Preformation and epigenesis converge to specify primordial germ cell fate in the early Drosophila embryo

A critical step in animal development is the specification of primordial germ cells (PGCs), the precursors of the germline. Two seemingly mutually exclusive mechanisms are implemented across the animal kingdom: epigenesis and preformation. In epigenesis, PGC specification is non-autonomous and depends on extrinsic signaling pathways. The BMP pathway provides the key PGC specification signals in mammals. Preformation is autonomous and mediated by determinants localized within PGCs. In Drosophila, a classic example of preformation, constituents of the germ plasm localized at the embryonic posterior are thought to be both necessary and sufficient for proper determination of PGCs. Contrary to this longstanding model, here we show that these localized determinants are insufficient by themselves to direct PGC specification in blastoderm stage embryos. Instead, we find that the BMP signaling pathway is required at multiple steps during the specification process and functions in conjunction with components of the germ plasm to orchestrate PGC fate.

Introduction Sexual reproduction enables multicellular organisms to transmit genetic information from one generation to the next. The process is initiated by the differentiation of the goniablasts produced by the asymmetric division of male and female germline stem cells (GSCs) and culminates with the fusion of male (sperm) and female (egg) gametes to produce the embryonic zygote. The GSCs of sexually mature animals do not arise de novo. Rather, they are derived from a special group of cells, the primordial germ cells (PGCs), that are set aside from the remaining somatic cells during a very early phase of embryonic development. Since the proper specification of PGCs is critical for progression of the reproductive cycle, mechanisms underlying their formation and fate specification have been a major focus of investigation [1][2][3][4][5][6].
Many characteristic features of PGC specification are broadly conserved in the animal kingdom. One of the key PGC specification steps in most animals is the downregulation of transcription. However, the mechanisms underlying the establishment and/or maintenance of transcriptional quiescence are diverse [6][7][8][9][10]. For instance, in mammals, progenitors of the cells that ultimately give rise to the germline undergo zygotic genome activation (ZGA) like other cells in the embryo; however, after the initial steps in PGC specification, transcription of somatic genes is largely terminated, and these cells begin to revert to an earlier, pluripotent state [6][7][8]. In worms, the cell lineage that produces the germline is set aside at the first division, and, when transcription commences in the remaining somatic cells at the 3-4 cell stage, it remains transcriptionally quiescent, as do the daughter cells in the lineage destined to become germline [9,10].
In flies, PGCs are formed after the onset of the minor wave of ZGA, which begins at nuclear cycle (NC) 8. During their formation, ongoing transcription is turned off by germ-cell less (gcl) [11,12] while genes activated during the major wave of ZGA (NC14) are kept off by combined action of polar granule component (pgc) [13][14][15] and nanos (nos) [16][17][18]. Broad downregulation of transcription is reflected in the phosphorylation status of the RNA polymerase II CTD domain. PhosphoSer2, a modification correlated with transcriptional elongation is largely absent in newly formed PGCs, while PhosphoSer5, a modification linked to initiation, is substantially reduced [14,17,19,20]. A second common feature is alterations in histone modification profiles. In worms and flies, histone H3meK4, a modification associated with active transcription, is largely absent in young PGCs. In both organisms, the germline determinant nos contributes to the inhibition of this histone modification [21]. Besides nos, suppression of the H3meK4 modification in flies requires the H3meK4 demethylase, Su(var)3-3, the Swi/Snf chromatin remodeling complex subunit, osa [14], and the transcriptional regulator pgc [13,14]. There are also alterations in the heterochromatic histone modification, H3meK9 [21]. Like worms and flies, mouse PGCs also display changes in histone modifications [20,22,23]. A third common feature is pausing the cell cycle. In worms, flies, and mammals, PGCs arrest cell cycle in G2 [17,20,[24][25][26]. In addition to these common characteristics, genes implicated in PGC specification in model invertebrates (flies or worms), like nos, vasa, and piwi, are also conserved in higher animals (mice, human etc.) [1,8,9]. While many of the characteristics of PGCs that distinguish them from the soma are widely shared amongst different animal species, there is one striking dichotomy, namely whether the mechanism driving specification is "epigenesis" or "preformation." In epigenesis, specification is non-autonomous and depends upon cell-cell signaling. In preformation, specification is autonomous and is driven by determinants that are localized in the presumptive PGCs. Mammals utilize epigenesis. In pre-implantation embryos, a combination of inductive signals from extraembryonic ectoderm and visceral endoderm acts to induce cells within the posterior epiblast to become PGCs [6,7,27]. Signaling appears to be at least a two-step process in which Wnt3 (Wingless 3) first primes cells in the epiblast. Once primed, the cells can respond to the BMP (bone morphogenetic protein) ligands Bmp4 and Bmp8b, which are secreted by the extraembryonic ectoderm, and Bmp2, which is secreted by the visceral endoderm [28][29][30][31]. These signals activate Smad transcription factors that set the transcriptional program required for PGC specification in motion [32][33][34]. By contrast, worms and flies are thought to employ an exclusively preformation mechanism involving localized cell-autonomous factors. In worms, localized PGC determinants segregate to only one of the daughter cells during each of the blastomere cell divisions that ultimately give rise to the two PGC founders, Z2 and Z3 [9,35]. In fly embryos, fertilization is followed by a series of rapid synchronous nuclear divisions which culminate in cellularization at the end of NC14 [36]. However, earlier, during NC9, several nuclei migrate to the posterior pole of the embryo and induce the formation of pole buds [37]. During bud formation, the centrosomes/microtubule network associated with each incoming nucleus triggers the release of localized PGC determinants from the posterior cortical cytoskeleton, and these factors are then incorporated into the newly formed PGCs during cellularization [37,38]. When these factors are not properly sequestered in the newly formed PGCs, PGC specification fails [12,[38][39][40].
The master determinant that orchestrates PGC formation and subsequent specification in flies is oskar (osk) [41]. osk mRNAs are localized to the posterior pole of the oocyte during oogenesis and translated during mid-oogenesis [42]. Osk protein then mediates the recruitment and assembly of key components of the germ plasm including Vasa, Tudor, Valois, and Aubergine proteins and pgc, gcl, and nos mRNAs. Once assembled at the posterior of the egg, the germ plasm is sufficient to induce the formation of PGCs. Illmensee and Mahowald showed that injection of pole plasm at the anterior of the embryo induces the formation of ectopic PGCs [43]. This finding was recapitulated by Ephrussi and Lehmann, who replaced the osk 3' UTR with a bicoid (bic) 3' UTR [44]. They found that pole plasm assembled by the ectopic Osk protein was sufficient to induce the formation of fully functional PGCs at the anterior of the embryo. Furthermore, overexpression of osk increases the number of PGCs and can also induce ectopic PGCs on the dorsal side of the embryo [45].
These and other findings have reinforced the idea that a preformation mechanism-relying exclusively on localized determinants-is responsible for PGC specification in flies. However, later in embryogenesis, PGCs are neither indifferent nor immune to extracellular signals. As is the case in other organisms, fly PGCs must migrate from their site of formation at the posterior pole through the embryo to coalesce with the somatic gonad precursor cells (SGPs) [46]. Their migration through the mesoderm is mediated by the Hedgehog (Hh) signaling pathway, and they are directed by potentiated Hh ligand produced by the SGPs [47][48][49]. During roughly the same time frame (stage 10-14 of embryogenesis), fly PGCs not only respond to BMP signaling but are dependent on BMP signals to maintain their identity. When the BMP receptor thickveins (tkv) is knocked down using the germline-specific nos-Gal4 driver, Vasa protein accumulation in the PGCs is disrupted while the assembly of the germline-specific spectrosome is compromised [50]. While hh directed PGC migration appears to be mediated by a novel nontranscriptional pathway [49,51], maintenance of PGC fate in stage 9-14 embryos depends upon canonical transcription factors downstream of BMP. Nos-Gal4 dependent RNAi knockdown of the Smad co-factor medea disrupts Vasa protein accumulation. Similarly, overexpression of smurf, a ubiquitin E3 ligase that targets phosphorylated (and active) Smad for degradation, induces defects in Vasa accumulation and spectrosome assembly [50].
These findings show that during mid-embryogenesis, in the period leading up to gonad coalescence, the BMP pathway not only helps maintain PGC identity but also functions to promote PGC differentiation into GSCs. This would suggest that, at least in this time frame, maternal determinants alone are not sufficient for either maintenance of PGC identity or promoting differentiation. If this is the case, a relevant question is whether localized maternal determinants are on their own sufficient for the specification of PGCs in pre-cellular blastoderm embryos or if they also require input(s) from the BMP signaling pathway even at this stage of development. Here we have addressed this question.

