Distinct cis-acting elements mediate targeting and clustering of Drosophila polar granule mRNAs

Subcellular mRNA localization is a prevalent mechanism for generating and maintaining the asymmetric distributions of proteins necessary for cellular and developmental polarity. In many organisms, including Drosophila, mRNA localization plays an essential and conserved role in germline specification (Houston, 2013; Schisa, 2012). In Drosophila, kinesin-mediated transport and localized translation of oskar (osk) mRNA initiates the formation of a highly specialized cytoplasm at the posterior of the oocyte – the germ plasm – that is both necessary and sufficient to induce formation of the primordial germ cells (pole cells) in the early embryo (Mahowald, 2001). Osk recruits proteins, including Vasa and Tudor, to assemble ribonucleoprotein (RNP) granules called polar granules (Lehmann, 2016). Numerous maternally synthesized mRNAs subsequently become localized to the germ plasm through their entrapment in polar granules. Several of these, including nanos (nos), polar granule component (pgc) and germ cell-less (gcl), have known roles in germline development (Hanyu-Nakamura et al., 2008; Jongens et al., 1994; Kobayashi et al., 1996).

These transcripts are produced by the ovarian nurse cells and are delivered to the oocyte en masse through a concerted contraction of the nurse cells. Initially, nos, gcl and pgc reside in separate, single-transcript RNP particles that disperse throughout the oocyte via diffusion facilitated by microtubule-mediated streaming of the oocyte cytoplasm (Forrest and Gavis, 2003; Little et al., 2015; Trcek et al., 2015). Localization begins as single-transcript RNPs become incorporated into developing polar granules. The first transcripts to be so incorporated serve as seeds that recruit additional like transcripts from the bulk cytoplasm, resulting in homotypic clusters. New seeding events and cluster growth continue into early embryogenesis (Niepielko et al., 2018). Because the different mRNAs are incorporated stochastically, polar granules are heterogeneous with regard to the assortment of mRNAs they contain (Little et al., 2015). Structured illumination microscopy (SIM) revealed that homotypic clusters of different mRNAs remain spatially distinct within the granule (Niepielko et al., 2018; Trcek et al., 2015). Ultimately, the polar granules become associated with embryonic nuclei that enter the germ plasm and accompany these nuclei as they bud from the posterior cortex to form pole cells (Lerit and Gavis, 2011). Co-packaging in polar granules facilitates the pole cell inheritance of mRNAs that encode proteins essential to germline development, viability and function.

mRNA localization is directed by cis-acting localization signals frequently found in the 3′ untranslated regions (3′UTRs) of transcripts. Sequences or structures within these signals are specifically recognized and bound by proteins that package the transcript into a localization competent, and often translationally repressed, RNP particle (Eliscovich and Singer, 2017). For those polar granule mRNAs characterized to date, the 3′UTR is both necessary and sufficient to direct localization (Gavis and Lehmann, 1992; Rangan et al., 2009). Despite their common destination, no similarity has been found among 3′UTRs of the various Drosophila germ plasm mRNAs (Gavis et al., 1996; Gavis and Lehmann, 1992; Jain and Gavis, 2008; Lécuyer et al., 2007). Deletion analysis of the nos 3′UTR showed that multiple localization elements (LEs) contribute to localization and wild-type germ plasm mRNA localization can be achieved by different combinations of these LEs (Gavis et al., 1996). nos LEs are thus distributed and display partial functional redundancy. Some nos LEs also exhibit additive or synergistic effects, with a specific 41 nucleotide sequence capable of generating substantial localization when present in three copies (Bergsten et al., 2001; Gavis et al., 1996; Gavis and Lehmann, 1992). The significance of such organizational complexity and functional redundancy, and whether it is common among Drosophila germ plasm transcript localization signals remains unknown. In Xenopus, germ plasm localization of nanos1 (also called Xcat2) mRNA requires two distinct LEs: a general element that direct RNAs to the mitochondrial cloud in a diffusion and entrapment process (Chang et al., 2004; Zhou and King, 1996); and a specialized element that targets nanos1 to germ granules within the cloud (Kloc et al., 2000). Whether different LEs can mediate unique aspects of germ plasm mRNA localization in Drosophila has yet to be elucidated.

To address these gaps, we used transgenic reporters, single molecule fluorescence in situ hybridization (smFISH), and quantitative image analysis to identify cis-acting elements that direct localization of two highly enriched and functionally important polar granule transcripts: gcl and pgc. These transcripts encode proteins required, respectively, for pole cell formation and specification of germline fate (Hanyu-Nakamura et al., 2008; Jongens et al., 1992; Lerit et al., 2017; Pae et al., 2017; Timinszky et al., 2008). Whereas previous studies of germ plasm localization signals were limited to qualitative assays (Gavis and Lehmann, 1992; Gavis et al., 1996; Rangan et al., 2009), we have now analyzed mRNA localization patterns quantitatively at the level of polar granules, with single-transcript resolution.

