Translational regulation enhances distinction of cell types in the nervous system

Reviewer #1 (Public Review):

This study seeks to understand how selective mRNA translation informs cellular identity using the Drosophila brain as a model. Using drivers specific for either neurons or glia, the authors express a tagged large ribosomal subunit protein, which they then use as a handle for isolating total mRNA and ribosome footprints. Throughout the study, they compare these data sets to transcriptional and ribosome profiles from the whole fly head, which contains multiple cell types including fat tissue, pigment cells, and others, in addition to neurons and glia. Using GO term analyses, they demonstrate the specificity of their cell-type-based ribosome profiling: known glial mRNAs are efficiently translated in glia and likewise in neurons as well. In further examining their RNAseq data set, they find that "neuronal" mRNAs, such as ion channels, are expressed in both neurons and glia, but are translated at higher rates in neurons. Based on this, they hypothesize that neuronal mRNAs are actively suppressed in glia, and next seek to determine the underlying mechanism. By meta-analysis of all mapped ribosome footprints, they find that glia have higher ribosome occupancies in the 5' leader of neuronal mRNAs. This is corroborated by individual ribosome occupancy profiles for several neuronal mRNAs. In 5'leaders containing upstream AUG codons, they find that the glial data sets show enrichment of ribosomes at these upstream start sites. They thus conclude that 5' leaders containing upstream AUGs confer translational suppression in glia.

Overall, the sequencing data sets generated in this study and their subsequent bioinformatic analyses seem robust and reliable. Their data echo the trends of cell-type specific translational profiles seen in previous studies (e.g. 27380875, 30650354), and making their data sets and analyses accessible to the broader scientific community would be quite helpful. The findings are presented in a logical and methodical manner, and the data are depicted clearly. The authors' results that 5' leaders facilitate translation suppression is well-supported in literature. However, they overinterpret their data by claiming that such suppression is key for maintaining glial/neuronal identity (it is even featured in their title), but do not present any evidence that loss of such regulation has any impact on cellular identity. In many places, the authors do not acknowledge possible biases in their analytical methods, or consider alternate explanations for their data. These weaken the manuscript in its current form, but many of these issues which I describe below, are rectifiable with modest effort.

1. The authors' data in Fig. 2-S1A-B shows substantial cell-to-cell variation in RpL3::FLAG expression. The authors do not consider that this variation may cause certain neuronal/glial types to be overrepresented in their datasets. A related point is that the authors do not discuss whether RpL3::FLAG is only present in the cell body or if it is also trafficked to the neuronal/glial processes where localized translation is known to occur (reviewed in 31270476).

2. The RNA-seq data set that they use to calculate translation efficiency (TE) only represents mRNAs associated with RpL3::FLAG, which is part of the large ribosome subunit. As the authors are likely aware, there are mRNAs on which the full ribosome moiety does not assemble and these are effectively excluded from this data set. Ideally, a more complete picture of the mRNA landscape can be obtained by 40S subunit profiling but I appreciate that this is technically very challenging. At a minimum, this caveat needs to be acknowledged.

How does the TPM of differentially regulated transcripts (such as those in Fig. 2H) compare between whole heads, neurons, and glia? Since the whole head RNA-seq data was not from an enriched sample, this might serve as a decent proxy for showing that the neuron/glia RNA-seq data sets are representative of RNA abundance.

3. The analysis in Fig. 2F shows that low abundance mRNAs in glia are further translationally suppressed, which the authors point out in lines 151-152. However, this data also shows that mRNAs with a 1:1 ratio in neuron:glia (which fall in the 0.5-1 and 1-2 bin) have a TE1; this suggests that on average, mRNAs that are equally abundant are translated equally efficiently. This is the opposite of the thesis presented in Fig. 2G-H where many mRNAs of equal abundance in neurons and glia are actually poorly translated in glia. How do the authors reconcile these observations?
It is also unclear from the manuscript whether all mRNAs were considered for the analysis in Fig. 2F or if some cutoff was employed.

4. Throughout the manuscript the authors favor a "translation suppression" model wherein glia (for example) actively suppress neuronal mRNAs, and this is substantiated in Fig. 3C showing higher ribosome occupancy on 5' leaders than in coding regions. However, they show no evidence that glial mRNAs (such as those indicated in Fig. 2B and 2-S2B) present a different pattern, say that of higher ribosome occupancy in CDS vs. 5' leaders. This type of positive control is a glaring omission from many of their analyses, including ribosome occupancy at upstream AUG codons (Fig. 4).
In order to make a broad case (as they do in the title) that differential translation regulation specifies multiple cell types, it is necessary to show the corollary: that glial mRNAs (repo, bnb, pnt, etc) are suppressed in neurons. There is an inkling of this evidence in Fig. 3-S1 where fat body mRNAs in neurons are shown to have low ribosome occupancy in the CDS regions and enhanced occupancy in the 5' leader region. This data is not quantified, nor is a control neuron mRNA shown as a reference for what the ribosome occupancy profile of an actively translated mRNA looks like in a neuron.

