells generated through PGC-like cells (PGCLCs) bear remarkable transcriptional similarities to in vivo counterparts [5][6][7] .
IVG platforms have provided valuable tools to dissect the transcriptional and epigenetic mechanisms underlying germline specification and subsequent gametogenesis. Recent studies using IVG-derived germ cells or primate embryos in vivo have revealed a substantial divergence in the origin of germ cells and transcriptional networks governing germ cell specification between mice and humans 1 . For example, in mice, core germ cell transcription factors, Prdm14, Blimp1 and Tfap2c, that are deployed by BMP4-induced TBXT, sufficiently establish germ cell fate 8,9 , whereas SOX17 and TFAP2C, deployed by EOMES and GATA2/3 make up the analogous transcriptional network and fate in humans 10,11 . Such divergence between mice and humans necessitates additional layers of caution in direct translation of IVG technologies to human infertility treatment and warrants careful scrutinization and functional validation of IVG-derived gametes in comparison to those developing naturally in vivo. Since ethical and legal constraints make research with human embryos difficult to impossible, IVG studies using model organisms that are phylogenetically close to humans is an important next step. The common marmoset (Callithrix Jacchus) is a new-world monkey that shares many biological characteristics with humans, and thus, has been widely used for biomedical research to bridge the gap between rodent models and clinical translation 12 . Marmoset embryo development, including implantation and formation of fetal membranes, is well conserved with that in humans, serving as a powerful surrogate model for human post-implantation development 13 . Moreover, the relatively short reproductive lifespan, small body size, and reasonable cost for breeding compared to other primates render the marmoset a tractable preclinical model for IVG. In particular, use of marmosets may enable functional validation of resultant IVG-derived gametes by fertilization and embryo transfer as well as vigorous validation of intermediary cellular derivatives by comparing them with their in vivo counterparts 12 .
In this study, we provide a highly efficient method to generate, expand and maintain Callithrix jacchus (cj)PGCLCs from cjiPSCs, which bear immunophenotypic and transcriptional similarities with pre-migratory stage cjPGCs.

Immunohistochemical characterization of pre-migratory cjPGCs
To validate germ cell generation in vitro, we must first have a precise understanding of the molecular features of cjPGCs in vivo. In particular, molecular characterization of early stage endogenous PGCs will be critical to guide the induction of PGCLCs, the first step of IVG, which appears to represent the pre-migratory PGCs in humans 7,14 . However, there is a dearth of information describing primate PGCs at stages before gonad colonization, primarily due to their scarcity. Therefore, we collected marmoset embryos from a triplet pregnancy at embryonic day (E)50 for immunofluorescence (IF) analyses and molecular analyses (Carnegie stage [CS] 11,19 somites, corresponding to ~E8. 5-9.0 in mice) (Fig. 1A, S1A). At this stage, TFAP2C + SOX17 + PDPN + cjPGCs were predominantly localized within the ventral portion of the hindgut endoderm and exhibited round nuclei with generally lower DAPI intensity (Fig. 1B, C). A few scattered cjPGCs were also seen in the adjacent hindgut mesenchyme outside of the basement membranes, suggestive of the initiation of active migration (Fig. 1B). Additional IF analyses revealed that cjPGCs were mostly non-proliferative (i.e., MKI67 -) and co-expressed pluripotency associated markers (e.g., POU5F1, NANOG), but were negative for SOX2 (Fig. 1D, E, S1B). Notably, cjPGCs did not express later germ cell markers (e.g., DDX4 and DAZL), that are typically strongly observed in testicular germ cells (i.e., prospermatogonia) (Fig.   S1C). IF analysis on cjPGCs showed increased global levels of histone H3 lysine 27 trimethylation (H3K27me3) and reduced global levels of histone H3 lysine 9 dimethylation (H3K9me2), consistent with germline epigenetic reprogramming that occurs in mice, cynomolgus monkeys and humans [14][15][16] (Fig. 1F).