Early embryonic overexpression of dpp results in a modest increase in PGC number due to enhanced mitosis
Studies by Dorfman and Shilo showed that newly formed PGCs in syncytial and cellular blastoderm embryos are not immune to the BMP signaling pathway [52]. Like their somatic counterparts on the dorsal side of the embryo, activated pMad accumulates to high levels in PGC nuclei, and this accretion depends on both the tkv receptor and its ligand, decapentaplegic (dpp) [53][54][55]. As would be expected from the accumulation of pMad in PGC nuclei, previous studies on the expression of dpp mRNA and the Dpp ligand have shown that though the dpp expression domain in early embryos is restricted to the dorsal side of the embryo, it encompasses the entire posterior (and anterior) pole of the embryo [56]. In germline clone embryos homozygous for a tkv mutation (FRTtkv 8 referred to as tkv mhenceforth), pMad is completely lost in the soma and in newly formed PGCs. When tkv membryos are fertilized by a WT tkv sperm, zygotic expression can partially mitigate defects in pMad accumulation both in the soma and in newly formed PGCs. This would indicate that unlike many other genes whose transcription is specifically turned off in PGCs, transcription of tkv is not [52]. The fact that tkv is amongst the few genes known to be transcribed in newly formed PGCs suggests that pMad accumulation may have an important function. On the other hand, it is also possible that pMad-dependent induction of downstream targets is blocked by the maternal determinant(s) that downregulate global transcription in newly formed PGCs. In this case, the PGCs would be immune to the normal activity of nuclear pMad.
To test whether newly formed PGCs are responsive to inputs from the BMP signaling pathway, we generated an excessive level of Dpp using twist-Gal4 to drive expression of a UAS-dpp transgene on the ventral side of the embryo. Consistent with the idea that PGCs retain the ability to respond to BMP signaling, likely via the phosphorylation and nuclear localization of pMad, we find that excess Dpp results in a modest but significant increase in the number of PGCs both in stage 4 syncytial blastoderm (S1A and S1B Fig and S1 Table) and in stage 5/6 i.e. late syncytial/cellular blastoderm stage embryos (S1C and S1D Fig) (26.4 PGCs/ embryo in twi-Gal4/UAS-dpp, n = 16; versus 21.9 PGCs per embryo, n = 20 in control; p = 0.0019 by ttest). A similar elevation in PGC count was also observed in syncytial /cellular blastoderm stage embryos when Dpp levels were enhanced using maternal tubulin-Gal4 (S1 Table; p = 0.027 by t-test).
Since WT PGCs divide only 0-2 times after they are formed and then cease division by the time the embryo cellularizes [57], the increase in the number of PGCs could be due to a failure to fully exit the cell cycle. If this idea is correct, the frequency of PGCs in mitosis should be elevated in twi-Gal4/UAS-dpp embryos. To test this possibility, we identified PGCs in mitosis using phospho-Histone 3 (pH3) antibody [58,59]. In WT, pH3 is detected relatively infrequently in PGCs in the period between their formation and the end of NC14 (Fig 1A and S2 Table; 0.7 pole cells per embryo, n = 18). By contrast, this number is considerably elevated (~4 fold) in twi-Gal4/UAS-dpp embryos as shown in Fig 1B (2.8 pole cells per embryo, n = 18; p = 1.6e-5 by t-test) (Fig 1B and S2 Table). This suggests that the Dpp signaling pathway promotes proliferation of newly formed PGCs.
To further analyze how excess Dpp might promote PGC division, we examined expression of Cyclin B (CycB). In WT PGCs, translation of cyclin B mRNA is specifically targeted for repression by the Nanos:Pumilio complex [25,26]. Therefore, only very low levels of CycB are detected in WT PGCs (Fig 1C). In contrast, in twi-Gal4/UAS-dpp embryos, CycB is readily detected in many PGCs (Fig 1D and S2