We find that the gcl 3′UTR contains multiple LEs that are widely dispersed, weak in isolation and partially functionally redundant. Although several regions in the pgc 3′UTR exhibit localization activity, we have delineated a 59-nucleotide segment of the pgc 3′UTR that is responsible for approximately half of the average pgc content of polar granules. Within this region, we identified a 6-nucleotide motif that is also found in the gcl and nos 3′UTRs, and for each mRNA, mutation of this motif significantly reduces its accumulation within polar granules. We demonstrate that, in addition to being functionally separable processes, the targeting of pgc transcripts to polar granules and their self-assembly into homotypic clusters are mediated by distinct classes of LEs, with the 6-nucleotide motif dedicated to clustering. Together, our results reveal complexity in the organization of regulatory elements comprising germ plasm mRNA localization signals that serves the targeting and self-recruitment mechanism for enrichment in the germ plasm.

To delineate LEs in the 3′UTRs of gcl and pgc, we used a transgenic reporter assay. 3′UTR sequences were inserted into a reporter construct containing the nos 5′UTR and green fluorescent protein (GFP) coding region under UAS control (Fig. 1A). All transgenes were inserted into the same chromosomal location and expression was induced with the maternal α-tubulin-GAL4 driver. Transgenic reporter mRNAs were detected by smFISH using probes complementary to the GFP-coding sequence. Consistent with previous results (Rangan et al., 2009), the intact gcl and pgc 3′UTRs both directed robust localization to the germ plasm in late-stage oocytes (data not shown) and early embryos (Fig. 1B,C). In contrast, the unregulated α-tubulin 3′UTR (tub) did not confer detectable posterior enrichment (see below), and gfp-tub3′UTR RNA was evenly distributed throughout the cytoplasm (Fig. 1D).

The efficacy of germ plasm localization was quantified using two approaches. First, we used a nearest-neighbor method to measure colocalization of reporter RNAs with a native germ plasm transcript, nos. This value allowed us to assess whether transcripts are targeted to and incorporated into polar granules (polar granule colocalization; Fig. 1E). Second, we quantified the average number of mRNA molecules per polar granule for a given mRNA species (polar granule mRNA content; Fig. 1F). This value primarily reflects the efficacy of mRNA recruitment into homotypic clusters (Niepielko et al., 2018). As expected, the gfp-pgc3′UTR and gfp-gcl3′UTR RNAs colocalized with nos and accumulated within polar granules similarly to their endogenous counterparts, indicating that they faithfully recapitulate the localization process (Fig. 1; Fig. S1). To account for random overlap due to the high density of particles within the imaging volume, we also measured the frequency with which gfp-tub3′UTR RNA appeared colocalized with nos. This value (18%) is consistent with previous measurements (Little et al., 2015) and provides a threshold above which transcripts detected as colocalized are considered to be co-packaged within the same granule (Fig. 1D,E).

Division of the gcl 3′UTR into two overlapping fragments, gcl(1-276) and gcl(247-525) (Fig. 2A), severely compromised its localization activity. Polar granule colocalization was observed for both the gfp-gcl(1-276) and gfp-gcl(247-525) reporter RNAs, but at greatly reduced frequency relative to gfp-gcl3′UTR (Fig. 2B,C). Polar granule mRNA content was also dramatically decreased, with only gfp-gcl(1-276) RNA accumulating to any significant extent (Fig. 2B,D; Fig. S2A-D). We therefore hypothesized that multiple sequences dispersed throughout the 3′UTR might be required for effective localization. To test this, we extended each fragment to include a larger region of the gcl 3′UTR [gcl(1-399) and gcl(150-525); Fig. 2A]. In each case, the extended fragments conferred significantly greater localization than their precursors, indicating that LEs reside in nucleotides 150-247 and nucleotides 276-399 (Fig. 2B-D; Fig. S2E,F). However, a reporter containing the central region (nucleotides 150-399) exhibited only weak accumulation (Fig. 2B-D; Fig. S2G). From these data, we conclude that nucleotides 150-399, although important, are insufficient to mediate robust germ plasm localization when isolated from the rest of the gcl 3′UTR. Consequently, we infer that additional LEs must reside in the terminal fragments encompassing nucleotides 1-150 and nucleotides 399-525. Notably, in nearly every case, colocalization of reporter mRNAs with nos correlated with their average polar granule mRNA content (Fig. 2C,D).

Because germ plasm transcript abundance can influence localization (Gavis and Lehmann, 1992; Jongens et al., 1994; Niepielko et al., 2018), we quantified reporter RNA levels in 0-1 h old embryos by RT-qPCR. Although we did find variation, there was no correlation between the level of a particular RNA and its localization efficacy (Fig. S2I). Indeed, gfp-gcl3′UTR, which had the highest polar granule mRNA content, was among the lowest in mRNA levels. Additionally, gfp-gcl(150-525) was more abundant than gfp-gcl(1-399), but was localized to a lesser degree. Thus, the observed behaviors of different reporter RNAs reflect the activity of the 3′UTR segments they contain.

