A critical but divergent role of PRDM14 in human primordial germ cell fate revealed by inducible degrons

PRDM14 is a crucial regulator of mouse primordial germ cells (mPGC), epigenetic reprogramming and pluripotency, but its role in the evolutionarily divergent regulatory network of human PGCs (hPGCs) remains unclear. Besides, a previous knockdown study indicated that PRDM14 might be dispensable for human germ cell fate. Here, we decided to use inducible degrons for a more rapid and comprehensive PRDM14 depletion. We show that PRDM14 loss results in significantly reduced specification efficiency and an aberrant transcriptome of human PGC-like cells (hPGCLCs) obtained in vitro from human embryonic stem cells (hESCs). Chromatin immunoprecipitation and transcriptomic analyses suggest that PRDM14 cooperates with TFAP2C and BLIMP1 to upregulate germ cell and pluripotency genes, while repressing WNT signalling and somatic markers. Notably, PRDM14 targets are not conserved between mouse and human, emphasising the divergent molecular mechanisms of PGC specification. The effectiveness of degrons for acute protein depletion is widely applicable in various developmental contexts.

Representative IF images of PRDM14 staining using two different antibody batches in Wk7 and Wk8 human female gonadal sections. hPGCs were marked by OCT4 or SOX17 expression. Nuclei were counterstained by DAPI. Note both nuclear and cytoplasmic PRDM14 in the upper sample (Wk8 male). (B) IF on sorted AP + hESCs and AP + cKIT + gonadal hPGCs. hPGCs and pluripotent hESCs were marked by OCT4. Nuclei were counterstained by DAPI. AP -cKITpopulation (gonadal soma) was used as a negative control. Similar results were obtained for a Wk9 male embryo. Scale bar is 100 μm.

Inducible degrons allow fast and reversible PRDM14 protein depletion
Most inducible loss-of-function approaches act at the transcriptional or post-transcriptional levels and often suffer from slow and incomplete protein removal [26][27][28][29] . Since PRDM14 is one of the core pluripotency factors in hESCs 13 , and exhibits dynamic changes during hPGCLC induction, conventional knockout approaches are unsuitable to study its functions in hPGCLCs. As PRDM14 expression commences within 24hrs of hPGCLC induction, we decided to use conditional degrons 30 to achieve fast, inducible and reversible PRDM14 depletion at the protein level.
Supplementary Figure 2. PRDM14-Venus and SOX2 expression during PGCLC specification. IF analysis of 12hrs and D5 EBs differentiated from PRDM14-AID-Venus fusion reporter cell line with or without cytokines ("+cyto" and "no cyto", respectively). Expression of PRDM14 and SOX2 is shown; hPGCLCs in "+cyto" EBs are marked by SOX17 or TFAP2C. Note that "no cyto" differentiation in the basal medium yields no hPGCLCs and SOX17 + cells likely belong to other lineages.
First, we tested AID-or JAZ-mediated PRDM14 depletion in hESCs by growing them with or without corresponding hormones and performing IF ( Fig.3B-C, Suppl. Fig.3). PRDM14 and Venus fully colocalized and were both reduced upon the addition of IAA or Cor in hormone-sensitive, but not in parental lines ( Fig.3B-C, Suppl. Fig.3A-D). Crucially, the AID system allowed reduction of Venus fluorescence to negligible levels, comparable to cells lacking the Venus knock-in (Suppl. Fig.3C). Time-course IF in hESCs revealed a rapid onset of PRDM14-AID-Venus depletion within 10 minutes of IAA treatment, reaching homogeneity within 25 minutes (Fig.3B, Suppl. Fig.3E). In the case of the JAZ degron, however, residual PRDM14-JAZ-Venus signal was detectable even after 2 days of Cor treatment (Fig.3C, Suppl. Fig.3D). Both systems are nevertheless reversible 24,32 , and IAA wash-off restored PRDM14 levels in less than 2 hours (Suppl. Fig.3F). Altogether, inducible degrons allow fast and reversible PRDM14 depletion, with AID being superior in terms of speed and efficiency.

