The gene expression landscape of the human locus coeruleus revealed by single-nucleus and spatially-resolved transcriptomics

  1. Department of Biostatistics, Johns Hopkins Bloomberg School of Public Health, Baltimore, MD, 21205, USA
  2. Lieber Institute for Brain Development, Johns Hopkins Medical Campus, Baltimore, MD, 21205, USA
  3. Department of Neuroscience, Johns Hopkins School of Medicine, Baltimore, MD, 21205, USA
  4. Department of Psychiatry and Behavioral Sciences, Johns Hopkins School of Medicine, Baltimore, MD, 21205, USA
  5. Department of Neurology, Johns Hopkins School of Medicine, Baltimore, MD, 21205, USA
  6. The Kavli Neuroscience Discovery Institute, Johns Hopkins University, Baltimore, MD, 21205, USA

Peer review process

Not revised: This Reviewed Preprint includes the authors’ original preprint (without revision), an eLife assessment, public reviews, and a provisional response from the authors.

Read more about eLife’s peer review process.

Editors

  • Reviewing Editor
    Michael Eisen
    University of California, Berkeley, Berkeley, United States of America
  • Senior Editor
    Michael Eisen
    University of California, Berkeley, Berkeley, United States of America

Reviewer #1 (Public Review):

Weber et al. collect locus coeruleus (LC) tissue blocks from 5 neurotypical European men, dissect the dorsal pons around the LC and prepare 2-3 tissue sections from each donor on a slide for 10X spatial transcriptomics. From three of these donors, they also prepared an additional section for 10x single nucleus sequencing. Overall, the results validate well-known marker genes for the LC (e.g. DBH, TH, SLC6A2), and generate a useful resource that lists genes which are enriched in LC neurons in humans, with either of these two techniques. A comparison with publicly available mouse and rat datasets identifies genes that show reliable LC-enrichment across species. Their analyses also support recent rodent studies that have identified subgroups of interneurons in the region surrounding the LC, which show enrichment for different neuropeptides. In addition, the authors claim that some LC neurons co-express cholinergic markers, and that a population of serotonin (5-HT) neurons is located within or near the LC. These last two claims must be taken with great caution, as several technological limitations restrict the interpretation of these results. Overall, there is limited integration between the spatial and single-nucleus sequencing, thus the data does not yet provide a conclusive list of bona fide LC-specific genes. The authors transparently present limitations of their work in the discussion, but some points discussed below warrant further attention.

Specific comments:

  1. snRNAseq:

a. Major concerns with the snRNAseq dataset are A) the low recovery rate of putative LC-neurons in the snRNAseq dataset, B) the fact that the LC neuron cluster is contaminated with mitochondrial RNA, and C) that a large fraction of the nuclei cannot be assigned to a clear cell type (presumably due to contamination or damaged nuclei). The authors chose to enrich for neurons using NeuN antibody staining and FACS. But it is difficult to assess the efficacy of this enrichment without images of the nuclear suspension obtained before FACS, and of the FACS results. As this field is in its infancy, more detail on preliminary experiments would help the reader to understand why the authors processed the tissue the way they did. It would be nice to know whether omitting the FACS procedure might in fact result in higher relative recovery of LC-neurons, or if the authors tried this and discovered other technical issues that prompted them to use FACS.

b. It is unclear what percentage of cells that make up each cluster.

c. The number of subjects used in each analysis was not always clear. Only 3 subjects were used for snRNAseq, and one of them only yielded 4 LC-nuclei. This means the results are essentially based on n=2. The authors report these numbers in the corresponding section, but the first sentence of the results section (and Figure 1C specifically!) create the impression that n=5 for all analyses. Even for spatial transcriptomics, if I understood it correctly, 1 sample had to be excluded (n=4).

  1. Spatial transcriptomics:

a. It is not clear to me what the spatial transcriptomics provides beyond what can be shown with snRNAseq, nor how these two sets of results compare to each other. It would be more intuitive to start the story with snRNAseq and then try to provide spatial detail using spatial transcriptomics. The LC is not a homogeneous structure but can be divided into ensembles based on projection specificity. Spatial transcriptomics could - in theory - offer much-needed insights into the spatial variation of mRNA profiles across different ensembles, or as a first step across the spatial (rostral/caudal, ventral/dorsal) extent of the LC. The current analyses, however, cannot address this issue, as the orientation of the LC cannot be deduced from the slices analyzed.

b. Unfortunately, spatial transcriptomics itself is plagued by sampling variability to a point where the RNAscope analyses the authors performed prove more powerful in addressing direct questions about gene expression patterns. Given that the authors compare their results to published datasets from rodent studies, it is surprising that a direct comparison of genes identified with spatial transcriptomics vs snRNAseq is lacking (unless this reviewer missed this comparison). Supplementary Figure 17 seems to be a first step in that direction, but this is not a gene-by-gene comparison of which analysis identifies which LC-enriched genes. Such an analysis should not compare numbers of enriched genes using artificial cutoffs for significance/fold-change, but rather use correlations to get a feeling for which genes appear to be enriched in the LC using both methods. This would result in one list of genes that can serve as a reference point for future work.

c. Maybe the spatial transcriptomics could be useful to look at the peri-LC region, which has generated some excitement in rodent work recently, but remains largely unexplored in humans.

