Identifying molecular instructions to hard-wire a sensory neuron’s synaptic connectivity

Reviewer #1 (Public review):

Summary:

The authors sequence the transcriptome of three sensory neurons from D. melanogaster to study the cell-cell and animal-animal variability in these cells, with a focus on cell adhesion molecules. The work reports useful cell-specific transcriptomics datasets that will be of great interest to those studying cell types, transcriptomes, neuronal development, and cell surface proteomes. The authors also report large numbers of knockdown data (gene-by-gene or in combinations) and report neuronal wiring and behavioral phenotypes. The manuscript is highly descriptive of the system studied - in a good way, but often over-speculates in rationale or conclusions.

Strengths:

The manuscript is data-rich. The single-cell transcriptomics datasets, not trivial to collect, are a major strength of the work and will prove useful to the field. Also, the biased expression of Dscam is interesting, even though the authors cannot pursue the mechanism or a function for this.

Weaknesses:

The study lacks depth (i.e., mechanism) in explaining observations.

Reviewer #2 (Public review):

Summary:

In this manuscript, dos Santos et al seek to identify cell-specific programs that drive neuronal wiring patterns. They focus on two chemosensory and mechanosensory neurons in the Drosophila nervous system, as they both display stereotyped connectivity in the ventral nerve cord. Single-neuron RNA sequencing identified cell surface molecules that distinguish the sensory neurons and may instruct their respective wiring patterns. They functionally test several of these candidates and observe miswiring phenotypes upon knockdown experiments. Additionally, they attempt to miswire the chemosensory neurons. Overall, this manuscript addresses an important question about how neurons identify appropriate synaptic partners through precise cell surface molecular codes. However, there are significant deficiencies in the experimental logic and rigor, and the manuscript can be very difficult to digest.

Strengths:

The use of two sensory neurons with stereotyped connectivity is a significant strength, as this enables the authors to identify genes that are required for wiring. Additionally, analyzing the transcriptomes of single neurons repeatedly could potentially be a robust approach to identifying cell-specific cell-surface molecules that drive wiring.

Weaknesses:

(1) The authors perform RNAseq for single identifiable neurons, as opposed to neuronal subclasses, which has been reported before. It would be beneficial to elaborate on the significance of using single neurons for answering the scientific question. This is briefly mentioned toward the end of one of the results subsections: "Repeated RNA sequencing of an identifiable neuron seeks to address the fundamental nature of variability in connectomics, axonal branching, and cellular identity." But this should be in the Introduction.

(2) The authors chose the P14 pupal stage for one of the analyses. It is not clear why this specific stage is chosen. Does pSc and aPa connectivity occur at this stage?

(3) This reviewer is confused as to why looking at differentially expressed CSMs between pupal and adult stages of two different neurons is useful. This does not seem like an appropriate comparison. This data might be better in the supplemental material, especially given the lack of precise age synchronization across pupal samples (as reported).

(4) It is very difficult to follow the logic because the manuscript seems to jump around between different results and lacks a compelling through line.

(5) "Single cell sequencing of the same neuron reveals transcriptome precision": What are the controls here? An aPa neuron is shown in Figure 3 as an example of a different neuronal subtype, but were other factors (e.g., lack of Repo expression) checked to ensure that samples were not contaminated?

(6) "However, whether any of these exon 6 or 9 splicing specificities are biologically significant can only be determined using exon 6 and 9 isoform-specific RNAi." The authors could alternatively use CRISPR techniques to target specific isoforms that they hypothesize might be important for neural wiring, enabling them to assess isoform-specific wiring defects.

