Reinvestigation of Classic T Cell Subsets and Identification of Novel Cell Subpopulations by Single-Cell RNA Sequencing
Curation statements for this article:-
Curated by eLife
Evaluation Summary:
This manuscript is of broad interest to readers interested in heterogeneity in immune cell populations with single-cell RNA sequencing, and for students of human T cell biology. It uses and reanalyses published single-cell RNA sequencing data dataset for this purpose. However, it does not adequately address major technical concerns, and therefore the interpretations are not robustly supported.
(This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. The reviewers remained anonymous to the authors.)
This article has been Reviewed by the following groups
Listed in
- Evaluated articles (eLife)
- Genetics and Genomics (eLife)
Abstract
Classic T cell subsets are defined by a small set of cell surface markers, while single-cell RNA sequencing (scRNA-seq) clusters cells using genome-wide gene expression profiles. The relationship between scRNA-seq clustered populations (scCPops) and cell surface marker–defined classic T cell subsets remains unclear. In this article, we integrated six bead-enriched T cell subsets with 62,235 single-cell transcriptomes from human PBMCs and clustered them into nine scCPops. Bead-enriched CD4+/CD45RA+/CD25− naive T and CD8+/CD45RA+ naive T cells were mainly clustered into their scCPop counterparts, while cells from the other T cell subsets were assigned to multiple scCPops, including mucosal-associated invariant T cells and NKT cells. The multiple T cell subsets forming one scCPop exhibit similar expression patterns, but not vice versa, indicating scCPop is a more homogeneous cell population with similar cell states. Interestingly, we discovered and named IFN signaling–associated gene (ISAG) high T (ISAGhi T) cells, a T cell subpopulation that highly expressed ISAGs. We further enriched ISAGhi T cells from human PBMCs by FACS of BST2 for scRNA-seq analyses. The ISAGhi T cell cluster disappeared on t-distributed stochastic neighbor embedding plot after removing ISAGs, whereas the ISAGhi T cell cluster showed up by analysis of ISAGs alone, indicating ISAGs are the major contributor of the ISAGhi T cell cluster. BST2+ and BST2− T cells showing different efficiencies of T cell activation indicate that a high level of ISAGs may contribute to quick immune responses.
Article activity feed
-
Evaluation Summary:
This manuscript is of broad interest to readers interested in heterogeneity in immune cell populations with single-cell RNA sequencing, and for students of human T cell biology. It uses and reanalyses published single-cell RNA sequencing data dataset for this purpose. However, it does not adequately address major technical concerns, and therefore the interpretations are not robustly supported.
(This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. The reviewers remained anonymous to the authors.)
-
Reviewer #1 (Public Review):
The authors re-analyze publicly available single-cell transcriptomes of 6 bead-enriched T cell subsets and show that unbiased single-cell clustering analysis identifies distinct subsets that are not defined in the 6 bead-enriched T cell subsets. They describe a "new" IFNhi T cell subset and characterize this subset based on expression of BST2 surface marker. There are several critical concerns with the significance, innovation, analysis approach and interpretation of data presented in this manuscript.
Significance: The main conclusion from the authors is that single-cell transcriptome-based clustering is better than protein-marker based classification of T cell subsets. Unfortunately, the 6 bead-enriched T cell subsets are not very well resolved in the first place. For example, CD4Th subset is called as one …
Reviewer #1 (Public Review):
The authors re-analyze publicly available single-cell transcriptomes of 6 bead-enriched T cell subsets and show that unbiased single-cell clustering analysis identifies distinct subsets that are not defined in the 6 bead-enriched T cell subsets. They describe a "new" IFNhi T cell subset and characterize this subset based on expression of BST2 surface marker. There are several critical concerns with the significance, innovation, analysis approach and interpretation of data presented in this manuscript.
Significance: The main conclusion from the authors is that single-cell transcriptome-based clustering is better than protein-marker based classification of T cell subsets. Unfortunately, the 6 bead-enriched T cell subsets are not very well resolved in the first place. For example, CD4Th subset is called as one subset, standard FACS analysis with 10-15 markers would resolve this subset into >10 distinct subsets like Th1, Th2, Th17, Th1/17, Tfh, Tfr, naïve and memory Treg, CD4-CTL, TEMRA, TCM, naïve etc. Unless the authors show that single-cell RNA-seq can resolve T cell subsets better than standard cytometry, I don't see the value/significance of this study.
