A call for a unified and multimodal definition of cellular identity in the enteric nervous system

Abstract

The enteric nervous system (ENS), the largest division of autonomic nervous system, is a tantalizing frontier in neuroscience. With the advent of single-cell transcriptomics, the ENS has been increasingly well-characterized. Precise functional mapping of enteric neuron diversity is critical for understanding ENS biology and disease, but technical barriers remain. We used different approaches to compare and contrast functional annotations of several independently-reported ENS datasets. Differential module scoring, co-expression and correlation analysis, unbiased biological function hierarchical clustering, data integration and label transfer highlighted substantial discrepancies stemming from an overreliance on transcriptomics data without adequate tissue validations. For understanding enteric neurons’ functional identity, it is imperative to expand tissue sources and incorporate technologies such as multiplexed imaging, electrophysiology, spatial transcriptomics, as well as comprehensive epigenome, proteome, and metabolome profiling. Harnessing human pluripotent stem cell models provides unique opportunities for ENS lineage tracing and offers unparalleled scalability and amenability to genetic and functional screens. We encourage a paradigm shift in our comprehension of ENS cellular and functional complexity by calling for large-scale collaborations and research investments.

Keywords: Comparative Analysis; Neurochemical Coding; Neuronal Classification.

Figure 1. Cross-dataset expression of primary enteric neuron cluster-specific markers.

(A) Total number and distribution of enteric neuron cluster annotations Fig. EV1A–D. (B) Venn diagram indicating shared markers used by each study. (C) Bar plot proportion of different functional annotations described for TAC1+ enteric neurons across primary ENS datasets.

Figure 2. Comparison of enteric neuron cell types and subtypes in primary enteric datasets.

(A) Schematic of 2-step neurochemical identity annotation of hPSC-derived enteric neurons. (B) Overall percentage of neurotransmitter-synthesizing neurons in mouse and human primary enteric neurons. (C, D) Schematic (C) and percentage (D) of neurons showing mono-and multi-neurotransmitter profiles in mouse and human primary enteric neurons. (E, F) RNA labeling representative images (E) and quantification (F) analysis of primary human stomach with probes against SYP, NOS1 and CHAT. Arrows indicate exclusive expression and arrowheads indicate colocalized expression of NOS1 and CHAT. (G) Immunostaining of primary human stomach with antibodies against NOS1, 5-HT, and HuC/D (top), and NOS1, GLS, and HuC/D (bottom).

Figure 3. Unbiased cross-dataset classification of primary enteric neurons using SingleCellNet.

(A) Schematics of unbiased label transfer using SCN (Tan and Cahan, 2019). (B–C) Reference primary enteric neuron scRNA-seq datasets of AR-mouse (B) and ST-human (C) were used to train SingleCellNet (Tan and Cahan, 2019). These models were then used for label transfer and cross-annotation in the other datasets. Please see Methods for more details. The abbreviations (consistent with the commonly used denotations in the field and in the original papers): IMN (inhibitory motor), EMN (excitatory motor), IN (interneuron), IPAN (intrinsic primary afferent), PSVN (putative secretomotor/vasodilator), and SN (sensory). The inclusion of “P” in the AR-mouse and AR-human indicates “putative” as originally termed.

Figure 4. Cross-dataset module scoring of primary enteric neuron clusters.

(A) Heatmap of the average module scores of AR-human, UM-mouse, and AR-mouse neuronal subtype transcriptional signatures in ST-human. Please see Methods for more details. Please see Methods for more details.

Figure 5. Comparative analysis of primary ENS neurons using GOBP hierarchical clustering.

Hierarchical clustering of primary enteric neuron clusters based on normalized enrichment scores of biological process gene ontology (GOBP) pathways. (A) Four main clusters each contain multiple enteric neuron clusters from different datasets. Blue boxes indicate closely clustered neuronal subtypes with matching functional annotation from two different datasets.

Figure 6. Transcriptional identities are not synonymous with functional identities in enteric neurons.