Inducible PRDM14 degradation reduces hPGCLC specification
Next, we addressed the importance of PRDM14 for hPGCLC specification by adding IAA or Cor at the onset of hPGCLC differentiation (D0), followed by measuring the induction efficiency on D4 by recording the percentage of NANOS3-tdTomato + AP + cells (Fig.4A). Notably, PRDM14 depletion using both approaches resulted in a significant reduction of hPGCLC induction efficiencies (by 70% and 30-60%, respectively) ( Fig.4B-C, Suppl. Fig.4A-C). The effect was more pronounced in the case of the AID system, presumably due to its higher efficiency and faster kinetics of PRDM14 depletion. Therefore, we focused on the AID system to study PRDM14 further. Crucially, the observed phenotype was replicated using a PRDM14-AID in another hESC line (Suppl. Fig.4D).
To further test AID efficacy in this context, we induced degradation of SOX17, a known hPGC and definitive endoderm (DE) regulator 3,20,34 . IAA addition fully abrogated hPGCLC and DE specification (Suppl. Fig.4E-G), confirming the key roles of SOX17 in these lineages. Notably, as a control, PRDM14 depletion had no adverse effect on the induction of definitive endoderm (DE) (data not shown), confirming the specificity of AID for our investigation.
To establish the time when PRDM14 is essential for hPGCLC specification, we performed a time-course of IAA supplementation. We performed depletion of PRDM14 by adding IAA starting on D0, D1 or D2, followed by analysis on D4. We noted a strong phenotype with reduction in hPGCLC specification (Fig.4D).
By contrast, the addition of IAA on D3 had no detectable effect on the number of NANOS3-tdTomato + AP + cells (Fig.4D). This suggests that either PRDM14 is dispensable after D2 or that its depletion at later timepoints does not affect hPGCLC numbers, although transcriptional or epigenetic consequences cannot be excluded. Since PRDM14 starts being expressed in many hPGCLCs on D2 (Fig.2) and its loss from D2 to D4 yielded similar results to D0-D4 depletion (Fig.4D), it is likely that PRDM14 is most critical around D2 of hPGCLC specification. Therefore, PRDM14 might act to consolidate the germ cell network established by SOX17 and BLIMP1 from D1 of hPGCLC induction 3, 5 . PRDM14 is highly expressed in human PGC-competent pluripotent cells, unlike in mice, where it is repressed at the equivalent stage 3,35 . Notably, however, PRDM14 depletion in 4i hESCs for 1 passage prior to hPGCLC induction (Suppl. Fig.4H) had no effect on hPGCLC specification (Suppl. Fig.4I), indicating no detectable involvement of PRDM14 in the maintenance of competence for hPGCLC fate. To determine if PRDM14 is required for the acquisition of competence, we induced competence for hPGCLCs in hESCs via premesendoderm (preME), which transiently displays hPGCLC potential 5 . However, when IAA was added to the preME medium and washed off before hPGCLC induction (Suppl. Fig.4H), we observed no effect on hPGCLC specification efficiency (Suppl. Fig.4J). Altogether, these data demonstrate that PRDM14 is important for hPGCLC specification but is dispensable for the acquisition and maintenance of the hPGCLCcompetent state.  Fig.3B. PRDM14-AID-Venus cells were grown with or without IAA for 1 day. To focus on pluripotent cells that would normally express PRDM14, fluorescence intensities were measured in the nuclei of NANOG-positive cells. Data show the mean of n=4 (2 independent experiments using 3 clones), **** P<0.0001 (twoway ANOVA followed by Sidak's multiple comparison test) and n=1 for "no TIR1" control. A cell line lacking TIR1 hormone receptor ("no TIR1") was used as a negative control for PRDM14-AID-Venus depletion.  The induction efficiency of untreated cells ("no IAA") was set to 1. Data show mean+/-SD for n=5-6 independent experiments per clone, ns -not significant, **** P<0.0001 (Two-way ANOVA followed by Sidak's multiple comparison test). (C) Quantification of flow cytometry results for PRDM14-JAZ-Venus hPGCLCs. The induction efficiency of untreated cells ("no Cor") was set to 1. Data show mean+/-SD for n=4 independent experiments per clone, ns -not significant, **** P<0.0001 (Two-way ANOVA followed by Sidak's multiple comparison test). (D) Time-course of PRDM14 depletion. IAA (500 µM) was added on indicated days of hPGCLC induction from PRDM14-AID-Venus clones. Note that IAA was not washed off until the end of differentiation (D4), when hPGCLC induction efficiency was assessed by flow cytometry. Data show mean+/-SD of n=3 (2 independent experiments using 2 clones), or n=2 for "no TIR1" control, ns -not significant, ** P<0.01, *** P<0.001 (two-way ANOVA followed by Sidak's multiple comparison test).  Fig.4B). The induction efficiency of untreated cells ("no IAA") were set to 1. Data show mean+/-SD for n=3-5 per clone; ns -not significant, **** P<0.0001 (Two-way ANOVA followed by Sidak's multiple comparison test). (D) PRDM14 depletion using AID in H9 hESC background also reduces hPGCLC specification. Note that this cell line lacks NANOS3-tdTomato reporter and hPGCLCs were identified by AP and CD38 staining. H9TIR10 is a parental cell line expressing TIR1 but lacking the AID-Venus KI. Data show mean+/-SD for n=3 independent experiments per clone, ** P<0.01, **** P<0.0001 (Two-way ANOVA followed by Sidak's multiple comparison test). (E) Epifluorescence photographs show colocalization of NANOS3-tdTomato and SOX17-AID-Venus fluorescence in D4 EBs; both signals are reduced upon IAA treatment. (F) Representative flow cytometry plots showing hPGCLC induction using the SOX17-AID-Venus cell line with or without IAA. (G) SOX17 degradation decreases definitive endoderm (DE) specification. Representative flow cytometry plots; DE cells are marked by CXCR4 expression. (H) Experimental workflow of PRDM14 depletion in competent hESCs, as well as during preME induction. PRDM14 role in competence maintenance was assessed by depleting it in competent (4i) hESCs, followed by washing off the IAA for PRDM14 restoration and inducing hPGCLCs (without IAA). To test if PRDM14 is required for competence acquisition, it was depleted during the generation of the preME competent cells. Auxin was subsequently removed and hPGCLCs were induced without IAA. (I) PRDM14-AID depletion in competent hESCs does not reduce hPGCLC induction therefrom. IAA was added for 1 passage (3 days) prior to hPGCLC induction and washed of before the onset of hPGCLC differentiation. hESCs grown without auxin were differentiated with or without IAA as above. Data show mean+/-SD for n=2-4 independent experiments per clone, **** P<0.0001 (Two-way ANOVA followed by Sidak's multiple comparison test). (J) PRDM14-AID depletion during the acquisition of hPGC-competent state via pre-mesendoderm (preME) does not reduce hPGCLC induction therefrom. IAA was added for 12 hours during preME generation from conventional hESCs and washed off before the onset of hPGCLC differentiation. preME induced without auxin was differentiated with or without IAA as above. Data show mean+/-SD for n=2 independent experiments (2 clones).

