In situ and transcriptomic identification of microglia in synapse-rich regions of the developing zebrafish brain

Microglia are brain resident macrophages that play vital roles in central nervous system (CNS) development, homeostasis, and pathology. Microglia both remodel synapses and engulf apoptotic cell corpses during development, but whether unique molecular programs regulate these distinct phagocytic functions is unknown. Here we identify a molecularly distinct microglial subset in the synapse rich regions of the zebrafish (Danio rerio) brain. We found that ramified microglia increased in synaptic regions of the midbrain and hindbrain between 7 and 28 days post fertilization. In contrast, microglia in the optic tectum were ameboid and clustered around neurogenic zones. Using single-cell mRNA sequencing combined with metadata from regional bulk sequencing, we identified synaptic-region associated microglia (SAMs) that were highly enriched in the hindbrain and expressed multiple candidate synapse modulating genes, including genes in the complement pathway. In contrast, neurogenic associated microglia (NAMs) were enriched in the optic tectum, had active cathepsin activity, and preferentially engulfed neuronal corpses. These data reveal that molecularly distinct phagocytic programs mediate synaptic remodeling and cell engulfment, and establish the zebrafish hindbrain as a model for investigating microglial-synapse interactions.

• Lines 68-69 -"highly homologous". I realize this is very picky but homology is not quantitative. "Homologous" or "highly similar" would be more appropriate.
• Lines 74-75 -I think the sentence is missing a word.
• Lines 79-80 -"whether synapse-associated microglia also exist in the fish". The way this is written is a little confusing, because the premise for synapse-associated microglia is not introduced prior to this statement. • Lines 81-83 -"Bulk transcriptomic sequencing in the fish has begun to uncover key information regarding microglial ontogeny, revealing that microglia populate the CNS in two waves 29,30." This is not an accurate statement. The evidence reported in the cited publications is based on fate mapping and lineage tracing strategies, not sequencing. • Lines 111-112 -the authors state that they found microglia within synapse-rich regions "as early as 7 dpf". I believe this is the earliest timepoint that the authors tested. If so, they should make that clear to avoid the possibility that people might interpret this statement to mean that none were evident prior to 7 dpf. • I'm a little uncomfortable with the designation "synapse-associated microglia". I understand that it is a convenient way to describe them following from the approaches in this manuscript but I worry that this description might, ultimately, be too narrow in terms of their cellular associations and functions. Perhaps the authors could consider another descriptor. • Lines 117-119 and Figure S1 -The authors label mpeg:EGFP fish with a 4C4 antibody and use the results to conclude that a majority of mpeg:EGFP cells in the CNS are microglia. The implication in the text is that the 4C4-negative cells are macrophages and, indeed, this is stated in the Figure S1 heading. I think this point needs some clarification to avoid confusion. First, I'm not convinced that the 4C4 antibody is a definitive microglia marker, so I'm not convinced that a CNS parenchymal mpeg:EGFP-positive 4C4-negative cell is not a microglia. Second, it is not apparent to me which cells the authors are referring to. Are these mpeg:EGFP-positive 4C4-negative cells in the parenchyma, or are they non-parenchymal, perivascular or meningeal macrophages? No less an authority than Ben Barres has said that all parenchymal macrophages are microglia, so I just don't want people to come away from this manuscript thinking that there is yet another class of brain macrophage. • The previous point also is relevant to the scRNA-seq cell clustering analysis, from which the authors identify a macrophage (JM3) cluster. I think it is important to know if these are likely nonparenchymal macrophages. • I would encourage the authors to double check their genetic nomenclature for consistency (e.g. mpeg:EGFP, mpeg:eGFP, mpeg-EGFP; 4C4, 4c4) and for conformity to guidelines available at ZFIN (e.g. mpeg should be mpeg1.1). • Wu et al 2020 argue that ameboid microglia are derived both from the rostral blood island (RBI) and the aorta-gonad-mesonephros (AGM) whereas ramified microglia are entirely from the AGM in zebrafish. I think this potentially provides some insight to mechanisms that determine the phenotypic and functional differences of the neurogenic-associated and synapse-associated microglia that these authors describe. It might be worth commenting on this in the Discussion.

