To gain an overview of the translation status, we first applied conventional Ribo-seq in the whole fly head,and successfully monitored footprint distribution at a single-codon resolution (Figure 1A, see ‘Materials and methods’ for technical details). The majority (96.2%) of ribosome footprints was mapped onto the annotated coding sequences (CDS), and its distribution displayed a clear 3-nt periodicity, reflecting the codon-wise movement (Figure 1B).

To compare transcriptome and translatome, we also performed RNA-seq from the same lysate (Figure 1A). As previously reported, the transcript level and the number of ribosome footprints did not always match, suggesting substantial posttranscriptional regulations (R² = 0.664; Figure 1C). For instance, while Shaker (Sh) and Trehalase (Treh), which encode a voltage-gated K⁺ channel and an enzyme that hydrolyzes trehalose, respectively, were similar regarding transcript levels, far more ribosome footprints were detected on Treh (Figure 1D and E). We therefore measured TE, ribosome footprints normalized by mRNA reads. TE was much higher for Treh than Sh (Figure 1F), and we found a striking genome-wide variability with more than 20-fold TE difference between the 5 and 95 percentiles (Figure 1G). Kyoto Encyclopedia of Genes and Genome (KEGG) pathway enrichment analysis revealed that the transcripts involved in fatty acid metabolism and proteasome are actively translated (Figure 1H). In contrast, ribosome proteins, as previously reported (Chen and Dickman, 2017; Cho et al., 2015), and proteins mediating neuronal ligand–receptor interactions were significantly enriched in the transcripts with low TE, suggesting translational suppression (Figure 1H). Indeed, many transcripts encoding ligand- or voltage-gated ion channels, G-protein coupled receptors (GPCR) showed remarkably low TE (Figure 1C and I). These results suggest translational regulations specific to neuronal transcripts in the fly head.

Because the translatome/transcriptome status of the whole heads was a mixed average of diverse cell types, such as neurons, glial cells, fat bodies, and muscles, we set up an experimental approach to dissect cell-type-specific translational regulations. By expressing epitope-tagged RpL3 (uL3 in universal nomenclature) (Chen and Dickman, 2017) under the control of UAS using the nSyb- or the repo-GAL4 drivers, we immunopurified the tagged ribosomes and associated mRNAs separately from neurons and glia, and performed Ribo-seq (Figure 2A). By immunohistochemistry, we confirmed that UAS-RpL3::FLAG on the third chromosome exhibited minimum leakage expression in the brain and did not display any apparent morphological defects upon expression using either driver compared to other insertions or constructs (Chen and Dickman, 2017; Huang et al., 2019; Thomas et al., 2012; Figure 2A, Figure 2—figure supplement 1A and B). The exogenously expressed RpL3::FLAG was highly concentrated in cell bodies but also detectable in neurites, consistent with the subcellular localization of the endogenous ribosome (Figure 2—figure supplement 1C and D).

Through the purification of FLAG-tagged ribosomes, we successfully profiled translatome from neurons and glial cells in the fly heads: footprints were found on 10,821 (78.4% of all the annotated genes) and 10,994 (79.7%) genes in neurons and glia, respectively, with decent reproducibility among the biological replicates (R² > 0.9, Figure 2—figure supplement 2A). The FLAG-tagged RpL3 in the corresponding cells far exceeded the endogenous RpL3, as RpL3 reads were 7.8 and 42.7 times higher in neurons and glia, respectively, compared to the wild-type whole-head samples (Figure 2—figure supplement 2B). The known marker genes were strongly enriched while non-target markers were depleted (Figure 2B, Figure 2—figure supplement 2C and D; Croset et al., 2018; Davie et al., 2018; Li et al., 2022), and the KEGG enrichment analysis showed significant enrichment of footprints on genes associated with the known functions of these cell types (Figure 2—figure supplement 2E). Interestingly, the KEGG analysis also revealed that neurons exhibit a greater extent of protein synthesis related to oxidative phosphorylation and mitochondrial ribosome proteins, while glial cells show higher expression of proteins associated with glycolysis (Figure 2—figure supplement 2E and F). These findings support the glia-neuron lactate shuttle hypothesis, a recently proposed concept of metabolic specialization (Mason, 2017; Volkenhoff et al., 2015). Furthermore, apart from the annotated CDS, we detected clustered ribosome footprints on Hsr-ω, previously annotated as a long non-coding RNA, strongly suggesting the synthesis of hitherto undescribed polypeptides (Figure 2—figure supplement 2G; Singh, 2022). Altogether, the combination of genetic labeling of ribosomes in selective cell types and Ribo-seq revealed the differential translatome profiles in the fly heads.

