Independent and coordinate trafficking of single Drosophila germ plasm mRNAs

Abstract

mRNA localization is a conserved mechanism for spatial control of protein synthesis, with key roles in generating cellular and developmental asymmetry. While different transcripts may be targeted to the same subcellular domain, the extent to which their localization is coordinated is unclear. Using quantitative single molecule imaging, we analyzed the assembly of Drosophila germ plasm mRNA granules inherited by nascent germ cells. We find that the germ cell-destined transcripts nanos, cyclin B, and polar granule component travel within the oocyte as ribonucleoprotein particles containing single mRNA molecules but co-assemble into multi-copy heterogeneous granules selectively at the posterior of the oocyte. The stoichiometry and dynamics of assembly indicate a defined stepwise sequence. Our data suggest that co-packaging of these transcripts ensures their effective segregation to germ cells. In contrast, compartmentalization of the germline determinant oskar mRNA into different granules limits its entry into germ cells. This exclusion is required for proper germline development.

INTRODUCTION

Intracellular localization of mRNAs is a widespread strategy for targeting the encoded proteins to regions of the cell where they function. mRNAs travel as ribonucleoprotein particles (RNPs), whose associated proteins direct their localization through interactions with molecular motors, cytoskeletal elements, and/or other cellular components^1,2. These RNPs can contain many mRNAs, several mRNAs, or only a single mRNA molecule, depending on the cell type, context, and possibly the mechanism of localization^3–8. How RNP composition is determined, how multiple mRNAs are co-assembled into RNPs, and where assembly occurs remain poorly understood.

The Drosophila germ plasm, whose assembly and function rely on localized mRNAs, provides an ideal system to address these questions. This specialized cytoplasm at the posterior of the embryo is necessary and sufficient for induction of the germ cell progenitors, the pole cells⁹. Numerous maternal mRNAs are enriched within the germ plasm and then inherited by the pole cells^10,11, where they direct production of proteins required for specification of germline fate and for germline development^12,13. Formation of the germ plasm is initiated during mid-oogenesis by the kinesin-dependent transport of oskar (osk) mRNA to the posterior of the oocyte. Osk protein produced from this localized RNA recruits the helicase Vasa (Vas) and other proteins, establishing a posterior domain of germ plasm assembly¹⁴. This initial step is essential for a second wave of mRNA localization during late stages of oogenesis, whereby additional osk as well as numerous other mRNAs including nanos (nos), polar granule component (pgc), and cyclin B (cycB) accumulate at the posterior^15–18. Live imaging of nos mRNA showed that its localization occurs as a consequence of diffusion within the oocyte cytoplasm and entrapment at the posterior in Vas-containing RNPs known as polar granules¹⁶. Later, during embryogenesis, these granules are segregated to the pole cells as those cells bud from the posterior of the embryo¹⁹.

The accumulation of numerous different mRNAs within the germ plasm via a common localization pathway and their ultimate destination in the pole cells begs the question of whether their trafficking is coordinated at any step. Moreover, whether the movement of multiple mRNAs to the posterior pole and ultimately into the pole cells is coordinated through co-packaging into common RNPs or whether they are organized into functionally distinct RNPs has not been investigated. To address these questions, we analyzed the mRNA composition of RNPs containing germ plasm mRNAs at different stages of oogenesis and embryogenesis by high resolution fluorescent in situ hybridization (FISH) and quantitative image analysis. Using nos as a paradigm, we show that localization occurs by the exclusive assembly of RNPs containing single mRNAs into multi-copy granules at the posterior cortex. Co-packaging of nos with cycB and pgc occurs selectively at the posterior, forming heterogeneous polar granules that co-transport germ plasm mRNAs into pole cells. In contrast, osk travels in multi-copy RNPs and is segregated from the other mRNAs at the posterior into compositionally distinct granules that can contain hundreds of osk transcripts. These granules must be excluded from the pole cells for germline development.

RESULTS

nos mRNAs travel as single-copy particles prior to posterior localization

To better understand how nos becomes assembled into polar granules, we performed fluorescence in situ hybridization (FISH) using a method capable of detecting individual mRNA molecules in Drosophila embryos^20,21 (Supplementary Video 1). We observed two classes of labeled objects: dim punctae at high density filling the embryo interior and bright objects near the cortical surface at the posterior (Fig. 1a,b). Both classes appear as diffraction-limited point sources (Supplementary Fig. 1a). Visualized in sections near the mid-sagittal plane, the bright objects occupy a cap-shaped volume covering the posterior at a depth up to ~5 μm beneath the cortex (Fig. 1b). The posterior cap extends ~25 μm from the posterior pole, measured as an arc along the cortical surface (Fig. 1a). Localization of nos is observed as an increase in fluorescent signal close to posterior pole, resulting from the enrichment of bright objects containing many mRNAs (i.e., localized granules) accompanied by a depletion of dim punctae (i.e., unlocalized particles) that contain fewer mRNAs (Supplementary Fig. 1b–d).

Figure 1. Measuring absolute mRNA content of nos granules.

(a,b) Confocal section near the cortical surface (a) or near the mid-sagittal plane (b) of an early, nuclear cycle (n.c.) 3 embryo (anterior left, dorsal up) labeled with nos probes (a; green box in inset shows region displayed in main panel). Right panels show magnified views of regions indicated by yellow (right upper panels) and red (right lower panels) boxes. Scale bars: 20 μm (left panels); 2 μm (right panels). (c,d) Detection efficiency of unlocalized nos particles in two-color FISH using nos probes with alternating fluorophores. Particles are detected in one channel (“Detection channel”) on the basis of separation from imaging noise, then intensity measurements are made in the second channel (“Sample channel”) at the same positions. Plots show 2D histograms where color indicates relative density of data points. Red horizontal and green vertical lines represent detection thresholds separating true objects from imaging noise. Percentages indicate fraction of objects found in the sample channel that fall above or below the threshold in the detection channel. (e) 2D histogram shows a weak correlation for normalized mRNA content of unlocalized particles. Cyan and red dashed lines indicate correlated and anticorrelated axes, respectively, through the point (1,1). (f) Cross-sections through histogram shown in (e) along correlated (cyan) and anticorrelated (red) axes. Circles indicate data points. A single Gaussian distribution fits experimental observations along the anticorrelated axis (red), whereas the distribution along the correlated axis is approximated by the sum (cyan line) of two Gaussians (blue lines) with means of 1 and 2 mRNAs (see also Supplementary Fig. 1). (g) Log-normal distribution of nos mRNA content in localized granules (separated from unlocalized punctae by vertical red line). (h) mRNA content as a function of distance from the posterior pole. Heat map indicates relative density of data points (see also Supplementary Fig. 1j).