PGCs maintain transcriptional quiescence in the presence of excess BMP signaling
Accompanying the shutdown of mRNA transcription in newly formed PGCs, the phosphorylation of Ser2 and Ser5 in the heptad repeats in the CTD (C-terminal domain) of the large Pol II subunit is substantially reduced [13,14,19,21,60]. However, after transcription is upregulated in PGCs, the level of both CTD modifications increases [19]. Since phosphorylation and nuclear localization of pMad modulates transcription in somatic cells [61], we wondered whether excess Dpp disrupts the global downregulation of transcription normally observed in WT PGCs. As the accumulation of pSer2 faithfully reports on transcriptional elongation, we examined the levels of pSer2 in the twi-Gal4/UAS-dpp and WT PGCs. However, there was no detectable difference between twi-Gal4/UAS-dpp and WT embryos (S2 Fig and S3 Table; p = 0.676744 by Fisher's exact test). Thus, PolII transcription in PGCs is not detectably increased by excess Dpp.

Vasa accumulation in newly formed PGCs depends upon the BMP signaling pathway
The increase in mitotic activity induced by excess Dpp indicates that PGCs are neither indifferent nor immune to BMP signaling. A plausible inference from this finding is that, as is the case in mammals, BMP signaling contributes to the specification of PGC identity in blastoderm stage fly embryos. To start exploring this possibility, we first examined Vasa protein accumulation in PGCs of two different genetic backgrounds. The first is a viable, partial loss of function (LOF) allele of the dpp ligand, dpp hr92 [62], while the second is a viable partial LOF allele of the tkv receptor, tkv 427 [63]. pMad levels have been shown to decrease in the embryos compromised for components of dpp signaling, including dpp itself [52]. Accompanying this reduction in pMad, there is a decrease in Vasa, a marker of germline identity (Fig 2B). For dpp hr92 , we found that 38% of the dpp hr92 PGCs had noticeably reduced levels of Vasa (n = 63) compared to only 8% for WT (n = 49) (S4 Table,  A similar result was obtained when we analyzed Vasa in PGCs of embryos produced by tkv mgermline clone mothers mated to WT fathers (tkv m-z+ ). We found 32.8% of tkv m-z+ PGCs (n = 48; p = 3.4e-5 by Fisher's exact test) showed reduced levels of Vasa compared to 2.1% of PGCs in the control syncytial blastoderm embryos (n = 47, S4 Table).
These results show that BMP pathway components impact accumulation of Vasa, a PGCspecific marker, early in embryogenesis. While this observation indicates that BMP signaling impacts PGC specification, there is a potential complication. In all three experiments, either the mother or the maternal germline was homozygous for a mutation in dpp or tkv. While there is no known role for the BMP pathway in the assembly of functional pole plasm during oogenesis, this remains a formal possibility. For this reason, we took two different approaches to ascertain whether the effects on Vasa are strictly zygotic, rather than a consequence of some undocumented role for BMP signaling in pole plasm assembly/functioning. In the first, we used two UAS:dppRNAi transgenes (dppi 33618 and dppi 25782 ) to knock down expression of dpp in blastoderm stage embryos. In these experiments, we mated mothers carrying a mat-tubulin-Gal4 transgene to fathers homozygous for each UAS-dpp RNAi transgene. Knockdown of dpp in pre-cellular blastoderm embryos disrupts Vasa accumulation in newly formed PGCs ( Fig  2D). We found that 31.7% of the PGCs (n = 145) had reduced Vasa compared to 2.9% of the PGCs (n = 136) in control embryos in which mat-tublin-Gal4 mothers were mated to fathers carrying an UAS:egfpi transgene (S4 Table; p = 0.0 by Fisher's exact test). Though somewhat less effective, the dppi 25782 transgene gave similar results (21.1% of the PGCs, n = 180) had reduced levels of Vasa (S4 Table; p = 1.0e-6 by Fisher's exact test).
In the second approach, we examined Vasa accumulation in PGCs of embryos homozygous for a strong loss-of-function mutation of dpp, dpp 4 . In order to identify homozygous dpp 4 blastoderm stage embryos, the dpp mutation was recombined with a twist protein null. Embryos collected from a cross between heterozygous dpp 4 twi/CyO parents were probed with a combination of Vasa and Twi antibodies. As was observed in the zygotic RNAi knockdowns, Vasa levels were reduced in PGCs of homozygous dpp 4 twi embryos. We found that 27.8% of the PGCs (n = 248) in embryos that had no detectable Twi protein had reduced levels of Vasa, compared to 0.9% of the PGCs (n = 216) in the Twi-positive siblings (S4 Table; p = 0.0 by Fisher's exact test).