  1. The comparison of snRNAseq data to published literature is laudable. Although the authors mention considerable methodological differences between the chosen rodent work and their own analyses, this needs to be further explained. The mouse dataset uses TRAPseq, which looks at translating mRNAs associated with ribosomes, very different from the nuclear RNA pool analyzed in the current work. The rat dataset used single-cell LC laser microdissection followed by microarray analyses, leading to major technical differences in terms of tissue processing and downstream analyses. The authors mention and reference a recent 10x mouse LC dataset (Luskin et al, 2022), however they only pick some neuropeptides from this study for their analysis of interneuron subtypes (Figure S13). Although this is a very interesting part of the manuscript, a more in-depth analysis of these two datasets would be very useful. It would likely allow for a better comparison between mouse and human, given that the technical approach is more similar (albeit without FACS), and Luskin et al have indicated that they are willing to share their data.

  2. Statements in the manuscript about the unexpected identification of a 5-HT (serotonin) cell-cluster seem somewhat contradictory. Figure S14 suggests that 5-HT markers are expressed in the LC-regions just as much as anywhere else, but the RNAscope image in Figure S15 suggests spatial separation between these two populations. And Figure S17 again suggests almost perfect overlap between the LC and 5HT clusters. Maybe I misunderstood, in which case the authors should better clarify/explain these results.

Reviewer #2 (Public Review):

The data generated for this paper provides an important resource for the neuroscience community. The locus coeruleus (LC) is the known seed of noradrenergic cells in the brain. Due to its location and size, it remains scarcely profiled in humans. Despite the physically minute structure containing these cells, its impact is wide-reaching due to the known neuromodulatory function of norepinephrine (NE) in processes like attention and mood. As such, profiling NE cells has important implications for most neurological and neuropsychiatric disorders. This paper generates transcriptomic profiles that are not only cell-specific but which also maintain their spatial context, providing the field with a map for the cells within the region.

Strengths:

Using spatial transcriptomics in a morphologically distinct region is a very attractive way to generate a map. Overlaying macroscopic information, i.e. a region with greater pigmentation, with its corresponding molecular profile in an unbiased manner is an extremely powerful way to understand the specific cellular and molecular composition of that brain structure.

The technologies were used with an astute awareness of their limitations, as such, multiple technologies were leveraged to paint a more complete and resolved picture of the cellular composition of the region. For example, the lack of resolution in the spatial transcriptomic platform was compensated by complementary snRNA-seq and single molecule FISH.

This work has been made publicly available and accessible through a user-friendly application such that any interested researcher can investigate the level of expression of their gene of interest within this region.

Two important implications from this work are 1) the potential that the gene regulatory profiles of these cells are only partially conserved across species, humans, and rodents, and 2) that there may be other neuromodulatory cell types within the region that were otherwise not previously localized to the LC

Weaknesses:

Given that the markers used to identify cells are not as specific as they need to be to definitively qualify the desired cell type, the results may be over-interpreted. Specifically, TH is the primary marker used to qualify cells as noradrenergic, however, TH catalyzes the synthesis of L-DOPA, a precursor to dopamine, which in turn is a precursor for epinephrine and norepinephrine suggesting some of the cells in the region may be dopaminergic and not NE cells. Indeed, there are publications to support the presence of dopaminergic cells in the LC (see Kempadoo et al. 2016, Takeuchi et al., 2016, Devoto et al. 2005). This discrepancy is further highlighted by the apparent lack of overlap per given Visium spots with TH, SCL6A2, or DBH. While the single-nucleus FISH confirms that some of the cells in the region are noradrenergic, others very possibly represent a different catecholamine. As such it is suggested that the nomenclature for the cells be reconsidered.

The authors are unable to successfully implement unsupervised clustering with the spatial data, this greatly reduces the impact of the spatial technology as it implies that the transcriptomic data generated in the study did not have enough resolution to identify individual cell types.

The sample contribution to the results is highly unbalanced, which consequently, may result in ungeneralizable findings in terms of regional cellular composition, limiting the usefulness of the publicly available data.

This study aimed to deeply profile the LC in humans and provide a resource to the community. The combination of data types (snRNA-seq, SRT, smFISH) does in fact represent this resource for the community. However, due to the limitations, of which, some were described in the manuscript, we should be cautious in the use of the data for secondary analysis. For example, some of the cellular annotations may lack precision, the cellular composition also may not reflect the general population, and the presence of unexpected cell types may represent the accidental inclusion of adjacent regions, in this case, serotonergic cells from the Raphe nucleus.

Nonetheless having a well-developed app to query and visualize these data will be an enormous asset to the community especially given the lack of information regarding the region in general.

Reviewer #3 (Public Review):

In this study, the authors present the first comprehensive transcriptome map of the human locus coeruleus using two independent but complementary approaches, spatial transcriptomics and single nucleus RNA sequencing. Several canonical features of locus coeruleus neurons that have been described in rodents were conserved, but potentially important species differences were also identified. This work lays the foundation for future descriptive and experimental approaches to understand the contribution of the locus coeruleus to healthy brain function and disease.

This study has many strengths. It is the first reported comprehensive map of the human LC transcriptome, and uses two independent but complementary approaches (spatial transcriptomics and snRNA-seq). Some of the key findings confirmed what has been described in the rodent LC, as well as some intriguing potential genes and modules identified that may be unique to humans and have the potential to explain LC-related disease states. The main limitations of the study were acknowledged by the authors and include the spatial resolution probably not being at the single cell level and the relatively small number of samples (and questionable quality) for the snRNA-seq data. Overall, the strengths greatly outweigh the limitations. This dataset will be a valuable resource for the neuroscience community, both in terms of methodology development and results that will no doubt enable important comparisons and follow-up studies.