Ectopic PRDM14 or desensitization to auxin rescues hPGCLC differentiation
To confirm the specificity of the observed phenotype, we attempted to rescue the endogenous PRDM14 depletion with ectopic PRDM14, using the ProteoTuner system, whose kinetics is comparable to that of AID 36 . PRDM14 transgene was fused to a destabilisation domain (DD) and thus continuously degraded by the proteasome but stabilised by Shield-1 ligand 36,37 . Interestingly, overexpression of PRDM14 on D0, but not on D1 could fully rescue hPGCLC specification efficiency and even enhanced hPGCLC specification compared to "no IAA" control ( Fig.5A). This suggests that the transient PRDM14 repression observed at the onset of hPGCLC specification (Fig.2B) is not prerequisite for germ cell induction.
We then asked if we could restore the endogenous PRDM14 levels by blocking auxin perception. To achieve this, we designed an AID-JAZ degron switch, where PRDM14-AID-Venus is degraded by IAA supplementation, while TIR1 is fused to JAZ and thus depleted by Cor (Fig.5B). IF analysis in hESCs confirmed that simultaneous administration of both IAA and Cor desensitised hESCs to IAA via TIR1 degradation and replenished the PRDM14-Venus pool (Fig.5C). In line with previous experiments, these cell lines also showed decreased hPGCLC specification efficiency upon IAA treatment (Fig.5D). Crucially, in the presence of both IAA and Cor, hPGCLC differentiation efficiency was significantly restored (Fig.5D). This confirms the causative role of PRDM14 depletion in hPGCLC specification phenotype and exemplifies the use of orthogonal degron approaches for rescue design. IAA and/or Shield-1 (Sh) were added on indicated days of differentiation to deplete the endogenous PRDM14 or stabilise the PRDM14-DD transgene, respectively. "Control" cell line lacks both the AID knock-in and the PRDM14-DD transgene. Data show mean+/-SD for n=4 (PRDM14-DD22) or n=2 (control) independent experiments, * P<0.05, ** P<0.01 (Two-way ANOVA followed by Sidak's multiple comparison test separately within "no IAA" or "D0 IAA" groups). (B) Scheme of the double AID/JAZ degron design. Endogenous PRDM14 is under the control of the AID degron as above, while the stability of TIR1 hormone receptor is controllable by the JAZ degron. (C) AID/JAZ validation by IF in hESCs. Simultaneous IAA and Cor supplementation depletes TIR1 and restores PRDM14, as shown by IF staining for PRDM14-AID-Venus and TIR1(V5-tagged). Nuclei were counterstained by DAPI. Scale bar is 100 µm. (D) Desensitisation to auxin via the AID/JAZ degron switch rescues hPGCLC induction efficiency. Data show mean+/-SD for n=4 (no TIR1 control) or n=11 (TIR-JAZ, 3 clones combined), **** P<0.0001 (Two-way ANOVA followed by Sidak's multiple comparison test).
GO analysis on genes derepressed in PRDM14-deficient hPGCLCs identified enrichment of terms related to WNT signalling, as well as heart and nervous system development (Fig.6G). A similar response occurs in hPGCLCs specified without BLIMP1 or TFAP2C Tang et al., 2015). Indeed, pairwise comparisons revealed a significant overlap between both up-and downregulated targets, suggesting potential combinatorial roles of PRDM14 with BLIMP1 and TFAP2C. Accordingly, 115 genes were downregulated in all three mutants (including germ cell markers UTF1, NANOG, AKAP12, NANOS1, and TRIM28), while 281 shared upregulated genes were mostly implicated in morphogenesis, WNT signalling, as well as cell migration and adhesion (Suppl. Fig.5F-G and Suppl. Table 11). the gene set, if compared against "PGCLC no IAA" control. (C) qPCR shows transcriptional rescue of tested DEGs by ectopic PRDM14-DD. hPGCLCs were differentiated with or without IAA and/or Shield-1 (Sh) and sorted as NANOS3-tdTomato + AP + cells; soma denotes the double-negative population from the same experiments. Data show gene expression relative to no IAA no Sh PGCLCs and normalized to GAPDH, mean+/-SEM of n=2 independent experiments. (D) Parental ("no TIR1") control for qPCR validation of RNA-seq (see Fig.6G). hPGCLCs were sorted as NANOS3-tdTomato + AP + cells, while soma denotes the double-negative population from the same experiments. Data show gene expression relative to no IAA hPGCLCs and normalized to GAPDH, mean+/-SEM of n=2-5 independent experiments, * P<0.05, ** P<0.01, *** P<0.001, **** P<0.0001 (multiple t-tests). (E) qPCR on unsorted PGC-competent (4i) PRDM14-AID hESCs cultured with or without IAA for 3 consecutive passages. The parental cell line lacking TIR1 is shown as a control. Data show mean+/-SEM of n=2 hormone-sensitive clones; n=1 for the parental control. Note downregulation of pluripotency genes and upregulation of pro-neuronal differentiation transcripts. (Supplementary Figure 5 is continued on the next page)