Transcriptional quiescence is partially compromised when the BMP pathway is disrupted
The loss of Vasa in newly formed PGCs suggests that the BMP pathway functions at an early step in PGC specification. If so, other hallmarks of newly formed PGCs might also be disrupted. One of these is the establishment of transcriptional quiescence. To assess the impact of the BMP pathway on downregulating RNA Pol II transcription, we used several approaches. In WT, the signature for transcriptional elongation (pSer2) is absent in syncytial blastoderm stage embryo PGCs while it is readily detected in somatic nuclei. As shown for two different embryos ( Fig 3B and 3C), newly formed PGCs (marked by Vasa protein) in dpp hr92 embryos accumulate levels of pSer2 approaching that of surrounding somatic nuclei. Quantification of pSer2-positive PGCs indicates that 23.9% of the syncytial blastoderm stage dpp hr92 PGCs (n = 67; p = 0.000597 by Fisher's exact test) and 30% (n = 120; p = 1.0e-6 by Fisher's exact test) of tkv 427 PGCs have nuclear pSer2, as compared to 5.1% for WT PGCs (n = 98) (Figs 3 and S3 and S5 Table).
To further test the effects of the BMP pathway on RNA polymerase II activity, we examined the PGCs of embryos laid by tkv germline clone mothers (tkv m-z+ ). While PGCs in the WT control (Fig 3) only infrequently had detectable levels of pSer2 (4.6%, n = 43), over forty percent (41.2%, n = 51) of PGCs in the tkv m-z+ blastoderm stage embryos had pSer2 (S5 Table; p = 2.70e-5 by Fisher's exact test). Moreover, those tkv m-z+ PGCs with the highest levels of pSer2 often showed evidence of Vasa depletion (carets, Fig 3B and 3C).
pSer2 is not the only marker of transcriptional activity that is altered in PGCs when components of the BMP pathway are not fully active. In WT, the chromatin marker of active transcription, the histone H3 modification, H3meK4, is upregulated in the somatic nuclei of syncytial blastoderm embryos, while it is almost completely absent in PGC nuclei. In contrast, this marker is readily detected in tkv 427 PGCs (51.7% of tkv 427 PGCs, n = 56; p = 0.0 by Fisher's exact test), as opposed to 2.9% of WT PGCs (n = 69) (S4 Fig). Furthermore, weaker but significant elevation in the H3meK4 signal was also observed in the PGCs from dpp hr92 embryos (22.2%, n = 54; p = 0.001077 by Fisher's exact test) (S5 Table). Taken together, these observations suggest that loss of BMP signaling in early embryos disrupts the proper downregulation of transcription in PGCs.

Genes normally silenced in newly formed PGCs are ectopically expressed when the BMP pathway is compromised
Previous studies have shown that the three maternal factors known to be responsible for repressing transcription in newly formed PGCs, Gcl, Nos, and Pgc, function at different times and target an overlapping set of genes. gcl functions during PGC cellularization. Gcl is responsible for shutting off transcription of genes that are activated in the minor wave of ZGA. Its targets include scute (sis-b), sis-a, and runt [11,12]. It also has a role in silencing the Sxl establishment promoter, Sxl-Pe. Pgc and Nos function after PGC formation but have different activities and gene targets. pgc inhibits the kinase, pTFb, that phosphorylates the Ser2 residue in the Pol II CTD domain. Its known targets include tailless (tll) and slow as molasses (slam) [13,14]. While it is not understood how nos inhibits Pol II transcription, its known targets include fushi-tarzu, even-skipped, and the Sxl-Pe promoter [17].
To test the effects of compromising BMP signaling on transcriptional activity, we selected tll, slam, and Sxl-Pe and examined their expression using a combination of fluorescent in situ hybridization (FISH) and single molecule FISH (smFISH). The tll and slam probes were directed against sequences in the corresponding mRNAs while the Sxl-Pe probe is homologous to a large intron in the Sxl-Pe transcript and thus only detects nascent mRNAs. As shown in Fig 4, we find that tll mRNA is expressed in a subset of dpp hr92 PGCs (Fig 4). While all WT embryos had no tll transcription in PGCs (n = 19), 70.4% of dpp hr92 embryos displayed ectopic tll transcription (n = 27; p = 1.0e-6 by Fisher's exact test). To further quantify tll mRNA expression in dpp hr92 PGCs compared to WT, we normalized tll staining intensity in PGCs with respect to adjacent somatic cells. As shown in the plot, there is a significant increase in the average level of tll signal in the dpp hr92 PGCs compared to WT (Fig 4H). In WT embryos, slam transcripts begin to appear at syncytial blastoderm stage, and the levels rise considerably in cellular blastoderm stage embryos. slam RNA also associates with Slam protein, which decorates membranes [64]. Consequently, slam probe directed against the mRNAs labels the extending membrane compartment of somatic nuclei/cells. slam is not, however, transcribed in PGCs and they are normally devoid of slam mRNA. Using smFISH, we detected slam expression in a subset of the dpp hr92 PGCs (Fig 5B: see asterisks). Approximately 26.3% of the dpp hr92 (n = 38; p = 0.000991 by Fisher's exact test) embryos have one or more PGCs that express slam mRNA (S6 Table). We also examined PGCs in embryos carrying the dpp RNAi line dppi 33618 (Fig 5C: see asterisks). We find that 35.7% of the dppi 33618 express slam (n = 14; p = 0.000852 by Fisher's exact test) as compared to 0% for WT embryos (n = 37). For Sxl-Pe, the frequency of dpp hr92 embryos that that have detectable transcripts is lower that for tll and slam. We find that 18.2% of the dpp hr92 embryos (n = 11; p = 0.47619 by Fisher's exact test) have Sxl-Pe transcripts, while Sxl-Pe transcripts are not observed in WT embryos (n = 10) (S6 Table). It seems likely that we detect Sxl-Pe transcripts much less frequently in dpp hr92 PGCs because the Sxl-Pe probe only

PLOS GENETICS
hybridizes to nascent transcripts, and transcription bursts are relatively infrequent. Furthermore, we have found that ectopic Sxl-Pe expression does show a slight sex bias in gcl embryos [12]. A similar bias may render detection of Sxl transcripts relatively rarer compared to the sex-nonspecific somatic genes (e.g. slam, tll).