We quantified the mRNA content of the unlocalized particles by comparing the number of particles detected by FISH to the number of mRNAs measured by absolute RT-qPCR. Counting dim punctae in the entire left or right halves of embryos and multiplying by two yields an estimate of 2.0 ± 0.5 x 10⁶ objects per embryo. Our ability to detect most dim objects was confirmed by FISH using nos probes alternating in fluorophore color along the length of the transcript. Plotting the distribution of fluorescence intensities for detected objects in each channel in two-dimensional (2D) histograms, we found that ≥97% of dim objects detected in one channel also appear in the second channel (Fig. 1c,d), indicating high detection efficiency. The total number of detected particles in the embryo (2.0 ± 0.5 x 10⁶) approximates the total number of nos mRNA molecules per embryo determined by RT-qPCR (2.5 ± 0.4 x 10⁶, Supplementary Fig. 1e). Thus, the average mRNA content of the imaged dim punctae is 1.3 ± 0.5 mRNAs. Since only ~4% of nos mRNA is localized²² (also see Methods), unlocalized particles, which constitute the vast majority of nos mRNA in embryos, largely represent single mRNAs.

Given the high density of unlocalized nos particles, we estimate that ~15% of unlocalized transcripts reside within the same imaging volume by random chance (Supplementary Fig. 1f–j and Methods). Co-localization detected in excess of this amount would thus indicate concerted co-packaging of multiple nos mRNAs. To estimate the fraction of unlocalized particles containing >1 nos transcript, we examined the fluorescence intensity distributions in 2D histograms after normalizing to the most frequently observed intensity in each channel. Because the number of probes decorating each mRNA molecule is drawn at random from a distribution about the mean number of probes, the two intensity measurements should be uncorrelated for particles representing single mRNAs. We observed a weak correlation with a clear skew toward higher mRNA content along the correlated axis (Fig. 1e,f, cyan) compared with the symmetric distribution along the anticorrelated axis (Fig. 1e,f, red). The distribution along the correlated axis is well matched by model in which 83% and 15% of detected punctae represent 1 and 2 mRNAs, respectively (Fig. 1f). Thus, the predicted rate of spurious co-localization within the same volume is similar to the observed frequency of punctae containing 2 mRNAs. We conclude that unlocalized nos exists as single mRNAs that are not co-packaged prior to localization.

nos localization results from the spatially restricted assembly of multi-copy granules

In two-color FISH experiments, normalized intensity values for each fluorophore in granules at the posterior pole are highly correlated, indicating that in contrast to unlocalized nos particles, these granules contain multiple nos transcripts (Supplementary Fig. 1k). Normalization to the intensity of single molecules observed in the dim punctae permits estimation of granule size in absolute mRNA numbers. Posterior cortical localization of nos results from the enrichment of granules containing >4 nos mRNAs (Fig. 1g,h, Supplementary Fig. 1k,l) that, as expected, do not form in osk or vas mutant embryos (Supplementary Fig. 1m,n). nos content of individual granules spans a wide range and can be >60 transcripts, with a mean of 16 (Fig. 1g,h). The variability in mRNA content (standard deviation = 12) is on the same order as the mean, indicating significant heterogeneity in mRNA content among localized granules (Supplementary Fig. 1o). A Poisson process that produces a mean of 16 mRNAs per granule by random arrival at sites of granule formation would predict a standard deviation of only four mRNAs. In support of a more complicated granule assembly process, we observed that localized nos content in these granules is not normally distributed and is better described by a log-normal fit characteristic of an exponential growth process in which growth rate scales with granule size (Fig. 1g; see Discussion).

The localization of nos occurs during late stages of oogenesis. To determine when and where the transition from single mRNAs to multi-copy granules occurs and the relationship between granule formation and posterior localization, we analyzed nos during this period (Fig. 2a). Synthesized by the ovarian nurse cells, nos enters the oocyte en masse when the nurse cells “dump” their contents at the end of stage 10 and becomes distributed throughout the oocyte by diffusion and the concurrent streaming of the oocyte cytoplasm (ooplasm)¹⁶. Consistent with previous analysis of nos mRNA labeled using the MS2/MS2 coat protein (MCP) system¹⁶, we observed continuous accumulation of nos in granules at the posterior of the oocyte beginning at stage 10 of oogenesis. nos-containing granules increase in both number and mRNA content up until the oocyte reaches maturity at stage 14 (Fig. 2b). Similarly to the early embryo, multi-copy granules were observed only within a depth of ~5 μm from the cortex and only within the localization domain (Supplementary Fig. 2a).

Figure 2. nos granule assembly occurs exclusively at the posterior oocyte cortex.

(a) Maximum z-series projections spanning 5 μm at the posterior oocyte cortex show the extent of nos localization at four different stages of oogenesis. Scale bars: 10 μm (stage 9, 10a); 20 μm (stage 12, 13). (b) Distributions of nos mRNA content per granule during late oogenesis (also see Supplementary Fig. 2). (c) nos content as a function of distance from the posterior pole. (d) Calibration of intensity distributions of nos*GFP granules in time-lapse movies (solid blue bars; arbitrary units) to FISH granules (red outline, absolute mRNA number). Inset depicts one frame of a nos*GFP time-lapse movie compared to a FISH image. (e) An individual granule (arrowhead) containing approximately 15 mRNAs travels ~3 μm in 10 sec while maintaining nos content. Frames correspond to times t = 4, 7, 10, 12, and 15 sec. Displacement (blue) is calculated as root-mean-square distance from the t = 0 position. Fluctuations in mRNA content (red) centered on ~15 mRNAs result from imaging noise (see also Supplementary Fig. 2).

The appearance of localized granules as diffraction-limited objects (Supplementary Fig. 1a) suggests that they correspond to individual higher order RNPs containing many mRNAs. To support the idea that the bright objects represent physically associated mRNA particles and not particles transiently corralled into the same volume, we compared our FISH data to data from live imaging of MS2-tagged nos labeled with MCP-GFP (nos*GFP). Localized nos*GFP granules have a spatial distribution and fluorescence intensities similar to those of nos granules detected by FISH, indicating that they likely represent the same entities (Fig. 2d). Previous live imaging experiments showed that nos*GFP granules undergo rapid, directed movement along microtubules at the posterior cortex.²³ By performing a chi square fit to match to the right sides of the fluorescence intensity distributions of nos granules detected by FISH and nos*GFP granules in live oocytes, we found that objects undergoing rapid, directed runs over several micrometers at ≥0.5 μm/s contain upwards of 10–20 nos mRNAs (Fig. 2d). Throughout each run, the intensity measurement remains unchanged, indicating that the detected mRNAs are constituents of the same granule (Fig. 2e, Supplementary Fig. 2b–d, Supplementary Video 2). Taken together, our results indicate that nos localization results from the incorporation of single transcript particles into multi-copy granules exclusively at the posterior cortex.