5. The cell-type specific ribosome profiling data sets in the manuscript are from mRNAs associated with 80s subunits that have been treated with cycloheximide during sample preparation. Cycloheximide, and many other translation inhibitors, are known to non-uniformly bias reads towards start codons (PMID: 22056041,22927429). This important caveat and its implications on the start-codon occupancy analysis in Fig. 4 are not acknowledged in the manuscript.
Again, the ideal resolution would be a ribosome profiling data set from 40S footprinting or harringtonine-treated samples (PMIDs: 32589966, 27487212, 32589964) to show the true accumulation of ribosomes at AUG codons. In the absence of such a data set, a comparative meta-analysis of the ribosome distribution around upstream and initiation AUG codons of differentially translated transcripts from neurons would be a useful control.

6. The authors chose Rhodopsin 1 (Rh1) as a model mRNA which is translated efficiently in neurons but suppressed in glia. Though the data in Fig. 2-S3B shows higher TE for Rh1 in neurons, the data in 5A show lower ribosome occupancy in the Rh1 CDS in neuron samples (at least in the fragment of the CDS visible). These data are somewhat contradictory.
Further, given that the neuron data are from all nsyb-positive cells but that Rh1 is expressed only in R1-R6 photoreceptors, it is unclear what motivated them to choose Rh1 as opposed to an mRNA that is more broadly expressed in neurons.

7. Similar to the heterogeneity in nsyb- and repo-GAL4 expression in Fig. 2-S1A-B, Fig. 5C shows substantial variation in the expression of the UAS-GFP reporter driven by tub-GAL4. This variable GAL4 activity makes the mRNA abundance data difficult to interpret. Also, since the authors presume that Rh1 mRNA is expressed in glia (it is not annotated in the RNA-seq analysis in Fig. 2-S2B), would Rh1-GAL4 not be a more apt driver?
These issues are further compounded by the lack of a cellular compartment marker (repo marks glial nuclei) which makes it impossible to determine which cell the mRNA signal is in. There are also no negative controls presented for the mRNA probes.
Most confoundingly though, the control reporter itself seems to show variable translation efficiencies from one cell to another, with high-GFP protein cells showing lower GFP mRNA and vice versa.
The mRNA:protein ratio may be easier to examine by using repo-GAL4 to specifically drive the Rh1-reporter expression in glia (such as in Fig. 5-S1A) rather than simultaneous expression in both neurons and glia using tub-GAL4.

Reviewer #2 (Public Review):

Summary:
The authors used next-generation sequencing approaches combined with ribosome trapping to investigate gene expression in neurons and glia in the heads of adult fruit flies. Ribosome footprinting was further used to investigate the translational efficiency (TE) of particular RNAs in these two tissues. The evidence convincingly demonstrated that translation of specific messages is repressed in glia while others are repressed in neurons. Further evidence suggests that cis-acting elements within the 5'UTR of neuronal transcripts cause the repression of translation in glia. For instance, a fluorescent reporter using the 5'UTR of Rhodopsin-1 is highly translated in neurons but fluorescence from this reporter is nearly undetectable in glia. Furthermore, pausing of ribosomes on start codons of upstream Open Reading Frames (uORFs) is seen on the 5'UTR of this and other messages in glia but not in neurons.

Strengths:
The main strength of the manuscript is its use of cutting-edge next-generation sequencing and bioinformatic approaches to investigate the tissue-specific translatome of Drosophila.

Weaknesses:
A minor weakness is that little insight is provided into the mechanism that leads to ribosome stalling on uORFs in glia but not in neurons. The manuscript could be improved by some discussion on potential pathways that might control the differential TE through uORF pausing.

Reviewer #3 (Public Review):

It is well established that there is extensive post-transcriptional gene regulation in nervous systems, including the fly brain. For example, dynamic regulation of hundreds of genes during photoreceptor development could only be observed at the level of translated mRNAs, but not the entire transcriptomes. The present study instead addresses the role of differential translational regulation between cell types (or rather classes: neurons and glia, as both are still highly heterogenous groups) in the adult fly brain. By performing bulk RNA-seq and Ribo-seq on the same lysates, the authors are able to compare the translation efficiency (TE) of all transcripts between neurons and glia. Many genes display differential TE, but interestingly, they tend to be the genes that already show strong differences at their mRNA level. The most striking observation is the finding that neuronal transcripts in glia display increased ribosome stalling at their 5' UTR, and in particular at the start codons of short "upstream ORFs". This could suggest that glia specifically employ a mechanism to upregulate upstream ORF translation, enabling them to better suppress the expression of the genes that have them. And neuronal genes tend to have longer 5' UTRs, perhaps to facilitate this type of regulation.

However, it is difficult to evaluate the functional significance of these differences because the authors provide only one follow-up experiment to their RNA-seq analysis. Venus expressed with the Rh1 UTR sequences may be displaying differential levels between glia and neurons, but I find this image (Fig. 5C) rather unconvincing to support that conclusion. There are no quantifications of colocalization or even sample size information provided for this experiment. And if there is indeed a difference, it would still be difficult to argue this is because of the 5' stalling phenomenon authors observe with Rh1, because they switched both the 5' and 3' UTRs.

I also find it puzzling that the TE differences between the groups are mostly among the transcripts that are already strongly differentially expressed at the transcriptional level. The authors would like to frame this as a mechanism of 'contrast sharpening'; but it is unclear why that would be needed. Rh1, for instance, is not just differentially expressed between neurons and glia, but it is actually only expressed by a very specific neuronal type (photoreceptors). Thus it's not clear to me why the glia would need this 5' stalling mechanism to fully suppress Rh1 expression, while all the other neurons can apparently do so without it.