Compromised BMP signaling correlates with inappropriate distribution of pole plasm components
In analyzing PGCs in embryos compromised for BMP signaling, we noted that Vasa protein was not always fully sequestered in newly formed PGCs (S5 Fig, caret). This led us to wonder if there were defects in the distribution of pole plasm constituents during PGC formation. To explore this possibility, we used smFISH to examine the localization of two pole plasm mRNAs, pgc and gcl, in dpp hr92 and dppi 33618 embryos. Fig 6 shows maximum intensity projections of pgc mRNA in WT (egfpi), dpp hr92 , and dppi 33618 embryos. In control embryos, pgc mRNA is efficiently incorporated into PGCs when they cellularize, and little, if any, escapes to the surrounding soma (Fig 6A). In contrast, in both dpp hr92 and dppi 33618 pre-cellular blastoderm embryos, incorporation of pgc mRNA into PGCs seems less efficient, and pgc mRNA is observed in the surrounding soma where it is associated with somatic nuclei (Fig 6B and 6C). The failure to properly sequester pgc mRNA in the newly formed PGCs is documented in the plots of individual embryos (Fig 6D). gcl mRNA is also not captured efficiently by PGCs in dpp hr92 and dppi 33618 .

PLOS GENETICS
The defects in sequestering pole plasm constituents during PGC formation could arise from a defect in the anchoring of the pole plasm RNAs and proteins to the posterior cortex and/or the premature release of these factors. Such defects should be apparent in pre-syncytial embryos prior to nuclear migration into the pole plasm. To assess this possibility, we analyzed the localization of pgc and gcl RNAs in young embryos before pole cell formation. Unlike in later stages, we found that pgc and gcl mRNAs are tightly associated with the posterior cortex in all young embryos prior to pole bud formation (n = 5 for each genotype; S6 Fig). Taken together, these data argue that BMP signaling is required for proper sequestration of pole plasm components in PGCS at step during, or more likely after, the release of the pole plasm from the posterior cortical surface.

BMP signaling is required to suppress the functioning of the terminal pathway
One of the characteristic phenotypes in embryos from gcl mutant mothers is a failure to properly sequester pole plasm components during PGC cellularization. As is observed when expression of Dpp is disrupted, pgc mRNA and Vasa protein are not properly incorporated into gcl mutant PGCs [12]. The failure to properly sequester the pole plasm during PGC formation appears to be due, at least in part, to the inappropriate activation of the terminal signaling pathway in gcl mutants. Studies by Pae et al. (2017) [65] showed that Gcl targets the terminal pathway receptor Torso for proteolytic degradation, thereby shutting down the pathway at the very posterior pole of the embryo. When the terminal signaling cascade is activated at the posterior pole, either in gcl mutants or by gain-of-function mutations in the Torso receptor or its downstream kinases, pole plasm components are not properly incorporated into PGCs when they cellularize [12]. These observations raise the possibility that the BMP pathway might be needed to help suppress the terminal signaling pathway. To test this directly, we probed WT and dpp hr92 embryos with antibodies against the di-phosphorylated form of the downstream ERK kinase, dpERK, which has been used as a diagnostic marker of pathway activation. WT PGCs have very little, if any, dpERK. In contrast, PGCs in dpp hr92 embryos accumulate readily detectable levels of dpERK (S7 Fig). We found that 66.0% (n = 47; p = 0.0 by Fisher's exact test) of the PGCs in syncytial blastoderm stage dpp hr92 embryos have dpERK, while dpERK is found in only 9.8% of the PGCs of equivalently staged WT embryos (n = 51).

Activation of the terminal pathway at the posterior induces Vasa loss
The presence of dpERK in PGCs when the BMP signaling pathway is compromised indicates that one of its functions in PGC specification is suppressing terminal signaling. To provide further evidence for this connection, we examined Vasa accumulation in PGCs in which the terminal signaling pathway was ectopically activated. For this purpose, we used a light activated SOS protein, optoSOS, to turn on the terminal pathway at the posterior pole of the embryo [66,67].  that pgc and gcl mRNAs are not properly incorporated into PGCs unless the terminal signaling pathway is shutdown during their cellularization, would provide additional support for the idea that this pathway is an important target for the BMP signaling pathway in PGC specification. On the other hand, it seems likely that the terminal pathway is not the only BMP target: we found that accumulation of dpERK either via optogenetic activation or expression of a constitutively active form of upstream kinase does not lead to inappropriate and/or precocious activation of somatic target genes such as tll [68].

Loss of dpp results in precocious invasive migration of PGCs
Hyperactivation of the terminal pathway in late syncytial or early cellular blastoderm embryos is known to induce an unusual change in the behavior of PGCs. Instead of remaining adhered to one another in a monolayer on the surface of the embryo, a subset of the PGCs lose adhesion and begin to invade the underlying soma. Since loss of BMP signaling induces ERK phosphorylation in PGCs, one might expect to observe evidence of invasive migration. As indicated in S7 Table, invasive migration is observed only infrequently in dpp hr92 embryos. Invading PGCs were observed in only 13.3% of dpp hr92 embryos (S7 Table; p = 0.484127 by Fisher's exact test). On the other hand, in the case of the strong hypomorph allele, dpp 4 , invasive migration was observed in nearly one half of the dpp 4 embryos (Fig 8 and S7

Dominant genetic interactions between osk and dpp
Our findings demonstrate that the early steps in PGC specification are disrupted when the functioning of the BMP signaling pathway is compromised. Factors important for the specification process are not properly segregated into newly formed PGCs, and critical steps in establishing PGC identity such as shutting down the terminal pathway and imposing transcriptional quiescence are disrupted. These steps are also known to be dependent upon the osk gene, which encodes the maternal determinant that orchestrates the assembly of the pole plasm during late stages of oogenesis [44,45]. In this case, one might anticipate that there would be synergistic genetic interactions between osk and the BMP pathway even though their critical functions are required in different contexts-osk in the mother and BMP signaling in the zygote.
To test this prediction, we reduced osk activity in the mother and BMP signaling in the zygote by mating osk A87 /+ heterozygous mothers to dpp hr92 homozygous fathers. All the embryos derived from this cross i.e. "dpp hr92 /+; osk/+" have a WT dpp gene trans to the

