Assigning transcriptomic class in the trigeminal ganglion using multiplex in situ hybridization and machine learning

Single cell sequencing has provided unprecedented information about the transcriptomic diversity of somatosensory systems. Here, we describe a simple and versatile in situ hybridization (ISH)-based approach for mapping this information back to the tissue. We illustrate the power of this approach by demonstrating that ISH localization with just 8 probes is sufficient to distinguish all major classes of neurons in sections of the trigeminal ganglion. To further simplify the approach and make transcriptomic class assignment and cell segmentation automatic, we developed a machine learning approach for analyzing images from multiprobe ISH experiments. We demonstrate the power of in situ class assignment by examining the expression patterns of voltage-gated sodium channels that play roles in distinct somatosensory processes and pain. Specifically, this analysis resolves intrinsic problems with single cell sequencing related to the sparseness of data leading to ambiguity about gene expression patterns. We also used the multiplex in situ approach to study the projection fields of the different neuronal classes. Our results demonstrate that the surface of the eye and meninges are targeted by broad arrays of neural classes despite their very different sensory properties but exhibit idiotypic patterns of innervation at a quantitative level. Very surprisingly, itch-related neurons extensively innervated the meninges, indicating that these transcriptomic cell classes are not simply labeled lines for triggering itch. Together, these results substantiate the importance of a sensory neuron's peripheral and central connections as well as its transcriptomic class in determining its role in sensation.