Drosophila embryos spatially sort their nutrient stores to facilitate their utilization

Animal embryos are provisioned by their mothers with a diverse nutrient supply critical for development. In Drosophila, the three most abundant nutrients (triglycerides, proteins, and glycogen) are sequestered in distinct storage structures, lipid droplets (LDs), yolk vesicles (YVs) and glycogen granules (GGs). Using transmission electron microscopy as well as live and fixed-sample fluorescence imaging, we find that all three storage structures are dispersed throughout the egg but are then spatially allocated to distinct tissues by gastrulation: LDs largely to the peripheral epithelium, YVs and GGs to the central yolk cell. To confound the embryo’s ability to sort its nutrients, we employ mutants in Jabba and Mauve to generate LD:GG or LD:YV compound structures. In these mutants, LDs are missorted to the yolk cell and their turnover is delayed. Our observations demonstrate dramatic spatial nutrient sorting in early embryos and provide the first evidence for its functional importance.


Introduction 38
After fertilization, animal embryos develop for extended periods without access to external 39 nutrients. In oviparous species, for example, the young animal gains access to additional 40 nutrients only after hatching when it can feed independently. Therefore, mothers provision 41 their embryos with nutrient reserves to enable embryonic development. In Drosophila, these 42 reserves include neutral lipids (triglycerides and sterol esters), proteins (yolk proteins) and 43 carbohydrates (predominantly glycogen) 1-4 . These nutrients provide both energy supplies and 44 carbon backbones for anabolic metabolism. Different nutrients are not present freely in the 45 cytoplasm, but are segregated from each other and packaged into distinct structures 5 . Neutral 46 lipids are present in the center of lipid droplets (LDs), ubiquitous cellular organelles in which a 47 single phospholipid layer surrounds a central hydrophobic core. Yolk proteins reside inside 48 membranous yolk vesicles (YVs), oocyte-and embryo-specific LROs (lysosome-related 49 organelles) delimited by a phospholipid bilayer. Glycogen forms so-called b particles 50 (carbohydrate chains attached to the priming protein Glycogenin) that assemble into larger 51 glycogen granules (GGs or a particles). These nutrients are utilized at different rates and at 52 different embryonic stages 2,3,6,7 . Most previous studies relied on biochemical analysis of whole 53 embryos and thus could not address the spatial organization of these nutrients in the embryo. 54 In this paper, we address this issue and find that GGs, LDs, and YVs undergo dramatic sorting in 55 the first few hours of embryogenesis and that this sorting is a prerequisite for proper nutrient 56 utilization. 57 Drosophila embryos are a syncytium for the first ~2.5hrs. During this period, the nuclei divide 58 13 times near synchronously before undergoing bulk cytokinesis. The first 8 divisions (stage 1 59 and 2) occur deep within the embryo, with a minority of nuclei staying in the interior and most 60 migrating to the periphery. Arrival of nuclei at the surface and formation of pole cells/germline 61 (stage 3) mark the beginning of the syncytial blastoderm. Over the next hour (stage 4), the 62 peripheral nuclei undergo the 9 th -13 th divisions 8 . Stage 5 encompasses a 1hr long interphase 63 where a simultaneous cytokinesis (cellularization) generates ~6000 diploid cells organized as an 64 epithelium at the embryo's periphery and one central, syncytial yolk cell. The epithelium gives 65 rise to all the tissues of the future larva and adult, while the yolk cell is a transient tissue that 66 functions as a yolk protein depot as well as a positional cue during the maturation of ecto-and 67 endoderm 8,9 . Gastrulation movements (stage 6) and germ-band extension (stages 7-10) then 68 transform this 2D epithelial layer into a complex 3D body plan. 69 The embryo initially shows little or no spatial segregation of nutrients, with LDs and YVs 70 homogeneously distributed 10 . In larvae and adults, in contrast, nutrients are unevenly 71 distributed, with specialized storage tissues dedicated to receiving and disseminating specific 72 nutrients (e.g., fat body tissue for fat storage). Nutritional specialization is already evident at 73 gastrulation when LDs and YVs are allocated to distinct cells: most LDs to the peripheral 74 epithelium; YVs exclusively to the yolk cell 11 . This distribution persists for the rest of 75 embryogenesis (Fig. 1a-c). Thus, one of the first organizational events in Drosophila 76 embryogenesis differentially sorts nutrients, foreshadowing nutrient handling by specialized 77 tissues in later life stages. 