Diversity amongst trigeminal neurons revealed by high throughput single cell sequencing

Abstract

The trigeminal ganglion contains somatosensory neurons that detect a range of thermal, mechanical and chemical cues and innervate unique sensory compartments in the head and neck including the eyes, nose, mouth, meninges and vibrissae. We used single-cell sequencing and in situ hybridization to examine the cellular diversity of the trigeminal ganglion in mice, defining thirteen clusters of neurons. We show that clusters are well conserved in dorsal root ganglia suggesting they represent distinct functional classes of somatosensory neurons and not specialization associated with their sensory targets. Notably, functionally important genes (e.g. the mechanosensory channel Piezo2 and the capsaicin gated ion channel Trpv1) segregate into multiple clusters and often are expressed in subsets of cells within a cluster. Therefore, the 13 genetically-defined classes are likely to be physiologically heterogeneous rather than highly parallel (i.e., redundant) lines of sensory input. Our analysis harnesses the power of single-cell sequencing to provide a unique platform for in silico expression profiling that complements other approaches linking gene-expression with function and exposes unexpected diversity in the somatosensory system.

Fig 1. Unbiased clustering identifies 13 classes of trigeminal sensory neurons.

(a) tSNE representation of 6998 STAMPS from Dropseq analysis of dissociated trigeminal ganglia separates clusters of sensory neurons (red) enriched in markers like Tubb3 and Scn9a from other associated cells including Epcam-positive epithelial cells (green) and cells producing components of myelin (e.g., Plp1 and Mbp, blue). (b) Trigeminal neuron STAMPs were re-clustered using a smart local moving algorithm to identify 15 potential classes of neurons. Nodes marked in black separated groups of neurons that did not express distinguishing markers; thus this analysis identified 13 clusters, which we have numbered according to their expression profiles and position in (c) the tSNE representation, where cells are color coded according to their cluster number.

Fig 2. Classes of trigeminal neurons are distinguished by differential expression of many genes.

(a) Heatmap showing the relative expression of a variety of markers in the trigeminal neuron STAMPS that have been ordered by cluster (numbered underneath the heatmap). Yellow represents high expression and purple low expression; the broad categories of neurons based on markers or function and referred to in the text are indicated above the heatmap. The list of genes used for this analysis is provided in S2 Table. (b) The expression profiles of select genes highlighted in this study are indicated in the tSNE-distribution of trigeminal neuron STAMPs with relative expression level color coded from grey to red. This representation emphasizes that genes like Nppb, Mrgpra3 and Mrgprd serve as markers of individual classes of neuron, whereas genes like Trpv1, Piezo2 and S100b are expressed in several different neural clusters.

Fig 3. In situ hybridization localization of Trpv1, Trpm8 and Piezo2 in trigeminal neurons validates important predictions from sc-transcriptome analysis.

Comparison of adjacent sections through the trigeminal ganglion (a, b) using single label ISH reveals (b) that a mixed probe for Trpv1, Trpm8 and Piezo2 labels essentially all the neurons detected in (a) using a probe to the highly expressed pan-neuronal marker Tubb3. (c) Three color, triple-label ISH demonstrates that Trpv1, Trpm8 and Piezo2 each mark large populations of neurons that do not express the other two transcripts, as predicted by Dropseq analysis (Fig 2b and S1 Fig). In keeping with the single cell data we also observed a subset of neurons that express both Trpv1 and Piezo2 (examples indicated by arrowheads) as well as neurons expressing both Trpv1 and Trpm8 (arrows); 165 of 718 Trpm8 and 1431 Trpv1 neurons across 29 sections were double positive. Scale-bars (a) 100 μm; (c) 50 μm.

Fig 4. ISH localization of Nppb, Mrgpra3, Trpv1 and Piezo2 highlight the difficulty of assessing penetrance of low to moderate level gene expression in sc-transcriptome data.

Representative sections through trigeminal ganglia subjected to three color, triple label ISH for (a) Nppb, Mrgpra3 and Trpv1 illustrate that all neurons expressing Nppb (blue) are also positive for Trpv1 (green); in contrast only a subset of Mrgpra3 neurons (red) co-express Trpv1; arrows indicate Mrgpra3 neurons not expressing Trpv1. (b) Trigeminal ganglion neurons expressing Mrgpra3 (red) all co-express Piezo2 (green); most Nppb (blue) also co-express Piezo2; arrowhead highlights a single Nppb positive, Piezo2 negative neuron. Note that Nppb and Mrgpra3 label almost completely non-overlapping subsets of neurons; scale bar 50 μm.

Fig 5. ISH localization markers reveals bias in representation of some cell classes in the Dropseq data.

