Teneurin-3 and latrophilin-2 are required for somatotopic map formation and somatosensory topognosis

Abstract

Somatotopy is a recurring organisational feature of the somatosensory system where adjacent neurons and their connections represent adjacent regions of the body. The molecular mechanisms governing the formation of such “body maps” remain largely unknown. Here we demonstrate that the cell surface proteins teneurin-3 and latrophilin-2 are expressed in opposing gradients in multiple somatotopic maps in the mouse, including within the dorsal horn of the spinal cord. Genetic manipulation of these proteins in spinal dorsal horn or sensory neurons distorts the somatotopy of neuronal connections and impairs accurate localisation of a noxious stimulus on the surface of the body. Our work provides the foundation for a molecular model of somatotopic map formation and insights into their function in the localisation of somatosensory stimuli or topognosis.

Introduction

The structure of neuronal connections can reveal their functional logic. One of the organising principles of neuronal circuits are “topographic maps” which maintain the spatial relationship between a group of neurons and their post-synaptic targets. The consequence of this in the context of sensory processing is that information concerning adjacent stimuli is communicated by adjacent neurons at multiple levels of a sensory system. The classical system for studying the development of such maps has been the retinotopic map, where it has been shown that map formation is driven by specific cell surface proteins and refinement by neuronal activity (1–8). However, our understanding of the development of other sensory maps, particularly in the somatosensory system, remains limited.

Somatotopic maps are a unique example of topographic maps where nearby neurons represent contiguous anatomical structures. Somatotopic organisation is characteristic of most vertebrate somatosensory regions, including but not limited to the dorsal horn of the spinal cord (DH) (9–11), dorsal column nuclei (12–14), ventrobasal thalamus (15, 16), and primary somatosensory cortex (17, 18). The rodent DH exhibits a mediolateral (ML) organisation that corresponds to the proximodistal (PD) axis of the body where the medial dorsal horn receives sensory inputs from distal regions (e.g. the paw), while the lateral dorsal horn corresponds to more proximal regions of the body (e.g. the shoulder/hip to midline) (9–11). The ML map is oriented orthogonally to the dorsal horn laminae that receive modality-specific sensory afferents (19, 20); the result is a columnar-like architecture where a vertical “columel” processes noxious, thermal, and light touch information originating in a particular position on the surface of the body (21, 22). This somatotopic map reflects both the neuronal activity induced by sensory stimuli (22–27), and the organisation of incoming sensory afferents (10, 11, 28).

Despite the discovery of somatotopic maps over a century ago and much speculation as to their role in internal representation of our bodies, the mechanisms that underlie their development remain poorly understood (17, 29–31). It is known that neuronal activity plays a critical role in the refinement of developing ST maps (32–34). On the other hand, afferent sensory axons show somatotopic organisation immediately upon entry into the DH, suggesting a contribution by hard-wiring molecular signals (28, 35–38). However, the identity of such cues has remained elusive: none of the classical neuronal connectivity molecules appear to be involved despite their extensive study. Here, we identify cell surface proteins teneurin-3 (Ten3) and latrophilin-2 (Lphn2) as being essential cues for the formation of the DH somatotopic map.

Ten3 and Lphn2 expression reflects the ST organisation of the DH

Recent work has shown that heterophilic repulsion mediated by Ten3 and Lphn2, in combination with Ten3-Ten3 homophilic attraction, establish a point-to-point topography of neuronal connections in the hippocampus (39–41). To determine whether Lphn2 and Ten3 could direct the development of DH connectivity, we examined their expression in the mouse lumbosacral DH at embryonic day 14.5 (e14.5), when cutaneous afferents begin their entry into spinal laminae I through IV (42, 43). There, we observed complementary mRNA gradients that spanned the limb somatotopic map: Lphn2 mRNA expression was high in the medial DH (subserving the distal limb) and low in the lateral DH, while Ten3 mRNA expression was high in the lateral DH (subserving the proximal limb) and low in the medial DH (Fig. 1A). Analysis of protein expression using an anti-Ten3 antibody (39) and Lphn2-mVenus knock-in mice (44) revealed similar mediolateral gradients (Fig. 1B).