78 How embryos spatially control the third major nutrient, glycogen, remains unknown. We 79 therefore visualized GGs using Periodic Acid-Schiff (PAS) staining and fluorescence microscopy 80 as well as a Glycogenin-YFP fusion. These approaches revealed GGs as highly dynamic, 81 undergoing both morphological changes and redistribution before cellularization. By 82 gastrulation, GGs are almost exclusively allocated to the yolk cell, like YVs. We also found that 83 in embryos lacking the LD protein Jabba, LDs are tightly associated with GGs and are 84 transported together to the yolk cell. These misallocated LDs are consumed slower than in the 85 wild type and persist through hatching. Perturbed LD turnover is likely the result of LD 86 misallocation, since delayed consumption is also observed in embryos in which LD 87 misdistribution results from inappropriate interactions with YVs. We conclude that early 88 nutrient sorting during Drosophila embryogenesis leads to an optimal nutrient allocation, 89 ensuring that nutrients are utilized efficiently. 90

Results 91
Lipid droplets and yolk vesicles are sorted to different tissues by cellularization 92 Previous studies had suggested that LDs and YVs are likely present throughout early Drosophila 93 embryos 11 . However, they had relied on fluorescence microscopy of whole mount samples. To 94 unambiguously determine the distribution of LDs and YVs, we analyzed the two organelles in 95 cross-sections using TEM. Both LDs and YVs were homogenously distributed throughout the 96 depth of the embryo ( Fig. 1e; in the lower left, LDs are false colored in yellow, YVs in blue). 97 Thus, early embryos start out with LDs and YVs intermixed. 98 In stage 2 embryos, myosin-II mediated contractions of the cortex lead to large-scale circular 99 flows 12,13 . At the periphery, cytoplasm flows from the pole along the cortex to the middle of the 100 embryo where it descends towards the interior and then flows along the long axis of the 101 embryo to reemerge at the poles 12 (cartooned in Supplementary Fig. 3 f). YVs are known to be 102 carried by this flow 12 . We injected embryos with an LD-specific dye and monitored them live by 103 confocal microscopy (Video S1). In stage 2, LDs flow in the expected pulsative manner along the 104 periphery. In subcortical planes (Video S2), individual particles can be followed for long 105 distances, allowing us to quantify the flow of several organelles by particle image velocimetry 14 106 (see below). Because our imaging conditions do not allow us to image the middle of the 107 embryo, we could not directly observe their flow in the embryo center. However, movies taken 108 at 40 µm depth (Video S1) are consistent with massive rearrangement of these interior LDs and 109 their flow towards the poles. 110 Previous fixed-embryo analysis had found that by Stage 3, LDs are highly enriched in the 111 peripheral ~40 µm of the embryo 10,15 , while YVs remain throughout. We see the same pattern 112 in our movies, where LDs enrich at the periphery during Stage 2 and remain enriched there 113 through Stage 5 (Video S1). During Stage 4, YVs deplete from this region 13 , and by stage 9, LDs 114 and YVs have segregated into the epithelial cells versus the yolk cell 11 , respectively, a 115 distribution that remains through the rest of embryogenesis (Fig. 1b,c) 10,15 . 116 In summary, YVs and LDs are intermixed in the early embryo and segregate from each other in 117 two steps. During stage 2, LDs enrich in the periphery, and during stage 4, YVs deplete from the 118 same region. As a result, the two nutrient stores are allocated to distinct cells by cellularization, 119 creating an early nutrient differentiation between the cells of the embryo (Fig. 1a). 120 121 Two novel, subcellular approaches to visualize glycogen granules by light microscopy 122 Does the third nutrient store (GGs) also undergo sorting? By electron microscopy, GGs appear 123 as large, weakly staining, membrane-less structures (Fig. 1f)   embryos show signal throughout the cytoplasm, but conventional imaging approaches do not 148 reveal fine structure (Fig. 1d). However, when adapted for fluorescence microscopy (Thomas 149 Kornberg, personal communication), fluorescent PAS (fPAS) signal revealed discrete granular 150 structures of <1-5µm diameter throughout the cytoplasm (Fig. 1g). When embryos were 151 pretreated with a-amylase to degrade glycogen, fPAS signal was largely abolished (Fig. 1h), 152 confirming that most fPAS signal in the early embryo represents glycogen. As an independent 153 test, we analyzed fPAS signal during oogenesis, where glycogen specifically accumulates in late-154 stage oocytes (Stage 13 and 14) 5,16 . fPAS signal recapitulates this pattern: only the oldest 155 oocytes were fPAS positive (Fig. 1j). We conclude that fluorescent imaging with PAS staining 156 labels glycogen, with sufficient contrast to resolve individual granules. 