To further examine translational regulation by calculating TE, we performed RNA-seq from the same immunoprecipitated complexes, similar to Translating Ribosome Affinity Purification (TRAP) (Heiman et al., 2008; Figure 2A, Figure 2—figure supplement 3A and B). Because this approach relies on the 80S-ribosome-mRNA complex, we may miss mRNA with little or no translation. Nevertheless, our transcriptome was similar to the sn-transcriptome data (Li et al., 2022; Figure 2—figure supplement 3C). We identified groups of genes undergoing neuron- or glia-specific translational regulations compared to the whole heads (Figure 2—figure supplement 4A). Genes mediating fatty acid metabolism and degradation, for example, were actively translated in the whole head, but showed lower TE in neurons or in glia (Figures 1H and 2C). Because many of these genes are highly expressed in the fat bodies (Dobson et al., 2018), these results suggest selective translational enhancement in the fat body. Strikingly, TE of genes involved in neuroactive ligand–receptor interaction was significantly higher in neurons but lower in glia (Figure 2C and D), suggesting cell-type-specific translational regulation of these genes.

This differential translational regulation was highlighted in the weak TE correlation between neurons and glia (R² = 0.534, Figure 2E). We found a genome-wide tendency that genes transcribed less in glia are further suppressed at translation (Figure 2F). Specifically, many functionally characterized neuronal genes, such as voltage- or ligand-gated ion channels, G-protein-coupled receptors, neuropeptides, and proteins for visual perception, showed particularly lower TE in glia (Figure 2E, G, and H. Figure 2—figure supplement 4C). For these genes, the distinction between neuronal and glial cells was much exaggerated at the level of translation than at transcription (Figure 2H). Consistently on the genome-wide scale, the inter-cell-type correlation became weaker in Ribo-seq data compared to in RNA-seq (R² = 0.59 vs. 0.81, Figure 2—figure supplement 3B). Altogether, these data indicate substantial contributions of translational regulation to shaping the cell-type-specific protein expression.

We next analyzed the distribution of ribosome footprints on the differentially translated transcripts (DTT). Fat-body-related genes showed lower TE in neurons compared to the whole head (Figure 2C). Among these genes, we found a remarkable ribosomal accumulation on the start codon specifically in neurons (Figure 3—figure supplement 1A and B), as if the first round of the elongation cycle was arrested in neurons. Through the reanalysis of the published RNA-seq data (Dobson et al., 2018), we found that mRNAs showing strong ribosomal accumulation on the start codons are highly abundant in the fat bodies (Figure 3—figure supplement 1C). On the other hand, DTTs suppressed in glial cells compared to neurons (defined as genes with more than 10 times higher TE in neurons than in glia, n = 161), we noticed that glial ribosome footprints were remarkably biased toward 5′ leaders (Figure 3A and B). Notably, this pattern was not obvious on the genome-wide scale (Figure 3A and B). The high 5′ leader/CDS ratio of ribosome footprints in glia was commonly observed on many transcripts with known neuronal functions, such as Rab3, Syt4, Arr1, and Syn (Figure 3D and E). Conversely, we observed accumulated ribosome footprints on the 5′ leaders of several glial marker genes specifically in neurons (Figure 3—figure supplement 2). Altogether, these results suggest that the translation of 5’ leaders in selective mRNAs differentiates protein synthesis among cell types.

We reasoned translational downregulation via upstream ORFs (uORFs) in the 5′ leaders in glia, as the translation of uORFs was reported to suppress that of the downstream main ORF (Ferreira et al., 2013; Zhang et al., 2019; Zhang et al., 2018). Consistent with this idea, metagene plot around the AUG codons on 5′ leaders revealed strong accumulation of footprints on the upstream AUG codons, similar to those observed on the initiation codon of CDSs (Figure 4A and B). We calculated the footprint accumulation score on each codon (defined as the ratio of footprints on each codon with surrounding –50/+50 nt), and found that upstream AUG and the near cognate codons (NUG or AUN) showed relatively high accumulation (Figure 4C). On the other hand, inside the annotated CDS, none of the codons exhibited such significant accumulation (Figure 4D). Consistently, we found that transcripts related to neuronal functions typically contain long 5′ leaders and many upstream AUG (Figure 4—figure supplement 1). We thus propose that glial cells suppress the translation of neuronal transcripts by stalling ribosomes on 5′ leader via uORF.