Transcriptomes of pre-migratory cjPGCs
Having identified cjPGCs residing in the hindgut endoderm by IF studies, we next set out to determine the transcriptome of endogenous cjPGCs. Given the scarcity of cjPGCs and the lack of reliable surface markers to isolate them, we first enriched cjPGCs by dissecting the posterior portions from two marmoset embryos at E50, followed by trimming of the amnion and yolk sac (Fig. 1A). These tissues were dissociated into single cell suspensions and subjected to high-throughput single-cell RNA-sequencing using 10x  Table 1. Analysis of differentially expressed genes (DEGs) in the cjPGC cluster revealed upregulated expression of potential germ cell specifier/regulator genes (e.g., DND1, KIT, PRDM1, SOX15, SOX17, TFAP2C), pluripotency associated genes (e.g., DPPA3, KLF4, NANOG, POU5F1, TFCP2L1, UTF1, ZFP42), mesoderm/endoderm associated genes (e.g., GATA4, TBXT) and other germ cell related markers (Fig. 2B-E).

Derivation of cjiPSCs through PBMC reprogramming
Our next goal was to derive cjiPSCs, from which germ cells could potentially be induced.
Three cell lines, 20201_6, 20201_7, and 20201_10 were established by reprogramming of peripheral blood mononuclear cells (METHODS). Although hematological chimerism is frequently observed in marmosets 21,22 , whole-exome sequencing confirmed that the established cjiPSCs originated from the intended PBMC donor (ID number, 38189) ( Fig.   S2A, B). CjiPSCs were initially established using conventional on-feeder (OF) culture conditions (see below), but were subsequently switched to feeder-free (FF) culture conditions (PluriSTEM for basal medium and iMatrix-silk for a substrate) for its ease of maintenance. Under these conditions, FF cjiPSCs could be stably maintained over multiple passages (more than 20 passages) when passaged every 4-6 days in the presence of Y27632, a ROCK inhibitor. FF cjiPSCs bore a high nuclear to cytoplasmic ratios, were tightly packed colonies with sharp borders and exhibited flat morphology, each of which are characteristic features of primate primed-state pluripotent cells (Fig. S2C). These cells were mycoplasma-free, exhibited normal 46, XY karyotypes, and uniformly expressed key pluripotency-associated genes ( Fig. S2D-G).
Notably, similar to FF culture, conventional OF cultures also allowed long-term propagation of cjiPSCs. However, OF cjiPSCs tended to differentiate at the center or periphery of colonies 4-5 days after passaging (Fig. S3A, B). Moreover, OF cjiPSCs required clump passaging because single-cell passaging led to inefficient colony formation to be maintained for more than 2 passages (Fig. S3C). Accordingly, OF cjiPSCs exhibited modest upregulation of mesodermal (e.g., T, EOMES, MIXL1) and endodermal genes (e.g., FOXA2, SOX17) compared to those maintained under FF conditions (Fig. S3D). Previous studies showed that inhibition of WNT signaling stabilizes primate iPSC/ESC cultures [23][24][25]  spontaneous differentiation, the effects were not as great as when PluriSTEM was used as a basal medium (Fig. S3B, F). Together, our findings reveal that we have identified an optimal culture protocol in both FF and OF conditions that allows stable propagation of cjiPSCs in an undifferentiated state and with a normal karyotype, thus serving as a foundation for directed differentiation towards the germline.