157 Live observation of YVs and LDs has revealed detailed information about their mechanism of 158 motion 11 . Because b particles, the subunits of GGs, contain the priming protein Glycogenin at 159 their center 17 , we employed a protein trap line 18 in which an additional coding YFP exon is 160 inserted into the endogenous Glycogenin (Gyg) gene. By confocal microscopy, we observed 161 structures ~1-5 µm in diameter in stage 2 embryos of this strain (Fig. 1i). Because YFP 162 fluorescence is destroyed by PAS staining, we could not directly compare PAS-stained GGs and 163 Glycogenin-YFP, but several lines of evidence strongly support that Glycogenin-YFP indeed 164 reveals GGs. First, size, distribution, and abundance of the YFP-positive structures fit the GGs 165 visualized by PAS staining (Fig. 1g). Second, the only known abundant embryonic structures in 166 this size range are YVs and GGs, and live imaging of embryos revealed that the autofluorescent 167 YVs are distinct from the YFP structures (Fig. 1i,i',i''). Third, during oogenesis, YFP structures 168 become distinct only in Stage 13 and Stage 14 oocytes (Fig. 1k,k'), just like GGs (Fig. 1j). Fourth, 169 in certain mutants (see below), LDs are arranged around GGs (as observed by TEM or fPAS), and 170 we observe the same association around Glycogenin-YFP granules (Fig. 3g'). Thus, Glycogenin-171 YFP reveals GGs. 172 As a final test, we performed in-vivo embryo centrifugation, a technique in which living 173 syncytial embryos are centrifuged to separate their components by density 19,20 . GGs as revealed 174 by PAS staining represent the densest fraction, opposite the low-density neutral lipid 175 ( Supplementary Fig. 1b); this fraction appears clear in bright light ( Supplementary Fig. 1a), 176 fitting a lipid free fraction. Similarly, in centrifuged Glycogenin-YFP embryos, YFP signal 177 accumulates at the very bottom of the embryos, even below the autofluorescent YVs 178 ( Supplementary Fig. 1b). In centrifuged oocytes, coalesced GGs also form a cap below tightly 179 packed YVs 5 . 180 181 Glycogen distribution changes during development leading to yolk cell allocation 182 By TEM, GGs are present throughout the embryo in stages 1 and 2, just like YVs and LDs (Fig.  183 1e). To determine whether GGs are also spatially sorted during embryogenesis, we performed 184 fPAS staining on embryos of different ages. In stage 1, glycogen was evenly distributed (Fig. 1G).

185
By stage 3, imaging in a subcortical plane shows a reduction in the number of GGs at the 186 embryo's surface (compare Fig. 2a to Fig. 1g). In stage 5, fPAS signal was absent from the 187 periphery; the embryo shown in Fig. 2B was imaged 40 µm below the surface (embryo is 188  And fPAS staining of embryos in stages 9 and 10 reveals that the majority of 212 glycogen is indeed in the yolk cell ( Supplementary Fig. 2b). Overall signal is much reduced at 213 that time, and by stage 11 we can no longer detect appreciable fPAS signal (Supplementary Fig.  214 2c), suggesting that glycogen is being turned over. Glycogenin-YFP is also restricted to the yolk 215 cell in later stages ( Supplementary Fig. 2c). 216 217 GG morphology changes during development 218 Our fPAS and Glycogenin-YFP time courses suggest that GG morphology changes as the embryo 219 develops. We therefore examined fPAS-stained embryos at higher magnification. Newly laid 220 embryos were characterized by discrete GGs evenly distributed within the focal plane (Fig. 2e).

221
In Stage 3, GGs were arranged much less homogeneously, forming clusters: frequently, multiple 222 GGs were in close contact with each other, and there was glycogen-free space between clusters 223 (Fig. 2f). This pattern implies that GGs not only move inwards (i.e., perpendicular to the focal 224 plane shown), but also within the plane towards each other. Consistent with that notion, some 225 of the GGs are no longer round, but appear as oblong or more complicated aggregates, 226 implying that GGs are coalescing. By stage 5, fPAS signal forms a single, largely homogeneous 227 mass ( Fig. 2g) that only retains remnants of the granular structure at its outer edges (Fig. 2c).