PRDM14 regulates target gene expression mainly through promoter binding
To identify direct targets of PRDM14 in hPGCLCs and hESCs, we performed chromatin immunoprecipitation with high-throughput sequencing (ChIP-seq). We took advantage of the PRDM14-AID-Venus line to use a ChIP-grade anti-GFP antibody for PRDM14-Venus immunoprecipitation, which proved more efficient than anti-PRDM14 antibodies (Suppl. Fig.6A). Importantly, auxin treatment led to a significant loss of signal (Suppl. Fig.6A), which allowed us to use ChIP-seq on IAA-treated hESCs as a control.
Note that the strongest ChIP-seq enrichment persisted in IAA-treated hESCs (clusters 1-2) with 1728/3319 hESC binding sites detectable in hESC+IAA, albeit with significantly lower signal ( Fig.7A and Suppl. Table   12). Such persistent regions might be more tightly bound, and/or require longer auxin exposure to eliminate PRDM14 binding completely.
Notably, PRDM14 in both hESCs and hPGCLCs predominantly binds within 1 kilobase (kb) of transcription start sites (TSS) (Suppl . Table 12), with ~30% of peaks spanning annotated promoters (Fig.7B), which agrees with a previous ChIP-seq in hESCs 13 . By contrast, PRDM14 binds predominantly to distal genomic regions in mice, and only 4-10% of peaks are located within 1 kb from the TSS in mESCs 16,38 .
Motif analysis 39 confirmed that the conserved PRDM14 motif was top-ranking in both hESCs and hPGCLCs (Fig.7C). Notably, TFAP2C motif was the second most enriched within hPGCLC-specific targets, suggesting cooperation between the two factors (Fig.7C), consistent with the overlapping targets from RNA-seq (Suppl. Fig.5F). We also found significant enrichment of BLIMP1 and SOX motifs in hPGCLCs, whereas unique hESC peaks were enriched in the OCT4-SOX2 motif (Fig.7C). Altogether, this indicates that PRDM14 cooccupies genomic targets with core TFs specific to hPGCs or pluripotent cells, respectively. Representative direct targets of PRDM14 were labelled. Upregulated genes were highlighted in orange and downreguated in blue. Note that downregulated PRDM14 targets form a separate co-expression cluster.
Next, we correlated ChIP-seq peaks mapping within 100 kb of the TSS with the corresponding changes in gene expression from RNA-seq (Fig.7D, Suppl. Fig.5H and Suppl. Table 12). In hPGCLCs, 314 peaks were associated with 168 (17%) upregulated genes and 93 peaks with 47 (10%) downregulated genes (Suppl. Table   12). GO analysis confirmed that genes related to WNT signalling, heart and brain development are among the direct PRDM14 targets (Suppl. Fig.6B). The hESC-specific targets included SOX2, ERBB3 and GBX2, while hPGCLC-specific peaks were found near SOX17, TFAP2C and NANOS3 among others (Fig.7E-F). Whereas PRDM14 binds to the regulatory elements of SOX17, PRDM14 depletion did not significantly affect SOX17 expression in hPGCLCs (Fig.6E and Suppl. Table 9) presumably because other factors sustain its expression.

Distinct molecular functions of PRDM14 in mouse and human
Next, we asked if the molecular roles of PRDM14 are conserved between mouse and human, by comparing the protein-coding transcriptomes of PRDM14-AID hESCs and hPGCLC, with the equivalent mouse Prdm14 -/cells 11 (Suppl. Fig.7A-B, Suppl. Table 13). PGC-competent mouse epiblast-like cells (mEpiLCs) showed very few DEGs since Prdm14 is repressed in these cells 11 . Only two genes (CDH4 and HS6ST2) were derepressed in both mESCs and hESCs, and 3 genes (DNMT3B, SPRY4 and FHL1) showed opposite changes, probably reflecting the broader differences between mESCs and hESCs 40   In mPGCs, PRDM14 initiates global epigenetic reprogramming, in part through the repression of Uhrf1 to allow passive DNA demethylation 11 ; this gene is also repressed in hPGCs and hPGCLCs 41 . A small subset of PRDM14-depleted proliferating SOX17 + cells failed to downregulate UHRF1, potentially suggesting a conserved regulation (Suppl. Fig.7C).
PRDM14 alone is sufficient to induce mPGCLCs fate 8 ; however, this was not the case for hPGCLC specification using either doxycycline-inducible or ProteoTuner systems (Suppl. Fig.7D), consistent with the observation that upregulation of SOX17 and TFAP2C precedes that of PRDM14 (Fig.2). Altogether PRDM14 evidently plays an important role shortly after the initiation of hPGCLC fate, but its function is distinct from that in mPGCs.