As the gcl 3′UTR appears to contain multiple independently weak LEs, we wondered whether they might act additively or synergistically in mediating localization. To test this, we generated transgenes containing two tandem repeats (×2) of the proximal (1-276), central (150-399) or distal (247-525) segments of the gcl 3′UTR (Fig. 3A). In each case, we observed substantial increases in both polar granule colocalization and average mRNA content for the ×2 transcripts when compared with their single counterparts. This was most dramatic for gfp-gcl(150-399)×2, which showed an 87% increase in polar granule mRNA content over gfp-gcl(150-399), and colocalized with nos at a frequency comparable with gfp-gcl3′UTR (Fig. 3B-D; Fig. S3A-H). These results cannot be attributed to differences in RNA levels; for example, gfp-gcl(150-399)×2 is present at a lower level than gfp-gcl(150-399) (Fig. S3I). The finding that multiple copies of one region can partly substitute for the loss of others in both polar granule targeting and homotypic cluster growth indicates that gcl LEs can function redundantly. Together, our data indicate that gcl germ plasm localization is mediated by a combination of multiple, widely dispersed, and partially functionally redundant LEs in its 3′UTR.

We took the same approach to delineate LEs in the pgc 3′UTR. A fragment containing the 5′ two-thirds of the pgc 3′UTR, pgc(1-273), conferred significant but not wild-type germ plasm localization, whereas the remaining one-third, pgc(255-392), had no activity (Fig. 4; Fig. S4A-D). Although pgc(255-392) is insufficient in isolation, sequences in this region are important for wild-type localization efficacy, as shown by the 30% decrease in polar granule colocalization and 50% decrease in polar granule mRNA content when they are deleted from the pgc 3′UTR (Fig. 4C,D). When the 5′ fragment was shortened to the first 150 nucleotides, localization competence was lost. However, the central nucleotide 150-249 segment was also unable to direct localization on its own (Fig. 4; Fig. S4E,F). Therefore, the proximal nucleotide 1-150, the central nucleotide 150-249 and distal nucleotide 255-392 regions all contain LEs that work together to mediate localization.

Several mRNAs that are localized during early Drosophila development are known to rely on LEs characterized by secondary structures (Bullock et al., 2010; Cohen et al., 2005; Jambor et al., 2014). We therefore sought to identify potential secondary structure motifs in the pgc 3′UTR that are evolutionarily conserved among drosophilids, reasoning that these might identify LEs (see Materials and Methods). This analysis revealed a predicted, conserved stem-loop encompassing nucleotides 184-243 (Fig. 5B). Deletion of this sequence, pgc(Δ184-243), had no effect on colocalization of the reporter RNA with nos (Fig. 5A-D). By contrast, the polar granule mRNA content for gfp-pgc(Δ184-243) was decreased by 49% when compared with gfp-pgc3′UTR (Fig. 5E; Fig. S4G).

To directly test whether the function of this 59-nucleotide region is mediated by the putative stem loop, we analyzed three additional sets of mutations (Fig. 5A,B). Substitution of nucleotides in the loop and bulge regions with random sequences that maintain the predicted structure in gfp-pgc(RL) had little or no effect on either polar granule colocalization or mRNA content (Fig. 5C-E; Fig. S4H), indicating that these sequences are dispensable. Two sets of mutations, within nucleotides 184-206 and nucleotides 215-243, which are each predicted to disrupt folding of the stem loop, reduced polar granule mRNA content by 15% and 35%, respectively, but had little or no effect on polar granule targeting (Fig. 5C-E; Fig. S4I,J). Surprisingly, although these mutations are predicted to restore the stem-loop structure when combined in pgc(Comp) (Fig. 5A,B), polar granule mRNA content was reduced by 49%, with little effect on polar granule colocalization (Fig. 5C-E; Fig. S4K). The failure of the compensatory mutations to restore mRNA content leads us to conclude that rather than secondary structure, primary sequences within nucleotides 184-206 and nucleotides 215-243 of the pgc 3′UTR together are responsible for nearly half of polar granule pgc content. The ability of these sequences to selectively affect polar granule mRNA content independent of polar granule colocalization suggests that distinct elements in the pgc 3′UTR mediate targeting to polar granules and the subsequent growth of homotypic clusters.

Using SIM, we recently showed that posteriorly localized RNAs can form two or more homotypic clusters within a polar granule. These likely occur through multiple, independent targeting events followed by recruitment of like transcripts (Niepielko et al., 2018). Thus, although polar granule mRNA content is determined primarily by homotypic cluster size (Niepielko et al., 2018), sequences such as those within pgc nucleotides 184-243 could affect mRNA content by contributing to these additional targeting events, as well as to homotypic cluster growth. To address this possibility, we performed SIM imaging on gfp-pgc3′UTR and gfp-pgc(215-243^mut) RNAs.