Major comments:

Overall, the discovery of some cells in the LC region that express serotonergic markers is intriguing. However, no evidence is presented that these neurons actually produce 5-HT.

Concerning the snRNA-seq experiments, it is unclear why only 3 of the 5 donors were used, particularly given the low number of LC-NE nuclear transcriptomes obtained, why those 3 were chosen, and how many 100 um sections were used from each donor. It is also unclear if the 295 nuclei obtained truly representative of the LC population or whether they are just the most "resilient" LC nuclei that survive the process.

The LC displays rostral/caudal and dorsal/ventral differences, including where they project, which functions they regulate, and which parts are vulnerable in neurodegenerative disease (e.g. Loughlin et al., Neuroscience 18:291-306, 1986; Dahl et al., Nat Hum Behav 3:1203-14, 2019; Beardmore et al., J Alzheimer's Dis 83:5-22, 2021; Gilvesy et al., Acta Neuropathol 144:651-76, 2022; Madelung et al., Mov Disord 37:479-89, 2022). It was not clear which part(s) of the LC was captured for the SRT and snRNAseq experiments.

The authors mention that in other human SRT studies, there are typically between 1-10 cells per expression spot. I imagine that this depends heavily on the part of the brain being studied and neuronal density, but it was unclear how many LC cells were contained in each expression spot.

Regarding comparison of human LC-associated genes with rat or mouse LC-associated genes (Fig. 2D-F), the authors speculate that the modest degree of overlap may be due to species differences between rodents and human and/or methodological differences (SRT vs microarray vs TRAP). Was there greater overlap between mouse and rat than between mouse/rat and human? If so, that is evidence for the former. If not, that is evidence for the latter. Also would be useful for more in-depth comparison with snRNA-seq data from mouse LC: https://www.biorxiv.org/content/10.1101/2022.06.30.498327v1.

The finding of ACHE expression in LC neurons is intriguing, especially in light of work from Susan Greenfield suggesting that ACHE has functions independent of ACH metabolism that contributes to cellular vulnerability in neurodegenerative disease.

High mitochondrial reads from snRNA-seq can indicate lower quality. It was not clear why, given the mitochondrial read count, the authors are confident in the snRNA-seq data from presumptive LC-NE neurons.

Author Response:

eLife assessment

This is a valuable initial study of cell type and spatially resolved gene expression in and around the locus coeruleus, the primary source of the neuromodulator norepinephrine in the human brain. The data are generated with cutting-edge techniques, and the work lays the foundation for future descriptive and experimental approaches to understand the contribution of the locus coeruleus to healthy brain function and disease. However, due to small sample size and the need for additional confirmatory data, the data only incompletely support the main conclusions presented here. With the strengthening of the analyses, this paper, and the associated web application, will be of great interest to neuroscientists working on arousal-based behaviors and neurological and neuropsychiatric phenotypes.

Thank you for the assessment and comments. Overall, the majority of the issues raised by the reviewers relate either directly or indirectly to limitations of the sample size that precluded further optimization of protocols and expansion of the dataset. We fully acknowledge the limited sample size in this dataset and aim to be transparent about the limitations of the study. This is the first report of snRNA-seq and spatially-resolved transcriptomics in the human locus coeruleus (LC). The LC is a very small nucleus, located deep within the brainstem, which is extremely challenging to study due to its small size, difficult to access location, and the very small number of norepinephrine (NE) neurons located within the nucleus, which were of prime interest for this study. We note that this study represents our initial attempt to molecularly and spatially characterize cell types within the human LC. We note that we did not have significant, established funding from extramural sources dedicated to this study, and tissue resources for the LC are difficult to ascertain, contributing to the small sample size in this initial study. We acknowledge that there are limitations in sample size as well as data quality. Findings from this study will be used to inform, improve, and optimize future and ongoing experimental design, as well as technical and analytical workflows for larger-scale studies. As brought up by one of the reviewers, this field is still in its infancy -- pilot experimentation in new brain regions is labor-intensive and these sequencing approaches remain costly. Moreover, due to the small size and difficulties in dissecting, tissue resources from the human brain in this area are a highly limited resource. Hence, notwithstanding limitations, in our view it is important to release the data for community access at this time. Specific responses to the reviewers’ comments are provided point-by-point in the following sections.

Reviewer #1 (Public Review):

Weber et al. collect locus coeruleus (LC) tissue blocks from 5 neurotypical European men, dissect the dorsal pons around the LC and prepare 2-3 tissue sections from each donor on a slide for 10X spatial transcriptomics. […] The authors transparently present limitations of their work in the discussion, but some points discussed below warrant further attention.

Specific comments:

  1. snRNAseq:

a. Major concerns with the snRNAseq dataset are A) the low recovery rate of putative LC-neurons in the snRNAseq dataset, B) the fact that the LC neuron cluster is contaminated with mitochondrial RNA, and C) that a large fraction of the nuclei cannot be assigned to a clear cell type (presumably due to contamination or damaged nuclei). The authors chose to enrich for neurons using NeuN antibody staining and FACS. But it is difficult to assess the efficacy of this enrichment without images of the nuclear suspension obtained before FACS, and of the FACS results. As this field is in its infancy, more detail on preliminary experiments would help the reader to understand why the authors processed the tissue the way they did. It would be nice to know whether omitting the FACS procedure might in fact result in higher relative recovery of LC-neurons, or if the authors tried this and discovered other technical issues that prompted them to use FACS.