Generation of primordial germ cell-like cells from cjiPSCs
Our next goal was to derive cjPGCLCs directly from cjiPSCs following the protocol established in humans and cynomolgus monkeys 7,23 . For this, we first treated FF cjiPSCs with a PGCLC induction cocktail (i.e., BMP4, LIF, SCF, EGF, Y27632) in GK15 [GMEM supplemented with 15% KSR] or aRB27 [advanced RPMI1640 and supplemented with 1% B27]) basal medium. Under these conditions, cjiPSCs formed aggregates with a markedly cystic appearance and did not generate SOX17 + TFAP2C + cjPGCLCs (Fig. S4A, B), suggesting that they may not have germline competency. Thus, we next turned our attention to OF cjiPSCs without WNT inhibition given prior success in humans 7 . Remarkably, upon floating culture with a PGCLC induction cocktail in GK15 or aRB27, ~3-4% PDPN + ITGA6 weak+ cells emerged as a distinct population starting at d4 of induction, although the frequency of such cells generally declined after d4 ( Fig. S4B-D). Sectioning of these aggregates at d4 revealed small clusters of PDPN + cells uniformly expressing cjPGC markers (TFAP2C, SOX17, PRDM1, NANOG and POU5F1), which was further confirmed by qPCR (Fig. S4B, E, F).
We posited that the relatively low induction efficiency of cjPGCLCs might be due to their tendency to differentiate under OF conditions. Therefore, we next utilized OF/IWR1 cjiPSCs for cjPGCLCs induction. Upon induction in floating culture, cjiPSCs readily formed tighter and more uniform size/shape aggregates compared to those induced from OF cjiPSCs (Fig. 3A, B). Moreover, under this condition, the induction efficiency of cjPGCLCs was significantly improved, with ~15-40% cells becoming PDPN + ITGA6 weak+ at d4 and d6 of induction ( Fig. 3C, S4G, H). Although variable across experiments, the median yield of PDPN + cells per aggregate was ~600 at d4 and d6, but declined thereafter (Fig. 3C). IF on sections of aggregates at d4 revealed multifocal large clusters of PDPN + cells uniformly expressing key early germ cell markers (e.g., SOX17, TFAP2C, PRDM1, POU5F1, NANOG) (Fig. 3D). Notably, this finding suggests that PDPN can serve as highly specific surface marker of cjPGCLCs that will allow for isolation of cjPGCLCs for downstream analyses. In support, qPCR of isolated PDPN + cjPGCLCs also expressed pluripotency-associated genes (i.e., POU5F1, NANOG), PGC specifier/early marker genes (i.e., SOX17, TFAP2C, PRDM1 and NANOS3) and lacked detectable SOX2 and late germ cell marker (i.e., DDX4, DAZL) (Fig. 3E), features similar to pre-migratory PGCs (Fig. 1D, 2E) 14 . Together these results indicate that our in vitro platform enables highly efficient and reproducible generation of cjPGCLCs.

2D cjPGCLCs expansion culture
We noted that cjPGCLCs induction from cjiPSCs via floating aggregates is somewhat time-consuming and limited in scalability. However, 2D expansion of PGCLCs that retain the cellular and molecular characteristics of PGCs would greatly enhance our ability to generate PGCLCs in a scalable manner that can be utilized, off-the-shelf, for downstream molecular and functional characterization. To accomplish this, we modified a culture method previously utilized to expand human (h)PGCLCs (Fig. 4A) 26 . Specifically, we cultured sorted d6 PDPN + cjPGCLCs on a STO-feeder layer in DK15 medium containing 2.5% FBS, SCF, FGF2 and Forskolin. Plated cjPGCLCs formed loosely arranged clusters, which increased in size and became confluent by expansion culture day (c)10 ( Fig. 4B). These cells expressed markers of early cjPGC/PGCLCs but did not possess late germ cell markers (i.e., DDX4, DAZL), suggesting that they retain the cellular state of cjPGCLCs ( Fig. 4C-E). Moreover, these cells could be passaged approximately every 10 days by dissociation and FACS-sorting of PDPN + cells and exhibited exponential growth at least until at c30 (Fig. 4F, G). Although marker expression pattern was largely unchanged during 30 days of expansion culture, DPPA3 showed modest upregulation, similar to hPGCLCs under expansion culture (Fig. 4H). ITGA6, which is a surface marker weakly expressed on cjPGCLCs also exhibited modest upregulation along the time course (Fig. 4H). These findings highlight the feasibility of 2D expansion culture of cjPGCLCs analogous to hPGCLCs.