We next asked whether the 5′ leader sequences of neuronal genes cause cell-type differences in translation. To this end, we focused on Rh1 (Rhodopsin 1, also known as ninaE), which encodes an opsin, also detecting stimuli of other sensory modalities (Leung et al., 2020; O’Tousa et al., 1985; Shen et al., 2011; Zuker et al., 1985). Consistently, active translation of Rh1 was almost exclusively observed in neurons (Figure 5A). Similar to other neuronal genes shown in Figure 3C, the distribution of ribosome footprints was distinct among neuronal and glial cells: they were heavily biased to 5′ leader in glia, with the striking accumulation on the putative uORFs composed only of the start and stop codons (Figure 5B and C).

To address the function of these sequences on differential translation, we constructed a transgenic reporter strain using the Rh1 UTR sequences under the control of UAS (Figure 5D), and directed gene expression ubiquitously using Tub-GAL4. While the reporter mRNA was detected both in neuronal and glial cells, the protein levels were much more heterogeneous and strikingly weak in glia (Figure 5E, Figure 5—figure supplement 1A). The control reporter strain (DeLuca and Spradling, 2018), on the other hand, exhibited more ubiquitous expression, with significantly higher fluorescent intensity in glia (Figure 5E and F). Driving the reporter expression using the nSyb- or repo-GAL4 further corroborated cell-type-specific suppression in glia (Figure 5—figure supplement 1B and C). Strikingly, when the six-base putative uORFs were mutated, the in vivo protein-to-mRNA ratio of the reporter was significantly increased in glia but not in neurons (Figure 5G and H). Based on these results, we propose that glial cells selectively suppress the protein synthesis of neuronal genes through uORF and thereby enhance the translatome distinction from neurons.

Because our cell-type-specific Ribo-seq and RNA-seq are based on immunoprecipitation of genetically tagged RpL3 (Chen and Dickman, 2017), the read counts could contain biases, such as underestimation of mRNA level with little or no translational activity, or over- and under-representation of certain cell types originating from the heterogeneous expression of the drivers.

The flies were reared in a mass culture at 24°C under the 12–12 hr light-dark cycles on the standard cornmeal food. The Canton-S strain was used as the wild-type. We utilized the following transgenic strains: w¹¹¹⁸;;GMR57C10-GAL4 (nSyb-GAL4; BDSC #39171), w¹¹¹⁸;;repo-GAL4 (BDSC #7415), y¹w¹¹¹⁸;;tublin-GAL4 (BDSC #5138), w¹¹¹⁸;MB010B (BDSC #68293), w¹¹¹⁸;;UAS-RpL3::FLAG (BDSC #77132) (Chen and Dickman, 2017), y¹v¹;UAS-Rh1-Venus (made in this study; see below), y¹v¹;UAS-m-Rh1-Venus (made in this study), w;UASz-GFP (a kind gift from Dr. Steven DeLuca) (DeLuca and Spradling, 2018). Females of the GAL4 drivers were crossed to males of the UAS effectors, and the F1 progenies were used for the experiments. Of note, although UAS-EGFP::RpL10Ab (Thomas et al., 2012) has been used to isolate ribosomes from specific cells, its expression using the repo-GAL4 caused lethality in our hands.

The crude (Figure 1) or the immunoprecipitated (Figure 2) lysate was prepared using the same protocol as described above, but without RNase digestion. RNA was purified using TRIzol-LS. The libraries were constructed in Azenta Japan Corporation, using the NEBNext Poly(A) mRNA Magnetic Isolation Module (E7760, New England Biolabs) and MGIEasy RNA Directional Library Prep kit (1000006386, MGI tech). Briefly, poly-A tailed mRNAs were enriched with the oligo dT beads, fragmented, and reverse-transcribed using random primers. After the second strand cDNA was synthesized, an adapter sequence was added. DNA library was PCR-amplified. The libraries were sequenced with DNB-seq (MGI tech) with an option of paired end reads for 150 bases.