Germ plasm mRNAs assemble into common but heterogeneous granules at the posterior

Because numerous mRNAs in addition to nos are localized to the germ plasm and incorporated into pole cells^10,11, we hypothesized that their localization might be coordinated through co-packaging into the same RNPs. To determine if and when co-packaging occurs, we examined the distributions of two additional germ plasm mRNAs, pgc and cycB, in pairwise combinations with nos. All three mRNAs are expressed at high levels in the nurse cells at mid-oogenesis (Fig. 3a–d). Around stage 10, prior to the onset of nurse cell dumping, they become enriched on colecemid-sensitive track-like structures, likely microtubules, that are best visualized at the cortical surface (Fig. 3c). Tracks are also detected that extend toward the ring canal junctions connecting nurse cells to the oocyte and to each other (Fig. 3d). Despite their similar distributions, we did not detect co-localization of nos, pgc, or cycB mRNAs prior to posterior localization, in either the nurse cells or bulk ooplasm (Supplementary Fig. 3a–d).

Figure 3. Co-localization of polar granule components begins during late oogenesis.

(a–d) Detection of nos, pgc, and cycB in nurse cells by FISH. (a,b) Confocal sections of nurse cells at stage 8. nos, pgc, and cycB are found in separate particles (also see Supplementary Fig. 3). In some instances, putative sites of nascent transcription cluster to one side of a nurse cell nucleus (arrowheads in b) and this is often associated with enrichment of single mRNAs in the cytoplasm closest to the transcription sites; additional sites are present in adjacent z sections. (c,d) Maximum z-series projections spanning 5 μm at stage 10b near the cortex (c) and near a putative ring canal leading to the oocyte (d) showing filamentous networks enriched for these RNPs. Blue: DAPI staining. Scale bars: 10 μm. (e–g) Maximum z-series projections over 5 μm of the posterior oocyte cortex. pgc and nos are first detected in posterior granules at stage 10a (e) and are often enriched in the same granule (f: stage 10b; g: stage 13). Scale bars: 5 μm (e,f); 10 μm (g). (h–j) 2D histograms of nos and pgc content in all detected particles at stage 10b (h), stage 12 (i), and stage 13 (j). Heat map indicates relative density of data points. Horizontal red and vertical green lines indicate thresholds separating localized granules and unlocalized particles. For all localized granules detected above the threshold in one channel (to the right of the green line for localized pgc; above the red line for localized nos), percentages indicate the fractions with zero content, low content (primarily unlocalized particles), or high content (corresponding to localized granules) in the second channel (as read from left to right for nos in green; from top to bottom for localized pgc in red). cycB behaves similarly to nos and pgc (see Supplementary Fig. 3).

Once they reach the posterior, pgc and cycB mRNAs, like nos, begin to form large granules (Fig. 3e–j). To determine whether nos and pgc can reside within the same polar granule, we calculated the distance in 3D space between nearest neighbors in the two channels. In the 2-color nos FISH experiments described above, >95% of localized granules detected in one channel possessed a partner in the other channel at a distance of <200 nm (<3 pixels in the xy plane). Given a density of about 0.7 granules/μm³, the expected co-localization rate due to random chance is ~10% (see Methods). Using 200 nm as the criterion for co-localization, we found that by late oogenesis 33% of nos granules contain pgc, and conversely 62% of localized pgc contains nos (Fig. 3j). The frequent overlap of pgc and nos granules suggests that nos and pgc mRNAs are physically linked. Similar results were obtained for nos and cycB (Supplementary Fig. 3e,f). We conclude that different germ plasm mRNAs co-assemble in granules exclusively at the posterior cortex.

Dual mRNA labeling experiments showed that a substantial fraction of localized granules contain only one of two different mRNAs regardless of the copy number of that mRNA (Fig. 3h–j, Supplementary Fig. 3f). In contrast, the vast majority of granules containing >5 molecules of nos or cycB either alone or combined together exhibit co-localization with granules of Vas, visualized by direct fluorescence of GFP-Vas (Fig. 4). The co-localization of germ plasm mRNAs with GFP-Vas occurs in the same spatial domain where we observe their enrichment, consistent with a model wherein Vas activity at the posterior is required to recruit single-copy mRNA particles into nascent granules. Moreover, a small fraction (6%) of localized GFP-Vas objects lack both nos and cycB (Fig. 4j) but likely contain other posteriorly localized mRNAs. The enrichment of granules completely lacking any given mRNA rules out a simple model of granule assembly by random selection of mRNAs from the unlocalized pool. Such a model predicts that, for granules of a given total size, the fraction containing at least one transcript from a particular gene must always be greater than the fraction containing zero copies of that mRNA (Supplementary Fig. 4). This is not observed in the dual mRNA labeling experiments, where the fraction containing zero copies is notably enriched (Fig. 3c, Supplementary Figs. 3f and 4). Instead, our data are consistent with a model in which 1) transcripts of a given gene are preferentially incorporated into granules through the formation of homotypic mRNA clusters, and 2) homotypic clusters containing different mRNAs associate in higher-order assemblies to form heterogeneous polar granules.

Figure 4. mRNA content of posterior granules is heterogeneous.

(a–g) Maximum z-series projections spanning 8 μm at the posterior cortex of a stage 13 GFP-Vas oocyte co-labeled with probes for nos and cycB. (a) Merge of all three channels. (b,c) Pairs of channels illustrate a high degree of co-localization between GFP-Vas and cycB (b) or nos (c). (d–g) An enlarged view of the boxed area in (a) shows heterogeneity of granule components. Scale bars: 10 μm (a); 0.5 μm (d–g). (h–j) 2D histograms of GFP-Vas fluorescence in arbitrary units (A.U.) and the number of nos (h), cycB (i), or nos+cycB (j) mRNAs per granule. Horizontal red lines indicate the threshold separating localized and unlocalized particles by mRNA content. Vertical green lines represent the threshold between readily distinguished GFP-Vas granules and imaging noise. 13% of GFP-Vas granules contain 0 nos mRNAs and 11% contain 0 cycB. Conversely, at least 88% of localized nos and 96% of localized cycB granules contain GFP-Vas. A small fraction (6%) of posteriorly localized GFP-Vas granules contain neither mRNA (j). We did not detect GFP-Vas over background levels in unlocalized nos and cycB particles, but cannot exclude the possibility that there is some Vas associated with these particles.