Supplementary Figure 7. Comparison of molecular functions of PRDM14 in mouse and human hPGCLCs. (A)
Venn diagrams comparing transcriptional changes in Prdm14 -/mouse PGCLCs and PRDM14-AID hPGCLCs (+IAA) relative to corresponding WT controls. P-values were obtained by pairwise hypergeometric tests. Examples of overlapping DEGs are shown. * In PRDM14-AID hPGCLCs PRDM14 is depleted only at the protein level, while the transcript is still present and even upregulated, presumably due to autoregulation. (B) 3D PCA plot showing distinct transcriptional response of mouse and human PGCLCs to the absence of PRDM14. Data from Suppl. Table 13 with logFC>1 were used. Note that mouse WT and KO samples separate by PC4, while human samples induced with or without PRDM14 separate along PC3. Par -parental control lacking TIR1; r1 and r2 -replicates 1 and 2. (C) IF analysis of UHRF1 expression in D4 PRDM14-AID EBs. Putative hPGCLCs are marked by SOX17, proliferating cells are marked by Ki-67. White arrows highlight an example mutant proliferating hPGCLC that retains UHRF1. Note that UHRF1 + SOX17 + cells in the "no TIR1" control (grey arrows) are Ki67-negative. (D) Representative flow cytometry plots showing that ectopic PRDM14 induced by either dox or Shield-1 does not induce hPGCLC induction without BMP2 and other cytokines ("no cyto"). Dox or Shield-1 were added on D0 of differentiation.

Supplementary Figure 8. Graphical summary of key findings.
Model summarising the context-dependent PRDM14 roles in human pluripotency and hPGCLC specification. Using PRDM14-AID-Venus we were able to map PRDM14 binding in hESCs and hPGCLCs, as well as deplete the protein with high time resolution. This showed that PRDM14 regulates target genes through promoter binding together with SOX2/OCT4 in hESCs or TFAP2C and BLIMP1 in hPGCLCs. PRDM14, TFAP2C and BLIMP1 are all downstream of SOX17. In hESCs, PRDM14 activates the expression of pluripotency genes and represses pro-differentiation genes. Similarly, in hPGCLCs, PRDM14 represses somatic transcripts, but activates germ cell-and pluripotency-related genes. Representative examples of activated and repressed genes are shown. When auxin is added PRDM14-AID fusion protein is recruited to the E3 ubiquitin ligase via an F-box auxin receptor TIR1. It leads to rapid protein ubiquitination and degradation, which allows time-resolved functional studies.