As detected by SIM, both gfp-pgc3′UTR RNA and gfp-pgc(215-243^mut) RNAs form multiple homotypic clusters within polar granules (marked with Osk-GFP), similarly to endogenous pgc (Fig. 6A). To confirm that detection of multiple clusters was not due to random overlap of particles in the imaging volume, we analyzed a non-localizing transcript, lost, which shows minimal colocalization with polar granules (Niepielko et al., 2018). Among the rare polar granules where overlap with lost was detected, the majority (94%) had only one lost particle (Fig. 6B,C). To determine how disruption of sequences in gfp-pgc(215-243^mut) affects polar granule mRNA content, we measured the frequency distribution of multiple homotypic clusters of gfp-pgc3′UTR and gfp-pgc(215-243^mut) RNA, and quantified the average number of mRNAs per cluster. Both the distribution and the average overall frequency of multiple homotypic clusters were indistinguishable for the two transcripts (Fig. 6B-D). Moreover, the average number of mRNA/cluster decreased by 38% for gfp-pgc(215-243^mut) when compared with gfp-pgc3′UTR, a value identical to the decrease in granule mRNA content (Fig. 6E, compare with Fig. 5E). Thus, the decrease in mRNA content resulting from perturbation of nucleotides 215-243 can be attributed to a reduction in average homotypic cluster size, defining the sequences within this region as a clustering element.

To further delimit the responsible sequences within nucleotides 215-243, we introduced sets of 6-nucleotide mutations across this region: pgc(216-221^mut), pgc(230-235^mut) and pgc(240-245^mut) (Fig. 5A). Whereas none of these mutations impacted polar granule colocalization (Fig. 5D), gfp-pgc(230-235^mut) behaved comparably to gfp-pgc(215-243^mut), decreasing polar granule content for the reporter RNA by 32% (Fig. 5E). Although flanking sequences make minor contributions, nucleotides 230-235 likely constitute the bulk of the homotypic clustering element.

A priori, sequences that function in homotypic clustering might be expected to be RNA specific. Therefore, we were surprised to find two matches or close matches to the pgc(230-235) sequence, CAAGUA, in both the gcl 3′UTR (UAAGUA and CAAGUU) and the nos 3′UTR (CAAGUC and CAAGUA) (Fig. S2J, Fig. S5B). Remarkably, mutation of the gcl sequences to generate gfp-gcl(364-369+419-424^mut) (Fig. 2A; Fig. S2J) resulted in a 37% decrease in polar granule mRNA content when compared with gfp-gcl3′UTR without affecting polar granule colocalization (Fig. 2B-D; Fig. S2H,I). Similarly, mutation of the nos sequences to generate gfp-nos(217-222+645-650^mut) (Fig. S5A,B) resulted in a 42% decrease in polar granule mRNA content when compared with gfp-nos3′UTR, with minimal effect on colocalization (Fig. S5C-E). These data suggest that the conserved motifs have a generalized function in mRNA clustering.

The ability of mutations that disrupt sequences within nucleotides 184-243 of the pgc 3′UTR to affect polar granule mRNA content without affecting polar granule colocalization suggests that distinct elements within the pgc 3′UTR mediate pgc targeting to polar granules and homotypic clustering. Indeed, the nucleotide 184-243 region is ineffective on its own, with even three tandem copies [gfp-pgc(180-249)×3] failing to confer localization (Fig. 7; Fig. S6D,E). Thus, whereas the gfp-pgc(180-249) and gfp-pgc(180-249)×3 mRNAs contain clustering elements, they may lack a requisite targeting element. The partial localization of gfp-pgc(1-249) suggests that such a targeting element resides within the first 180 nucleotides of the pgc 3′UTR (Fig. 7, Fig. S6C). To test this, we sought to rescue localization of gfp-pgc(180-249)×3 by adding back this 180-nucleotide region, generating gfp-pgc(1-180)+(180-249)×3 (Fig. 7A). Indeed, polar granule colocalization and mRNA content of gfp-pgc(1-180)+(180-249)×3 RNA were nearly comparable with that of gfp-pgc3′UTR (Fig. 7B-D; Fig. S6F). Using SIM, we confirmed that the number of homotypic clusters per granule and the number of mRNAs per cluster were minimally affected (Fig. 6). These data provide evidence for distinct targeting and clustering elements that together confer wild-type germ plasm localization. The fact that additional copies of the nucleotide 180-249 region can substitute for the loss of nucleotides 250-392 indicates that there is functional redundancy among pgc clustering elements.