228
This single GG mass occupies the center of the embryo (Fig. 2b). Consistently, TEM imaging of 229 stage 7 embryos reveals a large, fused glycogen structure (Fig. 2c, red colored area). why LDs are allocated differently and why the embryonic nutrient supply is spatially organized 246 at all is unknown. 247 As a potential inroad into this problem, we re-examined embryos lacking the LD protein Jabba, 248 one of the most abundant proteins on embryonic LDs 22 . Such embryos have normal triglyceride 249 content but display abnormal LD distribution in stage 4 22 . When we imaged fixed stage 1-2 250 embryos stained for LDs, wild-type embryos displayed the pattern expected from our TEM 251 analysis (Fig 1e); LDs were absent from many circular areas (presumably GGs and YVs) but 252 distributed evenly throughout the remaining spaces (Fig. 3a). The pattern in Jabba mutants 253 (two different alleles, Jabba DL and Jabba zl01 ) was dramatically different. Here most LDs were 254 accumulated in rings, with few LDs occupying the space between the rings (Fig. 3b, and data 255 not shown). This aberrant distribution was not an artefact of fixation, as we saw the same 256 pattern in embryos injected with LD dyes and imaged live ( Supplementary Fig. 3d, 0s panel). 257 The pattern in Jabba mutants suggests that LDs accumulate around YVs or GGs. Co-detection of 258 LDs and YVs revealed no association between them ( Supplementary Fig. 3e). In contrast, co-259 labeling of GGs and LDs showed that LDs were surrounding GGs (Fig. 3A'-B"). This finding was 260 further confirmed in TEM cross sections ( Supplementary Fig 4A). A similar, but less pronounced 261 association was already observed when Jabba dosage was reduced ( Supplementary Fig. 3c 1x  262 Jabba). We employed this observation to confirm LD-GG association in living embryos. We 263 injected LipidSpot610 into embryos from mothers expressing Glycogenin-YFP and either one or 264 two copies of the wild-type Jabba gene. LDs and GGs displayed minimal association in the 265 otherwise wild-type background (Fig. 3g), while many LDs were present at or near the surface 266 of GGs when Jabba dosage was reduced (Fig. 3g'). We conclude that when Jabba levels are 267 reduced, LDs associate with GGs. This association can already be observed in oocytes, as soon 268 as GGs are detected ( Fig. 3e-f'). Finally, we detected no association between LDs and GGs when 269 other LD proteins are missing, namely PLIN-2/LSD-2, Sturkopf/CG9186, or Klar (Supplementary 270 Fig. 3c). We conclude that Jabba uniquely suppresses inappropriate interactions between these 271 storage structures. 272 Although we occasionally observe LDs deep within GGs of Jabba mutants, for the most part the 273 LDs are arranged in a ring pattern with glycogen in the center (Fig. 3d"). In fact, LDs appear 274 embedded within the outer regions of the GGs; see, for example, Fig. 3d',d", where GGs display 275 indentations/areas of exclusion in the PAS signal ( Fig. 3d', arrowheads). These regions are filled 276 with LDs (Fig. 3d"). Rarely were such associations or indentations observed in wild type (Fig. 3c',  277 arrowheads). TEM confirmed a tight LD-GG association: in Jabba embryos, LDs appear to 278 Wild type embryo Lipid droplets  directly contact GGs, with glycogen bulging out between LDs ( Fig. 3h; Supplementary Fig. 3b). In 298 the wild type, such association is observed rarely (Fig. 1e; Supplementary Fig. 3a; Fig. 3i). This 299 association is strong enough that when Jabba embryos are centrifuged, the dramatic separation 300 of LDs and GGs into distinct layers at opposite ends of the embryo observed in the wild type 301 ( Supplementary Fig. 1b) is disrupted. In the Jabba mutant, glycogen signal is found intermixed 302 with the LD layer in the region of lowest density, and pockets of LD signal are present within the 303 glycogen layer in the highest-density region (Supplementary Fig. 1c). 304 305

Consequences of LD:GG interactions on LD motility 306
To determine if the LD/GG association in Jabba embryos affects LD motility, we labeled LDs by 307 injecting dyes into embryos, monitored their behavior live, and quantified their flow speeds by 308 particle image velocimetry 14 . In the wild type, LDs and acidic organelles flow faster than YVs 309 (Fig. 3j), presumably due to their smaller size. Jabba mutant embryos displayed the expected 310 ring-arrangement of LDs ( Supplementary Fig. 3d); these rings moved as a unit (Video S5), 311 reinforcing that LD:GG complexes are stable structures. Compared to wild type, they displayed 312 stuttered motion, minimal displacement ( Supplementary Fig. 3d), and lower average velocity 313 (Fig. 3j). In contrast, YVs and acidic organelles showed similar mobility between the two 314 genotypes (Fig. 3j). Thus, altered LD flow in Jabba embryos does not represent a general defect 315 in cytoplasmic streaming, but rather a specific disruption of LD motility, likely due to the much 316 larger size of LD:GG complexes relative to individual LDs. 317 318

Consequences of LD-GG interactions on LD allocation 319
During wild-type development, GGs and LDs are initially intermixed and homogeneously 320 distributed throughout the embryo (Fig. 1e, Supplementary Fig. 4a). LDs enrich at the periphery 321 by stage 3 (Video S1), and by the end of stage 5, GGs and LDs are segregated from each other 322 and allocated to different tissues. In the absence of Jabba, GGs and LDs form large composite 323 structures that are also distributed throughout the early embryo ( Supplementary Fig. 4B). But 324 because these composite structures travel together, segregation to distinct locations seems no 325 longer possible. Indeed, in our movies with labelled LDs (Video S5, S6), the LDs trapped in 326 GG:LDs complexes were far less mobile, engaging in delayed, stuttered motion and did not 327 enrich in the periphery as in the wild type (Video S1). For post-cellularization embryos, TEM 328 analysis revealed a massive redistribution of LDs in Jabba embryos relative to wild type (Fig. 4b-329 e), with fewer LDs in the peripheral epithelium and more in the yolk cell. In the wild type, 72% 330 of 510 LDs scored were present in the epithelial cells, while in Jabba embryos it was only 23% of 331 the 696 LDs scored. 332 In contrast, we only found minor differences in glycogen distribution between the two 333 genotypes. By fPAS staining, wild type and Jabba were very similar ( Supplementary Fig. 2E). By 334 TEM, the bulk of GGs in Jabba embryos were appropriately localized to the yolk cell, just like in 335 the wild type ( Fig. 4B-E). There were a few instances of small LD:GG complexes segregated into 336 blastoderm cells (Fig. 4E green arrowhead). We conclude that the composite LD-GG structures 337 in Jabba embryos are allocated like GGs, leading to LD mislocalization.  showing the border between the cellularized epithelium (right) and yolk cell (left). Scale bars = 344 5 µm. The green arrowhead (E) shows a YV attached to LD:GG complex which has 345 inappropriately localized to the epithelium. F,G) L1 larva stained with Oil Red O to detect LDs. 346 Scale bar = 80 µm. The Jabba larva's residual LDs are visible in its gut. 347 As an independent approach, we detected LDs by Nile red staining and fluorescence microscopy 348 in whole-mount embryos (Fig. 4a). Already in early stages, LD distribution looks different: signal 349 is diffuse throughout the embryo for wild type and granular in Jabba mutants, presumably 350 reflecting LD enrichment around GGs. At the beginning of gastrulation, signal in the mutant is 351 more prominent in the yolk cell, a pattern that becomes even more pronounced in later stages.