Discussion
Using two acute protein depletion strategies, combined with rescue, transcriptomic and ChIP-seq experiments, we demonstrate that PRDM14 is required for hPGCLC specification and represses somatic differentiation while promoting germline fate (Suppl. Fig.8). Strikingly, the molecular function of PRDM14 in the human germline has diverged significantly compared to mPGCs. Indeed, the sets of targets regulated by PRDM14 in the two species are vastly different (Suppl. Fig.7A-B) and, unlike in the mouse 8 , PRDM14 alone cannot induce hPGCLCs (Suppl. Fig.7D). The regulatory network for hPGC specification is altogether distinct from that in mice, with the recently established critical role of SOX17, and the notable repression of SOX2 3,20 . Our study of PRDM14 provides further insights on the divergence of the molecular basis of germ cell fate determination in mouse and human.
PRDM14 is specifically upregulated in the nucleus of nascent hPGCLC from D1 of germ cell induction, following the expression of SOX17 and TFAP2C (Fig.2). PRDM14 depletion strongly reduces the efficiency of germ cell specification (Fig.4) and results in an aberrant transcriptome of the PRDM14-deficient hPGCLCs (Fig.6). The effects of PRDM14 depletion during hPGCLC specification, resemble those observed upon the loss of TFAP2C or BLIMP1 (Suppl. Fig.5F). Furthermore, ChIP-seq analysis revealed high enrichment of TFAP2C and BLIMP1 motifs within hPGCLC-specific PRDM14 targets (Fig.7C), which suggests combinatorial roles of the three regulators during the induction of hPGC-specific genes and the repression of somatic markers. However, unlike in mPGCs 11,42 , many prominent germ cell genes, such as TFAP2C, PRDM1 and DND1, are not the targets of PRDM14 in hPGCLCs. Furthermore, while the PRDM14 motif is conserved, PRDM14 binds predominantly to gene promoters in hESCs and hPGCLCs ( Fig.7B and 13, 16, 38 ), but to distal regulatory elements in mice 16,38 . The notable lack of overlap might be dictated by different protein partners or by the divergence of the protein itself, although there is significant functional conservation of the protein, and human PRDM14 rescues the lack of its mouse orthologue in mESCs 15 .
The context-dependent roles of PRDM14 are also exemplified by distinct set of target genes in hESCs, where it cooperates with OCT4 and SOX2 to sustain pluripotency, and limits neuronal differentiation (Suppl. Fig.8,   Fig.6A, Suppl. Fig.5E and Fig.7C). This and other studies highlighted the divergent roles of PRDM14 in mouse and human pluripotent cells 13,15,16,38 . Furthermore, there are significant human-mouse differences during early embryo development, at the time of PGC specification 43,44 ; altogether these differences likely contribute to the species-specific modes of PGC specification. Notably, we found an even more pronounced transcriptional phenotype upon PRDM14 loss in hPGCLCs than in hESCs (Fig.6). Note that hESCs were, however, in a 'steady state' of self-renewal, while hPGCLCs were undergoing cell fate transitions at the time of PRDM14 depletion. The persistence of PRDM14 at some loci even after 3 days of IAA treatment (Fig.7A), is reminiscent of a significant (27%) retention of CTCF-AID peaks that were reported after 2 days of exposure to IAA 32 . Since most other inducible loss-of-function approaches display lower efficiency and slower kinetics, it is likely that they may result in even slower protein depletion from chromatin.
The distinct roles of PRDM14 in mouse and human PGCs, further illustrate the impact of the species-specific regulatory networks for PGC fate. For example, expression of SOX17 and SOX2 is mutually exclusive in germ cells of human and mouse, respectively. Whereas SOX17 is essential for hPGC fate 3 , SOX2 promotes mPGC specification and survival 45 . By contrast, repression of SOX2 is apparently prerequisite for the initiation of hPGCLC fate 1 , since ectopic expression of SOX2 abrogates hPGCLC specification 46 . In the absence of BMP signalling, which initiates hPGCLC specification, high expression of SOX2 and PRDM14 is maintained (Suppl. Fig.2). It is possible that PRDM14 expression in the hPGCLC-competent cells is downstream of SOX2. The repression of SOX2 upon the initiation of hPGCLC in response to BMP signalling might therefore explain a transient loss of PRDM14. Indeed, in hESC where SOX2 binds to PRDM14 promoter 47 , loss of SOX2 leads to PRDM14 downregulation 48 . The observed re-expression of PRDM14 in hPGCLC follows after the upregulation of SOX17 and BLIMP1 at the onset of hPGCLC specification (Fig.2 and Suppl. Fig.2), which suggests a molecular shift in the regulation of PRDM14 expression in germ cells.
In mPGCs, PRDM14 promotes DNA demethylation, in part by repressing Uhrf1, and initiates the reduction of H3K9me2 through Ehmt1 downregulation 9 . How the TF network might control the epigenetic resetting in hPGCs is less clear, but some essential features of epigenetic reprogramming are evident in hPGCs, including the repression of UHRF1 and EHMT2 41 . PRDM14-deficient hPGCLCs indicated derepression of UHRF1 (Suppl. Fig.7C) but not of EHMT2 in a subset of mutant hPGCLC. Further studies are required to establish how precisely the epigenetic resetting is initiated in hPGCs and a potential role for PRDM14, if any, in this critical process.
Our study shows the importance of the use of rapid and comprehensive PRDM14 depletion for the phenotype unmasking. A previous study of PRDM14 in hPGCLCs used a partial knockdown at a later time-point (D2 BLIMP1 + SOX17 + precursors), where the homogeneity and speed of PRDM14 depletion was not monitored 19 .
We demonstrate the utility of AID and JAZ degrons to deplete endogenous proteins to study human cell fate decisions. Simultaneous use of both degrons allows independent control of two proteins, and the construction of degron switches as shown here (Fig.5B-D). The knock-in of AID/JAZ together with a fluorescent reporter facilitates protein expression analysis and allows the use of the anti-GFP (or other epitope tags) antibody for IF and ChIP. Furthermore, tissue-specific TIR1 expression can allow spatial control over protein stability, as shown in Caenorhabditis elegans and Drosophila melanogaster without detectable side effects 49,50 .
Neither mouse nor human adult tissues express PRDM14 except in some types of cancer 14,51 , indicating the importance of studying the role of PRDM14 in normal embryogenesis. Inducible degrons offer a more precise control over proteins when studying the roles of critical TFs undergoing dynamic changes during normal and malignant development.
ChIP experiments. The study was supervised by MAS. The manuscript was written by AS and MAS with contributions from most authors.
Conventional hESCs (used for hPGCLC differentiation via pre-mesendoderm) were grown in Essential 8 (E8) 4 in plates pre-coated with 5 μg/ml vitronectin for at least 1 hour. Media were replaced every day. Cells were passaged in clumps using 0.5 mM EDTA in PBS. All reagents for E8 hESC culture were from Thermo Fisher Scientific.
For transgene inductions in hESCs or during differentiation, 1 μg/ml doxycycline (dox, Sigma) and/or 0.5 μM Shield-1 (Clontech) were added to media, where specified. For depletion of AID-fused proteins, auxin (indole-3-acetic acid sodium salt, IAA, Sigma) was used at 100 μM (in H2O) unless otherwise specified. For depletion of JAZ-fused proteins, coronatine (Cor, Sigma) was used at 50 μM (in DMSO) in all experiments. DMSO was used as vehicle control for experiments with Cor.
For definitive endoderm (DE) induction 2 , mesendoderm (ME) was first obtained from E8 cells by 36-hour exposure to ME medium (Suppl. Table 3); ME was then washed with PBS followed by 48-hour culture in the DE medium (Suppl. Table 4).

Flow cytometry and fluorescence-activated cell sorting (FACS)
Flow cytometry and FACS were performed as in 3 . At least 6 EBs were washed in PBS and dissociated with 0.25% Trypsin-EDTA for 10 min at 37°C. Cells were washed, resuspended in FACS buffer (3% FBS in PBS) and incubated with anti-AP and anti-CD38 antibodies specified in Suppl. Table 5 for 30-60 minutes at 4 ºC in the dark. After washing, the cells were resuspended in FACS buffer with 0.1 µg/ml DAPI and filtered through a 50 µm cell strainer. Flow cytometry was done using BD LSR Fortessa, while FACS was performed with Sony SH100 Cell Sorter. Data were analysed using FlowJo (Tree Star).