Triple label ISH of sections from the trigeminal using probes for Trpv1 (green) and Trpm8 (blue) together with (a) S100b, (b) Cd34 and (c) Mrgprd (red) support extensive segregation of these three markers from Trpv1 and Trpm8 in trigeminal neurons in keeping with their expression profiles in non-overlapping clusters of neurons. However, the ISH data illustrate discrepancies between the relative numbers of hybridizing cells with probes to S100b and Mrgprd and the representation of STAMPs expressing these markers in Dropseq data. Cell counts are quantitated in S3 Table and demonstrate that STAMPs expressing Mrgprd and S100b are underrepresented in the Dropseq data.

Fig 6. Expression of marker genes that differ between DRG and trigeminal ganglia.

(a) Expression profiles of genes in trigeminal Dropseq data are shown in tSNE plots. In the trigeminal ganglion, putative affective touch neurons (red circles) expressing high levels of the mechanosensory ion-channel Piezo2 but low levels of Nefh are marked by Cd34 and P2ry1 but not by Slc17a8 orTh. Similarly the Mrgpra3 cluster (C12, blue circles) is not enriched in Mrgprx1 expression. Black circles highlight a class of large diameter (Nefh, high), mechanosensory (Piezo2, high) neurons that co-express Calca. (b-d) Representative sections of DRG (upper panels) and the trigeminal ganglion were subjected to ISH. (b) Single label ISH confirms that Th is prominently expressed in many DRG neurons but is only found in a much smaller subset of trigeminal cells. (c) Double label ISH reveals extensive co-labeling of DRG neurons expressing Cd34 (red) with Th (green) but demonstrates that the majority of trigeminal Cd34 neurons do not contain detectable Th. (d) ISH demonstrates that Mrgprb4 is prominently expressed in a small subset of DRG neurons (arrows), with weaker expression in many additional cells (upper panel). By contrast, Mrgprb4 is only weakly expressed in a small subset of trigeminal neurons. Scale bars; (c) 50 μm; (b, d) 100 μm.

Fig 7. Trpm8 neurons have related expression profiles in the DRG and trigeminal ganglion.

(a-c) Representative sections of DRG (upper panels) and trigeminal ganglion (lower panels) probed by double label ISH illustrate co-expression of Trpm8 (red) with (a) Trpv1, (b) Foxp2 and (c) Gpr26 (green). Note the similarity of expression patterns between the different somatosensory ganglia: (a) there is limited overlap between Trpv1 and Trpm8 in both ganglia (arrows) with 28 double positive in 534 Trpv1 and 265 Trpm8 cells across 7 sections in the DRG (5% of Trpv1 and 11% of Trpm8 neurons); for comparison a higher double positive ratio was observed in the trigeminal ganglion (12% of Trpv1 and 23% of Trpm8 cells, Fig 3; these differences are significant (p < 10⁻⁴, two tailed z-test). (b) Foxp2 is expressed with high specificity but at a low level in Trpm8 cells in both types of ganglion; (c) Gpr26 is another good marker of Trpm8 cells in both ganglia but is also expressed in cells that do not contain the cooling sensitive ion-channel (arrowheads); scale bar 50 μm. (d) Expression profiles of Foxp2 and Gpr26 in STAMPs indicated in tSNE (upper panels) and violin plots (lower panels) demonstrate that Dropseq analysis closely matches the ISH localization data.

Fig 8. Co-expression patterns predicted by trigeminal Dropseq are found in DRG as well as trigeminal ganglia.

(a) Relative expression level of genes in positive STAMPs plotted against each other and color coded according to cluster highlights co-expression of Trpv1 and Trpa1 in cluster C8 and co-expression of Thy1 and Calca in cluster C6 with far less co-expression outside these cells (see also Fig 2 and S4 Fig). (b,c) Representative sections of trigeminal (upper panels) and DRG (lower panels) probed by in situ hybridization illustrate (b) that a subset of Trpv1 neurons (green) express high levels of Trpa1 (red). However, many neurons, strongly positive for Trpv1 are Trpa1 negative in both types of ganglion. Note that a large subset of neurons expressing Mrgprd (blue) express a lower level of Trpa1 and that a smaller subset are also positive for Trpv1 in both trigeminal and DRG. A separate group of cells that do not express Trpv1 or Mrgprd are weakly positive for Trpa1 (arrows in the trigeminal ganglion) as predicted from the sc-transcriptome analysis (see also S1 Fig); these neurons may be less prominent in the DRG. (c) Thy1 (green) and Calca (red) are co-expressed primarily in large neurons in both the DRG and trigeminal ganglion; scale bar 50 μm.