Maturation of cjPGCLCs into early prospermatogonia-like state
One of the functional features of PGCLCs is their capacity to further develop into more advanced germ cells 5,6,27,28 . Therefore, we next utilized a xenogeneic reconstituted testis culture that allows hPGCLCs to mature into prospermatogonia to determine if cjPGCLCs could similarly differentiate 5 . After expansion of cjPGCLCs for 30 days by 2D culture, we initiated an xrTestis culture by mixing sorted PDPN + cjPGCLCs with mouse fetal testicular somatic cells depleted of endogenous germ cells (Fig. 5A). After two days of floating culture, xrTestes formed tight aggregates, which were subsequently maintained by air-liquid interface (ALI) cultures (Fig. 5A, B). At day 15 of ALI culture, we observed reconstituted testicular cords surrounded by NR2F2 + interstitial cells in xrTestis cultures ( Fig. 5C). Notably, there were a number of TFAP2C + POU5F1 + NANOG + cjPGCLCs, which primarily localized peripheral to SOX9 + mouse-derived Sertoli cell nuclei (Fig.   5C). In addition, xrTestes maintained until day 30 of ALI culture revealed prominent proliferation of TFAP2C + germ cells, which forced SOX9 + Sertoli cells towards the center of the testicular cords (Fig. 5C). Remarkably, we found a few scattered DAZL + DDX4 + SOX17 + TFAP2C + SOX2cells, suggesting progression into early prospermatogonia ( Fig. 5C) 5 . Together, these data indicate that cjPGCLCs can be integrated in the testicular niche and are capable of further expansion and differentiation.

Transcriptome accompanying formation of cjPGCLCs
We next sought to define gene expression dynamics accompanying specification of cjPGCLCs by bulk RNA-sequencing (Fig. S5A). Unsupervised hierarchical clustering (UHC) classified the cells during cjPGCLCs induction largely into two clusters, one with FF, FF/IWR1 and OF cjiPSCs and the other with cjPGCLCs and OF/IWR1 cjiPSCs, which was also supported by Pierson correlation among clusters (Fig. 6A, B). The relative positioning of cjPGCLC samples in principal component (PC) space supports a step-wise developmental progression during the in vitro culture period (Fig. 6C). First, FF and FF/IWR1 cjiPSCs were intermingled and formed a discrete cluster that was most distinct from cjPGCLCs. There were no significant differences in gene expression between FF and FF/IWR1 cjiPSCs, suggesting that IWR1 does not significantly alter the cellular properties of FF cjiPSCs (Fig. 6C, S5B). Notably, OF and OF/IWR1 cjiPSCs were positioned closer to cjPGCLCs in PC space, with OF/IWR1 cjiPSCs being closest to d2 cjPGCLCs, consistent with their higher competency to differentiate into cjPGCLCs ( Fig. 6C). OF/IWR1 cjiPSCs bore gene expression signatures characteristic of primed-state pluripotency, similar to that seen in FF or OF cjiPSCs (Fig. S5C). Notably, while most key germ cell genes were not significantly upregulated, there is a modest upregulation of TFAP2C and PRDM14 in OF/IWR1 cjiPSCs, which might predispose their high germline competency (Fig. S5C).
Pair-wise comparison of gene expression revealed that genes were primarily upregulated as FF cjiPSCs transitioned to OF and OF/IWR1 cjiPSCs (Fig. S5D). GO terms among the enriched genes in OF and OF/IWR1 cjiPSCs included "protein destabilization" or "ubiquitin-dependent protein catabolic process" (Fig. S5D).
Expression of most of these genes was sustained until d2 cjPGCLCs, suggesting that changes associated with Ubiquitin-Proteasome System (UPS)-mediated protein turnover might confer a permissive cellular environment for cjPGCLCs specification (Fig. S5E, F). Clustering analysis of variably expressed genes across the developmental trajectory revealed 4 large clusters (Fig. 6D). Genes in cluster 1 represented those whose expression is overall high in cjiPSCs and downregulated as they differentiate into cjPGCLCs. Those genes were enriched with GO terms such as "inner cell mass cell proliferation" or "stem cell population maintenance," consistent with their pluripotent nature (Fig. 6D). Genes in cluster 2 were those upregulated along the trajectory and included key germ cell genes (e.g., DND1, NANOS3, PRDM1, SOX17, TFAP2C) and GO terms included "germ cell development." Moreover, GO terms such as "DNA methylation" or "histone methylation" were also seen, consistent with the dynamic epigenetic remodeling observed in developing PGCs. Genes in cluster 3 were those primary upregulated in 2D expansion culture cjPGCLCs and included an enriched GO term, "response to oxidative stress," which might suggest changes associated with culture adaptation. Finally, genes in cluster 4 were those transiently upregulated in d2 cjPGCLCs. Those genes included endoderm and mesoderm markers (e.g., EOMES, HAND1, MESP1, MIXL1, NODAL, SNAL1) and were enriched with GO terms such as "mesoderm formation" and "cellular response to BMP stimulus," suggesting that cjPGCLC induction may accompany transient somatic programs, as previously observed following PGCLC induction in other primates ( Fig. 6D-F) 7,23,29,30 .
We next evaluated the dynamics of gene expression associated with germ cell specification and development. We noted that key germ cell specifier genes (e.g., SOX17, TFAP2C, PRDM1, NANOS3) started to activate and SOX2 was swiftly extinguished in d2 cjPGCLCs (