osk multimerization in the nurse cells and anterior oocyte precedes assembly of large localized granules

In contrast to other germ plasm mRNAs, posterior localization of osk occurs in two mechanistically distinct phases. During mid-oogenesis, osk is transported to the posterior in RNPs containing the double-stranded RNA binding protein Staufen (Stau), which may link osk to kinesin and is required for osk translational activation and anchoring at the posterior cortex^24–26. In addition, osk mRNA forms higher order complexes that may play a role in repression of osk translation and coordination of osk transport^8,27,28. The osk mRNA content of these complexes, and where and when they form, has not been determined. During late oogenesis, osk is localized along with other germ plasm mRNAs by diffusion and posterior entrapment^15,29. We investigated the nature of osk RNP complexes during both phases of localization using methods described above for nos to confirm detection of single osk mRNAs (Supplementary Fig. 5a–d) and correspondence between objects observed in live and fixed oocytes (Supplementary Fig. 5e–h).

Like nos, cycB, and pgc, osk is highly expressed in nurse cells and can be found on track-like structures during mid-oogenesis (Supplementary Fig. 5i). However, osk exhibits several differences from nos. First, whereas nos is found uniformly in all nurse cells as single mRNAs, a large fraction of osk particles in nurse cells contain two mRNAs at levels that are twice as high as expected from random co-localization alone (Fig. 5 and Methods). Second, among nurse cells at the posterior of the cluster, the nurse cell connected exclusively to the oocyte and to no other nurse cells contains less osk overall, as well as a lower ratio of particles containing two osk mRNAs compared to other nurse cells (0.8 versus 0.4 particles/μm³; Fig. 5, Fig. 6b,c). These observations are explained by a mathematical model in which 2-copy particles are preferentially transported between nurse cells as well as between nurse cells and oocyte (Supplementary Note), suggesting that the formation of 2-copy particles enhances osk transport to the oocyte. Third, osk particles containing 2–4 mRNAs are heavily enriched in the ooplasm from mid-oogenesis onward and persist in the bulk cytoplasm of the early embryo. Colocalization of osk with Stau-GFP was first detected in the oocyte, coinciding with the formation of osk particles containing > 2 mRNAs at the anterior and persisting in localized granules (Fig. 5). These results show that, unlike nos and other germ plasm mRNAs, posterior localization of osk is preceded by the assembly in the oocyte of RNPs containing Stau and multiple osk mRNAs.

Figure 5. osk mRNA forms multimeric complexes upon entry into the oocyte.

(a) Low-magnification view of a GFP-Stau stage 10b egg chamber labeled with osk probes. Squares indicate regions magnified in (b-b″) and (c-c″). (b-b″) osk particles in anterior (b) and posterior (b′) nurse cells and at the anterior of the oocyte (b″). Scale bar: 2 μm. (c-c″) osk and GFP-Stau in a nurse cell (c), at the anterior (c′) and at the posterior (c″) of the oocyte. Scale bars: 50 μm (a); 2 μm (b,c). (d) Histograms of mRNA content for unlocalized osk particles in anterior (upper) and posterior (middle) nurse cells, and at the anterior of the oocyte (lower). Cyan lines indicate Gaussian distributions centered on integer numbers of mRNAs. Red lines indicate sums of Gaussians. In anterior nurse cells, fitting to two Gaussians yields 42% of particles containing 2 mRNAs, compared with 22% in posterior nurse cells.

Figure 6. Dynamics of osk granule assembly.

(a–g) Confocal images showing the distribution of osk mRNA throughout oogenesis. Z-series projections are shown in all but (d). (a) Germarium (outlined) with expression in presumptive nurse cells and accumulation in presumptive oocytes. (b,c) Stage 4 (b) and stage 6 (c) egg chambers displayed at saturating settings to illustrate osk expression in nurse cells and high concentration in oocytes. At all stages, anterior nurse cells consistently display higher levels of osk than posterior nurse cells. (d) Single image plane at stage 7 reveals the transient accumulation of osk in the center of the oocyte. (e) Stage 10a with strong osk enrichment at posterior pole in midsagittal cross-section. (f,g) Cortical views at stage 12 (f) and stage 14 (g). Scale bars: 10 μm (d,e); 20 μm (f,g). (h) Histograms of cortical osk particle intensities during mid to late oogenesis. (i) osk content as a function of distance from posterior pole. (j) Normalized distribution of mRNA content of all particles found in confocal z-stacks encompassing the posterior 20% of a stage 13 oocyte (top) and a similarly sized volume in the center of the oocyte (bottom) illustrating a reduction in the fraction of particles containing 2–4 mRNAs in the posterior domain.

We next examined the posterior cortical enrichment of osk once it enters the oocyte. We observed bright granules at the posterior cortex during stages 8 and 9 of oogenesis, but the high density of osk in the ooplasm through stage 8 (Fig. 6a–d) precluded us from determining if these granules assemble prior to their arrival at the posterior. By stage 9, osk in the bulk ooplasm is found in 1- to 4-copy punctae, with nearly all brighter granules near the posterior cortex (Fig. 6e–g), where they grow in both number and mRNA content through the end of oogenesis (stage 14). Like nos, the distribution of osk mRNA content in localized granules is heterogeneous and can be described by a log-normal distribution (Supplementary Fig. 5j,k). Total osk content per granule approaches a maximum of ~250 mRNAs during late oogenesis (Fig. 6h). From stage 10 onward, a large majority (>90%) of granules containing 50 or more osk mRNAs are found ≤5 μm beneath the cortex, within 30 μm of the posterior pole as measured along the surface of the cortex (Fig. 6i). Thus, large osk granules reside within the same domain where nos granules subsequently form. Particles containing small numbers (2–4) of osk mRNAs are depleted at the posterior cortex (Fig. 6j), suggesting that the large localized granules are assembled preferentially from the smaller osk RNPs. These results indicate that localization occurs through osk granule assembly in a unique manner distinct from other posteriorly localized mRNAs, with the formation of particles containing 2 osk mRNA molecules in nurse cells, the association of these upon entry to the oocyte to form particles containing 4 osk transcripts and, by late oogenesis, the assembly of large complexes that contain hundreds of mRNAs.

osk mRNA is continuously segregated from germ granules

The accumulation of osk at the posterior cortex in multi-copy granules concomitantly with nos and other germ plasm mRNAs during late oogenesis raised the possibility that osk might be incorporated into the same localized granules. We therefore performed dual FISH for nos and osk in oocytes with expressing either GFP-Vas or Osk-GFP. Granules of localized osk mRNA inhabit regions devoid of nos, GFP-Vas or Osk-GFP (Fig. 7a–j). Nearest neighbor distance measurements confirmed that 90% of particles containing nos mRNA co-localize with GFP-Vas or Osk-GFP. In contrast, <25% of the detected osk mRNA granules reside within 200 nm of a granule containing nos, GFP-Vas, and/or Osk-GFP (Fig. 7k). Rather, localized osk granules show 90% concordance with Stau-GFP (Fig. 5c), suggesting a role for Stau in the late localization of osk in addition to its earlier role in osk transport. These results indicate that osk mRNA is segregated from nos and other germ plasm RNAs by incorporation into compositionally distinct granules.