Thank you for these comments. We agree these are valid concerns in assessing the data quality and validity of the findings from the snRNA-seq dataset. We will respond to these concerns here to the best of our ability, but in some cases, we do not have definitive answers since comparison data are not yet available for this region. In particular, we were limited in resources for this initial study -- some of the results of the study and issues that we identified in attempting to molecularly profile cells in the human LC were surprising to us, and we intend to generate additional samples and troubleshoot these issues to improve data quality and increase recovery in future work. However, these experiments are (i) expensive, (ii) time- and labor-intensive, and (iii) the tissue for this region is limited and difficult to ascertain. Given the extremely small size of the LC, the tissue resource is quickly depleted. For this study, we had fixed resources and made best-guess decisions on how to proceed with the experimental design, based on our experience with snRNA-seq in other human brain regions (Tran and Maynard et al. 2021). However, the LC is a unique region, and our experiences with this dataset will guide us to make technical adjustments in future studies. Due to the limitations in the tissue resources and the lack of data currently available to the community, we wanted to share these results immediately while acknowledging the limitations of the study as we work to increase our resource availability to expand molecular and spatial profiling studies in this region of the human brain.

Regarding the reviewer’s concern that our choice to use FANS to enrich for neurons could have potentially led to more damage and contributed to the low recovery rate of LC-NE neurons and the mitochondrial contamination -- we do not have a definitive answer to this question, since we did not perform a direct comparison with non-sorted data. As noted above, our limited tissue resource dictated that we could not do both. We made the decision to enrich for neurons based on our previous experience with identifying relatively rare populations in other brain regions (e.g. nucleus accumbens and amygdala; Tran and Maynard et al. 2021). Based on this previous work, our rationale was that without neuronal enrichment, we could potentially miss the LC-NE population, given the relative scarcity of this neuronal population. The low recovery rate and relatively lower quality / contamination issues may be due to technical issues that lead to LC-NE neurons being more susceptible to damage during nuclear preparation and sorting. We agree that directly comparing to data prepared without NeuN labeling and sorting is reasonable, as the additional perturbations may indeed contribute to cell damage. As mentioned in the discussion, we do not have a definitive answer to the reasons for increased mitochondrial contamination and we suspect that multiple technical factors may contribute -- including the relatively large size and increased fragility of LC-NE neurons. We agree that systematically optimizing the preparation to attempt to increase recovery rate and decrease mitochondrial contamination are important avenues for future work.

b. It is unclear what percentage of cells that make up each cluster.

We will add this information in the clustering heatmaps or as a supplementary plot in a revised version of the manuscript.

c. The number of subjects used in each analysis was not always clear. Only 3 subjects were used for snRNAseq, and one of them only yielded 4 LC-nuclei. This means the results are essentially based on n=2. The authors report these numbers in the corresponding section, but the first sentence of the results section (and Figure 1C specifically!) create the impression that n=5 for all analyses. Even for spatial transcriptomics, if I understood it correctly, 1 sample had to be excluded (n=4).

This is correct. We will update the figures and text in a revised version of the manuscript to make this limitation (small sample size) more clear, and to further emphasize that the intention of this study is to provide initial data to help determine next steps and best practices for a larger scale and more comprehensive study on this region, especially given the limited availability of tissue resources and currently limited data resources available for this region.

  1. Spatial transcriptomics:

a. It is not clear to me what the spatial transcriptomics provides beyond what can be shown with snRNAseq, nor how these two sets of results compare to each other. It would be more intuitive to start the story with snRNAseq and then try to provide spatial detail using spatial transcriptomics. The LC is not a homogeneous structure but can be divided into ensembles based on projection specificity. Spatial transcriptomics could - in theory - offer much-needed insights into the spatial variation of mRNA profiles across different ensembles, or as a first step across the spatial (rostral/caudal, ventral/dorsal) extent of the LC. The current analyses, however, cannot address this issue, as the orientation of the LC cannot be deduced from the slices analyzed.

We understand the point of the reviewer. However, we structured the manuscript in this format due to our aims of creating a data resource for the community as well as being transparent about the limitations of our study. Our experiments began with the spatial experiments on the tissue blocks because this (i) helped orient ourselves to the region, and (ii) provided guidance for how best to score the tissue blocks for the snRNA-seq experiments to maximize recovery of LC-NE neurons. Therefore, we also decided to present the results in this sequence.

The spatial data also provides more information in that the measurements are from nuclei, cytoplasm, and cell processes (instead of nuclei only). This is one of the main differences / advantages between the platforms at this level of spatial resolution. As noted above, we were also working with a finite tissue resource -- if we ran snRNA-seq first and captured no neurons, the tissue block would be depleted. Due to the logistics / thickness of the required tissue sections for Visium and snRNA-seq respectively, running Visium first allowed us to ensure that we could collect data from both assays.

Regarding a point raised below on why we only ran snRNA-seq on a subset of the donors -- this was due to resource depletion and not enough available tissue remaining on the tissue blocks to run the assay. We have conducted extensive piloting in other brain regions on the amount (mg) of tissue that is needed from various sized cryosections, and the LC is particularly difficult since these are small tissue blocks and the extent of the structure is small. Hence, in some of the subjects, we did not have sufficient tissue available for the snRNA-seq assay.