Isolation of gonadal hPGC by FACS
Human genital ridges were dissected in PBS and separated from the mesonephros followed by dissociation with TrypLE Express (Life Technologies) at 37°C for 20-40 minutes (depending on the tissue size) with pipetting every 5 minutes. Cells were washed, resuspended in FACS buffer (3% FBS and 5 mM EDTA in PBS) and incubated with anti-AP and anti-cKIT antibodies specified in Suppl. Table 5 for 15 minutes at room temperature with 10 rpm rotation in the dark. Cells were then washed in FACS medium and filtered through a 35 µm cell strainer. FACS was performed with Sony SH100 Cell Sorter and data were analysed using FlowJo (Tree Star). Cell populations of interest were sorted onto Poly-L-Lysine Slides (Thermo Scientific) and fixed in 4% PFA for IF analysis. incubated with primary (Suppl. Table 5) and secondary antibodies as specified for hESC IF above, but 1 g/ml DAPI was added to the secondary antibody mixtures. Finally, the slides were washed in the wash buffer twice and in PBS once and mounted with Prolong Gold Antifade Reagent (Molecular Probes). The images were acquired using Leica SP5 confocal microscope and analysed using Fiji software.
For fluorescence intensity quantification a Fiji plugin written by Dr. Richard Butler (Gurdon Institute, University of Cambridge, UK) was used. It counts 30-300 µm² nuclei and measures fluorescence intensity in each channel. The data was then manually processed in Microsoft Excel to filter pluripotent cells (NANOG or OCT4 fluorescence intensity >50) where applicable and to calculate mean fluorescence intensities. Statistical analyses were performed using GraphPad Prism 7. Table 5. Antibodies used in the study.
All other plasmids, including donor vectors for homology-directed repair were generated using In-Fusion cloning (Clontech) according to manufacturer's recommendations, but scaling down the reaction to a total volume of 5 μl. Primers used for cloning are specified in Suppl. For knockins either electroporation or lipofection was used. For electroporation, ~250,000 trypsinised hESCs were resuspended in 600 μl PBS+Ca2++Mg2+ containing 50 μg gRNA plasmid and 50 μg homologous repair donor plasmid and electroporated in a 0.4-cm cuvette using Gene Pulser Xcell System (Bio-Rad) with a single 20 ms square-wave pulse (250 V). Lipofections were performed using 2 μg gRNA plasmid and 2 μg donor construct in OptiMEM medium (GIBCO) and Lipofectamin Stem reagent (Invitrogen) according to manufacturer's recommendations. The volume of lipofectamine in μl was equal to the total amount of DNA in μg. For lipofection of PiggyBac transgenes a total of 1-5 μg DNA/100,000 cells was used, with the amount of PiggyBac transposase (PBase)-encoding plasmid equal to that of PiggyBac-delivered transgenes combined (in μg). Transfected cells were seeded onto drug-resistant DR4 MEFs (SCI) in 4i medium (ROCKi added for the first 24 hours; selection was initiated 48 hours after transfection).
After selection, individual clones were picked, expanded and genotyped using PCR. For this, genomic DNA was first extracted from cell pellets in the lysis buffer (10mM Tris pH 8.0, 100mM NaCl, 10mM EDTA and 0.5% SDS; Proteinase K (PK) was added immediately prior to lysis at a final concentration of 0.2 mg/ml) at 56 ºC for 4 hours -overnight, followed by PK inactivation at 98 ºC for 10 minutes. The supernatant was used directly in genotyping PCR performed with LongAmp (NEB) or PrimeStar GXL (Clontech) polymerase according to the manufacturers' protocols. Primers used for genotyping are specified in Suppl. Table 6. In the case of transgenes expression, homogeneity and leakiness of induction was checked by IF.
For knockins (PRDM14-T2A-Venus, PRDM14-AID-Venus, PRDM14-JAZ-Venus and SOX17-AID-Venus) correct targeting in homozygous (by genotyping) clones was confirmed by Sanger sequencing. This was followed by the removal of the Rox-flanked puromycin resistance cassette by transient transfection with 1 μg of Dre-recombinase-encoding plasmid 6 . 2 days of hygromycin B selection (50 μg/ml) were followed by 2 days of negative selection using FIAU (200 nM) 7 . The obtained clones were again genotyped by PCR and sequenced to confirm antibiotic cassette excision. Venus intensity and homogeneity were then verified using Flow cytometry and IF.
For generation of cell lines that deplete PRDM14 upon IAA supplementation, AID-Venus was added at the Cterminus of PRDM14 using CRISPR (see above). This was followed by the addition of TIR1 transgene (pPB-CAG-OsTir1-myc-IRES-HygR or pPB-CAG-OsTir1-V5-T2A-Puro) by lipofection, along with pPBase. TIR1 cDNA and the 44-amino-acid AID sequence were kindly provided by Dr. Elphège Nora.
For generation of cell lines that deplete PRDM14 upon Cor supplementation, JAZ-Venus was added at the Cterminus of PRDM14 using CRISPR (see above). This was followed by the addition of COI1B transgene (pPB-CAG-HA-FboxOsTir1-OsCoi1b-IRES-HygR) by lipofection, along with pPBase. COI1B cDNA and the 43-amino-acid JAZ sequence were kindly provided by Dr. Ran Brosh.