PLOS GENETICS
hypomorphic (and homozygous viable) dpp hr92 allele. As a control, we used embryos obtained from osk A87 /+ mothers (osk/+) mated to WT males. We expected that if dpp signaling functions in collaboration with the osk-dependent specification pathway, simultaneous reduction in the activity of both may impact the formation and/or specification process.
While PGC formation appeared to be normal, there were defects in specification. While the accumulation and distribution of Vasa is perturbed in both osk/+ and dpp hr92 /+; osk/+, the frequency of defects are greater in dpp hr92 /+; osk/+ (Fig 8A, 8B, and 8C). In osk/+ embryos, 42.3% (n = 26) had reduced levels of Vasa, while in dpp hr92 /+;osk/+ embryos, reductions in Vasa were evident in 69.2% (n = 26) of the PGCs (S9 Table; p = 0.092935 by Fisher's exact test). The loss of Vasa is likely due, at least in part, to the failure to properly segregate germ plasm components into the PGCs during the process of cellularization. In this case, we would expect to observe the aberrant localization of other germ plasm components into the surrounding soma. Fig 9 shows the localization of pgc mRNAs in osk/+ and dpp hr92 /+;osk/+ embryos. The efficiency of incorporation of pgc mRNAs into osk/+ PGCs in this experiment (Fig 9A, 9C, and 9E) appears similar to WT (Fig 6A). In contrast, pgc mRNAs are not properly captured by the PGCs in dpp hr92 /+;osk/+ embryos (Fig 9B, 9D, and 9F) and instead spread into the soma. To extend these observations, we simultaneously hybridized osk/+ and dpp hr92 /+;osk/+ embryos with slam and pgc mRNA probes. As shown in Fig 10B and 10D, pgc mRNA is found associated with somatic nuclei that are expressing slam mRNAs (see asterisks). In this experiment, we also occasionally observed pgc mRNA associated with somatic nuclei expressing slam mRNA in the osk/+ control embryos (Fig 10A and 10C, caret)). However, this occurs much less frequently, indicating that mislocalization of pgc mRNA is greatly exacerbated by a modest reduction in dpp activity.
After PGC formation and their initial specification, they remain adhered to one another and cluster together in a relatively uniform monolayer overlying the posterior pole of syncytial and cellular blastoderm embryos. While this is also true for most of the osk/+ control embryos, we observed striking defects in PGC grouping at the posterior pole in dpp hr92 /+;osk/+ embryos, presumably from loss of adhesion (Figs 8B and S8). 56.7% of these embryos (n = 78; p = 0.000151 by Fisher's exact test) have one or more PGCs that are detached from the smFISH was performed using probes specific for pgc (green) and slam (magenta) on 0-4 hr paraformaldehyde-fixed (A/C) osk/+ and (B/ D) dpp hr92 /+;osk/+ embryos to assess spread of pole plasm from the posterior. Caret indicates that some somatic nuclei are exposed to pgc in osk/+ embryos. Asterisk highlights the more severe mislocalization of pgc and overlap of pgc and slam signals in dpp hr92 /+;osk/+ embryos. Images are representative maximum intensity projections Scale bar represents 10 μm.