Global DNA methylation in cjPGCLCs
Previous studies suggested that hPGCLCs showed only modest reduction in global 5mC levels with or without expansion culture, suggesting that these hPGCLCs represent germ cells immediately after specification that have not yet completed global DNA demethylation, a hallmark of mammalian PGC development 1 . Therefore, we next evaluated global 5mC levels in cjPGCLCs by whole genome bisulfite sequencing (WGBS). Similar to hPGCLCs 7,26 , d4 PGCLCs showed a slight but significant reduction in 5mC levels (mean, ~63%) compared with OF/IWR1 cjiPSCs (mean, ~75%) (Fig. 7A, B). Notably, cjPGCLCs in expansion culture exhibited further reduction in 5mC levels, bearing a 5mC level of ~50% at c30 (Fig. 7A, C). Thus, the dynamics of global 5mC levels during cjPGCLC induction and expansion is similar to that of humans 7,26 . To gain further insight into the regulation of global DNA methylation profiles in cjPGCLCs, we evaluated expression dynamics of genes related to DNA methylation. Among de novo DNA methyl transferases, DNMT3B, was highly expressed in cjiPSCs, but exhibited a sharp downregulation upon cjPGCLC induction (Fig. 7D). On the other hand, DNMT3A showed modest downregulation upon cjPGCLC induction, and DNMT3L was expressed only at low levels in all cells examined. Among the genes related to maintenance of DNA methylation, DNMT1 was expressed at a significant level in any of the cells analyzed, whereas UHRF1, which is responsible for the recruitment of DNMT1 into replication foci 18,31,32 , showed a marked reduction upon cjPGCLC induction (Fig. 7D). In cjPGCLC expansion cultures, DNMT3B expression was further downregulated whereas UHRF1 showed slightly higher expression than d4/6 cjPGCLCs (Fig. 7D). Among genes related to active DNA demethylation, TET1 was highly expressed in all cells whereas expression of TET2 and TET3 were very low. Thus, compared to cjiPSCs, cjPGCLCs at d6 or in expansion culture showed reduced but detectable levels of DNMT3B and UHRF1, which might serve as a basis for the modest reduction of global DNA methylation of these cells.
Overall, the expression pattern of epigenetic modifiers, including those related to DNA methylation, is similar to those observed in endogenous cjPGCs at E50 (Fig. 2E), further supporting the notion that cjPGCLCs derived by our protocol accurately resemble endogenous pre-migratory cjPGCs.