Our initial dissection of the pgc 3′UTR (Fig. 4) provided evidence that LEs within nucleotides 1-150 and nucleotides 150-249 work together to mediate localization. Given the evidence that one or more LEs within nucleotides 1-150 functions in mRNA targeting, we wondered whether multimerization of this sequence could promote stable polar granule interaction in the absence of a capacity for homotypic clustering. To achieve this, we made gfp-pgc(1-150)×3 (Fig. 7A). Strikingly, not only did this RNA very effectively colocalize with polar granules, it also showed dramatic accumulation within the polar granules (Fig. 7B-D; Fig. S6H). Consistent with the previous analyses, localization behavior could not be explained by transcript levels (Fig. S6I).

From these data, we derive several conclusions. First, mRNA localization is directed by both targeting and clustering elements. Without a targeting element, the number of clustering elements has little bearing, as exemplified by gfp-pgc(180-249)×3. Second, the 58% increase in polar granule mRNA content for gfp-pgc(1-180)+(180-249)×3 relative to gfp-pgc(1-249) indicates that the centrally located clustering elements can indeed act additively, but only in the context of a sequence that confers competency to interact with polar granules. Finally, either some pgc targeting elements can promote homotypic clustering when multimerized, or an additional weak clustering element is present in nucleotides 1-150.

All characterized mRNA localization mechanisms function through cis-acting localization signals. For transcripts such as nos, gcl and pgc, which localize by a diffusion and entrapment process, 3′UTRs are sufficient to direct localization (Gavis and Lehmann, 1992; Rangan et al., 2009) but the specific features responsible for their function have remained largely unknown. Here, we have used high-resolution imaging and quantitative analysis of mRNA localization at the level of polar granules to map and characterize the functions of LEs in the gcl and pgc 3′UTRs. Our results reveal parallels to nos, with robust localization of reporter RNAs requiring the function of multiple regulatory elements spread widely across the 3′UTR. Although weak in isolation, these LEs can function redundantly such that multiple copies of one can at least partly substitute for lack of another. Thus, complexity in localization signal organization appears to be a feature of Drosophila germ plasm transcripts. Our results also provide evidence for another layer of complexity, by showing that targeting to polar granules and formation of homotypic clusters are separable processes regulated by distinct classes of cis-acting elements.

Compound mRNA localization signals with elements that direct discrete intermediate steps of a localization pathway have been found in some actively transported mRNAs such as Drosophila bicoid and osk (Jambor et al., 2014; Kim-Ha et al., 1993; Macdonald and Kerr, 1998; Macdonald et al., 1993). These elements allow the sequential association of an mRNA with different transport machineries. The compound nature of Drosophila germ plasm mRNA localization signals also allows for a process involving multiple events, in this case the targeting of mRNAs to polar granules and the growth of homotypic clusters within the granules. These roles are potentially similar to roles played by LEs comprising the compound Xenopus nanos1 mRNA localization signal: the mitochondrial cloud localization element (MCLE) that directs nanos1 along with various RNAs to the cloud and the germ granule localization element (GGLE), which targets nanos1 to the germ granules. Like the pgc clustering elements, which function only when paired with a targeting element, the GGLE functions only in the presence of a mitochondrial cloud targeting element – i.e. the RNA must enter the cloud in order to associate with germ granules (Chang et al., 2004; Kloc et al., 2000). Intriguingly, different Xenopus germline RNAs exhibit different distributions within the mitochondrial cloud, with nanos1 sequestered within granules and other RNAs on the outside or associated with the matrix between granules (Kloc et al., 2002). This pattern hints that Xenopus germ granule assembly may also involve mRNA clustering and that the nanos1 GGLE might function similarly to the clustering elements found in Drosophila germ granule mRNAs.

The presence of multiple elements that act additively, and often redundantly, to confer localization is a feature of many RNA localization signals, regardless of the method of localization (Kloc and Etkin, 2005; Shahbabian and Chartrand, 2012). A multiplicity of partially redundant LEs within Drosophila germ plasm mRNA localization signals might be advantageous for several reasons. The presence of multiple clustering elements may counterbalance an inherently inefficient localization process as 4% or less of the various transcripts ultimately become localized to the germ plasm (Bergsten and Gavis, 1999; Little et al., 2015; Trcek et al., 2015). Given the large volume of the oocyte and the narrow posterior cortical domain in which germ granules form (Little et al., 2015), contact of the diffusing single-transcript RNPs with nascent polar granules is likely to be infrequent. The presence of multiple elements and cognate binding proteins would thus increase the likelihood of interaction.