Discussion
Drosophila melanogaster is one of the classic and most thoroughly studied examples of organisms that use a "preformation" mechanism for PGC specification [1]. One of the first indications that PGCs in flies are specified by preformation came from experiments nearly 100 years ago in which the germline determinants were inactivated by UV irradiating the posterior of the embryo [69][70][71]. The existence of localized and UV-sensitive determinants was subsequently supported by transplantation experiments. Okada et al. showed that the sterility of UV irradiated embryos could be rescued by injection of cytoplasm from unirradiated embryos into the posterior region of the UV treated embryos [72]. Importantly, they found that rescuing activity is localized within the donor; it is present at the posterior, but not the anterior pole. In complementary experiments, cytoplasm from the posterior pole was found to induce PGCs when injected into the anterior [43]. Moreover, these ectopic PGCs were able to generate a functional germline if reinjected into the posterior pole of similarly staged recipients. Subsequent experiments showed that both PGC formation and abdominal patterning requires genes that are active in the mother during oogenesis and that the key maternal factor for the formation of the pole plasm is encoded by osk [41,42,44,73]. Ectopic localization of osk mRNAs engineered by linking it to bcd 3' UTR is sufficient for the assembly of pole plasm at the anterior of the oocyte [44]. In pre-cellular blastoderm embryos, the ectopic pole plasm induces the formation of PGCs at the anterior end. Moreover, as was first observed by Illmensee and Mahowald [43], these PGCs are able to populate the adult germline if transplanted into the posterior of recipient embryos.
While these and other studies [45] clearly demonstrate that maternal determinants localized at the posterior pole of the fly embryo orchestrate PGC specification, they do not establish that the mechanism is exclusively preformation. There is a potentially significant caveat with the experiments showing that PGCs induced at the anterior (either by injection of pole plasm or by the osk-bcd 3' UTR) are capable of generating a functional adult germline: the ectopic PGCs were transplanted into the posterior of recipient embryos and thus are subject to much the same milieu as PGCs formed by the normal mechanisms. This leaves open the possibility that epigenesis might play a role in PGC specification. To address this issue, we have asked whether the BMP signaling pathway is involved in the specification of fly PGCs when they are formed during the pre-cellular blastoderm stages of embryogenesis. We selected this pathway for four reasons. First, this pathway is known to function in PGC specification in animals that rely on epigenesis [1]. Second, though dpp expression in the zygote is restricted to the dorsal side of the embryo by the Dorsal morphogen, both in situ hybridization and antibody staining experiments indicate that dpp expression encompasses the entire posterior pole [56]. Third, experiments by Dorfman and Shilo showed that the transcriptional activator, pMad, induced in receiving cells by the BMP signaling pathway is present in the nuclei of pre-cellular blastoderm PGCs [52]. Moreover, pMad accumulation depends not only on the Dpp ligand, but also maternal and zygotic sources of the Tkv receptor. Lastly, in previous studies we found that BMP signaling is required to maintain PGC identity and their differentiation during midembryogenesis in the period leading up to the coalescence of the embryonic gonad [50].
Our results show that maternal determinants are insufficient on their own for proper specification of PGCs and that this process is not exclusively cell autonomous as has long been thought. Instead, a hybrid of preformation and epigenesis is deployed to generate a full complement of functional PGCs. We find that when the BMP pathway is disrupted in pre-cellular blastoderm embryos, newly formed PGCs exhibit a variety of defects indicative of a failure in PGC determination. Moreover, the same phenotypes are observed in PGCs in embryos whose mothers are mutant in one or more of the three maternal factors, gcl, pgc and nos, that are known to be required for proper PGC specification. It is also worth noting that the functioning of the BMP pathway was not completely disrupted in our experiments. Thus, the possibility remains open that more dramatic or even some additional phenotypic effects might be observed under conditions where BMP signaling was entirely absent in pre-cellular blastoderm embryos. (For instance, dpp activity may affect PGC cellularization. It will also be of interest to assess if pole plasm anchoring is completely normal in pre-blastoderm embryos if BMP pathway is compromised).
Included in the defects that were observed are the partial loss of Vasa protein, a failure to downregulate Ser2 CTD phosphorylation, changes in the profile of histone modifications and the phosphorylation of the terminal (and EGFR) signaling pathway protein ERK. In WT PGCs, phosphorylation of Ser2 in the CTD domain of the large PolII subunit by pTfb is blocked by the Pgc protein [13][14][15]. This block is overridden in embryos homozygous for the partial loss of function dpp hr92 allele and in embryos produced by tkvgermline clone mothers. Moreover, two of the known targets for Pgc repression, tll and slam, are expressed in dpp hr92 PGCs. However, the misfunctioning of Pgc is not the only defect in establishing transcriptional quiescence. We also observed activation of the Sxl establishment promoter, Sxl-Pe. Previous studies have shown Sxl-Pe transcription is inappropriately turned on in PGCs in the progeny of gcl and nos mothers, but not in the progeny of pgc mothers [12,13,17]. Thus, compromising BMP signaling would seem to broadly impact transcriptional quiescence, leading to the misexpression of genes normally repressed in PGCs by the activity of several different factors. In the case of tll, for example, Pgc is not the only pole plasm component expected to play a role in its repression. tll transcription is activated in somatic nuclei at the anterior and posterior ends of the embryo by the terminal pathway. However, the terminal pathway is normally shut down in WT PGCs by Gcl protein, which mediates the degradation of the Torso receptor [65]. gcl function must also be disrupted either directly or indirectly when the BMP signaling pathway is compromised as we find that dpERK accumulates in PGC nuclei just as in gcl mutants.
These are not the only connections between the BMP pathway, Gcl, and the terminal signaling pathway. In gcl embryos, components of the pole plasm (including Vasa and pgc) are not properly captured by PGCs when they cellularize. Instead, they spread into the posterior region of the embryo and are found associated with somatic nuclei. A similar disruption in the proper distribution of pole plasm during PGC cellularization was found when the terminal pathway was upregulated by blocking degradation of Torso, or by a constitutive allele of the downstream MEK kinase [12]. Moreover, as shown here, we find that Vasa is lost from PGCs when the terminal pathway is optogenetically activated at the posterior of the embryo. In fact, the Vasa loss phenotype induced by optogenetic activation of SOS is quite similar to that seen when we disrupted the BMP pathway by RNAi knockdown of dpp and by dpp and tkv mutants. Taken together, these findings argue that one of the functions of the BMP pathway in PGC specification is to block the terminal signaling pathway. In this context, it is interesting to note that inhibition of EGFR-dependent signaling may be important for PGC specification in mammals. When ESC cells are cultured under conditions that promoted the formation of mesoderm lineages, a PGC-like identity could be induced by adding an inhibitor of the upstream kinase MEK to the culture media [74].
In flies there is a complicated relationship between the BMP and EGFR signaling pathways. In the case of the Torso-dependent terminal pathway, dpERK phosphorylation of Capicua in the anterior and posterior soma counteracts the repression of dpp by Dorsal [75,76]. In other developmental contexts, the relationship between BMP signaling and the cognate EGFR pathway is complex. In the wing disc, for example, BMP and EGFR signaling is reported to establish a positive feedback loop, reinforcing each other by promoting the synthesis of their respective ligands [77]. In other contexts, however, the interactions between the two pathways is different. Genome wide studies of dorsoventral patterning during embryogenesis indicate that BMP signaling both negatively and positively regulates the expression of components of the EGFR signaling pathway during embryogenesis [78]. EGFR, in turn, was proposed to temper rather than augment the BMP pathway by a dpERK-dependent phosphorylation of dSmad that results in its degradation. In studies on the patterning of the eye field and head epidermis in Drosophila, Chang et al. also proposed an antagonistic relationship between BMP and EGFR signaling [79]. High levels of Dpp were found to block EGFR signaling by inhibiting dpERK accumulation, while EGFR gain of function mutants suppress Dpp signaling. Though the mechanism for inhibiting dpERK accumulation was not uncovered, the same mechanism could be deployed to block the activation of the terminal pathway during PGC specification in pre-cellular blastoderm embryos. Alternatively, the mechanism of inhibition may be specific to the process of PGC specification. For example, studies by Pae et al. (2017) would predict that Torso degradation by Gcl should in itself be sufficient to eliminate both canonical and non-canonical activities of the terminal signaling pathway in PGCs [65]. Thus, it is possible that the BMP pathway might inhibit terminal signaling by potentiating Gcl activity either directly or indirectly.
That the BMP signaling pathway is required for the proper functioning of maternally deposited pole plasm components during PGC specification is also suggested by the dominant genetic interactions between osk and dpp. In these experiments, females heterozygous for an osk mutation were mated to males carrying the weak viable dpp allele, dpp hr92 . Reducing the dose of osk in the mother by itself appears to result in a minor perturbation of pole plasm sequestration in her progeny; however, this defect is substantially enhanced when the progeny are also heterozygous for dpp hr92 . In addition to failing to completely capture the pole plasm, dpp hr92 /+;osk/+ PGCs exhibit other abnormalities, including a novel loss of cell:cell adhesion and invasive migration. This synergistic interaction would argue that the BMP pathway collaborates with osk in the process of PGC specification, and in doing so serves to integrate preformation with epigenesis.
The fact that one of the classical models of preformation deploys a signaling pathway that is known to play a critical role in PGC specification in species that rely on epigenesis would seem to bolster the argument that epigenesis is the ancestral mode for generating this special cell identity. This view would be supported by the evolutionary history of osk and nos, genes that function cell autonomously in PGC specification in flies. The former is restricted to a subset of insects that utilize preformation in PGC specification and is thought to arise from the fusion of bacterial and eukaryotic sequences [80]. nos, by contrast, is conserved from worms to human and spans species that are classically identified as using either preformation or epigenesis for PGC specification.
For the embryo staining, we used white 1 embryos as a control in most instances. Wherever the experiment involved an RNAi knockdown strategy, we generated egfpi embryos by crossing maternal-tubulin-GAL4 virgins with UAS-egfpi males and used these as control.