Bruce Appel
Reviewer #2: Remarks to the Author: Silva et al present an elegant study analysing different subsets of microglia in larval and adult zebrafish brains. They combine single cell sequencing, regional bulk sequencing, immuno histochemistry and in situ hybridisation to analyse the variety of microglia. Based on this, they identify different populations and highlight two major populations, synapse associated microglia and neurogenic associated microglia. Although rather descriptive, this study is in my opinion highly relevant for the zebrafish field as it provides a detailed assessment of microglia sub populations which is clearly needed for future mechanistic studies. I did not identify any technical issues, the study is sound and all conclusions are backed up by strong data. My only recommendation would be to adjust Figure S1: provide images for the single channels and provide sufficient labelling (mpeg vs 4C4).
Reviewer #3: Remarks to the Author: Silva and co-workers describe 2 subsets of microglia in the developing zebrafish brain using scRNAseq. The paper further contributes to our understanding of microglia subsets in the developing CNS. Here a phagocytic, neuronal-corpse engulfing subtype in the optic tectum and a synapse-associated subtype in the hindbrain are reported.
The strength of the paper is that the functional properties of these microglia subsets are confirmed using Prosense for proteolytic activity and quantifying SV content for synapse engulfement. A weakness is the low number of cells subjected to scRNAseq, somewhat hampering the interpretation of the scRNAseq data, which is inherently less sensitive than bulk RNAseq, which in its turn lacks cellular resolution. For more robust clustering and marker genes etc, more cells would help.
The paper reads well and the data are clearly presented, several points for improvements are listed July 27, 2021 Re: In situ and transcriptomic identification of microglia in synapse-rich regions of the developing zebrafish brain (NCOMMS-21-16792) We sincerely thank all three reviewers for their detailed feedback, which we have carefully considered in revising our manuscript. This revision contains new data, including further characterization of 4C4 negative cells and morphologic analysis of microglia. We have also added several new bioinformatic analyses to extend our data, including comparison with a published human fetal dataset. As suggested by reviewers, the text has been corrected for clarity and consistency. We hope that in its revised form, this manuscript will be a useful resource to the zebrafish community and prompt further investigations of microglial-synapse interactions in this unique model organism.
A point-by-point response to each reviewer's comments follows below. Cell lineage studies, functional investigations and RNA-sequencing approaches have begun to uncover heterogeneity among microglia between brain regions and across developmental time and aging. The extent of this heterogeneity, how it is acquired and how it affects brain development and homeostasis remain as very important questions in the field. This manuscript advances our understanding of microglia heterogeneity by describing two distinct populations of microglia in the zebrafish brain. Although the work does not reveal how these microglia become different nor their specific roles in brain development and function, nevertheless I think this manuscript is an important contribution to the literature. In particular, the gene expression data, which I think is of high quality, will be a very rich resource for the field. For the most part, I think the data adequately support the claims made by the authors, that the statistical analyses are appropriate and that the methods and reagents are clearly described. My specific comments are relatively minor and mostly aim to increase clarity.
• Lines 68-69 -"highly homologous". I realize this is very picky but homology is not quantitative. "Homologous" or "highly similar" would be more appropriate.
• Lines 74-75 -I think the sentence is missing a word. • Lines 79-80 -"whether synapse-associated microglia also exist in the fish". The way this is written is a little confusing, because the premise for synapse-associated microglia is not introduced prior to this statement. • Lines 81-83 -"Bulk transcriptomic sequencing in the fish has begun to uncover key information regarding microglial ontogeny, revealing that microglia populate the CNS in two waves 29,30." This is not an accurate statement. The evidence reported in the cited publications is based on fate mapping and lineage tracing strategies, not sequencing. • Lines 111-112 -the authors state that they found microglia within synapse-rich regions "as early as 7 dpf". I believe this is the earliest timepoint that the authors tested. If so, they should make that clear to avoid the possibility that people might interpret this statement to mean that none were evident prior to 7 dpf. • I'm a little uncomfortable with the designation "synapse-associated microglia". I understand that it is a convenient way to describe them following from the approaches in this manuscript but I worry that this description might, ultimately, be too narrow in terms of their cellular associations and functions. Perhaps the authors could consider another descriptor. • Lines 117-119 and Figure S1 -The authors label mpeg:EGFP fish with a 4C4 antibody and use the results to conclude that a majority of mpeg:EGFP cells in the CNS are microglia. The implication in the text is that the 4C4-negative cells are macrophages and, indeed, this is stated in the Figure S1 heading. I think this point needs some clarification to avoid confusion. First, I'm not convinced that the 4C4 antibody is a definitive microglia marker, so I'm not convinced that a CNS parenchymal mpeg:EGFP-positive 4C4-negative cell is not a microglia. Second, it is not apparent to me which cells the authors are referring to. Are these mpeg:EGFPpositive 4C4-negative cells in the parenchyma, or are they non-parenchymal, perivascular or meningeal macrophages? No less an authority than Ben Barres has said that all parenchymal macrophages are microglia, so I just don't want people to come away from this manuscript thinking that there is yet another class of brain macrophage.
• The previous point also is relevant to the scRNA-seq cell clustering analysis, from which the authors identify a macrophage (JM3) cluster. I think it is important to know if these are likely non-parenchymal macrophages. • I would encourage the authors to double check their genetic nomenclature for consistency (e.g. mpeg:EGFP,mpeg:eGFP,4C4,4c4) and for conformity to guidelines available at ZFIN (e.g. mpeg should be mpeg1.1). • Wu et al 2020 argue that ameboid microglia are derived both from the rostral blood island (RBI) and the aorta-gonad-mesonephros (AGM) whereas ramified microglia are entirely from the AGM in zebrafish. I think this potentially provides some insight to mechanisms that determine the phenotypic and functional differences of the neurogenic-associated and synapse-associated microglia that these authors describe. It might be worth commenting on this in the Discussion.

