Decoding Cellular Mechanisms for Mechanosensory Discrimination

Abstract

Single-cell RNA-sequencing and in vivo functional imaging provide expansive but disconnected views of neuronal diversity. Here, we developed a strategy for linking these modes of classification to explore molecular and cellular mechanisms responsible for detecting and encoding touch. By broadly mapping function to neuronal class, we uncovered a clear transcriptomic logic responsible for the sensitivity and selectivity of mammalian mechanosensory neurons. Notably, cell types with divergent gene-expression profiles often shared very similar properties, but we also discovered transcriptomically related neurons with specialized and divergent functions. Applying our approach to knockout mice revealed that Piezo2 differentially tunes all types of mechanosensory neurons with marked cell-class dependence. Together, our data demonstrate how mechanical stimuli recruit characteristic ensembles of transcriptomically defined neurons, providing rules to help explain the discriminatory power of touch. We anticipate a similar approach could expose fundamental principles governing representation of information throughout the nervous system.

Keywords: Piezo2; Touch; functional imaging; mechanosensation; nociception; sensory coding; somatosensation; spatial transcriptomics; trigeminal system.

Conflict of interest statement

Declaration of Interests The authors declare no competing interests.

Published by Elsevier Inc.

Figures

Figure 1.. Aligning in vivo imaging of large ensembles of neurons with whole mount ISH

A) Example images of a dorsal view of the trigeminal ganglion showing red fluorescence in a live Tac1-tagRFP mouse (left panel, red) and whole mount tagRFP ISH from the excised ganglion (middle panel, green). Images were aligned using scaled rotation (right panel); note that only very few cells overlap but other non-aligned pairs of red and green neurons can be identified. B) Images were aligned by manual matching of cell pairs (left panel, arrows) to identify guideposts for Delaunay triangulation of the ISH image (middle panel) and construction of corresponding triangles in the live image (right panel). The aligned region is delimited by the triangles; affine transform of Delaunay triangles (see Figure S1) generates a continuous image (C) where a large area and many neurons are perfectly matched between the live image (red) and ISH (green). D) Imaging from a different mouse showing in vivo fluorescence of tagRFP (top left, red) and GCaMP6f calcium transients (green, see methods for details). The whole mount ISH counterpart (top right; tagRFP, cyan; GCamP6f, magenta) was aligned to the live image using tagRFP guideposts (bottom left). As a result, functional activity maps also matched the GCaMP6f expression (bottom right). Note that as expected only a subset of GCaMP6f expressing neurons responded to mechanical stimulation of the cheek; scale-bars = 100 μm; see also Figure S1.

Figure 2.. Trigeminal neuron responses to naturalistic mechanical stimuli identify clear functional categories

A) Heatmap showing the in vivo GCaMP6f responses from 1840 neurons responding to five types of mechanical stimulation as labeled (top) in 23 mice. Calcium transients were first normalized to the median pinch response in that animal and colored as indicated in the scale bar (inset). The five cell-categories (labelled colored boxes, with cell numbers) were assigned using an automated script and a set of thresholding rules. Note sharp boundaries in response profiles demarcate the categories highlighting robust differences in the tuning of these groups of cells. B) Representative example GCaMP-transients (ΔF/F fluorescence changes indicated by scale bars) for each category of cells showing responses of individual neurons to a series of air-puff durations, vibration frequencies, gentle brush strokes, vigorous hair-pulls and pinching (the order, approximate timing and repetitive nature is schematically represented by number scales or arrows). Note that responses to vibration are somewhat variable in all categories of LTMRs; see also Figure S2.