Immunostaining
Embryos were formaldehyde-fixed, and a standard immunohistochemical protocol was used for fluorescent-or DAB-visualized immunostaining as described previously [17]. The primary antibodies used were rabbit anti-Vasa . Fluorescent immunostaining employed Alexa-Fluor secondary antibodies used at 1:500 (ThermoFisher), and DNA was labeled using either DAPI (10 ng/mL, ThermoFisher Scientific) or Hoescht (3μg/ml, Invitrogen). For DAB staining, horse radish peroxidase (HRP) secondary antibodies (Jackson Immunoresearch) were used 1:1000. Stained embryos were mounted using Aqua Poly/mount (Polysciences) on slides. At least three independent biological replicates were used for each experiment.

Fluorescence in situ hybridization (FISH) and single molecule FISH (smFISH)
FISH was performed as previously described using probes specific to tll [68]. To quantify tll levels, we used an internal control within each embryo to normalize the intensity. Using Ima-geJ, we measured and averaged intensities of three randomly selected PGCs, and we also collected averaged intensities from three measurements per embryo within the somatic cells positive for tailless. The normalization was done according to the following: Normalized intensity = Averaged intensity from PGCs / averaged intensity from somatic cells. This normalization was done for each condition separately. These normalized intensities for individual embryos were plotted and presented in Fig 4. smFISH was performed as previously described using formaldehyde-fixed embryos [12]. All probe sets were designed using the Stellaris probe designer (20-nucleotide oligonucleotides with 2-nucleotide spacing). pgc and gcl, smFISH probes (coupled to either atto565 or atto647 dye, Sigma) were a gift from Liz Gavis. Sxl-Pe intronic probes (coupled to atto565 dye) were a gift from Thomas Gregor., and slam probes (coupled to Quasar 670) were produced by Biosearch Technologies. All samples were mounted using Aqua Poly/mount (Polysciences) on slides. At least three independent biological replicates were used for each experiment.

Optogenetic activation of ERK
OptoSOS and Hist-GFP WT control embryos were collected in the dark for 2 hours and then stimulated with blue light. Blue light stimulation was done at~1mW/cm2 at 450 nm for 2 hours using a custom-built panel of 30 LEDS placed~5cm from the embryos and enclosed in foil. After stimulation, Hist-GFP WT control and OptoSOS embryos were immediately pooled, decoronated, and placed in fixative. During fixation embryos remained in the blue light for 10 minutes to ensure continued activation. After fixation, embryos were stained for Vasa. Zstacks were taken of NC12-14 embryos (WT n = 30, OptoSOS n = 41). Images were processed in MATLAB, k-means clustering with 4 bins was applied to the max-projected Vasa images to segment the vasa positive cells (using the highest bin). The background staining of the embryo was calculated using the mean intensity of the middle 2 bins. This background was subtracted from the average intensity of the segmented vasa positive cells. These averages were calculated for each embryo in the optogenetic and hist-GFP groups before normalizing both averages by the mean of the hist-GFP group.

Microscopy and image analysis
NIKON-Microphot-SA microscope was used to capture images of DAB-stained embryos (40X). Imaging for all other smFISH and fluorescent immunostaining experiments was performed on a Nikon A1 inverted laser-scanning confocal microscope.
Images were assembled using ImageJ (NIH) and Adobe Photoshop and Illustrator software to crop regions of interest, adjust brightness and contrast, generate maximum-intensity projections, and separate or merge channels. To assess the mislocalization of the RNAs or protein in different genetic backgrounds compared to the control, we generated plot profiles using ImageJ. The posterior-most 75 μm of each embryo was plotted for comparison, and embryos from a single biological replicate are plotted in figures given that variation between fluorescence between replicates obscured the pole plasm distribution trends if embryos from all replicates were plotted together.

Statistical analysis
Using NC13/14 embryos, PGCs of each genotype were counted from the 1st Vasa-positive cell to the last through an entire z-volume captured at 2 micron intervals. These PGCs counts were analyzed using a Student's t-test. The same analysis was applied when counting either pH3-or CycB-positive PGCs in individual embryos for dpp gain of function experiments.
To compare numbers of PGCs with high or low levels of listed markers (Vasa, pSer2, H3meK4, dpERK), PGCs either positive or negative for each marker were counted from all identifiable PGCs of each embryo. The differences in marker levels were consistent, and PGCs could be easily classified into each category. Individual PGCs were counted by going through each slice of a Z stack and categorizing each cell (viewed throughout all relevant slices), and pairwise comparisons of these populations for each genotype were performed using Fisher's Exact test. For FISH/smFISH experiments, total number of embryos expressing tll, slam, or Sxl-Pe in PGCs were counted, and Fisher's Exact Test was used to test significance in the compared proportions of embryos positive for transcription in PGCs. Likewise, proportions of embryos displaying aberrant PGC behavior (lack of adhesion/invasion) were compared to control embryos using Fisher's Exact Test. Data were plotted and statistical analyses were performed using Microsoft Excel or R Project software.   Table. PGC specification and behavior in embryos simultaneously compromised for osk and dpp. (XLSX)