The authors should combine CITE-Seq with 20 markers and single-cell RNA-seq to definitively show that single-cell RNA-seq is a better at resolving T cell subsets than standard methods.
Innovation: There is very little innovation in this study. None of the subsets resolved are new, including the rare IFNhi subset (Tibitt at all, 2019, Seumois et al, 2020, Meckiff et al, 2020). The methodological approaches are not novel. Data set analyzed in a public resource that does not use the latest high-resolution 10x methods..
Methodology and Interpretation.
• Datasets used for analysis: The markers used to classify the 6 pre-defined subsets are not optimal. For example, TREG cells cannot be defined as CD4+CD25+ (additional CD127 marker is necessary to resolve properly). Similarly, naïve CD4 or CD8 T cells cannot be resolved by RA+ cells, additionally CCR7 is needed, otherwise TEMRA cells that express RA will be erroneously labelled as naïve cells. The biggest problem is that their initial cell classification based on markers is suboptimal and will lead to incorrect classification. This problem cannot be resolved by using this dataset for analysis.• Clustering analysis: Details need to be provided for rigor and reproducibility concerns. Number of genes used to generate the clusters, number of PCA, what resolution and perplexity were used? Published data on single cell analysis of PBMC usually showed clearer separation between cluster and cell subsets. Also, it is very conventional to provide a heatmap showing the number of signature genes. The manuscript only lists a few genes for each T cell subsets, are those gene the most differentiated genes? Also, proportion of cells between clusters and within cluster would need to be provided as a figure for clarity. As for example, the % of CD8 MAIT cells appears to be high, is this from the result of a specific enrichment? Or some sort of technical artefact?
• Activation, authors show that BST2+ cells get activated by showing increased expression of CD25. However, authors failed to show their claims about fast activation. BST2 is usually described to be expressed by regulatory T cells. One would suggest to show activation in a time course of CD25 induction or better other activation markers such as CD69, or CD154.
-
Reviewer #2 (Public Review):
Wang et al investigated classic T cell subsets with scRNA-seq data set from classic staining defined population groups, identified novel cell subpopulations, and further evaluated its function. In brief, they had found there are 9 groups of T cells with scRNA-seq from 6 classic staining defined groups, where they identified a new subpopulation IFNhi T cell group. They had further proved ISAGs are major contributor of IFNhi T cells with aid of FACS sorted scRNA-seq.
The conclusions of this paper are mostly well supported by their data and approaches,
but some of their observations and claim needed to be further extended and discussed.1), The 9 clusters of 6 bead-enriched T cells were identified with Seurat package. To make their conclusion more solid, the authors should use other Independent approach to …
Reviewer #2 (Public Review):
Wang et al investigated classic T cell subsets with scRNA-seq data set from classic staining defined population groups, identified novel cell subpopulations, and further evaluated its function. In brief, they had found there are 9 groups of T cells with scRNA-seq from 6 classic staining defined groups, where they identified a new subpopulation IFNhi T cell group. They had further proved ISAGs are major contributor of IFNhi T cells with aid of FACS sorted scRNA-seq.
The conclusions of this paper are mostly well supported by their data and approaches,
but some of their observations and claim needed to be further extended and discussed.1), The 9 clusters of 6 bead-enriched T cells were identified with Seurat package. To make their conclusion more solid, the authors should use other Independent approach to check whether their conclusion is robust or not.
2), They found Bead-enriched CD4 Naïve and CD8 Naïve were mainly clustered into their scCPop counterparts, while cells from the other T cell subsets were assigned to multiple scCPops including mucosal-associated invariant T cells and natural killer T cells. Their results indicate the different group assessments from protein staining comparing with RNA expression, which demonstrates importance of using joint molecule profiles (RNA expression, protein, and others together ) to define single cell status. Using joint profiles to define single cell status becomes possible now, such as Cite-seq and others, and it will be very interesting to discuss this point in the manuscript.