DISCUSSION
In contrast to relatively well characterized cj germ cell development at postnatal stage, there is a paucity of information regarding the transcriptomic and epigenomic properties of the early marmoset germline cells, primarily due to the inherent difficulty in recovering marmoset embryos. Previous studies demonstrated the presence of POU5F1 + NANOG + cjPGCs localized within the hindgut endoderm in an E50 embryo, similar to that observed in human and monkey embryos at equivalent stages 33 . We extended this study and further provided the first comprehensive immunophenotypic and transcriptomic profile of cjPGCs from E50 embryos (CS11) (Fig. 1, 2). We discovered that cjPGCs displayed immunophenotypic and transcriptomic features characteristic of endogenous PGCs of humans and old-world monkeys. For example, they expressed key primate germ cell specifier genes (SOX17, SOX15, TFAP2C PRDM1, PRDM14 [at low levels]), and lacked SOX2. In contrast, mouse germ cells highly express SOX2 but only transient express SOX17 immediately after specification 34,35 . Recent studies suggest that these features are also shared in rabbits and pigs, suggesting that the germline gene regulatory networks functioning in primates are more widespread evolutionary than that of rodents 36,37 . Importantly, cjPGCs at E50 (CS11) embryos are primarily pre-migratory (i.e., localized within the hindgut endoderm), and exhibited features of early PGCs (i.e., lack of DDX4 or DAZL expression) similar to PGCs of cynomolgus monkeys at the corresponding developmental stage (Fig. 1E, 2E) 14 . Interestingly, human/cynomolgus PGCs at the same chronologic age (E50) already colonize the gonads and upregulate DDX4/DAZL 14,38 . This finding is likely due to the overall delay in early post-implantation embryo development in marmoset and suggest that germ cell development is synchronized with overall embryo development rather than chronologic age 39 .
With immunophenotypic and transcriptomic characterization of cjPGCs in hand, we were now able to validate methods required to generate cjiPSCs that could subsequently be used to assess molecular events associated with germline induction. To this end, we first established various culture methods for cjiPSCs (Fig. S2, S3). Previous studies suggested that cynomolgus ESCs cultured in DK20 on MEF were prone to differentiate, but that inhibition of WNT signaling in these cultures stabilized the undifferentiated state 23 . Consistently, we found that the inhibition of WNT signaling by IWR1 stabilized the undifferentiated state of cjiPSC cultured on MEF (Fig. S3). In contrast, our newly established PluriSTEM-based FF cjiPSC culture method facilitated stable maintenance of an undifferentiated state, regardless of the presence or absence of IWR1 (Fig. S2C, S3D, S5B, C). This might be due to the inclusion of proprietary factors in the PluriSTEM base medium that support the undifferentiated state. Nonetheless, despite differential propensity towards differentiation, the cjiPSCs used in this study exhibited gene expression characteristics of primed-state pluripotency and could be maintained across multiple passages in all of above culture conditions. Our successful identification of culture conditions capable of generating and maintaining cjiPSCs allowed us to next compare their competency to differentiate into cjPGCLCs. Remarkably, we noted that FF cjiPSCs (with or without IWR1) had no germline competency whereas OF and OF/IWR1 cjiPSCs had modest and high germline competency, respectively (Fig. 3, S4). Transcriptomes of FF, OF, OF/IWR1 cjiPSCs aligned accordingly on the PC space, where FF cjiPSCs were the most distant from and OF/IWR1 cjiPSCs were the closest to cjPGCLCs. As FF cjiPSCs can be cultured stably without overt meso/endodermal differentiation despite their complete lack of germline competency (Fig. S3D), these findings suggest that the inhibition of precocious meso/endodermal differentiation by IWR1 is not the primary reason why OF/IWR1 cjiPSCs are superior to OF cijPSCs in cjPGCLCs induction. In support, transcriptome comparison across cjiPSCs cultured under different conditions revealed that meso/endodermal differentiation did not differ substantially between these conditions. Rather, a comparison of OF/IWR1 (vs OF) and OF (vs FF cjiPSCs) revealed that upregulation of a number of genes related to UPS protein catabolism, particularly those with E3 ubiquitin ligase complex known as Skp, Cullin, F-box containing (SCF) complex (e.g., UBE3A, FBXW7), correlated with increased germline competency (Fig. S5D). The vast majority of these genes continued to be expressed in d2 cjPGCLCs, which suggests their potential role in germ cell specification (Fig. S5E, F). In line with this, recent studies highlighted the critical role of the UPS system, and in particular FBXW7, in regulation of pluripotency and germ cell development [40][41][42] . Further mechanistic studies investigating the role of UPS and protein catabolism in germline competency and cjPGCLCs specification are warranted.
In this study, we provide evidence that cjPGCLCs can be derived from cjiPSCs through direct floating culture of OF/IWR1 cjiPSCs in the presence of PGCLC induction cocktail ( Fig.3) 7,23 . Under this condition, cjPGCLCs were induced in a highly efficient manner, with the number of PDPN + cjPGCLCs peaking at d4 (~600 cells/aggregate). The marmoset germline induction efficiency is higher than that of hPGCLCs induced under direct floating culture and similar to those induced through step-wise method (2D induction of incipient mesoderm-like cells [iMeLCs] by ACTIVIN A and WNT agonist, CHIR99021, followed by floating culture with PGCLC induction factors) (Fig. 3) 7 . Since the iMeLC induction step is not essential for robust PGCLC induction in cynomolgus monkeys, these results highlight differential requirements for WNT/NODAL/ACTIVIN signaling to prime PGCLC specification, or differential endogenous production by the aggregates themselves 23 . Regardless, transcriptomic and epigenetic features of cjPGCLCs are highly similar to those from humans and cynomolgus/rhesus monkeys and divergent from rodent PGCLCs 28,43 , suggesting conservation of the germ cell specification program among primates.
Interestingly, a previous study by Okano and colleagues failed to produce cjPGCLCs from cjiPSCs by floating culture using a PGCLC induction cocktail similar to ours 44 . This may in part be due to the relatively poor germline competency of cjiPSCs/ESCs used in the study, which were cultured under conventional OF condition without IWR1. To overcome the lack of induction, these authors employed an alternative approach in which cytokine-based induction was combined with over-expression of key PGC specifier transcription factors, PRDM1 and SOX17 44 . Although this approach allowed induction of PRDM1-Venus + cjPGCLCs, the efficiency was variable, with two cjESC lines showing 30-40% induction while cjiPSCs only induced PGCLCs at 1.7% efficiency. Moreover, DDX4 was upregulated in some of PRDM1::Venus + cells as early as d9-10. In humans and primates including marmoset, DDX4 is not expressed in pre-migratory PGCs in vivo (Fig. 1E, S2E) and is upregulated only after prolonged xrTestis/xrOvary culture in vitro, 5,6 suggesting that the induction method utilized in this study might not fully recapitulate the physiological germ cell developmental trajectory.
Whether over-expression of transcription factors can drive cjPGCLC formation from the OF/IWR1 cjiPSCs with high germ cell competency that we established in this study remains to be determined.
In summary, the in vitro platform described here enables efficient induction of cjPGCLCs from cjiPSCs, which will serve as a foundation for analyzing mechanisms of PGC specification in marmoset monkeys. Although the road ahead will likely be long,