We agree with the reviewer that spatial studies could, in future work, offer needed and important information about expression profiles across the spatial axes (rostral/caudal, ventral/dorsal) of the LC. Our study provides us with insight about optimizing the dissections for spatial assays, as well as bringing to light a number of technical and logistical issues that we had not initially foreseen. For example, during the course of this study and parallel, ongoing work in other small, challenging brain regions, we have now developed a number of specialized technical and logistical strategies for keeping track of orientation and mounting serial sections from the same tissue block onto a single spatial array, which is extremely technically challenging. We are now well-prepared for addressing these issues in future studies with larger numbers of donors and samples, e.g. spaced serial sections across the extent of the LC to make these types of insights. Due to the rarity of the tissue, limited availability of information in this region, and high expense of conducting these studies, we want to share this initial data with the community immediately. We also note that in addition to the 10x Genomics Visium platform, which lacks cellular and sub-cellular resolution, many new and exciting spatial platforms are entering the market, which may be able to address questions in very small regions such as the LC at higher spatial resolution.

b. Unfortunately, spatial transcriptomics itself is plagued by sampling variability to a point where the RNAscope analyses the authors performed prove more powerful in addressing direct questions about gene expression patterns. Given that the authors compare their results to published datasets from rodent studies, it is surprising that a direct comparison of genes identified with spatial transcriptomics vs snRNAseq is lacking (unless this reviewer missed this comparison). Supplementary Figure 17 seems to be a first step in that direction, but this is not a gene-by-gene comparison of which analysis identifies which LC-enriched genes. Such an analysis should not compare numbers of enriched genes using artificial cutoffs for significance/fold-change, but rather use correlations to get a feeling for which genes appear to be enriched in the LC using both methods. This would result in one list of genes that can serve as a reference point for future work.

We agree this is a good suggestion, and will add additional computational analyses to address this point in a revised version of the manuscript.

c. Maybe the spatial transcriptomics could be useful to look at the peri-LC region, which has generated some excitement in rodent work recently, but remains largely unexplored in humans.

We agree this is an excellent suggestion -- assessing cross-species comparisons related to convergence, especially, of GABAergic cell populations in the human LC is of high interest. We note that these types of extensions are exactly the reason why we have provided the publicly accessible web app (R/Shiny app, which includes the ability to annotate regions). We hope that others will use these apps for specialized topics they are interested in. As discussed above, we note that our initial dissections precluded the ability to keep track of the exact orientation of our tissue sections on the Visium arrays with respect to their location within the brainstem, so definitive localization of this region across subjects is difficult in our current study. However, it is possible, for example, to investigate whether there is a putative peri-LC region that is densely GABAergic that is homologous with the GABAergic peri-LC region in rodents. We also raise attention to a recent preprint by Luskin and Li et al. (2022), who apply snRNA-seq and spatially-resolved transcriptomics to molecularly define both LC and peri-LC cell types in mice -- in a revised version of our manuscript, we will extend our computational analyses of inhibitory neuronal subtypes in our data (Supplementary Figures 13, 16) to directly compare with those identified in this study in more detail. As noted above, we we have now developed a number of specialized technical and logistical strategies for keeping track of orientation of sections from the tissue block onto a single spatial array, and we feel that combined with optimized dissection strategies for this region and the guide of RNAscope for GABAergic markers on serial sections, that annotating the peri-LC region on spatial arrays in future studies will be possible.

  1. The comparison of snRNAseq data to published literature is laudable. Although the authors mention considerable methodological differences between the chosen rodent work and their own analyses, this needs to be further explained. The mouse dataset uses TRAPseq, which looks at translating mRNAs associated with ribosomes, very different from the nuclear RNA pool analyzed in the current work. The rat dataset used single-cell LC laser microdissection followed by microarray analyses, leading to major technical differences in terms of tissue processing and downstream analyses. The authors mention and reference a recent 10x mouse LC dataset (Luskin et al, 2022), however they only pick some neuropeptides from this study for their analysis of interneuron subtypes (Figure S13). Although this is a very interesting part of the manuscript, a more in-depth analysis of these two datasets would be very useful. It would likely allow for a better comparison between mouse and human, given that the technical approach is more similar (albeit without FACS), and Luskin et al have indicated that they are willing to share their data.

As noted above, we plan to extend our comparisons with the dataset from Luskin and Li et al. (2022) in a revised version of the manuscript, which will provide a more in-depth cross-species comparison. In addition, we also note that there are some additional recent studies using TRAPseq of LC-NE neurons in a functional context, i.e. treatment vs. control experiments or in model systems (e.g. Iannitelli et al. 2023), which provide new opportunities for understanding disease context using in-depth cross-species comparisons. By providing our dataset and reproducible code, we will enable others to adapt and extend these types of comparisons (i.e. TRAPseq of LC-NE neurons or LC snRNA-seq following functional manipulations or in the context of disease or behavioral models) in the future.

  1. Statements in the manuscript about the unexpected identification of a 5-HT (serotonin) cell-cluster seem somewhat contradictory. Figure S14 suggests that 5-HT markers are expressed in the LC-regions just as much as anywhere else, but the RNAscope image in Figure S15 suggests spatial separation between these two populations. And Figure S17 again suggests almost perfect overlap between the LC and 5HT clusters. Maybe I misunderstood, in which case the authors should better clarify/explain these results.