352
By stage 14, LD signal is absent from the yolk cell and present everywhere else, while in Jabba 353 embryos this pattern is reversed (Fig. 4A). We conclude that in Jabba embryos the bulk of LDs 354 are indeed mislocalized to the yolk cell. 355 In late-stage embryos, LD signal in Jabba embryos was not only restricted to the yolk cell but 356 also displayed increased intensity. This difference even persisted post-hatching. LD staining 357 using Oil Red O revealed strong signal in the gut lumen (the location of the yolk cell remnant) of 358 newly hatched Jabba larvae (Fig. 4f), unlike wild type (Fig. 4f). Thus, Jabba mutants not only 359 mislocalize their LDs, but fail to consume them properly. 360 361 LD interactions with YVs also lead to their mislocalization and persistence 362 LD consumption in the mutant might be impaired because Jabba is important for LD breakdown 363 or because the yolk cell is not equipped to metabolize this high concentration of LDs. To 364 distinguish between these possibilities, we took advantage of a recent report that embryos 365 lacking the Mauve protein display an interaction between LDs and YVs 23 . Mauve is a resident 366 protein on lysosome related organelles (LROs) important for their maturation. YVs are a type of 367 LRO and in the absence of Mauve display several phenotypes, including an association with LDs 368 23 . Might these YV-LD interactions affect nutrient sorting? 369 As strong disruption of Mauve severely impairs early embryonic development 23 , we generated 370 mothers transheterozygous for two weak alleles, mauve Rosario and mauve 3 . These embryos 371 displayed LDs association with YVs ( Fig. 5a arrow), as well as the YV size heterogeneity and 372 autofluorescence variability reported for stronger allele combinations (Fig. 5a) 23 , but no obvious 373 association between GGs and LDs (Fig. 5b) 23 . 374 The mauve Rosario/3 YV-LD association was an exciting opportunity to test our model that LDs are 375 missorted during cellularization if they are associated with a structure destined for yolk cell 376 deposition, allowing us to utilize YVs instead of GGs. When the mutant embryos were stained 377 for LDs (Fig. 5c), signal in stage 1 was clustered instead of diffuse like in wild type (Fig. 4a). At 378 stage 9, LDs signal was predominately in the yolk cell, but also present in the other tissues (Fig.  379 5c). At stage 14, LD staining was clearly enriched in the yolk cell (Fig. 5a) relative to wild type 380 (Fig. 4a). Thus, mutations in two unrelated genes, Jabba and mauve, show that when LDs 381 inappropriately interact with other nutrient structures, they are mislocalized to the yolk cell and 382 are turned over more slowly.  in Jabba mutants. Note that the timeline is not scaled to absolute developmental time but is 395 meant to show the brief life of nutrients through synthesis, distribution and consumption when 396 compared to other developmental events. Mislocalized LDs in Jabba mutants are speculated to 397 negatively affect completion of embryogenesis or the larvae itself. 398 Glycogen is a major energy store in animals, uniquely capable of providing energy rapidly, both 401 aerobically and anaerobically. While glycogen's functions are well investigated in adult tissues, 402 its role during embryogenesis is less understood. In this study, we developed new tools to 403 determine the spatial distribution of glycogen in Drosophila oocytes and embryos. Consistent 404 with previous biochemical and electron microscopic analysis, we find that glycogen stores 405 accumulate late in oogenesis and are organized into large, membrane-less structures. These 406 GGs are evenly distributed throughout the early embryo and undergo two types of transitions 407 during syncytial stages. First, they are displaced from the subcortical region during stage 4, 408 leading to their accumulation in the center and allocation to the yolk cell. Second, 409 simultaneously, GGs fuse into larger and larger structures, so that by the end of stage 5, most 410 glycogen is present in a large superstructure in the yolk cell. LDs and YVs also start out with an 411 even distribution and are then specifically allocated. After stage 5, YVs are restricted to the yolk 412 cell, like GGs, while LDs are predominately sorted to the peripheral epithelial cells. In two 413 different mutant conditions, we find physical interactions between LDs and either GGs or YVs.