Fig. 1. Complementary expression of Ten3 and Lphn2 in the DH and DRG.

(A) RNAscope for Ten3 and Lphn2 mRNA in the lumbosacral dorsal horn at e14.5 (left). Quantification of RNAscope puncta and fluorescence intensity (right, n = 4 embryos). (B) Immunostaining for Ten3 and Lphn2 (anti-GFP antibody) in the lumbar dorsal horn and adjacent dorsal root entry zone (DREZ) of Lphn2-mVenus knock-in mice at e14.5 (left). Quantification of fluorescence intensity in the DH (right, n = 3 embryos) (C) RNAscope for Ten3 and Lphn2 in lumbar dorsal root ganglia at e14.5 (left) and quantification of RNAscope puncta (right, n = 3 embryos). (D) Hypothesized interactions governing DH somatotopic map formation based on the expression and previously described roles of Ten3 and Lphn2. All data shown as mean ± SEM. Scale bars in A, B = 100 μm, C = 50 μm. Axis labels in this and subsequent figures: M, medial; L, ventral; D, dorsal; V, ventral.

Although there is no somatotopic organisation of DRG neurons within individual ganglia, different ganglia innervate distinct patches of skin termed “dermatomes” (9, 45, 46). Correspondingly, we did not observe any spatial bias to Ten3 or Lphn2 expression in individual ganglia along either the dorsoventral or mediolateral axes (fig. S1A). On the other hand, we did observe an inverse relationship between Ten3 and Lphn2 mRNA levels in individual DRG neurons, where those that expressed a high level of Ten3 mRNA, also expressed low levels of Lphn2 and vice versa (fig. S1B). Additionally, DRG neurons that predominantly express either Ten3 or Lphn2 could be identified across all major sensory neuron subtypes (fig. S1C).

Reflecting the dermatome organisation, ganglia enriched in distal hindpaw-innervating neurons, such as L3–5, tended to have high Lphn2 and low Ten3 levels of expression. Conversely, ganglia enriched in more proximally-projecting neurons, such as L1–2 and L6, tended to have low Lphn2 and high Ten3 levels of expression (Fig. 1C). Thus, the level of sensory neuron expression of Lphn2 and Ten3 correlated with DH Lphn2 and Ten3 gradients and the somatotopic map of limb: distal limb sensory neurons that innervate the Lphn2-high medial DH express high levels of Lphn2, and proximal sensory neurons that innervate the Ten3-high lateral DH express high levels of Ten3. A similar correlation could be made between the protein gradients in the DRG axons located in the dorsal root entry zone (DREZ) immediately outside the DH and the gradients in the DH (Fig. 1B, quantified in fig. S1D). Next, to identify if Ten3 and Lphn2 gradients align with somatotopic maps in other somatosensory regions, we examined Ten3 and Lphn2 expression in the ventrobasal (VB) thalamus (postsynaptic to projection neurons in the DH) and uncovered a similar gradient that aligned with the VB somatotopic map (fig. S1E). Finally, to examine whether such Ten3/Lphn2 gradients are evolutionarily conserved we examined the expression of TENM3 and ADGRL2 mRNAs, encoded by the human homologues of mouse Ten3 and Lphn2, in the human fetal dorsal horn and observed similar mediolateral gradients (fig. S1F). Therefore, Ten3 and Lphn2 are expressed in gradients along the proximodistal axis of somatotopic maps in a conserved manner across the somatosensory system.