Figure 7. osk mRNA is continuously segregated from germ plasm components.

(a–e) Maximum z-series projection spanning 5 μm at the posterior cortex of a stage 13 GFP-Vas oocyte and labeled with osk and nos probes. (f–j) Cortical image of a stage 13 Osk-GFP oocyte. (b–e, g-j) High magnification of boxed regions, showing osk mRNA occupying regions devoid of germ plasm components. Scale bars: 10 μm (a,f); 1 μm (b–e, g-j). (k) Distribution of distances between nearest neighbors for each pair of particles. Median distance between osk and nos, GFP-Vas or Osk-GFP is 400 nm, compared to 75 nm for nos and either GFP-Vas or Osk-GFP.

Localization of osk to germ granules impairs germ cell formation

The segregation of osk from nos is maintained into embryogenesis (Fig. 8a). Prior to pole cell formation, nos granules dissociate from the cortex and accumulate around the centrosomes of posterior nuclei through dynein-dependent transport on astral microtubules. This trafficking promotes incorporation of nos along with the nuclei into pole cells as they bud from posterior of the embryo¹⁹ (Fig. 8b,c). These granules retain other germ plasm mRNAs, as evidenced by the continued co-localization with cycB (Fig. 8d). Localized osk granules also dissociate from the cortex, but do not accumulate around posterior nuclei or become enriched in pole cells (Fig. 8b). By the onset of cellularization, osk granules are nearly absent from the posterior (Fig. 8c), whereas nos and other germ plasm mRNAs are highly enriched in pole cells (Fig. 8c–f; refs. ^10,11).

Figure 8. osk localization to germ plasm impairs germline development.

(a) Confocal section at the cortex of an early (n.c. 2) wild-type (WT) embryo labeled with nos and osk probes, showing the continued segregation of nos and osk. (b) Enrichment of nos on astral microtubules and depletion of large osk granules in a n.c. 9 embryo. (c) nos is highly enriched in pole cells of a blastoderm (n.c. 14) embryo, whereas osk is largely absent. nos displayed at lower contrast, and osk at high contrast, compared to (a) and (b). (d,e) Co-localization of nos and cycB during n.c. 3 (d) and n.c. 11 (e). (f) High magnification view of nos and cycB granules. (g–i) FISH analysis of osk-bcd3′UTR (OB) embryos with bcd and osk probes. (g) Z-series projection of the anterior cortex show co-localization of bcd and osk-bcd3′UTR mRNAs in granules at n.c. 3. (h) Granules containing osk are excluded from the vicinity of presumptive ectopic pole cells at n.c. 10. (i) bcd particles are enriched in ectopic pole cells at n.c.14, which are largely devoid of osk. Most osk granules are confined to a region basal to the blastoderm layer. (j,k) FISH analysis of oskΔi1,2-nos3′UTR embryos (OΔN) with osk and nos probes. Z-series projections of the posterior cortex at n.c. 4 (j) and n.c. 14 (k) show co-localization of oskΔi1,2-nos3′UTR and nos through pole cell formation. Scale bars: 10 μm (a–e, g-k); 2 μm (f). (l) Left: Anti-Vas immunofluorescence in wild-type (upper) and oskΔi1,2-nos3′UTR (lower) embryos. Right: box plot showing mean (red line), 25th and 75th quartiles (blue boxes), and range (black lines) of Vas-positive pole cell numbers (n=40 wild-type and 59 oskΔi1,2-nos3′UTR embryos). (m,n) Anti-Vas immunostaining reveals a reduction in Vas-positive cells in the gonads of late-stage oskΔi1,2-nos3′UTR embryos.

To determine whether osk is actively excluded from pole cells, we analyzed osk-bcd3′UTR embryos in which osk is mislocalized to the anterior by replacement of the osk 3′ untranslated region (3′UTR) with the bicoid (bcd) 3′UTR mRNA localization signal³⁰. Osk protein produced at the anterior of osk-bcd3′UTR embryos results in ectopic germ plasm assembly and the induction of functional pole cells at the anterior³⁰. Prior to this ectopic pole cell formation, a large fraction of granules at the anterior cortex contain both osk and bcd mRNAs (Fig. 8g). During pole cell formation, we unexpectedly observed the recruitment of granules containing exclusively bcd into regions surrounding prospective anterior pole cell nuclei whereas osk-containing granules are clearly excluded from the same region (Fig. 8h). By the end of the blastoderm stage only bcd is highly enriched in ectopic pole cells even though large osk granules persist at the anterior beyond the time when endogenous osk at the posterior has largely disappeared (Fig. 8i). These results indicate that osk is effectively excluded from germ plasm granules that are segregated to the developing germline.

To determine the functional relevance of this exclusion, we examined the consequence of forcing osk localization to nos-containing polar granules. We generated an osk transgene that lacks features required for osk localization³¹ and whose 3′UTR is substituted by the nos 3′UTR, which directs posterior localization and incorporation into pole cells³² (UAS-oskΔi1,2-nos3′UTR). oskΔi1,2-nos3′UTR mRNA is highly co-localized with nos in polar granules, resulting in the segregation of osk mRNA to pole cells (Fig. 8j,k). Strikingly, both the amount of Vas in the pole cells and pole cell number are significantly reduced in oskΔi1,2-nos3′UTR embryos compared to control embryos (Fig. 8l). In contrast, no defects were observed in embryos expressing gfp-nos3′UTR mRNA, which was similarly localized to polar granules and incorporated into pole cells (Supplementary Fig. 6). Moreover, gonads of oskΔi1,2-nos3′UTR embryos contain few germ cells (Fig. 8m,n). This decrease is not due simply to overexpression of Osk protein, since embryos bearing 6 copies of a wild-type osk transgene develop an excess of pole cells³³. These results indicate, therefore, that the inclusion of osk in germ plasm granules destined for pole cells is detrimental to pole cell induction and germline development.