In our view, the most likely scenario is that the 5-HT neurons come from contamination from the dorsal raphe nucleus based on spatial separation from the RNAscope images, which we agree are more definitive. As mentioned above, since we do not have definitive documentation for the tissue sections in terms of orientation, it is difficult to say with clarity that the regions are the dorsal raphe and which sub-portion of the dorsal raphe they are. This initial study has now allowed us to optimize and improve our dissection strategy and approaches for retaining documentation of the orientation of the tissue sections from their intact position within the brainstem as they move from cryosection to placement on the array, which will enable us to better annotate regions with definitive anatomical information with respect to the rostral/caudal and dorsal/ventral axes in future experiments. Given that there are reports in the rodent that 5-HT markers have been identified in LC-NE neurons (Iijima 1993; Iijima 1989), and taking into account the technical limitations in our study, we felt that it was premature to definitively conclude in the manuscript that we were sure these signals arose from the dorsal raphe. We will update this language in a revised version of the manuscript to ensure that these limitations are clear (referring to Supplementary Figures S14-15, S17).

Reviewer #2 (Public Review):

The data generated for this paper provides an important resource for the neuroscience community. The locus coeruleus (LC) is the known seed of noradrenergic cells in the brain. Due to its location and size, it remains scarcely profiled in humans. Despite the physically minute structure containing these cells, its impact is wide-reaching due to the known neuromodulatory function of norepinephrine (NE) in processes like attention and mood. As such, profiling NE cells has important implications for most neurological and neuropsychiatric disorders. This paper generates transcriptomic profiles that are not only cell-specific but which also maintain their spatial context, providing the field with a map for the cells within the region.

Strengths:

Using spatial transcriptomics in a morphologically distinct region is a very attractive way to generate a map. Overlaying macroscopic information, i.e. a region with greater pigmentation, with its corresponding molecular profile in an unbiased manner is an extremely powerful way to understand the specific cellular and molecular composition of that brain structure.

The technologies were used with an astute awareness of their limitations, as such, multiple technologies were leveraged to paint a more complete and resolved picture of the cellular composition of the region. For example, the lack of resolution in the spatial transcriptomic platform was compensated by complementary snRNA-seq and single molecule FISH.

This work has been made publicly available and accessible through a user-friendly application such that any interested researcher can investigate the level of expression of their gene of interest within this region.

Two important implications from this work are 1) the potential that the gene regulatory profiles of these cells are only partially conserved across species, humans, and rodents, and 2) that there may be other neuromodulatory cell types within the region that were otherwise not previously localized to the LC

Weaknesses:

Given that the markers used to identify cells are not as specific as they need to be to definitively qualify the desired cell type, the results may be over-interpreted. Specifically, TH is the primary marker used to qualify cells as noradrenergic, however, TH catalyzes the synthesis of L-DOPA, a precursor to dopamine, which in turn is a precursor for epinephrine and norepinephrine suggesting some of the cells in the region may be dopaminergic and not NE cells. Indeed, there are publications to support the presence of dopaminergic cells in the LC (see Kempadoo et al. 2016, Takeuchi et al., 2016, Devoto et al. 2005). This discrepancy is further highlighted by the apparent lack of overlap per given Visium spots with TH, SCL6A2, or DBH. While the single-nucleus FISH confirms that some of the cells in the region are noradrenergic, others very possibly represent a different catecholamine. As such it is suggested that the nomenclature for the cells be reconsidered.

We appreciate the reviewer’s comment, and are aware of the reports suggesting the potential presence of dopaminergic cells in the LC. We initially had the same thought as the reviewer when we observed Visium spots in the spatial data with lack of overlap between TH, SLC6A2, and DBH as well as single nuclei in the snRNA-seq data with lack of overlap between TH, SLC6A2, and DBH. This surprising result was exactly why we performed the smFISH/RNAscope experiment with these three marker genes. Given known issues with read depth and coverage in the 10x Genomics assays, we wanted to better understand if this was a technical limitation in the sequencing coverage, or rather a true biological finding. The RNAscope data showed very clearly that nearly every cell body we looked at had co-localization of these three marker genes. We included an image from a single capture array of one tissue section in Supplementary Figure 11, but could, in a revised version of the manuscript, provide additional examples to illustrate how conclusive the images were by visualization. As such, we were quite convinced that the lack of overlap on Visium spots and in single nuclei in the snRNA-seq data was more likely related to technical issues with sequencing coverage, rather than a biological finding. We also note that we checked for the presence of the dopamine transporter, SLC6A3, and as can be appreciated in the iSEE web app for the snRNA-seq data or the R/Shiny web app for the Visium data, there is virtually no expression of SLC6A3 in the dataset, which in our view provides additional evidence against the possibility that there are substantial quantities of dopaminergic cells in this human LC dataset. We will include supplementary plots showing the lack of SLC6A3 expression in a revised version of the manuscript.

The authors are unable to successfully implement unsupervised clustering with the spatial data, this greatly reduces the impact of the spatial technology as it implies that the transcriptomic data generated in the study did not have enough resolution to identify individual cell types.

The reviewer is correct -- this is a fundamental limitation of the 10x Genomics Visium platform, i.e. the spatial resolution captures multiple cells per spot (e.g. around 1-10 cells per spot in human brain tissue). We note that new spatial platforms now provide cellular resolution (e.g. Vizgen MERSCOPE, 10x Genomics Xenium, 10x Genomics Visium HD), which will help address this in future work. However, many of these cellular-resolution in situ sequencing platforms have the limitation that they do not quantify genome-wide expression, and instead require users to select a priori gene panels to investigate. This is a problem if no genome-wide reference datasets are available. Hence, despite the limited spatial resolution of the Visium platform, this dataset is useful precisely for helping investigators choose gene panels for higher-resolution platforms or higher-order smFISH multiplexing.

We also applied spatial clustering (using BayesSpace; Zhao et al. 2021) to attempt to segment the LC regions within the Visium samples in a data-driven manner as an alternative to the manual annotations, which was unsuccessful (and hence we relied on the manually annotated regions for downstream analyses) (Supplementary Figure S5). However, this is a different application of unsupervised clustering, which is separate from the task of identifying cell types.