Reverse-transcription quantitative PCR (qPCR)
Total RNA was extracted using RNeasy Mini Kit (QIAGEN) from unsorted cells or using Arcturus PicoPure (Thermo Fisher Scientific) from at least 1,000 sorted cells. cDNA was synthesized using QuantiTect Reverse Transcription Kit (QIAGEN). qPCR was performed on a QuantStudio 12K Flex Real-Time PCR machine (Applied Biosystems) using SYBR Green JumpStart Taq ReadyMix (Sigma) and specific primers (Suppl. Table 6). The ΔΔCt method was used for quantification of gene expression. Statistical analyses were performed using Microsoft Excel and/or GraphPad Prism 7.

RNA-sequencing
RNA-seq was performed on PRDM14-AID-Venus competent 4i hESCs and hPGCLCs induced therefrom. Two biological replicates were used for each condition (Suppl. Table 7). For IAA-sensitive cells, two clones (cl11 and cl21) from the same hESC passage or the same hPGCLC induction were used as replicates. For the parental ("no TIR1") control, the same cell line was used at different passages or inductions to yield two independent replicates. 10,000 AP+ 4i hESCs or 10,000 NANOS3-tdTomato+AP+ hPGCLCs (with the exception of hPGCLC cl21 replicate, where 3,000 cells were used) were sorted directly into 100 μl of extraction buffer from PicoPure RNA Isolation Kit (Applied Biosystems) for subsequent total RNA extraction according to manufacturer's protocol. RNA was stored at -80 ºC and its quality and quantity were checked by Agilent RNA 6000 Pico Kit with Bioanalyzer (Agilent Technologies) and Qubit (Thermo Fisher Scientific). RNA-seq library was prepared from 10 ng input RNA using the end-to-end Trio RNA-seq library prep kit (Nugen) following the manufacturer's protocol but omitting the AnyDeplete step. In short, the protocol contains the following steps: DNAse treatment to remove DNA from RNA; first strand and second strand cDNA synthesis to produce the reverse complement of the input RNAs; cDNA purification using Agencourt AMPure XP beads (Backman Coulter); single primer isothermal amplification (SPIA) to stoichiometrically amplify cDNAs; enzymatic fragmentation and end repair; sequencing adaptor (index) ligation; product purification using AMPure beads; library amplification (4 cycles were used); library purification using AMPure beads. Libraries were then quantified by qPCR using NEBNExt Library Quant Kit (NEB) for Illumina on QuantStudio 6 Flex Real-Time PCR System (Applied Biosystems). Fragment size distribution and the absence of adapter dimers was checked using Agilent TapeStation 2200 and High Sensitivity D1000 ScreenTape. Finally, RNA-seq libraries were subjected to single-end 50 bp sequencing on HiSeq 4000 sequencing system (Illumina). 24 indexed libraries (including samples from this work and 8 others) were multiplexed together and sequenced in two lanes of a flowcell, resulting in >30 million reads per sample. Table 7. Summary of RNA-seq samples. Note that two different IAA-sensitive clones (cl11 and cl21) were used as replicates. The samples of the parental cell line (Par) lacking TIR1 were obtained in two replicates (reps).

RNA-seq analysis
The library sequence quality in demultiplexed fastq files was checked by FastQC (v0.11.5) 9 and the lowquality reads and adaptor sequences were removed by Trim Galore (v0.4.1) 10 using the default parameters. The pre-processed RNA-seq reads were mapped to the human reference genome (UCSC GRCh38/hg38) using STAR (2.6.0a) 11 (--outFilterMismatchNoverLmax 0.05 --outMultimapperOrder Random --winAnchorMultimapNmax 100 --outFilterMultimapNmax 100) guided by the ENSEMBL (Release 87) gene models. Read counts per gene were extracted using TEtranscripts 12 and normalized by DEseq2 in R 13 . Differential expression analysis was also performed using DEseq2. The resulting gene expression table (Suppl. Table 9) was used for downstream analyses in Microsoft Excel and R. Pearson's correlation analysis was performed using the R cor command. Unsupervised hierarchical clustering was performed using the R hclust function with (1-Pearson's correlation coefficient) as distance measures. The R prcomp function was used for Principal component analysis (PCA). Gene set enrichment analysis was performed using the GSEA software by the Broad Institute 14 . Gene Ontology (GO) analysis was performed using DAVID 15 .
For comparative differential expression analysis of PRDM14 depletion in hPGCLCs and TFAP2C and PRDM1 KOs from 16 the samples were identically pre-processed and mapped to the human reference genome (UCSC GRCh38/hg38), sequencing reads were re-normalised together using DEseq2 to generate a new integrated gene expression table (Suppl. Table 11). A similar procedure was performed for comparisons with mouse Prdm14 KO from 17 , where processed reads were mapped to the mouse reference genome (UCSC GRCm38/mm10), and high-confidence human-mouse "one-to-one" orthology assignments for protein-coding genes were obtained from ENSEMBL (Release 87) (Suppl. Table 13). Venn diagrams were plotted using VennPainter 18 and p-values for two overlapping datasets were calculated using a generalized hypergeometric test for multiple samples 19 . For the co-expression analysis of differentially expressed genes, the gene-based Pearson's correlation coefficients were calculated for PRDM14-AID-Venus competent hESC and hPGCLC samples using R. All pairwise correlations r < 0.8 between genes were removed. The matrix was imported as an adjacency matrix into the R igraph package 20