DISCUSSION

Our quantitative analysis of germ plasm localized mRNAs has revealed several intriguing features about the localization process and the coordinate regulation of their integration into the pole cells. We show that nos, pgc, and cycB mRNAs are transported within the oocyte as single mRNAs and are co-packaged into granules specifically at the posterior cortex concomitant with localization. Thus, localization serves not only to concentrate these transcripts at the posterior, but generates large, multi-copy polar granules to coordinate the efficient incorporation of these transcripts into the pole cells.

Polar granules are heterogeneous with respect to both the amount of a particular mRNA and the combination of different mRNAs. While nos mRNA content of polar granules varies over a large range, there is a tendency toward higher values fitting a log-normal distribution. Log-normal distributions are often associated with exponential growth processes, whereby the rate at which an object grows is proportional to the size of the object^34,35. Thus, a log-normal distribution suggests that large granules grow at faster rates compared to small ones as assembly is accelerated through positive feedback. Additionally, we find that for granules containing both nos and pgc or nos and cycB, the quantities of the two different mRNAs are somewhat correlated and there is a greater fraction of granules completely lacking one species of mRNA entirely than granules containing just a few copies of that mRNA. Together, these data suggest that for each type of mRNA, cooperative interactions generate homotypic RNPs that then 1) accelerate the recruitment of additional mRNAs of the same type, and 2) facilitate granule assembly by promoting interactions with similarly sized homotypic clusters of other mRNAs. These results also predict the existence of dedicated molecular pathways, one to form homotypic clusters, and another to assemble homotypic clusters of many different transcripts into higher-order granules. These higher-order granules may form by fusion of smaller homotypic granules and indeed fusion of granules labeled with GFP-Vas is observed by live imaging²³. Alternatively, clusters of different mRNAs may grow alongside each other on a predefined granule scaffold. The localization of C. elegans germ granules – P granules – occurs via a phase transition in which soluble RNP components condense at the posterior of the embryo³⁶. It is interesting to consider whether formation of homotypic clusters occurs by a condensation of single transcript RNPs mediated by RNA-binding proteins.

In contrast to other posteriorly localized RNAs, which travel as single molecules, osk forms oligomeric complexes beginning in the nurse cells. Previous studies indicated that reporters containing the osk 3′UTR can hitchhike on wild-type osk mRNA by 3′UTR-mediated dimerization^8,37. Hitchhiking is not required for osk transport, but co-packaging may be important for translational repression of osk prior to localization^27,28. Consistent with this, multi-copy RNPs are competent for localization by both kinesin-dependent transport during mid-oogenesis and diffusion/entrapment during late stages of oogenesis.

Previous ISH-immuno-electron microscopy analysis of stage 10 oocytes showed co-localization of osk with Stau, but not with Osk protein in polar granules³⁸ and our results indicate that osk/Stau RNPs are continuously segregated from the germ plasm granules. This physical separation has functional consequences. Whereas co-packaging of nos, pgc, and cycB coordinates their transport to posterior nuclei and consequent segregation to the pole cells, osk is specifically excluded. Although we do not know why targeting of Osk to polar granules appears to alter their function, it is clearly detrimental to germline development. It will be interesting to determine whether osk/Stau granules contain other mRNAs whose functions are required specifically during oogenesis or early embryogenesis but must be excluded from pole cells.

METHODS

Fly stocks

The following mutants, mutant combinations, and transgenic lines were used: y w^67c23 (ref. ³⁹); osk^A87 (ref ⁴⁰); vas^PD/vas^D1 (ref. ^41,42); nos^BN (ref. ⁴³); nos-(ms2)₁₈ (ref. ⁴⁴); osk-(ms2)₆ (ref. ⁴⁵); hsp83-MCP-GFP (ref. ¹⁶); gfp-stau (ref. ⁴⁶); gfp-vas (ref. ⁴⁷); osk-gfp (ref. ⁴⁸); osk-bcd3′UTR (ref. ³⁰). UASp-oskΔi1,2-nos3′UTR was created by standard molecular biology methods and introduced into y w^67c23 animals by P element-mediated transformation (ref. ⁴⁹). The UASp-gfp-nos3′UTR transgenic line was kindly provided by W. Eagle. Expression was elicited in heterozygous females using the maternal triple Gal4 driver (MTD-Gal4, ref. ⁵⁰).

Single molecule FISH

FISH was performed as previously described²⁰ using oligonucleotides 20 nt in length and complementary to transcripts nos-RA (CG5637; 75 probes), cycB-RA (CG3510; 48 probes), pgc-RA (CG32885; 48 probes), and osk-RA (CG10901; 99 probes). Sequences are listed in Supplementary Table 1. Custom oligonucleotides with 3′ amine modification were obtained from Biosearch Technologies, conjugated to either Atto 565 (Sigma 72464) or Atto 633 (Sigma 43429) dye, and purified by HPLC as previously described⁵¹. Intact ovaries were dissected from well-fed 2–4 day old females and processed for FISH as described²³. Imaging was performed using a 63x HCX PL APO CS 1.4 NA oil immersion objective on a Leica TCS SP5 laser-scanning confocal microscope equipped with GaAsP “HyD” detectors in photon counting mode, with pixels of 76 x 76 nm and z spacing of 340 nm. For each probe set, laser power was adjusted to optimize separation of signal and noise. The same settings were used repeatedly for all samples treated with a given probe set. For each experiment, image stacks were acquired from 3–5 different egg chambers/oocytes or embryos at each developmental stage shown. Each stack contained 20,000–100,000 particles, depending on the volume imaged and the size of the egg chamber/oocyte/embryo. Laser power exhibited fluctuations of ~10% between experiments; however, normalization to single mRNA intensity nullifies any effect in our measurements of absolute mRNA amount. For egg chambers/oocytes, z-series represent approximately half the thickness of the tissue starting from near the interface between the follicle cells and the nurse cells/oocyte. For embryos, image stacks extend from the cortical surface to near the midsagittal plane.

As controls for goodness of single particle detection, we compared the densities of punctae in early embryos from wild-type females and from females heterozygous or homozygous for the maternal RNA-null nos^BN allele. As expected, heterozygous embryos contain half as many particles as in wild-type (0.39 ± 0.01 versus 0.75 ± 0.02 particles/μm³, n = 4 embryos each). In homozygous nos^BN embryos, the number of particles/volume exceeding the “difference of Gaussians” detection threshold⁵² was <1% of the number detected in wild-type. Similarly for osk, embryos from females heterozygous for the mRNA-null allele osk^A87 contain half the number of particles of wild-type embryos (0.24 ± 0.03 versus 0.41 ± 0.03 particles/μm³ in wild-type, n = 4 embryos each). These results are similar to the two-fold reduction in particle number observed for maternally supplied hunchback (hb) mRNA in hb deficiency heterozygotes²⁰ and support previous findings^20,21 that the FISH method effectively distinguishes true objects from imaging noise.