The sample contribution to the results is highly unbalanced, which consequently, may result in ungeneralizable findings in terms of regional cellular composition, limiting the usefulness of the publicly available data.

We acknowledge the limitations of the work due to the small/unbalanced sample sizes. As mentioned above for Reviewer 1, this was an initial study in this region -- results of which will inform our (and hopefully others’) experimental design and approach to molecular profiling in this difficult to access brain region. Overall, this study was executed with finite tissue and financial resources and was intended to uncover limitations and help develop best practices and design workflows for future studies with larger numbers of donors and samples. Given the limited data availability for this brain region, we wanted to make this dataset available for the research community immediately. In addition, we note that making this genome-wide dataset available will help inform targeted gene panel design for higher-resolution platforms (e.g. 10x Genomics Xenium).

This study aimed to deeply profile the LC in humans and provide a resource to the community. The combination of data types (snRNA-seq, SRT, smFISH) does in fact represent this resource for the community. However, due to the limitations, of which, some were described in the manuscript, we should be cautious in the use of the data for secondary analysis. For example, some of the cellular annotations may lack precision, the cellular composition also may not reflect the general population, and the presence of unexpected cell types may represent the accidental inclusion of adjacent regions, in this case, serotonergic cells from the Raphe nucleus.

We agree, and have attempted to explain these limitations in the manuscript. We will clarify the language regarding the interpretation of the annotated cell populations and unexpected cell types, and the limited sample sizes, in a revised version of the manuscript.

Nonetheless having a well-developed app to query and visualize these data will be an enormous asset to the community especially given the lack of information regarding the region in general.

Reviewer #3 (Public Review):

[…] This study has many strengths. It is the first reported comprehensive map of the human LC transcriptome, and uses two independent but complementary approaches (spatial transcriptomics and snRNA-seq). Some of the key findings confirmed what has been described in the rodent LC, as well as some intriguing potential genes and modules identified that may be unique to humans and have the potential to explain LC-related disease states. The main limitations of the study were acknowledged by the authors and include the spatial resolution probably not being at the single cell level and the relatively small number of samples (and questionable quality) for the snRNA-seq data. Overall, the strengths greatly outweigh the limitations. This dataset will be a valuable resource for the neuroscience community, both in terms of methodology development and results that will no doubt enable important comparisons and follow-up studies.

Major comments:

Overall, the discovery of some cells in the LC region that express serotonergic markers is intriguing. However, no evidence is presented that these neurons actually produce 5-HT.

The reviewer is correct that we did not provide any additional evidence to show that these neurons actually produce 5-HT. As noted above in the response to Reviewer 1, in our view, the most likely explanation is that these neurons are from dorsal raphe contamination on the tissue section. However, due to technical and logistical limitations in this study, we could not definitively say this because we did not clearly track the orientation of the tissue sections, and we did not have remaining tissue sections from all donor tissue blocks to repeat RNAscope experiments. For some of the donors, where we had remaining tissue sections to go back to repeat RNAscope experiments after completion of the snRNA-seq and Visium assays, we could see clear separation of the LC region / LC-NE neuron core from where putative 5-HT neurons were located (Supplementary Figure 15). However, we did not have sufficient tissue resources to map this definitively in all donors, and the orientation and anatomy of each tissue block were not fully annotated.

Due to the lack of clarity, and the fact that there have been reports that LC-NE neurons express serotonergic markers (Iijima 1993; Iijima 1989), we felt that it was premature to definitively declare that these putative 5-HT neurons that we identified were definitively from the raphe. We will clarify the language around this discrepancy in a revised version of the manuscript to ensure that these limitations are clearly described.

Concerning the snRNA-seq experiments, it is unclear why only 3 of the 5 donors were used, particularly given the low number of LC-NE nuclear transcriptomes obtained, why those 3 were chosen, and how many 100 um sections were used from each donor. It is also unclear if the 295 nuclei obtained truly representative of the LC population or whether they are just the most "resilient" LC nuclei that survive the process.

As discussed above for Reviewer 1, the reason we included only 3 of the 5 donors for the snRNA-seq assays was due to the tissue availability on the tissue blocks. We will clarify the language in a revised version of the manuscript to make this limitation more clear. We will also include additional details in the Methods section on the number of 100 μm sections used for each donor (which varied between 10-15, approximating 60-80 mg of tissue).

The LC displays rostral/caudal and dorsal/ventral differences, including where they project, which functions they regulate, and which parts are vulnerable in neurodegenerative disease (e.g. Loughlin et al., Neuroscience 18:291-306, 1986; Dahl et al., Nat Hum Behav 3:1203-14, 2019; Beardmore et al., J Alzheimer's Dis 83:5-22, 2021; Gilvesy et al., Acta Neuropathol 144:651-76, 2022; Madelung et al., Mov Disord 37:479-89, 2022). It was not clear which part(s) of the LC was captured for the SRT and snRNAseq experiments.

As discussed above for Reviewer 1, a limitation of this study was that we did not record the orientation of the anatomy of the tissue sections, precluding our ability to annotate the tissue sections with the rostral/caudal and dorsal/ventral axis labels. We agree with the reviewer that additional spatial studies, in future work, could offer needed and important information about expression profiles across the spatial axes (rostral/caudal, ventral/dorsal) of the LC. Our study provides us with insight about optimizing the dissections for spatial assays, as well as bringing to light a number of technical and logistical issues that we had not initially foreseen. For example, during the course of this study and parallel, ongoing work in other, small, challenging regions, we have now developed a number of specialized technical and logistical strategies for keeping track of orientation and mounting serial sections from the same tissue block onto a single spatial array, which is extremely technically challenging. We are now well-prepared for addressing these issues in future studies with larger numbers of donors and samples in order to make these types of insights.