(A) Schematic illustrating the contrast between the assumed clustering of functional classes of enteric neurons in transcriptomic datasets and the organization suggested by observed transcriptional profiles.

Figure 7. Enteric neuron identity should be defined based on multiple biological features.

(A) Multimodal characterization is essential for understanding enteric neuron function. Accurate classification requires integrating transcriptomic data with additional biological features, such as morphology, electrophysiology, metabolic and signaling profiles, and interactions with other cell types.

Figure EV1. Expression of cluster-specific markers across primary enteric neuron clusters.

(A–D) UMAPs of enteric neurons generated from the original datasets of UM-mouse (A), AR-mouse (B), AR-human (C), and ST-human (D). “?” refers to the cluster labeled as “ENC11” or “?” in Morarach et al (Morarach et al, 2021b). (E–H) Dot plot of cluster-specific markers originally used for UM-mouse, (E) in UM-mouse, (F) in AR-mouse, (G) in ST-human, (H) in AR-human. (I–L) Dot plot of cluster-specific markers originally used for ST-human, (I) in UM-mouse, (J) in AR-mouse, (K) in ST-human, (L) in AR-human. (M–P) Dot plot of cluster-specific markers originally used for AR-mouse and AR-human, (M) in UM-mouse, (N) in AR-mouse, (O) in ST-human, (P) in AR-human. (Q, R) Bar plot proportion of different functional annotations described for (Q) NOS1+ and (R) PENK+ enteric neurons across primary ENS datasets.

Figure EV2. Neurochemical identities in primary enteric neuron cell types and subtypes.

(A) Percentage of single-neurotransmitter-producing enteric neurons in primary datasets. (B) Percentage of double-neurotransmitter-producing enteric neurons in primary datasets. (C–F) Distribution of neurochemical identities in primary mouse (UM-mouse (C), AR-mouse (F) and human (ST-human (D), AR-human (E)) enteric neuron subtypes.

Figure EV3. SingleCellNet unbiased label transfer and classification of primary enteric neurons.

(A–J) Reference primary enteric neuron scRNA-seq datasets of mouse (UM-mouse (A, G, I), AR-mouse (H, J)) and human (ST-human (C, E), AR-human (B, D, F)) were used to train SingleCellNet (Tan and Cahan, 2019). These models were then used for label transfer and cross-annotation in the other datasets. Please see Methods for more details.

Figure EV4. Cross-dataset module scoring, Spearman correlation of transcriptional signatures and Harmony integration of primary enteric neuron clusters.

(A) Heatmap of the average module scores of ST-human and AR-human neuronal subtype transcriptional signatures in UM-mouse. (B) Heatmap of the ST-human, AR-human and UM-mouse neuronal subtype transcriptional signatures in AR-mouse. (C) Heatmap of the average module scores of ST-human and UM-mouse neuronal subtype transcriptional signatures in AR-human. (D, E) Heatmap matrix of Spearman correlations based on expression of 100 anchor features shared significantly variable genes (or anchor features) between (D) primary human (ST-human and AR-human) and (E) primary mouse (UM-mouse AR-mouse) enteric neuron subtypes. (F) Schematic representation of Harmony integration of UM-mouse and AR-mouse datasets. (G, H) Distribution of cells derived from UM-mouse and AR-mouse datasets (G) and their respective broad functional annotations in each Harmony cluster (H). (I) Schematic representation of Harmony integration of ST-human and AR-human datasets. (J, K) Distribution of cells derived from ST-human and AR-human datasets (J) and their respective broad functional annotations in each Harmony cluster (K). Please see Methods for more details.

Figure EV5. Hierarchical clustering and comparative analysis of primary ENS neurons.

Hierarchical clustering of primary enteric neuron clusters based on normalized enrichment scores of biological process gene ontology (GOBP) pathways. (A) Average enrichment scores of clusters are shown in Fig. 5A. (B, C) Five representative pathways that show higher enrichment scores in clusters 3 (B) and cluster 4 (C), respectively. Please see Methods for more details.