Additionally, the inclusion of multiple clustering elements within an mRNA molecule may create multivalency that can control clustering either through RNA-RNA interactions or through interactions with RNA-binding proteins, which may themselves be multivalent. Each newly recruited mRNA would in turn provide additional interaction surfaces, allowing for continuous cluster growth (Little et al., 2015). This is consistent with our observation that every reporter RNA that produces large homotypic clusters is predicted to include at least three clustering elements. Furthermore, RNAs with only one predicted clustering element form homotypic clusters that contain, on average, only slightly more than two transcripts. As pole cell survival correlates with germ plasm inheritance (Slaidina and Lehmann, 2017), the additive activity of multiple targeting and clustering elements together would ensure that a sufficient quantity of each RNA makes it into the germline to facilitate proper development.

Self-association of germ plasm transcripts in homotypic clusters occurs exclusively within the polar granules, never in the bulk cytoplasm (Little et al., 2015). Thus, there must be a permissive switch upon interaction of a single-transcript RNP with a polar granule that allows for the recruitment of additional like-transcripts, and the subsequent growth of homotypic clusters. The nature of this switch remains unknown, but it is notable that multiple RNA helicases are present in the polar granules, including Vasa, Belle and Me31B (Hay et al., 1988; Johnstone et al., 2005; Thomson et al., 2008). These factors might direct remodeling of single-transcript RNPs upon their interaction with a polar granule, thereby unmasking previously unexposed clustering elements.

Evidence for localization signal sequences or structures shared among Drosophila germ plasm mRNAs has been lacking, despite their common destination. Whether many different targeting elements are recognized degenerately by one or more trans-acting factors or whether different sets of proteins can all mediate polar granule targeting remains an unresolved issue. In contrast to sharing the ability to incorporate into polar granules, the capacity for nos, pgc and gcl transcripts to form spatially segregated homotypic clusters within polar granules (Little et al., 2015) predicts that each harbors unique clustering elements. Unexpectedly, at least one clustering element is shared by the pgc, gcl and nos 3′UTRs. This suggests that in addition to providing multivalency, clustering elements may act combinatorially to generate an RNA-specific signature. Analyzing the clustering behavior of reporters with chimeric 3′UTRs containing combinations of pgc, gcl and/or nos LEs may provide insight into the contributions of these elements to RNA self-recruitment.

The following Drosophila melanogaster alleles and transgenes were used: y¹, w^67c23 (Bloomington Stock Center 6599) as the wild-type strain; and osk-GFP (Sarov et al., 2016). The matα4-GAL-VP16 driver (Bloomington 7063) was used to express all UASp-gfp-3′UTR transgenes.

The gfp-coding sequence was PCR amplified from pEGFP-N1 (Clontech) and inserted into a pBS-SK plasmid containing the nos 5′UTR at the position of the nos ATG. A fragment containing the nos polyadenylation signal and 460 bp of 3′ genomic DNA was excised from a genomic nos plasmid (Gavis and Lehmann, 1992) and inserted downstream of the gfp-coding sequence. pgc 3′UTR sequences were PCR amplified from y¹, w^67c23 genomic DNA. The pgc 3′UTR we isolated was missing 12 bp (nucleotides 297-308) when compared with the annotated sequence on FlyBase. gcl 3′UTR sequences were amplified from cDNA clone LD23660 (Drosophila Genome Resource Center). The tubulin 3′UTR sequence was PCR amplified from a Bluescript plasmid containing nucleotides 1-198 of the alphaTub84B 3′UTR (Theurkauf et al., 1986). Each of these sequences was inserted 3′ to the gfp-coding sequence. For germline integration, we modified the pattB insertion vector by inserting the GAL4-binding site cassette from pUASp (nucleotides 114-734) into the pattB insertion vector (www.flyc31.org/sequences_and_vectors.php) between EcoRI and XbaI to generate pattB-UASp. The nos-gfp-3′UTR sequences were then cloned into pattB-UASp and integrated into the attp40 site via injection and phiC31-mediated recombination.

Mutations generated within pgc 3′UTR nucleotides 184-243 with altered nucleotides underlined:

wild type, CAAATGTTTGCTTTCGTGAAAACTCGCATTGTTTTGTCACTCTACCAAGTAATCAATTTG; pgc(RL), CAAATGTTTGCTTTCGTGAAAACTTCGGTTTTGTCACAACCTTAAGTAATCAATTTG; pgc(184-206^mut), ACCCTGTTTAAGGTCGTGCGCGCTCGCATTGTTTTGTCACTCTACCAAGTAATCAATTTG; pgc(215-243^mut), CAAATGTTTGCTTTCGTGAAAACTCGCATTGCGCTGGCACTCTACCCCTTAATCAAGGGT; pgc(216-221^mut), CAAATGTTTGCTTTCGTGAAAACTCGCATTGCGCCAGCACTCTACCAAGTAATCAATTTG; pgc(230-235^mut), CAAATGTTTGCTTTCGTGAAAACTCGCATTGTTTTGTCACTCTACACCTGCATCAATTTG; and pgc(240-245^mut), CAAATGTTTGCTTTCGTGAAAACTCGCATTGTTTTGTCACTCTACCAAGTAATCACGGGTC.