Figure 3.. Multiplexed ISH aligned to functional imaging maps cellular responses to transcriptomic class

A) Activity maps for 48 responding neurons in a region of the trigeminal ganglion shown in Figure S1C were color coded to distinguish their functional categories. B) Multiplex wholemount ISH was performed and images for 6 probes were aligned to the functional image using tagRFP guideposts; the overlaid 6 images were colored as indicated in the upper panel. Boxes (A) and (B) depict representative cells with a variety of distinct functional responses (A, air-puff cell, M, mixed responder B, brush cell and H, HT-cell) and gene expression patterns; scale bar = 100 μm. C) Activity map (left column) and detailed gene expression patterns for the cells boxed in (A) and (B) illustrate our strategy for assigning cell class to functionally characterized mechanosensory neurons (see also Figure S3). D) Map of transcriptomic classes of the responding cells; note in this imaging field 46 from 48 neurons were classified using this panel of genes (see also Figure S4 for details); the color code for classes is shown in the lower panel. E) GCaMP-transients for the cells boxed in (A) and (B) with transcriptomic class indicated. ΔF/F fluorescence changes are indicated by scale bars and traces are color coded according to transcriptomic class (as specified, D).

Figure 4.. Sweeping assignment of transcriptomic class to function reveals a cellular basis for the detection and coding of mechanosensation at the periphery

A) Heatmaps showing normalized mechanical responses from transcriptomically classified neurons tested with the five stimuli (indicated above the heatmaps) grouped according to transcriptomic class. Data display is as in Figure 2A and functional category color coding is shown in (C); note that mechanical stimuli broadly activate all transcriptomic classes of neurons with the exception of C1/2 cooling sensors (see Figure S5A). Nonetheless, clear functional differences are immediately apparent with most classes selectively tuned to either brush or noxious stimuli. By contrast C4 Aβ-LTMRs are diverse and include the vast majority of the specialized air-puff and vibration cells. B) Example GCaMP-transients of each cell class (ΔF/F fluorescence changes indicated by scale bars; traces color coded according to functional category) highlight their different tuning characteristics. Differential vibration sensitivity and calcium transient dynamics are also class dependent (see Figure S5B-D). C) Quantification of response category across cell-class; data were normalized to the number of responding neurons tested for each class (numbers of class responders / responding cells tested for that class by ISH are also indicated).

Figure 5.. Calretinin (Calb2) expression defines a subtype of Aβ-LTMRs selectively tuned to air-puff and vibration

A) UMap representation of snRNA sequence data (Nguyen et al., 2019) showing that Calb2 expression is restricted to a transcriptomically defined subgroup of C4 Aβ-LTMRs (dark blue); the expression of Calb2 (yellow) is overlaid on the classified cells; see also Figure S6. B-C) Calb2-Aβ-LTMRs are highly selective for air-puff and vibration; B) sample images showing ISH localization of Calb2 (green) overlaid on category specific spatial activity maps (red). Arrowheads highlight the Calb2-expressing cells with air-puff and vibration response profiles; note only a subset of air-puff or vibration cells express this marker and that many Calb2 cells are not stimulated, likely reflecting their projections to different areas of the head and neck. C) Heatmap showing the response profiles of all 37 responding Calb2-Aβ-LTMRs identified by ISH after functional imaging in 10 mice. D) Representative fluorescence micrographs of Calb2-Aβ-LTMR termini in the skin illustrating that these neurons form circumferential endings around small hairs. Neurons were labeled by tdT-expression (Calb2-Cre/Ai9 RCL-tdT mice with appropriate recombination, see also Figure S6); signal is shown in black and contrasts with the weaker autofluorescence of hairs allowing morphology to be clearly discerned. These neuronal termini are not limited to the cheek (left panel) but are also prominent in the nearby whisker pad (center left) and back skin (right panels); scale bars = 100 μm.

