Transcriptomically reserved identities for fly olfactory projection neuron

Original article:
Li, H. et al. Classifying Drosophila Olfactory Projection Neuron Subtypes by Single-Cell RNA Sequencing. Cell 171,1206.e22–1207 (2017).

The complexity of the nervous system has fascinated scientists for centries. We knew that injuring different part of the brain or spinal cord leads to distinctive function impairment, and we observed neurons with different shape and size.

Since the rise of molecular biology, people have been longing to find an explanation of this diversity that fits the central dogma: Genes corresponding to these unique cells have been identified, and people hypothesize the diversity is hardwired in genetic program and executed so that the position, number, and connectivity are precisely regulated. Nevertheless, whether gene expression is the cause or consequence of functional diversity in the nervous system remains largely an unanswered question. In the recent boom of single cell RNA-seq studies, many researchers have tried to solve the myth by the newly invented power, and this article from Li et among those attempts.

The story is quite concise and frank, in the sense of exposing the questions raised by observations and their attempt to address them, which are not always successful, to the reader.

Based on previous understanding of olfactory projecting neuron in fly, they tried and dismissed general clustering method and developed their own. The techinical advance in data collection and generation like single cell RNA-seq could drive the development of analytic tools, but before the tools become mature, it is inevitable to apply what one already know to retrospectively decide whether the analysis was done right.

Are the identities identified in the study real? The authors again turned to their prior knowledge and showed the clustering identities corresponded to known populations of projection neurons. Then, with the powerful genetic tools in fly, they identified a gene marking a population that looked like the DC2 neurons, and another, which mapped to 4 clusters, and two of them might represent DC3 and VA1d neurons in vivo. Trying to find more genes that separate these two clusters was not completely fruitful: Genes found were able to distinguish the populations of interest, but they were also expressed in other projection neurons, which greatly complicated the identification of each subtypes when marker genes overlap a lot. They in turn proposed a model of combinatorial gene barcodes to delineate the functionally different population, but they did not show how well this model reflects the real world..

The authors also made efforts to establish the link between transcriptomic identity and function. They found a transcription factor that was not known to regulate neural function before and was expressed in a subpopulation of projection neurons, the adPNs, that project to a defined brain region. The authors showed that when the transcription factor was mis-expressed in the neurons that would otherwise not express it, the neurons with mis-expression targeted to the same region as adPN did, which corroborated the role of this transcription factor as a determinant of neurite guidance. This is a neat way to show the link between transcriptome diversity and function, but it also seems like that the current analysis is not ready yet to identify this kind of functionally important gene for each single cluster because the authors was not able to but picked a transcription factor that is expressed in almost half of their clusters.

Finally, the authors collected projection neurons from different stages in development and showed that the diversity of single cell transcriptome peaked at the time when connection was forming. Though it is not surprising that the diversity is transient because unnecessary diversity could be too costly to maintain after connections had been formed, this is the first study actually showing that as far as I know.

The take-home messages here are:

  1. Because conclusive answer is rare, try your best to make educated guess in analysis.
  2. Frame a coherent hypothesis according to both significance and the questions that need to be addressed. (It is cliché, but in many articles, the struggle to achieve this is just packed too well for me to realize that it did happen.)
  3. Customizing methods is tedious, but it could pay off as additional novelty and might not be avoidable at the end of the day.
PhD Student

A graduate student interested in developmental biology, neurobiology and bioinformatics.