Immunostaining

For pole cell counting, anti-Vas immunofluorescence was performed on 1–3 hr old embryos as described⁵³ using 1:2000 rabbit anti-Vas (gift of P. Lasko) and 1:1000 Alexa Fluor–568 goat anti-rabbit secondary antibody (Molecular Probes). Images were taken at 2.5 μm intervals throughout the embryo posterior with a Leica TCS SP5 confocal microscope using a 20x objective. For gonad analysis, anti-Vas staining of 6–16 hour old embryos was followed by 1:2000 biotin goat anti-mouse (Jackson ImmunoResearch) secondary antibody, amplification with Vectastain reagent (Vector Labs), and peroxidase immunohistochemistry. Embryos were mounted in Aquapolymount and imaged using Nomarski optics.

Absolute RT-qPCR mRNA quantification and comparison to image particle counts

RT-qPCR was performed on individual embryos isolated from embryos 0–30 min after egg deposition as described²⁰. The following PCR primers were used: nos (5′-GTGCGAGTCACCAGCAAACGGACG-3′ and 5′-CGTAGGACATGCGACCGAGATCATCG-3′); osk (5′-CAACGAAAGGGGCGTGGTGCG-3′ and (5′-CGCTGCCGACCGATTTTGTTCCAG-3′). DNA standard curves were generated using the nos N5 (ref. ⁵⁴) or Blue-osk⁵⁵ cDNA plasmids; RNA standard curves were generated using template mRNA prepared by in vitro transcription of the corresponding plasmid with the mMessage mMachine T7 kit (Ambion). Fractional recovery of mRNA from processed embryos is estimated at about 30% (ref. ^20,21).

The mean number of nos transcripts measured by RT-qPCR (2.5 ± 0.4 x 10⁶, n = 6 embryos) was compared to the total number of particles detected in tiled confocal stacks representing the left or right halves of 4 embryos (2.0 ± 0.5 x 10⁶). Previous work has shown that 4% of nos mRNA is posteriorly localized²². Thus, although these transcripts are underrepresented in the total number of particles counted but not in the RT-qPCR measurement, a discrepancy of 4% is significantly smaller than the inherent measurement errors (16% for RT-qPCR, 25% for particle counting). The discrepancy between the number of particles and the number of transcripts reflects the probability that multiple mRNAs occupy the same imaging volume by random chance (see below). We conclude that most unlocalized nos particles observed by imaging correspond to individual mRNAs. The intensity distributions and densities of cycB and pgc are highly similar to those of nos in our experiments, consistent with the idea that the unlocalized transcripts do not form multi-copy complexes.

The mean number of osk mRNAs measured by RT-qPCR (1.9 ± 0.6 x 10⁶, n = 6 embryos) was compared to the total number of particles counted in tiled confocal stacks covering the left or right halves of 4 embryos (0.9 x 10⁶ ± 0.4 osk particles per embryo). From these measurements, the average osk particle contains 2.1 ± 1.1 mRNAs. Previous work has estimated that 18% of osk is found at the posterior pole²², a substantially larger fraction than for nos. Examining particle intensity and spatial distributions, we estimate that at least 95% of particles are not localized at the posterior pole. Since these must account for 82% of osk mRNA, we find that the average unlocalized particle contains (0.82 x 1.9) / (0.95 x 0.9) = 1.8 mRNAs. This value is consistent with the estimated frequency of co-localization of 2 or more osk mRNAs (see below and Supplementary Fig. 5d). In these estimates, 40% of particles contain 1 mRNA, 50% contain 2 mRNAs, and 10% 4 mRNAs. This yields a mean mRNA content per particle (0.4 x 1) + (0.5 x 2) + (0.1 x 4) = 1.8. We conclude that FISH is capable of distinguishing individual osk mRNAs.

For all FISH results, the fixation, in situ hybridization, and mounting processes lead to reduction in embryo length by up to 15% and compression along the imaging axis by 30%. Therefore, living embryos will exhibit distances that are larger, and densities that are lower, than the values we report for fixed samples.

Estimation of spurious co-localization

If individual mRNAs are scattered within a volume at random, then two or more mRNAs may occupy the same confocal imaging volume purely by chance. Such spurious co-localization will result in mis-detection of two mRNAs as a single object with an mRNA content of two mRNAs, erroneously appearing to represent a multi-copy particle. The rate of spurious co-localization depends on the density of mRNAs. To determine whether the observed frequency of nos co-localization exceeds the frequency expected for random chance, we utilized two methods: first, we developed a simple Poissonian model of random co-localization; second, we constructed pair-correlation functions.

Model of random co-localization

A simple model of random co-localization can predict the frequency of spurious co-localization. The model requires choosing an effective volume occupied by diffraction-limited spots in confocal microscopy, with the assumption that when random co-localization occurs, two or more mRNAs inhabit the same position within an imaging volume and are always mis-detected as 1 particle (and never correctly detected as 2 particles). In reality, our algorithm fails gradually, rather than abruptly, to discriminate between two closely spaced particles as the distances between the particles decrease. To bias our estimates toward the hypothesis that packaging does indeed occur for unlocalized nos, we intentionally used small effective volumes and chose high mRNA densities at which we fail to distinguish individual mRNAs.

In the first estimate, we assumed that diffraction limited spots have a volume of $\frac{4}{3}\mathrm{\pi }(0.3\times 0.3\times 0.9\mathrm{\mu }\mathrm{m})=0.35\mathrm{\mu }{\mathrm{m}}^3$ where, as observed previously with the same imaging protocol, the radius along the z-axis is about 3-fold higher than in the xy plane of the image²⁰. At a density of 0.8 particles/μm³ x 1.3 mRNA/particle = 1.0 mRNA/μm³, unlocalized nos mRNA occupies, on average, 1.0 mRNA/μm³ x 0.35 μm³/mRNA = 35% of available volume. With this mean occupancy rate, the Poisson distribution was used to predict the fraction of imaging volumes containing n = 0, 1, 2… mRNAs (Supplementary Fig. 1f), assuming that detected particles contain an integer number of mRNAs. Of the total volumes with n > 0 (that is, for all detected particles), the fractions with n = (1, 2, 3) are p = (0.83, 0.15, 0.02) and the fraction of volumes with n ≥ 4 is negligible (Supplementary Fig. 1f inset). This predicts that about 1 – 0.83 = 17% of detected nos punctae represent more than one mRNA that co-localize by random chance. This prediction is consistent with the distribution of normalized intensities of 55,744 particles in a 2D histogram (Fig. 1e,f): the distribution along the correlated axis can be approximated by the sum of a pair of Gaussians centered on means of 1 and 2 mRNAs with standard deviations of 0.28 and $\sqrt{2\times {0.28}^2}=0.40\text{mRNAs}$, since variances of the intensities of two mRNAs are additive. With these fits, the ratio of the area under the Gaussian with mean = 2 to that with mean = 1 mRNA in Fig. 1f is 20% ± 3%, or 9291 ± 1932 co-localized particles. This is close to the prediction from the Poisson model of p(n = 2) / p(n = 1) = 0.15 / 0.83 = 18%. Moreover, the predicted probabilities closely recapitulate the measured mean particle content: [1 mRNA x p(n=1)] + [2 mRNA x p(n=2)] + [3 mRNA x p(n=3)] = 0.83 + 0.3 + 0.06 = 1.2 mRNA/particle, similar to the measured value of 1.3 mRNA/particle. This supports the idea that co-localization of two or more unlocalized nos mRNAs arises by random chance alone.