-
Reviewer #3 (Public Review):
T lymphocytes are essential players of adaptive immunity and the extent of their functional, transcriptomic, and immunophenotypic heterogeneity is still unknown. Wang et al. reanalyzed previously published single-cell RNAseq data of sorted T cell populations from human peripheral blood (Zheng et al. Nature Communications, 8:14049, 2017). They described their composition and overlaps, and highlight a subpopulation of T cells with high interferon-responsive gene expression. They identified BTS2 as a candidate surface marker to enrich them, only to show low sensitivity and specificity.
The manuscript shows significant weaknesses:
- Analyzing the heterogeneity within sorted populations of T cells is not a novel approach. The authors should cite previous attempts by single-cell RNA sequencing to characterize T cell …
Reviewer #3 (Public Review):
T lymphocytes are essential players of adaptive immunity and the extent of their functional, transcriptomic, and immunophenotypic heterogeneity is still unknown. Wang et al. reanalyzed previously published single-cell RNAseq data of sorted T cell populations from human peripheral blood (Zheng et al. Nature Communications, 8:14049, 2017). They described their composition and overlaps, and highlight a subpopulation of T cells with high interferon-responsive gene expression. They identified BTS2 as a candidate surface marker to enrich them, only to show low sensitivity and specificity.
The manuscript shows significant weaknesses:
- Analyzing the heterogeneity within sorted populations of T cells is not a novel approach. The authors should cite previous attempts by single-cell RNA sequencing to characterize T cell heterogeneity (PMIDS: 30664737, 31371561, 30958799, 31227543, 29858286, 29352091, 30737144, 29434354, etc.).
- Regarding the dataset, the authors used a previously published dataset from 2017 of a single donor (Zheng et al. Nature Communications, 8:14049, 2017). Several donors would be needed in order to test the statistical significance and robustness of the results.
- The technology has largely improved since the publication of this dataset (in terms of sequencing depth, multimodal analysis). Sequencing depth, in particular, is an important parameter in cluster definition (Heimberg Cell Syst 2016, PMID 27135536). The authors cluster single-cell data without justifying the robustness of the clustering and discussing the role of sequencing depth. Newer datasets have now been published with better resolution and discrimination between T cell populations. Considering the "blob" appearance of the data, one could also argue that discrete clustering is somewhat arbitrary.
- Single-cell clusters overlap between sorted populations. The authors fail to discuss the various intricated technical explanations 1) cell sorting impurity 2) imperfect protein-RNA expression correlation 3) arbitrary cluster borders in the single-cell data in a tSNE "blob" 4) role of the depth of the sequencing.
- Pseudotime analysis: the authors misinterpret transcriptome overlap as temporal dynamics. Considering the highly polyclonal repertoire of peripheral T cells, the studied T cells do not represent temporal evolutions of single clones. As an example, it is well known that circulating Tregs do not differentiate from other circulating T cells and cannot be put in a trajectory.
- Their authors discuss the discovery of cytotoxic CD4 and CD8 T cells. These cells have however been previously described (Patil et al. Science Immunol 2018 PMID 29352091). Moreover, the interpretation that these might represent "convergent differentiation" is inappropriate: CD4 and CD8 T cells are restricted by MHC I and II, respectively, and therefore these cytotoxic cells are not a single population. It merely represents a shared expression of cytotoxic genes.
- scIFNhi cells are been previously described in various datasets [REF]. the authors should address and/or discuss the various intricated technical explanations: do they represent technical artifacts (sample processing/cellular stress)? Are they reproduced in different donors? Different conditions? The authors identified BTS2 as a candidate surface marker to enrich them, only to show low sensitivity and specificity, so its significance is unclear. The authors claim a higher rate of T cell activation in BST2+ cells but the in vitro experiment does not have replicates and therefore cannot be statistically interpreted.
- The manuscript is sometimes unclear. It is not clearly stated that the dataset represents human cells (it is only mentioned in the method section). Importantly, an effort should be made to recognize the difference between the functional classification of T cells, the classification based on surface markers, and the classification based on clustering by single-cell transcriptomics. Cytotoxic T cells, helper, T cells memory T cells are functional concepts that can only be evaluated by experimentation (cell transfer, chimeras, infection models, etc). However, the authors infer functions to T cell clusters (memory, naïve, effector, etc.) based on a few pre-selected markers in the RNAseq data without further investigations or whole transcriptome comparison (enrichment in gene expression signatures for examples).
-
-