Ten3 and Lphn2 are required for the somatotopy of sensory afferents

Based on the observed expression patterns in the DH and DRG, as well as the previously described functions of these proteins (40), we hypothesized the following (Fig. 1D): (1) repulsion from Ten3 in the DH acting on Lphn2 in DRG axons directs distal limb afferents to the medial DH, (2) a combination of repulsion from Lphn2 in the medial DH acting on Ten3 in the DRG axons and attraction from Ten3 in the lateral DH acting on Ten3 in DRG axons cooperate to direct proximal limb afferents to the lateral DH. To test our hypothesis, we first challenged DRG axons with purified Lphn2^Lec (the Teneurin-binding lectin domain) and Ten3 proteins. We found that both Lphn2^Lec and Ten3 are repulsive to DRG axons in vitro (fig. S2), thereby demonstrating that Ten3 and Lphn2 are capable of guiding DRG axons. Next, we generated conditional knockouts (cKOs) of Ten3 and Lphn2 in the DRG and DH (fig. S3A; see methods), hereby referred to as Ten3^ΔDRG, Lphn2^ΔDRG, Ten3^ΔDH, and Lphn2^ΔDH. To assess the distribution of sensory afferents along the mediolateral extent of the DH, we co-injected cholera toxin subunit B (CTB) and wheat germ agglutinin (WGA) subcutaneously to the hindpaw or dorsal midline. These preferentially label myelinated (e.g. innocuous touch) and unmyelinated (e.g. pain and temperature) sensory afferents, respectively (11, 34, 47).

First, we examined distal limb afferent inputs to the DH, which we hypothesized to be guided by Ten3 (DH) → Lphn2 (DRG) repulsion. We therefore examined these afferents in Ten3^ΔDH and Lphn2^ΔDRG cKO mice compared to littermate controls by injecting CTB/WGA into the hindpaw and quantified the distribution of the labelled axons along the ML and superficial-deep (SD) axes of the DH (fig. S3B). In wild-type mice, the paw representation is disproportionally large and duplicated at the most medial extreme of the dorsal horn (11, 22, 34, 48). In our controls, we saw the expected wild-type distribution of afferents however in Ten3^ΔDH cKO mice we observed a dramatic shift in the ML distribution of labelled afferents (Fig. 2, A to C and fig. S3C). Meanwhile, CTB and WGA-labelled axons remained in register along the ML axis (fig. S3D), and there was no shift in the distribution of CTB and WGA-labelled axons along the orthogonal SD axis (fig. S3, E to G). The Lphn2^ΔDRG cKO mice exhibited a milder but significant shift along the ML axis and no change in the SD axis distribution or in the ML alignment between CTB and WGA-labelled afferents (Fig. 2, D to F and fig. S3, H to L).

Fig. 2. Altered somatotopy of distal sensory afferents in Ten3 and Lphn2 cKOs.

(A and D) Schematic of lost and preserved interactions in (A) Ten3^ΔDH cKOs and (D) Lphn2^ΔDRG cKOs (top) and the observed shift in the ST distribution of afferents in the cKOs (bottom, filled lines) compared to controls (bottom, dotted lines). (B and E) Representative images of CTB and WGA labeled afferents in (B) Ten3^ΔDH cKOs (insets showing WGA near lateral edge of DH in cKOs) and (E) Lphn2^ΔDRG cKOs compared to controls. (C and F) Quantification of the distribution of CTB and WGA signal across the ML axis of the DH in cKOs and control littermates (C) n = 4 controls, 4 cKOs, (F) n = 4 controls, 3 cKOs. All data shown as mean ± SEM. Two-way mixed ANOVA was performed and genotype x bin interaction terms are reported. ***P ≤ 0.001; **P ≤ 0.01; *P ≤ 0.05. Scale bars = 100 μm.