Figure 6.. Piezo2 differentially confers mechanosensitivity to LTMR classes

A) UMap representation of snRNA sequence data (Nguyen et al., 2019) showing broad expression of Piezo2 in almost all classes of trigeminal neurons; note that only a limited subset of C1/2 and C7–12 class neurons express this gene; by contrast it is a prominent marker of C6 and C13 nociceptors. B) Representative spatial activity maps showing the functional categories of mechanosensory neurons in equivalent areas of the dorsal trigeminal ganglion in control and Piezo2-KO recordings. Cellular response magnitude (see methods) is indicated by brightness and functional category by color; note how the functional response spectrum is narrowed by Piezo2 knockout. C) Example images showing ISH images (green) aligned to activity maps as indicated by color coding (top four panels) as well as the classified cells (lower left) in Piezo2-KO. D) Heatmaps showing mechanosensory responses for the three classes C3, C4 and C5 that normally encompass LTMRs after Piezo2-KO. Functional category (colored bars) and number of cells of a given class are indicated. Note that whereas almost all C4 and most C5 cells are silenced by Piezo2-KO many C3 neurons appear transformed from brush cells in control animals (Figure 4) to HT-cells (see Figure S7 for quantitation and detail). E) Representative GCaMP-transients illustrating the lack of responses of C3 - C5 neurons to gentle mechanical stimuli but robust activation of select neurons by hair-pull and pinch after Piezo2-KO. F) Box and whisker plot comparing the mechanical tuning of C3-cLTMRs with C6- and C13-HTMRs in control and Piezo2-KO recordings. Points represent the ratio for the maximum response of individual cells to pinch or hair-pull to gentle brush or air-puff (plotted on a natural log scale). Note that in control animals (black) C3-responses are generally saturated by low threshold stimuli with ln(ratio) near 0 and are thus distinct from HT-cells. After Piezo2-KO (red) most mechanosensitive C3 neurons are HT-cells; * indicates p < 10−20, Welch’s t-test with Holm-Šídák correction for multiple tests. As a consequence of this functional transformation stacked bar graphs (G) show that the proportion of C3 cells (green) amongst HT-cells nearly triples after Piezo2-KO; scale-bars = 100 μm.

Figure 7.. Transcriptomic class dependent roles for Piezo2 in mechanical nociception

A) Typical multiplex ISH images from sections of the trigeminal ganglion show robust expression of Piezo2 in both C13 and C6 neurons that represent mechanonociceptors. Top panel shows co-expression of Piezo2 (green) with Mrgprd (magenta) in C13 non-peptidergic nociceptors; bottom panel identifies a subset of C6 Aδ-HTMRs by their co-expression of S100b (red) and Calca (blue). Also shown (middle panels) are the individual images; arrowheads highlight a subset of nociceptors expressing Piezo2; scale bar = 100 μm. B) Heatmaps and (C) GCaMP-transients showing mechanosensory responses for mechano-nociceptors after Piezo2-KO. Note the large proportion of C13 HTMRs and their very homogeneous response profiles. Indeed quantitation (stacked bar graph, D) reveals that after Piezo2-KO, C13 neurons are now the dominant class of HT-cells and that unexpectedly the proportional representation of C6 Aδ-HTMRs was reduced by a factor of 2.5 relative to the control; see also Figure S7. (E-G) Individual hair pull responses were analyzed to evaluate the receptive field size of classified neurons to this stimulus both in control cells and after Piezo2-KO. (E) Shows the location of hair pulls across the cheek during a typical recording from a control animal and (F) the corresponding hair pull calcium transients of all neurons in that animal (grey overlaid traces). Pink dots represent computer determined maxima and blue bars are the corresponding peristimulus time histograms demonstrating that the different hair pulls can be reliably identified from the data. (G) The number and magnitude of calcium transients in every hair pull responsive neuron was used to calculate its receptive field index (RFI); low RFI indicates a narrow receptive field, see Quantification and Statistical analysis for details. Shown are the RFI distributions for the major classes of mechanosensor in control (grey bars) and after Piezo2-KO (pink bars). Note the receptive fields of brush sensitive C3 and C5 neurons were broader than for C6 and C13 HT-cells in control cells; after Piezo2-KO C3 and C5 neurons had lower RFIs (p = 4.5 × 10−8 and 2.2 × 10−4, respectively) whereas C6 and C13 RFIs were unaffected (p = 0.22 and 0.07, respectively).