A comparative single-cell transcriptomic atlas for diverse populations of vertebrate hair cells
Curation statements for this article:-
Curated by eLife
eLife Assessment
This important study provides an accessible comparative atlas of gene activity in sensory hair cells across four vertebrate species. The evidence is convincing, supported by comparative analyses and experimental validation, although technical differences between datasets and unclear definitions of cell maturity limit some interpretations. The resource will be of interest to researchers studying hearing, balance, sensory cell development and regeneration, and evolutionary biology.
This article has been Reviewed by the following groups
Listed in
- Evaluated articles (eLife)
Abstract
Mechanosensitive hair cells vary widely in morphology and regenerative capacity across vertebrate organs and species. To investigate their underlying transcriptomic diversity, we integrated human, mouse, chicken, and zebrafish single-cell and single-nucleus RNA sequencing datasets and assembled a cross-species atlas of hair cells spanning organs, developmental stages, and species. Analysis of 29 hair cell populations, encompassing the major cochlear, vestibular, and lateral-line hair cell types, identified approximately 5,000 genes enriched in at least one hair cell population compared to supporting cells from the same organs. Unsupervised clustering of these hair cell-enriched (HCE) genes defined species-, organ-, and hair cell state–associated cohorts as well as broadly conserved hair cell-enriched programs. Using an AUC-based scoring framework, we further defined 884 pan hair cell-enriched (pan-HCE) genes with elevated expression in most developing and/or mature hair cell populations, including genes implicated in deafness, mechanotransduction, and synaptic transmission, along with genes not previously linked to hair cell function. Independent analysis of developing hair cells using the same metrics stratified pan-HCE genes based on when they are first enriched and identified an additional 97 genes that are transiently enriched. We used the pan- and developing HCE gene sets to assess transcriptional similarity between baseline hair cell states and hair cells produced during avian hair cell regeneration and in mouse cochlear organoids, as well as hair cell-like populations produced by fibroblast reprogramming. HCE gene sets with different developmental dynamics identified young vs. more mature hair cells when projected onto independent single-cell RNA sequencing datasets from developing zebrafish, mouse, and human. We provide a web-based resource of all HCE metrics and expression profiles, enabling future exploration of vertebrate hair cell gene expression across organs, species, and experimental contexts.
Article activity feed
-
eLife Assessment
This important study provides an accessible comparative atlas of gene activity in sensory hair cells across four vertebrate species. The evidence is convincing, supported by comparative analyses and experimental validation, although technical differences between datasets and unclear definitions of cell maturity limit some interpretations. The resource will be of interest to researchers studying hearing, balance, sensory cell development and regeneration, and evolutionary biology.
-
Reviewer #1 (Public review):
Summary:
"A comparative single-cell transcriptomic atlas for diverse populations of vertebrate hair cells" by Basu et al. aims to define pan-hair cell enriched (HCE) genes across hair cell subtypes from four vertebrate species through integration of published scRNA-seq, SMART-seq, and single-nucleus RNA-seq datasets. Pan-HCE genes are weighted using AUC values, where AUC is used to identify genes enriched in hair cells relative to supporting cells. The resulting dataset is made publicly available, allowing researchers to compare gene expression across species. The dataset integration and creation of a public resource are clear strengths for the inner ear community.
Strengths:
This manuscript presents a valuable cross-species resource and a substantial integration effort.
Weaknesses:
Weaknesses addressed by …
Reviewer #1 (Public review):
Summary:
"A comparative single-cell transcriptomic atlas for diverse populations of vertebrate hair cells" by Basu et al. aims to define pan-hair cell enriched (HCE) genes across hair cell subtypes from four vertebrate species through integration of published scRNA-seq, SMART-seq, and single-nucleus RNA-seq datasets. Pan-HCE genes are weighted using AUC values, where AUC is used to identify genes enriched in hair cells relative to supporting cells. The resulting dataset is made publicly available, allowing researchers to compare gene expression across species. The dataset integration and creation of a public resource are clear strengths for the inner ear community.
Strengths:
This manuscript presents a valuable cross-species resource and a substantial integration effort.
Weaknesses:
Weaknesses addressed by the authors in the text include the lack of human cochlear scRNA-seq comparisons to mouse/chicken cochlear hair cells. While the rationale that the datasets were not available at the time of investigation is acceptable, the use of human cochlear datasets would be the ideal experimental comparison to mouse and chicken cochlear hair cells.