Bruce Appel
Reviewer #1, response: We sincerely appreciate Dr. Appel's detailed feedback, and the comment that "this manuscript is an important contribution to the literature. In particular, the gene expression data, which I think is of high quality, will be a very rich resource for the field." We hope that as revised, this manuscript will be a useful resource and a baseline for future mechanistic studies. We address each point below: Figure S1 -The authors label mpeg:EGFP fish with a 4C4 antibody and use the results to conclude that a majority of mpeg:EGFP cells in the CNS are microglia. The implication in the text is that the 4C4-negative cells are macrophages and, indeed, this is stated in the Figure S1 heading. I think this point needs some clarification to avoid confusion. First, I'm not convinced that the 4C4 antibody is a definitive microglia marker, so I'm not convinced that a CNS parenchymal mpeg:EGFP-positive 4C4-negative cell is not a microglia. Second, it is not apparent to me which cells the authors are referring to. Are these mpeg:EGFPpositive 4C4-negative cells in the parenchyma, or are they non-parenchymal, perivascular or meningeal macrophages? No less an authority than Ben Barres has said that all parenchymal macrophages are microglia, so I just don't want people to come away from this manuscript thinking that there is yet another class of brain macrophage. Response: We agree that this should have been more precisely addressed in our manuscript and that the title of Fig. S1 was misleading. We now include new data characterizing mpeg-GFP+ 4C4 neg cells by their morphology and location relative to vessels and the brain borders (Reviewer Fig. 1 Fig. S1 to make it clear that 4C4 is specific, but not entirely sensitive.

The previous point also is relevant to the scRNA-seq cell clustering analysis, from which the authors identify a macrophage (JM3) cluster. I think it is important to know if these are likely non-parenchymal macrophages.
Response: This is an important point. We have added additional supplementary analyses and additional discussion to the text, clarifying that more targeted analyses will be required to definitively identify this macrophage subset (Reviewer Fig. 2). Briefly, cluster JM3 is consistent with macrophages (lack of p2ry12, hexb, and csf1ra) and could potentially include perivascular, meningeal, or circulating macrophages, as these animals were not perfused and the meninges were not removed. However, of the recently proposed mammalian BAM markers with fish homologs, none clearly segregated as expected and in some cases were not detected at all. F4/80/adgre10). As such, we cannot definitively identify these macrophages based on existing datasets. We now note this in the discussion and propose as a future direction targeted examination of these subsets, which might be best accomplished by bulk RNAseq of purified microdissected meningeal macrophages vs. microglia. The remaining points are addressed in chronological order: • Lines 68-69 -"highly homologous". I realize this is very picky but homology is not quantitative. "Homologous" or "highly similar" would be more appropriate.
Response: Thank you for the suggestion, we have changed this to "homologous".
• Lines 74-75 -I think the sentence is missing a word.
Response: Corrected, thank you. • Lines 79-80 -"whether synapse-associated microglia also exist in the fish". The way this is written is a little confusing, because the premise for synapse-associated microglia is not introduced prior to this statement.
Response: We agree. This section has been substantially re-written in response to this and subsequent comments, and we hope the logic is now clearer.
• Lines 81-83 -"Bulk transcriptomic sequencing in the fish has begun to uncover key information regarding microglial ontogeny, revealing that microglia populate the CNS in two waves 29,30." This is not an accurate statement. The evidence reported in the cited publications is based on fate mapping and lineage tracing strategies, not sequencing.
Response: Thank you for this correction, we have now re-written this section to clarify the relationship between ontogeny and function (see reviewer point below), and have updated our citations in a manner that we hope will be more accurate.
• Lines 111-112 -the authors state that they found microglia within synapse-rich regions "as early as 7 dpf". I believe this is the earliest timepoint that the authors tested. If so, they should make that clear to avoid the possibility that people might interpret this statement to mean that none were evident prior to 7 dpf.
Response: This statement has been corrected.