smFISH probe sets of 20-nucleotide oligonucleotides with 2-nucleotide spacing, complementary to nos (CG5637; 63 oligos), to the nos-coding region (34 oligos), to pgc (CG32885; 52 oligos), to gcl (48 oligos), to lost (48 oligos) or to gfp (32 oligos) were designed using the Stellaris Probe Designer. Oligonucleotides were obtained from Biosearch Technologies with a 3′ NHS ester modification and conjugated to Atto 565 (Sigma 72464) or Atto 647N (Sigma 18373) dyes, then purified via HPLC as described previously (Raj et al., 2008). smFISH was performed as previously described (Little et al., 2015) on 0-1 h old embryos. Embryos were mounted in Prolong Diamond medium (Fischer Scientific P36965) and cured for 3-4 days at room temperature in the dark prior to imaging.

Confocal imaging was performed on a Nikon A1-RS laser-scanning confocal microscope using a 60×1.4 NA oil immersion objective at either 2× or 3× optical zoom, with pixels of 102×102 nm or 72×72 nm, respectively. Confocal sections were acquired with 16× line averaging. Super-resolution images were taken on a Nikon Structured Illumination Microscope using a 100×1.49 NA oil immersion objective, with pixels of 33×33 nm. For each embryo, a z series of 21 slices were taken, with a step size of 150 nm. In each experiment, laser power and gain were adjusted to avoid signal saturation while maximizing separation of signal and noise. Embryos selected for imaging were oriented such that the germ plasm was facing the cover slip. To minimize fluorescent signal distortion, all images were taken within 5 µm of the embryo cortex. When the same transgene is shown in more than one figure, the same accompanying confocal image is used each time to allow comparison across figures.

Colocalization of nos with mRNAs in the germ plasm reflects their incorporation into polar granules (Little et al., 2015; Niepielko et al., 2018). This metric is therefore a proxy for colocalization of reporter mRNAs with polar granules. All analysis of confocal images was carried out on single sections. In MATLAB (MathWorks), a region minimally outlining the germ plasm of each embryo was specified by a manually drawn polygon (Fig. S1E). The spotDetector algorithm (Aguet et al., 2013) was then used to detect and assign the x, y coordinates of every mRNA signal in each channel (∼500-1000 particles per embryo) (Fig. S1E,F). Distances were calculated from each point in the nos channel to the nearest point in the reporter mRNA channel. Functional colocalization is defined as a distance less than 300 nm, on the basis of average polar granule radius (Amikura et al., 2001; Houston, 2013; Illmensee and Mahowald, 1974). The colocalization frequency in an embryo is the percentage of total RNA molecules within the germ plasm that fall below this threshold (Fig. S1G). Colocalization frequencies are displayed as the mean value measured from 20-30 embryos per genotype±s.e.m.

The absolute mRNA content per granule was determined by normalization of granule fluorescence intensity to fluorescence per single-copy particle (Little et al., 2015). For each embryo, sequential images were collected under two conditions: one (high gain) optimized for the unlocalized, single-copy RNPs, which results in saturation of the germ plasm fluorescent signal; and one (low gain) optimized for the germ plasm signal.

The average fluorescence intensity of single-copy RNPs was determined by manually demarcating an ROI in the bulk cytoplasm of the high gain image containing >500 particles. A spot detection algorithm (Little et al., 2015, 2011) was then used to assign the x, y coordinates and measure the fluorescence intensity of every RNP in that region, using a threshold set manually to eliminate false positives. A 13×13 pixel mask was fitted to every detected RNP and the intensity of every pixel of that grid for all particles was averaged to obtain an intensity value for an idealized single-copy RNP. To obtain a conversion factor for normalization of germ granule fluorescence intensity, the same ROI was then analyzed similarly in the low-gain image. The intensity of every RNP in the low-gain image was plotted versus its intensity in the high-gain image and a linear regression analysis was used to fit a line to the data. The slope of this line yields the conversion factor.

Average fluorescence intensity of the germ granules was obtained from the low-intensity image. A polygon was drawn manually to bound the germ plasm, which was demarcated using endogenous nos RNA, and the spot detection algorithm was used to detect all RNPs within the germ plasm. The data were fit to a Gaussian for measurement of fluorescence intensity. Intensity values were multiplied by the conversion factor, and divided by the intensity of the idealized unlocalized RNP to generate an estimation for the number of transcripts within each polar granule. These values were then averaged for the germ plasm of a given embryo. The germ plasm of 10 embryos was quantified per genotype to generate the average polar granule mRNA content.