-
Reviewer #2 (Public review):
In this manuscript, the authors investigate transcriptional diversity and similarities between mechanosensory hair cells across different species to provide an atlas of hair-cell-enriched genes.
Their analysis encompasses publicly available and a few newly generated single-cell RNA-seq data from human, mouse, chick and zebrafish, in total comprising 43 populations of hair cells from different organs (cochlea, vestibular sense organs, lateral line) and different stages. They first compare the transcriptome of each hair cell population with that of supporting cells from the same data to find hair-cell-enriched genes. To do this, they use a score (AUC) that reflects how well a gene separates both cell populations; this approach identifies about 5000 genes enriched in at least one of 29 mechanosensory …
Reviewer #2 (Public review):
In this manuscript, the authors investigate transcriptional diversity and similarities between mechanosensory hair cells across different species to provide an atlas of hair-cell-enriched genes.
Their analysis encompasses publicly available and a few newly generated single-cell RNA-seq data from human, mouse, chick and zebrafish, in total comprising 43 populations of hair cells from different organs (cochlea, vestibular sense organs, lateral line) and different stages. They first compare the transcriptome of each hair cell population with that of supporting cells from the same data to find hair-cell-enriched genes. To do this, they use a score (AUC) that reflects how well a gene separates both cell populations; this approach identifies about 5000 genes enriched in at least one of 29 mechanosensory populations. Grouping these genes by their enrichment pattern gives 22 clusters of co-regulated genes, many species- or organ-specific and a few shared by most hair cells.
Compared with previously published lists of hair cell genes, they define 884 "pan-HCE" genes enriched in most hair cells across species. This set is enriched for stereocilia and deafness genes. In situ hybridization is presented to support enrichment for 7 of 10 tested genes.
Adding 14 immature populations splits the list into early-and-sustained, transient and late genes. The authors then use these to assess the maturity of hair cells from cochlear organoids, regenerating chicken basilar papilla, reprogrammed fibroblasts, and developing zebrafish, mouse and human tissue. All data are released through a Shiny app and the gEAR portal.
Overall, this is a useful resource for the community, presented in an exploitable format. It is timely; the inclusion of many datasets is very valuable, and the comparison of HCs vs SCs from each organ is a good approach. I have some concerns regarding the clarity in describing the analysis, whether some findings could reflect differences between datasets, and how cell 'maturity' is defined.
Major comments:
(1) The data were generated by different labs, different methods and sequencing platforms. Hierarchical clustering of the data splits cells mainly by species. It is not clear to me that the authors have ruled out that this indeed reflects biology or could be due to technical differences. This needs to be discussed. Is it possible to show that the results hold for data generated with the same platform?
(2) The calculation of enrichment scores needs clarification. In the methods, it says that AUC values below 0.5 were set to 0.5 and then supporting cell-enriched genes were set to 1 − AUC. How do supporting cell-enriched genes like ID2 end up with values like 0.11 (Figure 2C)? Please describe the calculation step by step; how were skipped genes because of low expression handled?
(3) It is not clear how duplicated genes were handled when assigning orthologues; the approach allows one-to-many. How were they combined into a single score? The results say that each human gene was linked separately to each ortholog?
(4) The reference gene list for known hair cell genes seems to overlap with genes reported to be enriched; this seems to be a circular argument. Among the hand-curated list are stereocilia-associated genes, which are then reported to be enriched. Is it possible to test enrichment using gene sets that were not used to choose the threshold?
(5) The authors need to define how they classify cells as mature, maturing, and immature - there are no definitions at all. Provide these and appropriate references. Mouse P7 cells are classified as mature, which they are not. Later data depend on this classification, so it needs to be justified properly.
(6) The authors assess maturity of hair cells in organoids; it seems that this depends on the % of genes in each set (mature, maturing, immature) that pass the cutoff. Has this been compared to a random or control set?
(7) Verification of hair cell enrichment by HCR/RNAscope: This verification is not very convincing for most genes tested. 30% of the genes tested fail, while genes described to be HC-specific are also expressed in surrounding cells. E.g. Figure 5C: both Apba1 and Pcsk5 seem to be expressed in hair cells as well as the surrounding cells; Figure 5D: both genes are not expressed in all hair cells. Figure 5 supplement 2B: the images of the organ of Corti are too low resolution and magnification to identify hair cells reliably. It is not possible to see if these genes are indeed expressed in hair cells. If the authors claim HC enrichment, they need to provide quantification of expression levels e.g. HCs vs SCs and show negative controls.