• I'm a little uncomfortable with the designation "synapse-associated microglia". I understand that it is a convenient way to describe them following from the approaches in this manuscript but I worry that this description might, ultimately, be too narrow in terms of their cellular associations and functions. Perhaps the authors could consider another descriptor.
Response: Thank you for this suggestion. We appreciate the risk of overinterpretation when naming these subsets, while also recognizing the utility of a nomenclature. We have changed the descriptor to "synapticregion associated microglia" to focus on location rather than function, but preserved the acronym SAM for simplicity. The title has also been revised to read: "In situ and transcriptomic identification of microglia in synapse-rich regions of the developing zebrafish brain".
• I would encourage the authors to double check their genetic nomenclature for consistency (e.g. mpeg:EGFP,mpeg:eGFP,4C4,4C4) and for conformity to guidelines available at ZFIN (e.g. mpeg should be mpeg1.1).
Response: Apologies for this oversight, all genetic nomenclature was reviewed and corrected for consistency.

Wu et al 2020 argue that ameboid microglia are derived both from the rostral blood island (RBI) and the aortagonad-mesonephros (AGM) whereas ramified microglia are entirely from the AGM in zebrafish. I think this potentially provides some insight to mechanisms that determine the phenotypic and functional differences of the neurogenic-associated and synapse-associated microglia that these authors describe. It might be worth commenting on this in the Discussion.
Response: Thank you for this comment. Wu et al identified ccl34b.1 as a marker of a subset of ameboid phagocytic microglia derived from the rostral blood island (RBI) and aorta-gonad-mesonephros (AGM). This marker is also a high confidence hit in our NAM signature, suggesting that NAMs are very likely similar to the subset they identified, and derived from RBI/AGM. It is reasonable to assume that SAMs represent a subset of the ramified, AGM-derived, ccl34b.1-negative cells described by Wu et al., although we also identify several additional clusters that they were not able to resolve by bulk-sequencing. A link between ontogeny and function may exist, but would require further investigation. We have now added this topic to our discussion.

Silva et al present an elegant study analysing different subsets of microglia in larval and adult zebrafish brains.
They combine single cell sequencing, regional bulk sequencing, immuno histochemistry and in situ hybridisation to analyse the variety of microglia. Based on this, they identify different populations and highlight two major populations, synapse associated microglia and neurogenic associated microglia.
Although rather descriptive, this study is in my opinion highly relevant for the zebrafish field as it provides a detailed assessment of microglia sub populations which is clearly needed for future mechanistic studies. I did not identify any technical issues, the study is sound and all conclusions are backed up by strong data. My only recommendation would be to adjust Figure S1: provide images for the single channels and provide sufficient labelling (mpeg vs 4C4).

Reviewer #2, response:
We thank the reviewer for their positive comments and hope that this manuscript will indeed help to promote future mechanistic studies.
My only recommendation would be to adjust Figure S1: provide images for the single channels and provide sufficient labelling (mpeg vs 4C4) Response: Thank you for this comment, we now present single channel images (Reviewer 2, Fig.  3). In addition, we have performed additional characterization to better define the identity of 4C4 negative cells, which we hope will be informative. All of these data are now in manuscript Figure S1.

