Ancient somatosensory circuit architectures employ flexible molecular strategies

Deep beneath the waves, a skate embryo is quietly rewriting what scientists thought they knew about how nervous systems are built. In a study published in iScience, a team led by Hyemin Lee and Jeremy S. Dasen of NYU Langone Health turned to the little skate, Leucoraja erinacea, an early-diverging cartilaginous fish whose lineage split from ours hundreds of millions of years before the first tetrapod walked on land. Their question was deceptively simple: when vertebrate nervous systems preserve the same circuit architecture across vast evolutionary distances, do they do so using the same molecular blueprint? The answer, it turns out, is a resounding no—and that answer may force a rethink of how sensory circuits evolve.

The vertebrate somatosensory system is a natural laboratory for studying circuit evolution. In mammals, touch, pain, and body-position sensing are carried by more than a dozen genetically distinct subtypes of dorsal root ganglion (DRG) neurons, the sensory cell bodies clustered beside the spinal cord. Three principal classes—mechanoreceptors for touch, nociceptors for pain, and proprioceptors for muscle feedback—transmit modality-specific information to spatially organized layers of the spinal cord. During mammalian development, each class preferentially expresses one of three neurotrophin receptors: ntrk1 in nociceptors, ntrk2 in mechanoreceptors, and ntrk3 in proprioceptors. Signals through these receptors, triggered by target-derived growth factors from the skin and muscle, sculpt neuronal identity, connectivity, and survival. Whether this target-dependent strategy for circuit assembly was present in the common ancestor of all jawed vertebrates, or is a mammalian innovation, was unknown.

Elasmobranchs—skates, sharks, and rays—are the earliest vertebrate lineage in which primary somatosensory neurons are encapsulated in DRG, making them an ideal entry point for reconstructing the ancestral state. Anatomical work over decades suggested their somatosensory system is comparatively simple: DRG soma of uniform size, bimodal distributions of axon diameter, and, strikingly, an apparent absence of unmyelinated C-fibers, the pain-carrying neurons that constitute the majority of DRG cells in mammals. Prior studies of aquatic sensory systems had focused on specialized modalities such as the lateral line, electroreception, or embryonic Rohon-Beard neurons, leaving the evolutionary history of canonical touch and pain circuits unexplored. The little skate, whose embryos develop in transparent egg cases and whose adults walk on the seafloor with fin-based locomotion, offered a tractable system for closing that gap.

The team combined bulk and single-nucleus RNA sequencing with 10x Genomics Xenium spatial transcriptomics on skate embryos at stages 31 and 32, when fins are becoming mobile. Using PIPseq, a microfluidics-free nuclear capture method, they profiled 22,724 nuclei from pooled spinal cords and individually isolated DRG. A custom 300-gene Xenium probe set, designed from the sequencing data and enriched for fate determinants, channels, and receptors, resolved expression patterns in intact tissue across pectoral and pelvic segments. Roughly 90 percent of targeted genes yielded spatially informative signals, and bulk RNA-seq differential expression analysis between spinal cord and DRG showed largely concordant spatial localization, validating the approach.

The spinal cord results were strikingly conservative. Re-clustering of spinal nuclei identified 29 clusters, many with molecular profiles matching mammalian subtypes: motor neurons marked by chat, mnx1, and slit3; dorsal excitatory dI1 interneurons marked by lhx2 and lhx9. Spatial transcriptomics assigned major classes—ventral and dorsal interneurons, progenitors, floor plate cells, glia—to stereotyped positions. Most importantly, the dorsal spinal cord showed a laminar organization deeply reminiscent of mammals. Superficial layers expressed lmx1a, lmx1b, and tlx3b, markers of the dI5 neurons that receive nociceptive input in mammals. More ventral populations expressed lhx2, lhx9, and barhl2, markers of mechanosensory-receiving dI1 neurons. Deeper layers expressed satb2, otpa, and palvb. Because skate DRG neurons express vglut1 (slc17a7) while spinal neurons largely do not, the researchers could use a VGLUT1 antibody to trace sensory afferent terminals. By stage 32, VGLUT1-positive projections enriched the dorsal spinal cord and extended toward the LHX2-positive dI1-like population—evidence that modality-specific targeting of sensory axons to spinal layers is an ancestral vertebrate feature.

The DRG results, by contrast, were anything but conservative. Skate sensory neurons segregated into broad mechanosensory-like and nociceptor-like classes, but the neurotrophin receptor code defining them had been radically reconfigured. Medially positioned runx3-positive neurons co-expressed ntrk3, as mammalian proprioceptors do—but they also co-expressed ntrk1, the mammalian nociceptor receptor, along with nociceptive markers including trp channels (trpv1, trpc4, trpa1) and tac1. Meanwhile, ntrk2a-positive neurons expressed the mechanoreceptor channel piezo2, consistent with conservation of the ntrk2 mechanosensory association. In mammals, this combination is incompatible: ectopic expression of ntrk3 in ntrk1-positive neurons disrupts nociceptor identity and drives proprioceptive features. In skate, the two programs coexist within single neurons. A fourth population expressed ntrk2a alone. Transcription factor profiling added further layers: broadly expressed factors such as tlx3b and pou4f3; ntrk1/3-restricted factors including runx3, skor2, ebf3a, and casz1; ntrk2a-restricted factors including tbx2b, onecut2, and etv4; and overlapping factors such as shox2 and runx1. Nine TFs were expressed in both DRG neurons and specific dorsal spinal populations, raising the possibility that shared transcriptional programs contribute to wiring specificity between connected classes.