(8) A new single-cell dataset from human embryos has just become available; van der Valk et al., 1026, Nat. Neuroscience. This dataset fills an important gap in their current analysis, and it would be very good to integrate it.
(9) The figure legends are often a bit short and do not explain the figures well; please review and expand.
-
Reviewer #3 (Public review):
Summary:
The authors generated a cross-species single-cell transcriptomic atlas of inner ear and lateral line organs across four species (human, mouse, chicken, zebrafish) and 11 developmental stages. By comparing mechanosensitive hair cells to surrounding supporting cells, they identified hair cell-enriched (HCE) genes, conserved transcriptional programs, and stratified developmental markers. They used these core and temporally stratified gene sets to evaluate hair cell identity and maturation state in models of avian regeneration, mouse cochlear organoids, and transcription factor-reprogrammed fibroblasts. A publicly available web-based resource (HRP Hair Cell Gene Explorer and gEAR portal) was created to facilitate field-wide data exploration.
Strengths:
(1) Comprehensive Multi-Species Benchmark:
Cross-spe…
Reviewer #3 (Public review):
Summary:
The authors generated a cross-species single-cell transcriptomic atlas of inner ear and lateral line organs across four species (human, mouse, chicken, zebrafish) and 11 developmental stages. By comparing mechanosensitive hair cells to surrounding supporting cells, they identified hair cell-enriched (HCE) genes, conserved transcriptional programs, and stratified developmental markers. They used these core and temporally stratified gene sets to evaluate hair cell identity and maturation state in models of avian regeneration, mouse cochlear organoids, and transcription factor-reprogrammed fibroblasts. A publicly available web-based resource (HRP Hair Cell Gene Explorer and gEAR portal) was created to facilitate field-wide data exploration.
Strengths:
(1) Comprehensive Multi-Species Benchmark:
Cross-species analysis spanning 43 hair cell populations and 11 developmental stages establishes a rigorous reference dataset for benchmarking hair cell identity across development, cell biology, and regenerative models.
(2) Biological Relevance:
The pan-HCE gene set captures 25% of all known human deafness genes, confirming its strong predictive value for hair cell-specific function and disease mechanisms.
(3) Methodological Rigor & Robust Thresholding:
The use of area under the receiver operating characteristic curve (AUC) provides a platform-independent metric. Testing four calculation strategies (combining 29 mechanosensitive vs. 23 mature populations with unweighted [ES1] vs. species-balanced [ES2] weighting) and taking their intersection ensures robust gene selection without species bias.
(4) Temporal Stratification & Application:
Categorizing pan-HCE genes into early, sustained (ES-HCE), late (L-HCE), and transiently enriched (T-HCE) sets enables tracking of maturation state across different animal and cellular models, proving effective in evaluating regenerating avian hair cells, mouse cochlear organoids (Lgr5^+ lineage), and Gfi1/Atoh1/Pou4f3/Six1-reprogrammed fibroblasts.
(5) Experimental Validation:
Key pan-HCE genes were independently validated across species using spatial transcriptomics, RNAscope, and Hybridization Chain Reaction (HCR).
(6) Community Utility:
Broad accessibility via an interactive ShinyApp and gEAR portal integration facilitates customized field-wide analysis.
Weaknesses:
Weaknesses were considered minor.
(1) Interspecies Homology Resolution:
Machine learning-based clustering (MetaNeighbor) showed that factors such as age, species, and sensory modality often override subtype identity, failing to resolve fine-grained cross-species homologies for specific subtypes (e.g., distinguishing mouse inner vs. outer hair cells relative to chicken tall vs. short hair cells).
(2) Species Imbalance & Mapping Artifacts:
Despite weighting schemes (ES2), teleost genome duplications (1-to-many ortholog mapping in zebrafish) complicate cross-species signal normalization. This remains an overarching issue in many fields when performing cross-species analysis.
(3) Different mechanisms of regeneration:
Converted fibroblasts induced only a percentage of pan-HCE genes (28.7%) and very few transient genes (9.3%). This could reflect incomplete programming towards a fully mature native hair cell state, as the authors suggest. It is also possible that direct programming via transcription factor expression or Lgr5+ cell conversion could employ mechanisms distinct from those used in normal development or in regenerative species.
-
-
-