Reviewer #3 (Remarks to the Author):
Silva and co-workers describe 2 subsets of microglia in the developing zebrafish brain using scRNAseq. The paper further contributes to our understanding of microglia subsets in the developing CNS. Here a phagocytic, neuronal-corpse engulfing subtype in the optic tectum and a synapse-associated subtype in the hindbrain are reported.
The strength of the paper is that the functional properties of these microglia subsets are confirmed using Prosense for proteolytic activity and quantifying SV content for synapse engulfement. A weakness is the low number of cells subjected to scRNAseq, somewhat hampering the interpretation of the scRNAseq data, which is inherently less sensitive than bulk RNAseq, which in its turn lacks cellular resolution. For more robust clustering and marker genes etc, more cells would help. Reviewer #3, response: We sincerely appreciate the reviewer's detailed feedback. We first focus on the question regarding the number of sequenced cells and the resolution at which our single-cell data was analyzed, then address each subsequent point chronologically.

The reviewer notes:
A weakness is the low number of cells subjected to scRNAseq, somewhat hampering the interpretation of the scRNAseq data, which is inherently less sensitive than bulk RNAseq, which in its turn lacks cellular resolution. For more robust clustering and marker genes etc., more cells would help. In point #7 the reviewer adds: Overall, the number of microglia in the scRNAseq data set is not very high, 3529. This might preclude a more robust identification of subcluster marker genes. JM0, the largest cluster, did not have a regional enrichment signature. If it is indeed a common microglia subset, as proposed by the authors, it is difficult to imagine why it was not represented in the bulk RNAseq data. In 2C, cells from clusters 0 and 1 are somewhat mixed, were the data possibly over/underclustered or is the # of cells too low for better separation?
Response: We agree that more cells would have been ideal. The data presented is from all CD45+ cells isolated from 13 fish (10 juvenile, 3 adult). We recovered 9043 total cells that passed our quality control thresholds (6666 juvenile and 2377 adult), of which 3539 juvenile cells and 2080 adult cells were mpeg1.1+.
To limit potential microglial activation from prolonged time ex vivo, we prioritized speed but recovered fewer cells. However, as we discuss further below, we feel that the conclusions put forth in this manuscript are well supported at this level of resolution, and are further supported by our parallel analysis using bulksequencing data.
To address the reviewer's point regarding clustering resolution, we have added new analyses (Reviewer 3,  Fig, 4, from Manuscript Fig.  S2H-I). First, we examine our dataset at multiple clustering resolutions (0.1, 0.3*, and 0.5). Clustering and differential gene expression analysis suggests that there is a meaningful biological difference between clusters 0 and 1 that is statistically significant at this number of sequenced cells. While it is possible to merge clusters 0 and 1 at a low enough resolution (0.1), doing so leads to loss of relevant information. We chose the resolution shown in the manuscript (0.3), for several reasons. First, at that resolution, differential expression analysis between clusters 0 and 1 suggest a clear separation: 337 genes that were up-or down-regulated by at least 15%, including 50 genes up-or down-regulated by at least 40% (including genes expressed in at least 10% of a cluster; p< 0.001; Rev. 3 Fig. 4B). Second, feature plots of DE genes showed a clear gradient in expression between clusters 0 and 1, particularly in genes associated with OT by bulk sequencing (e.g. bzw2, g0s2; Rev. 3 Fig.  4C). Third, our bulk sequencing eigengene analysis (Manuscript Fig. 4B) indicates a high degree of overlap between OT microglia (which are predominantly in neurogenic regions) and cluster 1, and little overlap with cluster 0. Fourth, RNA velocity analysis with scVelo predicts that cluster 0 may be a precursor state to cluster 1 (Reviewer-only Figure Fig. 4D) (Bergen et al 2020). This strengthens our confidence that a clustering resolution separating clusters 0 and 1 is a better representation of the neurogenic-associated microglial signature.
Notably, the main conclusions of the paper would remain largely unchanged by combining clusters 0 and 1. For example, 625 out of 800 differentially regulated genes between clusters 4 and 0/1 are shared regardless of whether clusters 0 and 1 are pooled or separated. Separating clusters 0 and 1 does not change the overall gene signature observed in neurogenic niche-associated and synaptic-region associated microglia, but rather emphasizes the observed heterogeneity in gene expression within the non-synaptic region associated microglial population.
The reviewer correctly points out that we were unable to identify the source of microglial cluster 0. There are multiple possible explanations for this. Cluster 0 may represent a microglial state found throughout the brain, and therefore not enriched in markers for any one brain region found in our bulk sequencing analysis. It is also possible that cluster 0 microglia are found in a brain region not included in our bulk sequencing experiment. We have noted this in our results and discussion.