Classification of sensory neuron types applying single-Cell RNA sequencing

In 1968, the Canadian psychologist from McGill University Ronald Melzack described pain as being multidimensional and complex, with sensory-discriminative, affective-motivational and cognitive-evaluative components 1. Such definition may be a hint as to why the biological meaning of pain has been an (to date) unceasing matter of debate since Aristotle, who thought pain is just a passion of the soul as opposed to pleasure. Is pain a similar sense to touch or vision or rather a mere trigger of emotions shaping the behaviour of individuals?

If we look at pain as a sense we imply the existence of a dedicated set of neurons signalling separately for different modalities of pain (e.g. thermal or inflammatory), leading to a protection response against damage (with the notable exception of neuropathic pain). However, if we do consider pain an emotional trigger we assume no distinct dedicated population of neurons conveying different pain modalities is needed. It would be a matter of intensity and signal integration instead of specificity. Any given sensory pathway (sense organs) may participate if the stimulus is strong enough, as determined by the number of impulses or firing frequency of the neuron.


The current picture suggests a mixture of both theories may contribute to the physiology of pain indeed. It has been unequivocally shown there are specific population of primary afferent neurons, which can be “labelled” because of the expression of established markers (mainly receptors and ion channels) activated by pain stimulus. This conclusion is based in the assessment of different knockout mice strains in which selected receptors or ion channels (like TRPV1, Nav1.9 or Nav1.8 ion channels, among others) have been deleted and the lack of different modalities of pain could be confirmed. Likewise, we know there are subsets of peripheral neurons not involved in pain detection (or silent), which can be recruited and sensitized when the noxious stimulus is “strong” enough. These events do occur as a consequence of molecule secretion by the neurons primary involved in pain detection showing a high degree of plasticity.

Despite extensive effort aiming to profile these populations of peripheral neurons, most pharmacological tools developed and tested by pharmaceutical industry targeting ion channels or receptors differentially expressed in these subpopulations have failed to alleviate pain, so far. Therefore, experts on the field are starting to question the current knowledge we have to this regard, saying perhaps we still have a too simplistic vision about the identity and/or functional properties of relevant populations.

An ambitious attempt to clarify this issue has been championed by Sten Linnarsson and Patrik Ernfors from the Karolinska Institute in Sweden2. Taking advantage of single-cell transcriptomics (the inspection of the whole set of mRNA expressed in a given cell) and the idea that specific neuronal identity will be reflected on its RNA expression pattern, they have used an unbiased single-cell sampling RNA-seq from 799 cell bodies from lumbar DRGs (to include proprioceptive neurons) to profile them. More than one million reads were mapped to roughly 3600 different genes in each cell, thus giving an idea on how powerful this type of approach is. In essence, they use the expression levels of thousands of genes from each sensory cell to get a sort of genetic fingerprint, which can be then used to make a classification according to their expression pattern. Their analysis led to the identification of five main clusters of cells from which surprisingly, one of them could be classified as non-neuronal cells, based on its markers (109 cells in total). The remaining ones were classified as either Neuronal-type (622 cells) or unresolved identity (68 cells).

What did they get from the neuronal-type population? Four different groups could be clearly identified from which three had been previously well characterized: The NF cluster (neurofilament heavy chain expression) corresponding to myelinated DRG neurons (involved in light touch detection and proprioception). The PEP cluster previously associated with peptidergic nociceptors expressing neuropeptides, which are critical in inflammatory processes and thermal sensitivity. The NP cluster was associated with the previously known nonpeptidergic nociceptors (no neuropeptide secretion), whose contribution to pain physiology is less clear. A first unexpected result came from the existence of a fourth cluster comprising neurons with predominance of tyrosine hydroxylase (TH) protein expression. TH is an enzyme present in the metabolic pathway leading to production of dopamine, which is in turn a precursor of adrenaline and noradrenaline. In the PNS, TH is mainly present in sympathetic neurons and even though it has been shown to be present in DRG cells, it was unknown that its expression could be a distinctive feature from a DRG subpopulation. Looking at other proteins present in this population, they conclude they may be involved in both noxious and pleasant touch detection.

Another novelty brought by the work is these clusters are not homogeneous when looking at their “fingerprint”, and can be further split down into subgroups with important divergence in expression pattern. All but the TH population (quite homogeneous as a whole) can be split down further into subgroups based in subtle gene expression differences, including five different ones for the NF cluster (NF1-NF5), two for the PEP cluster (PEP1 and PEP2) and three for the NP cluster (NP1-NP3), all of them being responsible for slightly different (or sometimes redundant) pain modality detection and conduction.

The role assigned to each subpopulation matches quite well indeed the previous knowledge we had about them in a broad way. This assumption is based in, 1) in vivo validation by using combinatorial (double and triple) immunohistochemical analysis with available antibodies, thus confirming their findings at the protein level (this is important as mRNA transcription doesn’t necessarily means we get protein), and 2) RNA profiling of the populations searching for the expression of well-known signalling pathway components.

In summary, the authors show single cell RNA sequencing can be reliably used to distinguish complex mixture of cell populations and may be useful to profile other neuronal types like the ones from the autonomous nervous system from which still little information is available. They show discrete transcriptional states in different classes of neurons are responsible for detection of different sensory modalities related to pain (supporting the selectivity theory), but they also suggest critical ion channels and receptors expressed in different class neurons will have different activation threshold and signal quality, thus supporting that integration of activity in ensembles of different neuronal classes is also important in pain physiology. But maybe more important is they have generated an open and comprehensive catalogue of gene expression data, meaning any researcher in the field can check the population in which a gene of interest is expressed. In mi case, (as some of you know already from the first post I wrote for MI in May 2014), I am interested in the role of HCN2 in neuropathic pain. Based on our studies, we know the expression of the channel in a subpopulation of neurons expressing Nav1.8 voltage-gated channel is critical for neuropathic pain, but when looking at the expression pattern of HCN2 on the gene database, which can be found on their web page, it turned out most HCN2 is expressed in TH positive neurons, indicating HCN2 expressed in this subpopulation could be relevant for other modalities of pain unknown as yet to us. This finding may lead to new and exciting ideas about my own research and so probably will be for others too, guaranteed.


  1. Melzack R, Casey KL. Sensory, motivational, and central control determinants of pain: a new conceptual model. In: The Skin Senses, edited by Kenshalo D. Springfield, IL: C. C. Thomas, 1968, p. 423–439
  2. Usoskin D., Saiful Islam, Hind Abdo, Peter Lönnerberg, Daohua Lou, Jens Hjerling-Leffler, Jesper Haeggström, Olga Kharchenko, Peter V Kharchenko & Sten Linnarsson & (2014). Unbiased classification of sensory neuron types by large-scale single-cell RNA sequencing, Nature Neuroscience, 18 (1) 145-153. DOI:

Leave a reply

Your email is never shared. Required fields are marked.


You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>