We next examined proximal afferent inputs to the DH, which we hypothesize are guided by a combination of Lphn2 (DH) → Ten3 (DRG) repulsion and Ten3 (DH) → Ten3 (DRG) attraction. We therefore examined these afferents in Ten3^ΔDH, Lphn2^ΔDH, and Ten3^ΔDRG cKO mice and compared them to littermate controls by injecting CTB/WGA subcutaneously at the dorsal midline. Like for the distal limb afferents, we observed a dramatic shift in the ML distribution of proximal afferents in the Ten3^ΔDH cKO mice (Fig. 3, A to C and fig. S4A). However, this was accompanied by a decrease in the ML alignment of the CTB with the WGA-labelled afferents (fig. S4B), and an expansion in the SD distribution of both classes of afferents (fig. S4, C to E). Lphn2^ΔDH cKO mice exhibited a comparatively mild but significant shift along the ML and SD axes, without a consequence on the alignment of CTB with the WGA-labelled afferents (Fig. 3, D to F and fig. S4, F to J). Unexpectedly, the Ten3^ΔDRG cKO mice did not exhibit any significant changes in the targeting of afferents in the DH (Fig. 3, G to I and fig. S4, K to O), indicating that Ten3 in the DRG may be dispensable for correct somatotopic and laminar targeting. Importantly, Ten3 (DH) → Lphn2 (DRG) repulsion is likely intact in Ten3^ΔDRG cKO mice, suggesting that this repulsion alone may be sufficient to separate the distal and proximal inputs along the ML axis. Together, our cKO experiments demonstrate that Ten3 and Lphn2 are essential for the correct somatotopic organisation of primary sensory afferents in the DH, with the severity of miswiring phenotypes depending on the ablated gene and targeted tissue.

Fig. 3. Altered somatotopy of proximal sensory afferents in Ten3 and Lphn2 cKOs.

(A, D and G) Schematic of lost and preserved interactions in (A) Ten3^ΔDH cKOs, (D) Lphn2^ΔDH cKOs, and (G) Ten3^ΔDRG cKOs (top) and the observed shift in the ST distribution of afferents in the cKOs (bottom, filled lines) compared to controls (bottom, dotted lines). (B, E and H) Representative images of CTB and WGA labeled afferents in (B) Ten3^ΔDH cKOs (insets show more medially located CTB/WGA in cKOs), (E) Lphn2^ΔDH cKOs, and (H) Ten3^ΔDRG cKOs compared to controls. (C, F and I) Quantification of the distribution of CTB and WGA signal across the ML axis of the DH in cKOs and control littermates (n = 4 controls and 4 cKOs). All data shown as mean ± SEM. Two-way mixed ANOVA was performed and genotype x bin interaction terms are reported. ***P ≤ 0.001; **P ≤ 0.01; *P ≤ 0.05. Scale bars = 100 μm.

Computational modelling of somatotopic map formation in the DH

To better understand our observed phenotypes and gain further insights into the developmental programs that produce the DH somatotopic map, we sought to construct a comprehensive computational model of DH somatotopic map formation. Over the past 40 years, several computational models of topographic map formation in the visual system have been developed and iteratively improved with experimental data. We first sought to determine if these theoretical models of topographic map formation could be applied to somatotopic maps. We therefore adapted the Koulakov model (8, 49, 50) (see methods) as it is a relatively abstract model without excessive constraints regarding the cellular mechanisms. It also effectively predicts the observed effects of genetic manipulations (51), and incorporates guidance cues, competition, and activity-dependent refinement into a single unified model (8). After adapting this model to the DH somatotopic map, we then performed Bayesian Simulation-Based Inference (SBI) using the Sequential Monte Carlo method for Approximate Bayesian Computation (ABC-SMC) to identify parameter values for our model that fit both the observed control and cKO data in an unbiased manner (Fig. 4A; see methods).

Fig. 4. Computational modelling of DH somatotopic map formation.

(A) Simplified schematic of model parameters and ABC-SMC procedure. (B) Correct distal → medial and proximal → lateral mapping from one representative simulation under wild-type conditions. Simulated ML density of terminal zones in the DH from 10 DRG axons selected across the PD axis. (C) Observed and simulated tracer injections for controls and cKOs across all genotype/injection site combinations examined. Average distributions of each condition are shown. Simulated distribution taken from the average of 10 independent simulations.