414
In both cases, LDs are misallocated to the yolk cell. In Jabba mutants, a portion of these LDs fail 415 to be consumed during embryogenesis and persist into larval stages; in mauve mutants 416 persistence through embryogenesis is milder, consistent with less severe mislocalization of LDs. 417 We conclude that mislocalizing LDs early in embryogenesis affects subsequent LD consumption. 418 GGs in Drosophila oocytes and embryos are akin to the a particles that mediate long-term 419 glycogen storage in many tissues, including muscle, liver, and fat body 17,24 . Their large size 420 (which must correspond to thousands of b particles 5 as opposed to ~30 in liver 24 ) likely protects 421 against glycogen breakdown and facilitates their physical movement for differential allocation.

422
How a particles assemble and disassemble remains unclear. Proposed binding agents holding 423 neighboring b particles together include covalent links between glycogen chains 25,26 or 424 Glycogenin molecules at the surface of the b particles 27 , in addition to the Glycogenin dimer in 425 their cores. We speculate that the amorphous mass of glycogen in stage 5 embryos represents 426 partial dissolution of GGs into b particles, as a prelude to enzymatic breakdown of glycogen. 427 Consistent with this notion, fPAS staining in stages 9-10 becomes very weak (Supplementary 428 Fig. 2b,c) and biochemical measures of glycogen levels show the same drop 2 . This timing for 429 glycogen depletion overlaps with a proposed switch in embryonic metabolism from 430 carbohydrate-based to triglyceride-based energy production 2 . Disassembly into b particles 431 might provide enhanced access for the cytosolic glycogen phosphorylase, responsible for most 432 glycogen turnover in embryos 28 . 433 Going forward, Drosophila oocytes and embryos should be a powerful model for unraveling the 434 mechanism of the conversion between a and b particles. GGs are large enough to be followed 435 by light microscopy, their assembly and disassembly occurs quickly (within ~2hrs or less), and 436 Glycogenin-YFP allows live imaging of these processes. In fact, to our knowledge this is the first 437 example of live imaging of glycogen in any system. 438 Although during oogenesis the three major nutrient stores are made at different at times and 439 through different mechanisms 5,11 , they all start out intermixed and homogenously distributed in 440 the early embryo. Nutrient sorting starts in stage 2 when myosin-II driven cortex contractions establish large-scale cytoplasmic flows throughout the embryo. Flow speeds, as estimated from 442 the behavior of YVs, are sufficient to spread out the interior nuclei along the entire anterior-443 posterior axis 12 . As LDs flow faster than YVs (Fig. 3j), these flows should be able to transport 444 most LDs from the center to the poles. If these LDs are somehow captured at the periphery, 445 reducing their return to the embryo center, it explains their enrichment in the periphery by 446 stage 3. Consistent with an important contribution from cytoplasmic flow, the LD-GG 447 aggregates in Jabba mutants flow with reduced speeds and LDs fail to enrich at the embryo 448 surface (Video S4). Thus, we propose that this cytoplasmic flow promotes the first step of 449 nutrient sorting, when LDs accumulate peripherally. 450 By stage 4 and 5, the nuclei at the embryo surface set up an array of radially oriented 451 microtubules that traverse a ~40 µm peripheral zone 10 . These microtubules are proposed to 452 push YVs into the interior 13 , and GGs may be displaced by the same mechanism, as they 453 deplete from this zone at the same time. LDs, in contrast, move bidirectionally along these 454 microtubules, employing cytoplasmic dynein and kinesin-1 29,30 ; this motion confines them to 455 the microtubule zone 15 . We propose that MTs keep LDs in the peripheral zone, while they push 456 GGs and YVs inward, resulting in the second sorting step. Our analysis of Jabba and mauve 457 mutants reveals that successful sorting also requires nutrient stores stay separated. When LDs 458 are tightly associated with either GGs or YVs, sorting fails, and LDs are misallocated to the yolk 459 cell. 460 Presumably the nutrients stored in GGs, YVs, and LDs all support the metabolic needs of the 461 developing embryo. Why then are they allocated differently? One reason might be the different 462 properties of their breakdown products. YVs' amino acids and GGs' glucose are water soluble, 463 making diffusion through gap junctions or membrane transporters viable options for 464 dissemination. Thus, the yolk cell can serve as a hub for glucose and amino acid distribution, as 465 it remains connected via cytoplasmic bridges to the blastoderm through stage 9 21,31 and 466 expresses numerous nutrient transporters. In contrast, free fatty acids (FAs) generated from LD 467 breakdown are poorly water-soluble and potentially toxic 32,33 ; they are typically immediately 468 channeled into specific intracellular pathways 34 . For example, efficient FA transfer from LDs to 469 mitochondria requires proximity and direct contact 35-37 . Thus, allocation of LDs predominately 470 to the periphery would allow efficient local energy production. Intriguingly, during zebrafish 471 embryogenesis, LDs are initially highly enriched in the future yolk sac but are imported into the 472 embryo proper via cytoplasmic bridges and actin-myosin based motility 38,39 . Thus, the zebrafish 473 embryo may also depend on a local LD supply to support its dividing cells. 