(7) In the section "The set of cell surface receptors required to wire up the pSc mechanosensory neuron": Several previous subsections of the Results use RNAseq to identify molecules expressed in pSc neurons across different stages. It's unclear why the authors did not start with the identified list of candidate cell surface receptors identified in their RNAseq experiments.

a. Were any of the genes screened the same as those identified by the authors as differentially expressed in pSc mechanosensory neurons, either across developmental stage (pupa vs. adult) or across neuronal subtype (pSc vs. Gr59d)? If so, it would be helpful to state this here. (They do mention later on that five CSMs identified were more highly expressed in pSc than aPa. However, changes in expression across developmental stages within the pSc neuron would still be helpful to comment on, especially since the authors identified greater transcriptomic differences across developmental stages than they did between different neuronal subtypes.)

b. The 39 genes not expressed in pSc neurons served as their negative control, but the average axonal targeting grade was 2.3 (between moderate and severe). This calls into question the use of this method as an appropriate measure of whether a gene expressed by pSc neurons is truly required for proper axon targeting; there seems to be a strong probability of significant off-target effects. Performing a global knockdown and cell-specific rescue could potentially complement these experiments and serve as a stronger indicator of candidate receptors' roles in pSc-specific axon targeting.

(8) It seems as though the purpose of the experiments described in the last results subsection ("Re-wiring the Gr59 chemosensory neuron") is to redirect the Gr59d neuron toward the pSc neuron's axonal targeting phenotype. However, the authors do not state whether they were able to do so effectively (i.e., whether or not there were significant differences between the rewired Gr59d neuron and the pSc neuron). This leaves the story unfinished.

(9) At the end of the discussion, the authors state that "...if a Gr59d chemosensory neuron is functionally rewired to a pSc mechanosensory circuit, activation of the Gr59d neuron using a bitter tastant molecule should elicit a grooming (mechanosensory) response...". The authors should attempt this experiment, especially given that they have developed the PXGS technique.

Author response:

We are pleased that the reviewers found the repeated single-neuron sequencing and the finding of less than 1% transcriptomic variability to be original and striking, valued the single-neuron Dscam isoform repertoires and the scale of the functional screen, and judged the evidence solid to compelling. We provide below our provisional response and an outline of the revisions we plan.

Overall plan: We intend to submit a revised version that addresses the public reviews and the recommendations to the authors. Because our conclusions rest on data already in the manuscript, the revisions are clarifications, added analysis of existing data, tempered language, and improved figures, rather than new experiments. Given the focused nature of these revisions, we would be happy for the editors to assess the revised version without re-involving the reviewers.

One factual note for the Assessment and public reviews: The morphological RNAi screen comprised 213 cell-surface receptor genes; the figure of “140 genes” in one public review is the number that produced strong-to-severe phenotypes (Grade 3–5 at >40% penetrance), not the number screened. We will make this unambiguous in the revised text.

Main changes in the revision:

(1) We will explain that the RNAi screen was performed blind and independent of the RNA sequencing experiments. This was intentional, so that functional perturbation and transcriptomic identity would serve as independent lines of evidence, but could be compared with each other.

(2) We will revise the Methods and Results to clarify how the morphological RNAi screen and behavioral subset should be interpreted, including conservative treatment of the negative-control distribution and mild-to-moderate phenotypes.

(3) We will soften language that overstated certainty. Differentially expressed molecules are now described as prioritized candidates and convergent evidence, not definitive determinants.

(4) We will reframe Gr59d/PXGS experiments as morphological rewiring and ectopic branching, and no longer imply a pSc-like conversion or functional rewiring.

(5) We will add a limitations paragraph addressing RNAi off-target/background concerns, the absence of direct aPa functional testing, and the need for future mechanistic validation.

(6) We will disclose or remove any figure panels that overlap with the companion PXGS manuscript and revise legends/labels to make it more clear.

We hope these revisions make the logic of the study clearer and align the strength of the claims with the evidence.

Below is our more detailed (provisional) response (not sure if this is required at this stage):

Response to the eLife Assessment:

Clearer articulation of the experimental logic. The Assessment’s central request (Reviewer 2) concerns the relationship between the transcriptomic experiments and the functional screen. The two were performed independently on purpose; the RNAi screen was assembled from a comprehensive survey of the literature rather than from the results of our differentially expressed genes from single-cell RNA sequencing. Thus, the RNAi screen was performed and graded blind in parallel with the single cell sequencing, with the gene identities unmasked only after both were complete. This was intentional, so that the sequencing (i.e., which molecules differ between neurons) and the screen (which molecules are functionally required) would provide mutually unbiased corroborating evidence rather than self-referential support/circular reasoning. We will state this more explicitly in the Introduction, in the Results where the screen is introduced, and in the Methods.