At the start of every experiment, an image of TetraSpeck beads (ThermoFisher, T7279) was taken and used in conjunction with Nikon software to generate alignment correction parameters that were applied to germ plasm images. RNA particle detection and image quantification was carried out using a custom Matlab program as previously described in detail (Niepielko et al., 2018). In summary, N-SIM images were filtered by a balanced circular difference-of-Gaussian with a center radius size of 1.2 pixels and surround size of 2.2 pixels. We clipped a 13×13 pixel mask centered around each image punctae and fit a Gaussian distribution in three consecutive z slices and as well as in x-y to identify the x-y-z locations of candidate Osk protein and mRNA particles. True particles were determined by intensity thresholds set manually to eliminate false positives. To determine whether Osk and RNA particles are colocalized, we first calculated distances in x-y for all particle pairs. Next, we selected colocalized pairs based on the following criteria: (1) two particles must be within a z distance four slices, which accounts for chromatic aberration while eliminating pairs that may colocalize in x-y but cannot be in the same granule due to the size limitation of germ granules; and (2) a colocalized particle pair must also be within a distance limit of 200 nm in x-y. A conservative distance was chosen based on the average size of a germ granule that was previously used to calculate colocalization frequency among germ plasm mRNAs (Little et al., 2015). The number of discrete RNA particles that satisfy the colocalization criteria for each Osk particle was recorded. For our analysis, any colocalized RNA particle was considered to be a homotypic cluster, even if it contained a single transcript.

RNA was isolated using the RNeasy mini kit (Qiagen) from 0-1 h old embryos. RNA (∼1 µg) was reverse transcribed to cDNA using the Quantitect Reverse Transcription kit (Qiagen). qPCR was run using an Applied Biosystems 7900HT standard 96-well qPCR instrument. The following TaqMan Gene Expression probes from ThermoFisher were used: gcl (Dm01812234, 4351372); pgc (custom - DmAJHSOQG, 4441114); egfp (Mr04097229, 4331182); and rpl7 (Dm01817653, 4351372). All experiments used a CT threshold of 0.6613619. Three technical replicates were quantified from each cDNA pool. DNA standard curves were generated for each probe set using serial dilutions of plasmids containing target sequences. CT values from each assay were fit to the standard curves, corrected for the experimentally determined doubling efficiency of each probe, and normalized to the housekeeping gene rpl7. Doubling efficiencies were as follows: pgc (87.5%), gcl (92.7%), gfp (97.2%) and rpl7 (88.5%). Finally, reporter mRNA expression levels were normalized to endogenous pgc or gcl expression in 0-1 h old wild-type embryos. Expression levels are displayed as mean±s.d.

The D. melanogaster pgc 3′UTR sequence was obtained from FlyBase (flybase.org/). To obtain pgc 3′UTRs from representative species across the Drosophila phylogenetic tree, the extended pgc gene regions and coding sequences from D. melanogaster, D. sechellia, D. erecta, D. ananassae, D. willistoni and D. virilis were downloaded from FlyBase. A custom Python script ‘three_prime_detector.py’ based on the SeqIO and AlignIO modules from the Biopython library (Cock et al., 2009; available at biopython.org/) was used to extract each 3′UTR sequence by first generating a Clustal W 2.0 (Larkin et al., 2007) alignment of the extended gene region and coding region, and then selecting the nucleotide sequence in the extended gene region beginning downstream of the stop codon and extending the length of the D. melanogaster pgc 3′UTR+50 bp. Local secondary structures conserved among the different pgc 3′UTR sequences were predicted using RNApromo (Rabani et al., 2008), a tool that predicts structural motifs common to a set of RNA sequences, in conjunction with the ViennaRNA package (Lorenz et al., 2011). The output of RNApromo is a log-likelihood score for each instance of secondary structure motif discovered in a set of sequences. For each motif, the sum of the individual log-likelihood scores is a measure of its relative conservation. To identify ‘true positive’ structures, we used a custom Python script ‘random_seq_generator.py’. to generate 100 randomly shuffled pgc 3′UTR sequences for each of the six species. RNApromo was used to produce a dummy dataset of conserved local secondary structures from these shuffled sequences. We then used a custom R script ‘cutoff_secondary_score.R’ to calculate a ‘secondary score’ for each motif by first summing the log-likelihood scores of each secondary structure motif found in the random dummy dataset and then dividing by the square root of the length of that motif to control for increased likelihood due to length. Secondary scores were sorted in descending order and a threshold was set at the 95% percentile score. Secondary structures in the true dataset with log-likelihood scores above this threshold (<5% chance of being a false positive) were considered further. Two secondary structures in different regions of the pgc 3′UTR were identified whose log-likelihood scores exceeded the cutoff score. The functional significance of the top scoring motif was experimentally tested (Fig. 5). Custom scripts provided upon request.

We thank S. Little and J. Lee for assistance with the MatLab scripts used to quantify transgene colocalization and intensity phenotypes. We thank G. Laevsky for assistance with microscopy. We thank S. Little, C. Ruesch, P. Schedl and J. Tamayo for critical comments on the manuscript.