474 What are the consequences of mislocalizing LDs to the yolk cell? Our data indicate that a 475 fraction of these LDs persist through the end of embryogenesis. Although Jabba mutant 476 embryos are viable 22 , their progression through embryogenesis has not yet been analyzed in 477 detail. Recent work on embryonic glycogen metabolism suggests that even minor disruption of 478 LD metabolism has the potential for widespread effects on embryogenesis. Embryos that either 479 lack glycogen reserves or are unable to access them display widespread changes in their 480 metabolome as well as hatching delays 28 . Since fat contributes roughly 10x the energy of 481 glucose (derived from glycogen) during Drosophila embryogenesis 3 , even the modest retention 482 of LDs in Jabba mutants might have prominent effects on development. 483 LDs and GGs co-exist not only in embryos, but also in mature tissues (e.g., muscle, intestinal 484 epithelia, liver, fat body) 17,40,41 . It is conceivable that in embryos inappropriate interactions 485 between LDs and GGs are particularly harmful because of the large size of GGs and the 486 extensive cytoplasmic streaming which presumably leads to many encounters between these 487 organelles. It will therefore be interesting if mechanisms to keep LDs and GGs apart are specific 488 to embryos or also important in other cells. 489 By devising novel imaging methods for glycogen storage structures, we have shown that 490 Drosophila embryos dramatically reorganize their nutrients by cellularization, with distinct 491 nutrients sorted into separate nascent tissues. The embryo employs multiple mechanisms to 492 get its nutrient stores to the correct location, including cytoplasmic streaming, preventing 493 inappropriate interactions, and microtubule-dependent transport. We also provide the first 494 evidence that correct spatial allocation of LDs is necessary for their efficient consumption. 495 Together, these observations suggest that embryos need to achieve an optimal nutrient 496 allocation to support subsequent steps in development (cartooned in Fig 5d) and that the 497 spatial allocation of nutrients is essential to fully understand embryonic metabolism. The 498 importance of this allocation is particularly remarkable as these nutrient stores only exist 499 transiently and are consumed by the end of embryogenesis. 500

Methods 501
Origin of fly strains 502 Oregon R was used as the wild-type strain. Jabba DL and Jabba zl01 were generated previously in 503 the lab and are strong loss-of-function alleles with no Jabba protein detected in early 504 embryos 22 . Df(2R)Exel7158/CyO carries a large deletion that encompasses Jabba and is used to 505 reduce Jabba dosage; for simplicity, embryos from mothers carrying this deletion are referred 506 to as 1x Jabba. For TEM analysis, the core facility was given ten appropriately staged embryos per genotype 524 per experiment, and then chose which were imaged based on staining success. 525 Exclusion criteria for imaging embryos were predetermined. Embryos not of the stage of 526 interest, determined to have expired during preparation or image acquisition, or which were 527 imaged in the incorrect orientation/focal depth were excluded. 528 529 Periodic acid Schiff (PAS) and LD staining 530 Embryos were collected on apple juice plates for the desired time range and dechorionated 531 with 50% bleach and fixed for 20 min using a 1:1 mixture of heptane and 4% formaldehyde in 532 phosphate-buffered saline (PBS). To detect GGs, embryos were devitellinized using 533 heptane/methanol and subsequently washed three times in 1xPBS/0.1% Triton X-100. Embryos 534 were incubated first in 0.1M phosphatidic acid (pH 6) for 1hr and then in 0.15% periodic acid in 535 dH2O for 15min. After one wash with dH2O, embryos were incubated in Schiff's reagent (Sigma-536 Aldrich) until the embryos went from uncolored, to pink, to red (~2 minutes To determine the distribution of LDs and GGs in centrifuged embryos, in-vivo centrifugation was 559 performed as described 20 , followed by fixation and simultaneous LD/GG detection as above. For 560 analyzing follicles, ovaries were dissected from females maintained on yeast at 25°C overnight. 561 Samples were then fixed with 4% formaldehyde in PBS for 15 min, washed in 1xPBS/0.1%Triton 562 X-100, and simultaneously stained for LDs and GGs as above. 563 564

Live imaging 565