Additional controls in the RNAi screen. We will treat the 39 genes that were identified in our single-cell RNA sequencing to be not expressed in the pSc neuron as a randomized negative control set in our RNAi experiments. We will state more explicitly the empirical RNAi false positive rate for a miswiring phenotype is 6/39 = 15%, likely due to RNAi off-target effects. We will also more clearly state that our claims about cell surface receptor functions are restricted to strong-to-severe phenotypes at high penetrance reproduced by at least two independent RNAi lines and corroborated independently (differential expression and/or single neuron qPCR).

A more complete characterization of the re-wiring. We will state more clearly that mis-expressing the pSc-enriched cell surface receptors within Gr59d neurons partially shifts the arbour toward a pSc-like pattern (e.g., increased ectopic branching), and does not reproduce the full anatomical wiring, and that functional/behavioral re-wiring was not tested.

Response to Reviewer 1:

Reviewer 1 found the work valuable and data-rich, and the Dscam expression bias interesting. They noted over-confident language and asked how rigorously the differentially expressed genes were identified.

Over-confident language. We will rewrite the two flagged sentences. The claim that the ~10 differentially expressed molecules are “likely the most important” will become a correlational statement, while also noting the lack of an aPa-specific Gal4 driver for direct testing. Our sentence that, “Our RNA sequencing data is biologically inadequate without a functional characterization of each molecule within the specific neuron” will be replaced with a clearer statement that gene expression data can nominate candidates, and functional perturbation of each gene is required to demonstrate necessity and sufficiency (i.e., biological function); which is exactly why we paired the RNA sequencing with an independent RNAi screen.

Rigor of the differential-expression calls. We will more clearly state the statistical criteria in the Results (absolute log2 fold change ≥ 2 and Benjamini–Hochberg-adjusted p < 0.05). We will also note the small replicate numbers for the pooled pSc versus aPa comparisons, and emphasize that the central gene calls are independently supported by the blind RNAi screen and, for five genes, by single neuron qPCR. The full statistical workflow is in the Methods.

Response to Reviewer 2:

Reviewer 2 considered the findings potentially important but raised concerns about the experimental logic, the rigour of the screen, the completeness of the re-wiring, figure quality, and overlap with our PXGS companion paper. We will address each.

Experimental logic. Beyond the design of our independent, blinded RNAi screen described above, we will add to the Introduction the rationale for sequencing single identified neurons (rather than subclasses) along with the two-pronged strategy, and add a summary paragraph at the start of the Discussion.

Developmental stage choices. We will clarify our justification for the P14 pupal stage (the period when the mechanosensory neuron is actively elaborating its arbour while also enabling dissection). We will also clarify the rationale and caveats for comparing the pupal pSc neuron with the adult Gr59d neuron (i.e., the wiring occurs at the pupal stage, but the pupal Gr59d neurons could not be isolated at sufficient quality; the pSc pupal samples are less age-synchronized, so we simply used the comparison to identify the genes shared with the adult comparison).

Transcriptome precision controls. We will state that the ten single pSc neurons passed the same quality controls for neuronal markers (elav, nSyb) and glial markers (Repo, moody < 20 reads) as all single-neuron libraries, which argues against any contamination by the attendant glial cell, and the aPa transcriptome is used as a different identity comparison.

Off-target rate. As stated above, we will add the false positive rate for RNAi and restrict our confidence claims to those genes/cell surface receptors with multiple lines of evidence (e.g., strong phenotype, multiple RNAi lines, RNA sequencing, etc).

Rewiring completeness and the behavioral prediction. As stated above, we will clarify that true re-wiring of the Gr59d neuron requires a future experiment, where a bitter tastant stimulus would elicit a grooming response.

Response to Reviewer 3:

We thank Reviewer 3 for judging our work to be fundamental in significance and the evidence compelling, with no major criticisms. Our clarifications above will further reinforce our hypothesis that the differential expression of specific cell surface receptors “do, in fact, control synaptic patterns,” which the reviewer highlighted.

We are grateful for the reviewers’ time and for eLife’s model. We believe the planned revisions substantially clarify the experimental logic and tighten the claims, and we look forward to submitting the revised version.