We first confirmed that this model was able to accurately reproduce the proximal → lateral and distal → medial connectivity between DRG and DH neurons under wild-type conditions (Fig. 4B and fig. S5A). We then confirmed we were able to closely recapitulate the observed distributions of CTB/WGA along the ML axis in both controls and all cKOs examined experimentally above (Fig. 4C). Inspection of the parameter values (table S1) reveals that the best fitting model incorporates a prominent role for activity-dependent refinement of connectivity, consistent with previous experimental results (32–34). Furthermore, disrupting activity dependent refinement in our model produced qualitatively similar broadening of DRG axon terminal zones and disruptions to DH somatotopy as what is observed experimentally (fig. S5, B to D). Our model also indicates that there is strong competition between DRG axons for limited space in the DH along the ML axis, consistent with previous results showing an expansion of the saphenous nerve afferent termini when the afferents of the sciatic nerve are eliminated (52, 53). We recreated this experiment in silico by ablating proximal limb DRG neurons and observed a similar expansion of the remaining axons to fill the available ML extent of the DH (fig. S5, E and F). Additional inferences of our model are that Ten3 (DH) → Lphn2 (DRG) repulsion plays a more prominent role than Lphn2 (DH) → Ten3 (DRG) repulsion or Ten3 (DH) → Ten3 (DRG) attraction, and that there is an effect on both guidance and activity-dependent refinement, particularly in the Ten3^ΔDH cKOs. Taken together, despite being developed for a different sensory system being patterned by unrelated molecular cues, theoretical models of topographic map formation in the visual system can be applied to developing somatotopic maps. We also show that our observed cKO data are consistent with the model that gradients of Ten3/Lphn2 cooperate to specify the DH somatotopic map.

Ten3^ΔDH cKOs exhibit an altered somatotopic distribution of neural activity and behavioural responses

Given the observed miswiring phenotypes, we sought to examine if this miswiring was accompanied by alterations in the function of DH circuitry. Towards this end, we selected the cKO with the most severe miswiring phenotype (Ten3^ΔDH) and examined the consequence of injecting the hindpaw with formalin, a noxious stimulus that causes a somatotopically restricted induction of the neuronal activity marker c-Fos in DH neurons (23). Following injection of formalin into the distal regions of the hindpaw, both Ten3^ΔDH cKO mice and control littermates displayed a significant induction of c-Fos expression in the DH ipsilateral to the injection (fig. S6, A and B). While the control littermates showed increased c-Fos expression predominantly in the somatotopically appropriate medial region of the DH, in line with the innervation pattern from distal limb DRG afferents, the Ten3^ΔDH cKO mice exhibited a greater number of c-Fos-expressing cells in the lateral DH (Fig. 5, A to D). Matching the more widespread induction of c-Fos, cKO mice also had more c-Fos-expressing cells overall on the side ipsilateral to the injection when compared to control littermates (fig. S6, C to E).

Fig. 5. Altered ST of neural activity and behavioral responses induced by a noxious stimulus in Ten3^ΔDH cKOs.

(A) Representative images of immunostaining for Fos in the DH induced by formalin injection to the hindpaw of Ten3^ΔDH cKOs and contols. (B-D) Quantification of Fos positive cells in Ten3^ΔDH cKO mice and control littermates (n = 5 controls and 5 cKOs). (B) Distribution of Fos positive cells across the ML axis of the DH. Two-way mixed ANOVA test was performed and genotype x bin interaction term is reported along with Benjamini-Yekutieli corrected post-hoc p-values for individual bin comparisons. (C) 2D kernel density plots showing their distribution of Fos positive cells in the DH. Hotelling T² p-value 0.03. (D) Average ML position of Fos cells. Welch’s t-test was performed. (E) Quantification of nocifensive licking behavior in response to formalin injection to the hindpaw near the digits, depicted as a percent of total linking time per limb region (n = 8 controls and 10 cKOs). Two-way mixed ANOVA was performed and genotype x location interaction term is reported. Post-hoc Mann-Whitney tests were performed for each limb location. All data shown as mean ± SEM. ***P ≤ 0.001; **P ≤ 0.01; *P ≤ 0.05. Scale bars = 100 μm.