For live imaging involving dye injections, a previously published procedure was followed 14 . In 566 short, embryos were collected on apple juice plates for the desired time, hand-dechorionated, 567 transferred to a coverslip with heptane glue, desiccated, and placed in Halocarbon oil 700. 568 Embryos were then injected with BODIPY 493/503 (1mg/mL), LysoTracker Red (1mM) or 569 LipidSpot 610 (1000x) and imaged on a Leica Sp5 confocal microscope. 570 For live imaging of Glycogenin-YFP and YV autofluorescence, embryos were collected on apple 571 juice plates for the desired time, hand-dechorionated, transferred to a coverslip with heptane 572 glue, covered with Halocarbon oil 27, and imaged. To improve signal, flies were kept in the 573 dark, and light exposure during embryo preparation was kept to a minimum. 574 575 TEM 576 Embryos were collected from 7-to 14-day-old flies, dechorionated in 3% sodium hypochlorite, 577 and washed extensively with distilled water. Embryos were fixed in 578 4%paraformaldehyde/2%gluteradlehyde/PBS with an equal volume of heptane added. The vials 579 were shaken then left on an agitator for 20 minutes. After fixation, embryos were washed 580 extensively with 1×PBS/0.1% Triton X-100, then transferred onto a piece of double-sided tape, 581 adhered, then submerged with 1×PBS/0.1% Triton X-100. The embryos were then gently hand 582 rolled using fine forceps until the vitelline membrane was removed. Embryos were transferred 583 to a small glass vial. The embryos were then fixed a second time with 584 4%paraformaldehyde/2%gluteradlehyde/PBS, excluding the heptane, for 30 minutes. Embryos 585 were then washed three times with 0.2 M sucrose in 0.1M cacodylate buffer. They were 586 washed an additional 3 times in 0.1 M sodium cacodylate before post fixation in 1% osmium 587 tetroxide for 2 hours followed by uranyl acetate enhancement in 0.5% uranyl acetate overnight 588 at 4 °C. Specimen were washed and then dehydrated in a graded ethanol series, transitioned to 589 propylene oxide and embedded in an Epon/Araldite resin. Thin sections were stained with 590 0.3% lead citrate and imaged on a Hitachi 7650 transmission electron microscope using an 11 591 MP Gatan Erlanshen CCD camera. TEM work was conducted at the Electron and Cryo 592 Microscopy Resource in the Center for Advanced Research Technologies at the University of 593 Rochester. 594 595

Quantification of TEMs 596
To quantify the association of LDs and GGs, LDs were manually identified based on their 597 appearance and size (diameter of 0.3-0.75µm), while GGs were manually identified based on 598 the staining pattern and diameter (2-7µm). The two structures were labeled associated if the 599 distance between them was less than 30nm (~2pixels). 600 601 Particle Image Velocimetry 602 We performed PIV as described in 14 . The motion of acidic organelles and LDs was captured 603 simultaneously in the same embryos. Embryos were collected, staged, mounted on a coverslip, 604 and co-injected with BODIPY 493/503 and Lysotracker Red as described above. Per genotype, 605 three embryos were imaged at 25°C. Timeseries were captured by confocal microscopy at a 606 rate of 1 frame per 30 seconds, within a superficial plane of the embryo. The raw timeseries 607 were then analyzed, finding the 8 th nuclear division by first finding the division where nuclei 608 arrive at the periphery (the 9 th division), then going back one contraction cycle. 10 sequential 609 frames were taken from this division starting at the period of the highest lipid droplet motion 610 determined empirically. The signal from the embryos was then isolated from these 10 frames 611 using a mask. These processed frames were then fed into a PIV algorithm based on OpenPIV, a 612 python-based PIV implementation, generating 9 frame transitions per timeseries. The PIV 613 analysis script used is available as supplemental data from our previous publication 14 . The 614 output vectors for each transition were then processed to remove any vectors which failed to 615 meet or exceeded the empirically determined vector boundaries. Then directionality was 616 removed, the vectors were averaged across the transition, pixels were converted to microns, 617 and these averages were plotted on violin plots to show that pulses were being captured. 618 To perform PIV analysis for YVs, the same procedure was followed, with the exception that 619 embryos were not injected, and embryos were illuminated with a 405nm laser and yolk 620 autofluorescence was captured by collecting emission in a 410-500nm window. 621 622 Statistics 623 All statistics were done using Graphpad Prism. P values were calculated using 2-tailed, unpaired 624 Student's T-tests. At least 3 embryos were used per genotype. For the PIV statistical 625 comparisons 3 embryos were used per genotype contributing 9 transitions per embryo, thus n = 626 27, at least, per genotype. 627 The questions we sought to answer with statistical tests were "Are the Jabba LD:GG complexes 628 flowing slower then wild-type singular LDs?". Having received a positive result from this 629 question, we next asked "Is the flow generally slower in Jabba than wild-type or is the 630 diminished LD:GG speed due to complex formation?" for which we used acidic organelles and 631 YVs as unclustered controls. These are binary questions seeking to determine if two populations 632 of related numbers (flow speeds of organelles in each genotype) were different, thus we chose 633 to use Student's t-tests. 634