DECLARATION OF INTERESTS
The authors declare no competing interests.        Table S1. Differentially expressed genes (DEGs) among cell clusters in Figure 2B. Figure 6D. Table S3. DEGs from pairwise comparisons in Figure S5D. Table S4. Primers used in this study.

Collection of marmoset embryo samples
Marmosets were housed at the Southwest National Primate Research Center (SNPRC), Texas Biomedical Research Institute, an AAALAC accredited institution. All procedures were reviewed and approved by the Texas Biomedical Research Institute IACUC (1772CJ). Marmosets at the SNPRC were maintained under standardized husbandry conditions as described previously 45 . For breeding, marmosets were housed in male-female monogamous pairs. Females received an unsedated transabdominal ultrasound monthly until pregnancy was confirmed with a GE Logiq portable ultrasound machine. Females were habituated to manual restraint and received positive reinforcement during the procedure. After a pregnancy was detected (<30 days estimated gestational age), pregnancy progression was assessed every 14 days. The gestational age of embryos was estimated with crown-rump length, assessed via ultrasound, which has previously been found to reliably estimate gestational age in marmosets to within ±3 days 46,47 .
Embryos at E50 were recovered from the uteri obtained through hysterectomy performed under full anesthesia. First, the endometrium was exposed by dissection of the serosa and myometrium at the lateral side of the explanted uterus. Then the exposed endometrium was carefully opened along the cervix-to-fundus direction to approach the uterine cavity, from which embryonic sacs were recovered and collected into dishes containing RPMI 1640 medium. Three embryos (Carnegie stage 11) were isolated from embryonic sacs and photographed. After removal of the amnion and yolk sac, the posterior portions of the embryos were dissected and used in histologic analysis or single cell RNA-sequencing.