In addition to being able to induce somatotopically restricted c-Fos activity, formalin injection elicits highly accurate licking of the injection site. Thus, using this paradigm we sought to examine the functional impact of somatotopic miswiring on the ability to locate a stimulus on the surface of the body, or topognosis. We first examined the precision of topognosis in Ten3 constitutive knockout (Ten3^KO) mice (39, 54). Compared to control littermates, Ten3^KO mice spent less time licking regions of the limb close to the injection site and more time licking regions of the limb farther from the injection (fig. S6, F to G), potentially indicating a deficit in their ability to precisely locate the stimulus. Meanwhile, the total duration of licking, an indication of stimulus intensity, was not different between KO and control mice (fig. S6H). However, Ten3^KO mice also showed defects in the ability to remove an adhesive tape placed on their hindlimbs, suggesting broad deficits in sensorimotor coordination (fig. S6, I and J). We next assessed topognosis in Ten3^ΔDH cKO mice given the magnitude of their anatomical miswiring phenotype. Similar to the Ten3^KO mice, Ten3^ΔDH cKO mice exhibited a reduction in time spent licking areas close to the injection and increase in time licking areas farther from the injection site (Fig. 5E and fig. S6L), indicating they also showed impaired topognosis. Again, this phenotype was not accompanied by any significant changes in total licking duration (fig. S6K). Finally, unlike the Ten3^KO mice, Ten3^ΔDH mice performed normally in the adhesive removal assay (fig. S6, M and N), indicating a lack of defects in sensorimotor coordination.

Discussion

Our study provides foundational insights into the molecular logic of somatotopic map formation. First, our experimental observations and modelling argue that the somatotopic map formed between peripheral sensory neurons and those in the spinal DH is driven by graded expression of Ten3 and Lphn2. This complements previous studies on their role in establishing of neuronal connections through their binary or all-or-nothing attractive and repulsive signalling. Additionally, we also provide direct functional evidence of the importance of somatotopic maps in the ability to locate a stimulus on the surface of the body.

Several parallels and distinctions between our work and that on the formation of retinotopic maps can also be made. Like in retinotopic maps, past studies and this work suggest that axon-axon interactions such as pre-ordering prior to arrival in target area and inter-axon competition, as well as activity-dependent refinement, play an important role in the establishment of the DH somatotopic map. However, the nature of the miswiring phenotypes observed by us differs significantly from those following Eph/ephrin signalling manipulations in the retinocollicular system. While Eph/ephrin signalling loss largely result in the formation of additional compact terminal zones of retinal axons in the superior colliculus (4, 5, 55), we generally observe broadening and/or shifting of sensory axon terminal zones in the ML axis of the DH, without the formation of any additional ones, suggesting the potential for fundamental cellular/molecular differences between these processes. Of particular interest would be the significance of Ten3/Lphn2 and Eph/ephrins gradient co-expression, the latter of which has been implicated in somatosensory cortex somatotopic map formation afferent thalamic axon connectivity (56, 57). Finally, the presence of Ten3/Lphn2 gradients in many regions of the developing nervous system raises the possibility that they pattern all somatotopic maps within the somatosensory system, and more generally, circuits whose organising principle requires the preservation of spatial relationship between neurons and their targets (58).

Although we cannot rule out other DH functions of Ten3 contributing to the impaired topognosis phenotype, the unchanged duration of licking in response to formalin in Ten3^ΔDH cKOs suggests that stimulus intensity and location can be dissociated, echoing the idea of parallel motivational/affective and discriminatory systems underlying nociception (59). Importantly, our data provide some of the only functional evidence supporting the importance of somatotopic maps in somatosensory function since Penfield’s original experiments involved stimulating cortical neurons and do not allow any inferences regarding the functional requirement of somatotopic maps (17). Our genetics-based precise manipulation of a somatotopic map provides a new potential inroad into questions such as map plasticity, the integration of somatotopic somatosensory maps with somatotopic motor maps, or those of other sensory modalities, and their overall function in representing our bodies and whole body action planning (31).

Supplementary Material

Materials and Methods

Figs. S1 to S6

Table S1

References (60–73)