To test whether skate sensory neurons depend on peripheral targets the way mammalian ones do, the team performed unilateral fin-bud ablations at stages 27 to 28 and let embryos develop for 14 to 28 days. About 90 percent survived normally, minus the ablated fins. The effects were selective and temporal. Fin-innervating motor neurons, marked by FOXP1, degenerated rapidly: 65 to 70 percent of pectoral and 30 to 40 percent of pelvic motor neurons were lost by stage 31, while axial-muscle motor subtypes were spared. DRG neurons were initially resilient but by stage 32 showed losses exceeding 70 percent in rostral pectoral segments. Crucially, the researchers identified SMAD phosphorylation as an early readout of target-derived signaling. In chick embryos, pSMAD labeled roughly 30 percent of brachial DRG sensory neurons; in skate, pSMAD appeared by stage 30, enriched in RUNX3-negative, ntrk2a-like mechanosensory subtypes. After limb ablation in chick, pSMAD dropped by about 70 percent on the ablated side; after fin ablation in skate, by about 53 percent. SMAD phosphorylation thus marks a conserved early response of sensory neurons to limb- and fin-derived cues in both species.

Fin ablation also revealed a biphasic model of sensory specification. The initial segregation of ntrk2a-positive and ntrk1/3-positive neurons into medial-lateral populations occurred normally without fins, indicating that broad sensory class identity is established target-independently—likely reflecting the birth order of neurogenin2 versus neurogenin1 lineages described in mammals. But downstream maturation failed. Expression of etv4 and ret, genes regulated by limb-derived signals in mice, was markedly depleted. Runx3, instead of restricting to ntrk1/3-positive populations, expanded throughout the DRG, with the RUNX3-positive fraction of ISL1/2-positive neurons rising from 53 to 82 percent on the ablated side. Central connectivity collapsed in parallel: VGLUT1-positive sensory terminals were confined to the most superficial spinal layers, with significant reductions across all dorsal laminar bins and the greatest losses in deeper regions. Target-derived cues, the data show, couple molecular specification with the fidelity of dorsal afferent targeting—an ancestral mechanism now demonstrated in a cartilaginous fish.

The findings carry implications beyond developmental biology. The long-standing debate over whether cartilaginous fish experience pain has often cited the absence of C-fibers as evidence for limited nociceptive capacity. This study shows skate DRG neurons express canonical mammalian nociceptive markers—trp channels, tac1, calca—and that recipient spinal neurons marked by npy1r are enriched in superficial layers, features consistent with nociceptive processing resembling mammalian thinly myelinated Aδ-fibers that mediate rapid first pain. Equally provocative is the apparent absence of dedicated muscle proprioceptors. Skate runx3-positive neurons express nociceptor markers but lack piezo2 and etv1, determinants essential for mammalian proprioceptor development. The authors hypothesize that muscle-based proprioception may be a later evolutionary innovation, emerging with terrestrial locomotion, and that ancestral fish relied on alternative mechanosensory feedback—such as the intraspinal piezo2-positive neurons recently described in adult zebrafish, or cutaneous afferents modulating motor output through deep dorsal spinal circuits.

What emerges is a principle with broad reach: conserved circuit architecture is compatible with malleability in receptor usage and fate-determinant deployment. The layered spinal cord, the modality-specific targeting of sensory axons, and the reliance on retrograde target-derived signaling are all deeply conserved. The molecular programs implementing sensory identity—the neurotrophin receptor code, the deployment of runx3, the combinatorial TF logic—have been extensively reconfigured across vertebrate evolution. This decoupling of architecture from specification provides a general mechanism by which sensorimotor systems accommodate innovation while preserving function. Whether the mixed molecular identities of skate sensory neurons represent retention of an ancestral program or a lineage-specific innovation remains open, as does the full diversity of mature sensory types, which embryonic profiling likely underrepresents. Definitive linkage of molecularly defined types to modality-specific physiology and behavior will require functional assays—electrophysiology, activity imaging, behavioral analyses—building on the molecular and spatial foundation this study establishes. But the conceptual shift is already clear: the stability of somatosensory organization across vertebrates rests not on a frozen genetic blueprint, but on a flexible one.

Subject of Research: Evolutionary neuroscience of somatosensory circuits in an early-diverging vertebrate

Article Title: Ancient somatosensory circuit architectures employ flexible molecular strategies

Article References: Lee, H., Frazel, P. W., Singer-Freeman, E., Cavanagh, A. E., Shin, H. D., Rice, K., Selvaraj, S., Alu, M., Kim, H., Loomis, C., Liddelow, S. A., Baek, M., & Dasen, J. S. (2026). Ancient somatosensory circuit architectures employ flexible molecular strategies. iScience, 29(11), Article 117919. https://doi.org/10.1016/j.isci.2026.117919

Image Credits: AI Generated

DOI: 10.1016/j.isci.2026.117919

Keywords: neuroscience, evolutionary biology, somatosensory system, dorsal root ganglion, spatial transcriptomics, elasmobranchs, developmental neurobiology, iScience, Ancient, somatosensory, circuit, architectures