The authors mention that in other human SRT studies, there are typically between 1-10 cells per expression spot. I imagine that this depends heavily on the part of the brain being studied and neuronal density, but it was unclear how many LC cells were contained in each expression spot.

The reviewer is correct that we did not include this information in the manuscript. We attempted to apply a computational method to count nuclei contained in each gene expression spot based on analyzing the histological H&E images (VistoSeg; Tippani et al. 2022), which we have developed and previously applied in data from the dorsolateral prefrontal cortex (DLPFC) (Maynard and Collado-Torres et al. 2021). Based on the segmentation using this workflow we observe that the counts in this region are similar to what we observed in the DLPFC, i.e., typically between 1-10 LC cells per expression spot, with approximately 1-2 LC-NE neurons (which are characterized by their large size) per expression spot. However, these analyses had several technical issues related to the images themselves, the relatively large size and pigmentation of LC-NE neurons, and parameter settings that had been optimized for different brain regions. We are currently optimizing this analysis workflow for these images to provide more accurate estimates of cell counts per spot to give readers additional context on the number of nuclei per spot in the annotated LC regions and outside the LC regions in a revised version of the manuscript.

Regarding comparison of human LC-associated genes with rat or mouse LC-associated genes (Fig. 2D-F), the authors speculate that the modest degree of overlap may be due to species differences between rodents and human and/or methodological differences (SRT vs microarray vs TRAP). Was there greater overlap between mouse and rat than between mouse/rat and human? If so, that is evidence for the former. If not, that is evidence for the latter. Also would be useful for more in-depth comparison with snRNA-seq data from mouse LC: https://www.biorxiv.org/content/10.1101/2022.06.30.498327v1.

We will investigate this question and discuss this in updated results in a revised version of the manuscript.

The finding of ACHE expression in LC neurons is intriguing, especially in light of work from Susan Greenfield suggesting that ACHE has functions independent of ACH metabolism that contributes to cellular vulnerability in neurodegenerative disease.

We thank the reviewer for pointing this out. We were very surprised too by the observed expression of SLC5A7 and ACHE in the LC regions (Visium data) and within the LC-NE neuron cluster (snRNA-seq data), coupled with absence of other typical cholinergic marker genes (e.g. CHAT, SLC18A3), and we do not have a compelling explanation or theory for this. Hence, the work of Susan Greenfield and colleagues suggesting non-cholinergic actions of ACHE, particularly in other catecholaminergic neurons (e.g. dopaminergic neurons in the substantia nigra) is very interesting. We will include references to this work and how it could inform interpretation of this expression in a revised version of the manuscript (Greenfield 1991; Halliday and Greenfield 2012).

High mitochondrial reads from snRNA-seq can indicate lower quality. It was not clear why, given the mitochondrial read count, the authors are confident in the snRNA-seq data from presumptive LC-NE neurons.

We will include additional analyses to further investigate and/or confirm this finding (e.g. comparing sum of UMI counts / number of detected genes and mitochondrial percentage per nucleus for this population to confirm data quality) in additional supplementary figures in a revised version of the manuscript.

References

  • Greenfield (1991), A noncholinergic action of acetylcholinesterase (AChE) in the brain: from neuronal secretion to the generation of movement, Cellular and Molecular Neurobiology, 11, 1, 55-77.

  • Halliday and Greenfield (2012), From protein to peptides: a spectrum of non-hydrolytic functions of acetylcholinesterase, Protein & Peptide Letters, 19, 2, 165-172.

  • Iannitelli et al. (2023), The neurotoxin DSP-4 dysregulates the locus coeruleus-norepinephrine system and recapitulates molecular and behavioral aspects of prodromal neurodegenerative disease, eNeuro, 10, 1, ENEURO.0483-22.2022.

  • Iijima K. (1989), An immunocytochemical study on the GABA-ergic and serotonin-ergic neurons in rat locus ceruleus with special reference to possible existence of the masked indoleamine cells. Acta Histochema, 87, 1, 43-57.

  • Iijima K. (1993), Chemocytoarchitecture of the rat locus ceruleus, Histology and Histopathology, 8, 3, 581-591.

  • Luskin A.T., Li L. et al. (2022), A diverse network of pericoerulear neurons control arousal states, bioRxiv (preprint).

  • Maynard and Collado-Torres et al. (2021), Transcriptome-scale spatial gene expression in the human dorsolateral prefrontal cortex, Nature Neuroscience, 24, 425-436.

  • Tippani et al. (2022), VistoSeg: processing utilities for high-resolution Visium/Visium-IF images for spatial transcriptomics data, bioRxiv (preprint).

  • Tran M.N., Maynard K.R. et al. (2021), Single-nucleus transcriptome analysis reveals cell-type-specific molecular signatures across reward circuitry in the human brain, Neuron, 109, 3088-3103.

  • Zhao E. et al. (2021), Spatial transcriptomics at subspot resolution with BayesSpace, Nature Biotechnology, 39, 1375-1384.

  1. Howard Hughes Medical Institute
  2. Wellcome Trust
  3. Max-Planck-Gesellschaft
  4. Knut and Alice Wallenberg Foundation