In the above estimate, mis-detection of 2 mRNAs as 1 occurs when the two centroids are separated by ≤0.6 μm in the xy plane. In reality, the algorithm becomes gradually less successful at discriminating two objects as the distance between centroids decreases. Thus, the degree of purely spurious co-localization might be smaller than estimated above. We can make a second estimate that compensates for this effect by using a smaller volume of a diffraction limited spot of $\frac{4}{3}\mathrm{\pi }(0.25\times 0.25\times 0.75\mathrm{\mu }\mathrm{m})=0.2\mathrm{\mu }{\mathrm{m}}^3$. In this case, the mRNA will occupy 20% of the available volume, leading to a predicted spurious co-localization frequency of 10%. Given our estimate from Fig. 1f the co-localization of unlocalized nos mRNAs is ~20%, the remaining 10% of co-localization would occur for non-random reasons, i.e. deliberate packaging of two or three nos mRNAs into the same RNP. However, we note that mRNAs are not randomly scattered with equal probability at all locations throughout the embryo interior. Examination of image stacks shows that a large fraction (possibly as high as 50%) of the embryo is devoid of mRNAs Supplementary (Fig. 1g), likely because of the presence of organelles, yolk granules, or other structures. This results in more dense crowding of mRNAs than expected (Supplementary Fig. 1g). Thus, the effective density of nos mRNA is >1/μm³, increasing the probability of spurious co-localization. Indeed, in two-color FISH images where one channel is rotated 90° relative to the other, the probability of co-detection of unlocalized particles is only about 10% (Supplementary Fig. 1h), consistent with the idea that nos is excluded from some regions of the embryo. If nos were excluded from 40% of the embryo volume, this would effectively increase the density of unlocalized material to 1 mRNA/0.6 μm³ = 1.7 mRNA/μm³. This density, if used in combination with a smaller particle volume, predicts mean mRNA content of 1.7 mRNA/μm³ x 0.2 μm³/mRNA = 34%, again giving rise in the Poisson model (Supplementary Fig. 1i) to values similar to the observed distributions in Fig. 1g. Thus, for nos, random co-localization of individual mRNAs in the same confocal imaging volume is the most parsimonious explanation for the fraction of unlocalized punctae that contain 2 mRNAs.

The same reasoning can be used to assess the likelihood that unlocalized osk mRNAs are found in dimeric complexes: the density of unlocalized osk particles is lower than for nos, at 0.4/μm³. Our imaging and analysis algorithm is capable of distinguishing nearly all particles at this density²⁰. osk mRNA is present at a density of 0.4 particles/μm³ x 1.8 mRNAs/particle = 0.7 mRNA/μm³. At this density, mRNA is expected to occupy 25% of the available volume, and so the predicted rate of co-localization of 2 or more mRNAs by random chance is 12%, 5-fold lower than the 60% estimated from observations (Supplementary Fig. 5d). Since random co-localization dramatically fails to account for the fraction of unlocalized punctae containing >1 mRNA, we conclude that unlocalized osk is deliberately packaged into higher-order RNP complexes containing 2 or more mRNAs.

Pair-correlation analysis

We constructed pair-correlation functions by measuring the density of nos and pgc as a function of distance from a set of “reference” nos particles. In determining nos densities, we considered 18% of detected nos particles to represent 2 mRNAs at the same coordinates. We observed no significant change in density of either nos or pgc as a function of distance from reference particles (Supplementary Fig. 1j), as expected if co-localization in bulk cytoplasm occurs by random chance alone.

mRNA content of localized granules

Particle fluorescence intensities and offsets were estimated by three methods: 2D elliptical Gaussian fitting⁵², summation of fluorescence in a circular area with radius of 3 pixels (for amplitude) and in a ring with inner radius 6 pixels and outer radius 9 pixels (for offset), and 3D summation as previously described for sites of zygotic gene expression²⁰. The fluorescence intensity value for a single mRNA molecule was estimated by two methods: 1) constructing an average image of an unlocalized single-copy particle and either fitting an elliptical Gaussian to obtain an amplitude and offset or summing fluorescence in 2D; 2) analyzing unlocalized particles within a volume of cytoplasm in the interior of the oocyte or embryo below the posterior cortex. In this region, particle density decreases approaching the posterior pole (Supplementary Fig. 1b), and particle density and fluorescence are roughly linearly related. The slope of a line fit to a plot of total intensity as a function of particle number yields fluorescence per particle²⁰. Normalization of granule intensity to fluorescence per single-copy particle yields absolute mRNA content per granule. In pairwise comparisons, the three analysis methods yielded mRNA counts that differed by a maximum of 20% for granules localized at the posterior cortex. This measurement error was considerably higher (up to 60%) for granules positioned less than 5 μm from the posterior pole.

Time-lapse movies of oocytes expressing nos*GFP or osk*GFP were obtained as previously described²³ from n = 3 (nos*GFP) or n = 4 (osk*GFP) oocytes. Granules were tracked using semi-automated custom software²³. Average velocities were calculated for 26 (nos) and 22 (osk) granules. Relative fluorescence intensities were obtained from >250 granules each via 2D summation, and histograms of granule intensities were fit to histograms of FISH intensities of a subset of granules found in a region of the posterior cortex comparable to the region visualized by time-lapse imaging. Histograms of granules observed in time-lapse images do not show a clear threshold separating true particles from imaging noise, so the fit was performed to the right side of the FISH distribution under the assumption that dim particles are not visible in time-lapse images. Fitting was performed to minimize chi square error over a range of bin sizes and minimum and maximum intensity values.