Adaptor sequences were removed using Fastp (Chen et al., 2018), and the reads that matched to the non-coding RNA were discarded. The remaining reads were mapped onto the Drosophila melanogaster release 6 genome. Mapping was performed using STAR (Dobin et al., 2013). PCR-duplicated reads were removed by referring to the unique molecular identifiers. The number of uniquely mapped reads are as follows:

DNA fragments containing a minimal hsp70 promoter (hsp70Bb) (DeLuca and Spradling, 2018), the 5′ leader or the mutated 5′ leader of Rh1-RA, the first 15 bases of the CDS of Rh1-RA, the Venus yellow fluorescent protein gene, and the 3′ UTR of Rh1-RA were synthesized and cloned into the pBFv-UAS3 plasmid (Addgene #138399). The sequence of the resultant plasmid is provided in the Supplementary file 3. The plasmid was then injected into y¹ v¹ P{nos-phiC31}; P{CaryP}attP40, and their progenies were screened for a v+phenotype. A single transformant was crossed to y¹ cho² v¹; Sp/CyO balancer to establish a transgenic line.

Immunohistochemistry (Figure 2A, Figure 2—figure supplement 1) was performed as previously described with minor modifications (Kanno et al., 2021). Briefly, dissected male fly brains were fixed in 2% paraformaldehyde in PBS for 1 hr at room temperature, washed three times with PBST (0.1% Triton X-100 in PBS), blocked with 3% goat serum in PBST for 30 min, then incubated with the primary antibody solution at 4°C overnight (mouse anti-FLAG (1:1000; Sigma-Aldrich; F1804), mouse anti-RpS6 (1:200; Cell Signaling; 54D2), and rat anti-elav (1:20; DSHB; 7E8A10)). Subsequently, the brains were washed three times with PBST, incubated with the secondary antibody solution at 4°C overnight (anti-mouse Alexa Fluor 488 (1:400; Invitrogen; A11001), anti-mouse Cy3 (1:1000; Jackson ImmunoResearch; 115-166-003), and anti-rat Cy3 (1:200; Jackson ImmunoResearch; 112-166-003)), washed three times with PBST, and mounted with 86% glycerol in PBS.

Fluorescent in situ hybridization, combined with immunohistochemistry, was performed in a similar manner to Yang et al., 2017 with several modifications (Figure 5E–H). Dissected male fly brains were fixed in PBS containing 3% formaldehyde, 1% glyoxal, and 0.1% methanol for 30 min at room temperature, followed by three quick washes with PBT (0.5% Triton X-100 in PBS). Consistent with the previous study, addition of glyoxal to the fixative improved the FISH signal (Yao et al., 2021). The buffer was then exchanged to the wash solution (10% Hi-Di Formamide [Thermo Fisher Scientific; 4311320] in 2× saline sodium citrate) and was incubated at 37°C for 5 min. Subsequently, the brains were incubated with the custom-made Stellaris Venus or GFP probes (100 nM; see Supplementary file 3 for the sequences; LGC BioSearch Technologies) and the primary antibody (mouse anti-Repo [1:100; DSHB; 8D12]) in the hybridization buffer (10% Hi-Di Formamide in hybridization buffer [Stellaris RNA FISH Hybridization Buffer, SMF-HB1-10]) at 37°C for 16 hr. The probes and the antibody were then removed by washing the samples quickly three times with preheated wash solution at 37°C, followed by three washes for 10 min at room temperature. Blocking was performed with 3% normal goat serum in PBT for 30 min at room temperature. The secondary antibody (Cy3 goat anti-mouse [1:2000; Jackson ImmunoResearch; 115-166-003]) was then added and was incubated at 4°C overnight. The samples were washed once quickly, three times for 20 min and once for 60 min with PBT, and then mounted in 86% glycerol in 1× Tris–HCl buffer (pH 7.4).

Tissues to detect native GFP or Venus signals (Figure 5—figure supplement 1) were prepared as follows: dissected brains were fixed in PBS containing 3% formaldehyde, 1% glyoxal, and 0.1% methanol for 30 min at room temperature, followed by one quick wash and three washes for 10 min with PBT. The samples were then mounted in 86% glycerol in 1× Tris–HCl buffer (pH7.4).

Imaging was done on the Olympus FV1200 confocal microscope with GaAsP sensors. A ×100/1.35 silicone immersion objective (UPLSAPO100XS, Olympus) or ×30/1.05 silicone immersion objective (UPLSAPO30XS) was used. Scan settings were kept constant across specimens to be compared.