Marmoset peripheral blood mononuclear cell collection and reprogramming to cjiPSCs
Marmoset whole blood was collected into Na-heparin vacuum tubes, mixed with an equal Quantitative reverse transcriptase PCR was used to determine whether the cjiPSCs had cleared the Sendai virus reprogramming factors. cjiPSCs were collected as described above for passaging and pelleted. The cell pellet was resuspended in 0.5 ml TRIzol, and total RNA was isolated with a Direct-zol RNA MiniPrep kit (Zymo Research). Two separate assays were performed to ensure that the cjiPSCs were free of mycoplasma contamination. First, the cjiPSC colonies were stained with DAPI to assess the presence of extranuclear DNA characteristic of mycoplasma infection. Second, cjiPSCs were harvested with 0.5 mM EDTA in PBS and pelleted at 400 g for 5 minutes.
Genomic DNA was isolated from the pellet with a QIAamp DNA mini kit (Qiagen).
Genomic DNA was screened with a LookOut® Mycoplasma PCR Detection Kit

Culture of cjiPSCs
For feeder free cjiPSC culture, the cjiPSCs (C6 and C10) were cultured on xeno-free Genes in the heatmap were hierarchically clustered according to the Euclidean distance, scaled by row, and then visualized with pheatmap. Gene ontology enrichment was analyzed with DAVID v.6.8.

Bulk RNA-seq library preparation
CjiPSCs cultured in the presence or absence of feeders, and in the presence or absence of IWR1, were collected. To minimize the contamination with feeder cells, >30 colonies of cjiPSCs cultured on feeder layer were randomly picked under an inverted microscope and pooled before isolation of total RNA. Total RNA was extracted with an RNeasy Plus Micro Kit (#74034, QIAGEN). RNA-seq libraries were made using a SMRT-Seq HT plus kit (#R400748, Takara) according to the manufacturer's protocol. Briefly, total RNA was quantified with a Qubit instrument, and RNA integrity was verified with a TapeStation. Then 1 ng RNA was used for cDNA conversion with a one-step first-strand cDNA synthesis and double-stranded cDNA amplification protocol. cDNA was purified with AMPxp beads, its concentration was measured with a Qubit, and its quality was verified with a TapeStation. Next, 2 ng cDNA was used for library construction. Libraries were dual indexed and pooled according at equal molecular concentrations. Subsequently, 100-base pair reads were sequenced on the Illumina Nextseq 2000 platform.

Bulk RNA-seq data analysis
Raw fastq files were demultiplexed with bcl2fastq2 (v.2.20.0.422). Barcodes and adapters were removed with Trimmomatic (v.0.32). Fastq files were mapped to the Callithrix jacchus (calJac4) reference genome with STAR (v.2.7.10a). The raw gene count table was generated with featurecounts, and weakly expressed genes were filtered with edgeR with the filterByExpr function with default parameters. Briefly, the raw counts were normalized to library size, and then genes with counts per million (CPM) above 10 were included in downstream analysis. DEGs were analyzed with edgeR

Bisulfite-sequencing and analysis
For collection of cjiPSCs for methylome analyses, colonies were picked under a microscope, then collected into a 1.5 ml tube for lysis. PGCLC aggregates were digested with 400 µl 0.25% trypsin for 15 min. Then 100 µl FBS was used to stop digestion, and the lysates were pipetted well to obtain a single cell suspension. The dissociated cells were stained with APC-conjugated anti-human PDPN and BV421-conjugated anti-human/mouse CD49f (ITGA6), and then the PDPN + ITGA6 weak+ fraction was sorted for methylome analyses.