Platonic neurons: Some thoughts about neuronal cell types or Metaphysics as a Guide to Neurons


Idea was that this could be a short opinion piece / overview article (not an Annual Review) with some novel ideas/synthesis. I jotted down some initial thoughts after conversations with David Anderson at Caltech in early 2015. Additional input from another connectomics/cell type person (e.g. Kording, Seung, Ascoli), someone specialising in quantitative approaches to speciation (e.g. Chris Jiggins or David Stern) or one of the people doing DropSeq type work would be interesting.


Attempts to define neuronal cell type have resulted in substantial debate between some enthusiasts and an equal measure of exasperation for those less exercised by these issues. Indeed a good number have thrown up their hands, in a response somewhat reminiscent of US Supreme Court Justice Potter Stewart when faced with the need to define another category regarded as highly subjective:

I shall not today attempt further to define the kinds of material I understand to be embraced within that shorthand description ["hard-core pornography"], and perhaps I could never succeed in intelligibly doing so. But I know it when I see it, and the motion picture involved in this case is not that.

But in spite of this, thinking about cell types from a philosophical and a practical point of view is something that could be useful for most experimental neuroscientists. In this review we will make 3 points about a successful definition of cell type. First, cell types are as much about scientists as about the brain; they are a way of helping us to organise our observations about cells, so that we can aggregate related observations from different experiments and communicate them to our colleagues. Second, the correlation between multiple co-occurring neuronal properties is crucial for any sensible definition of cell type (otherwise you have a blade of grass paradox); a type is only useful if we will see it again – thinking about neurons as snowflakes will probably not help us understand brain circuits and behaviour. Third, that it should be possible to come up with a general strategy to defining cell type that is both experimentally useful and captures our intuitive response to the statistical regularities in neuronal properties.

Starting from these ideas, we propose a definition of neuronal cell type based on an analogy with gene orthology defined for whole genome data. We consider briefly how this could be defined using cell properties and what datasets might prove experimentally amenable in the short and long term.

A philosophical digression

As a young teenager, a charismatic teacher introduced one of us (GJ) to some of the philosophical approaches to language and meaning. He particularly contrasted the abstract approach exemplified by Plato with the utilitarian approach of Wittgenstein. With apologies to professional philosophers, this is what I remember.

Plato articulated a Theory of Forms in which non-physical, ideal forms capture the essence of the objects that we see. Thus an ideal cup might capture all of the “cupness” of the different cups that we interact with. Interestingly Plato argued that this idealised cup had a real existence, even if we were not privileged to experience it directly. In the context of neuronal cell types, this view would posit that a particular cell type is a real phenomenon that is worth striving to understand. Indeed the dichotomy between the ideal form and its physical instantiations might actually suggest some useful lines of thought. For example we can think that any individual neuron might be minted from such an ideal form by a somewhat noisy but statistically characterisable process. So Plato’s ideal form might drive us to consider classes of generative statistical model that would capture the regularities of neuronal types.

To many scientists schooled in palpable realities, the Theory of Forms probably sounds rather nebulous and not particular useful. And in the context of neuronal cell types, we could say the ideal form sounds great in theory, but it will not be very useful if we have no idea how it might be derived in practice. In comparison Wittgenstein’s linguistic philosophy (aka ordinary language philosophy) is about as down to earth as it gets. For example Wittgenstein wrote [3]:

For a large class of cases of the employment of the word ‘meaning’—though not for all—this word can be explained in this way: the meaning of a word is its use in the language

In the context of neuronal cell type, we can say that the approach of Wittgenstein seems useful because it emphasises communication and convention. And this I think is an important point, because any cell type definition that does not enable scientific communication is doomed to irrelevance.

If a community agrees on the definition of a type then they can exchange experimental data and conclusions; this is clearly vital for scientific progress. Now the problem is that where conventions of usage arise organically over time within a research community, they can be difficult to set down precisely for a newcomer or in a way that might allow a machine to reason for them. Furthermore, they may drift or not be used equivalently. The alternative – organised attempts to define conventions turn out to be extremely difficult – witness the Petilla Interneuron meeting (Ascoli 2008). In the absence of any higher truth each expert’s opinion can carry equal weight and so building consensus can be challenging. So perhaps Wittgenstein is not as useful as he seems. Indeed it seems that this area of philosophy was already largely defunct when I met it in 1999 11 One prominent critic Ernest Gellner with a significant role in its demise thought as follows: [A]t that time the orthodoxy best described as linguistic philosophy, inspired by Wittgenstein, was crystallizing and seemed to me totally and utterly misguided. Wittgenstein’s basic idea was that there is no general solution to issues other than the custom of the community. Communities are ultimate. He didn’t put it this way, but that was what it amounted to. And this doesn’t make sense in a world in which communities are not stable and are not clearly isolated from each other. Nevertheless, Wittgenstein managed to sell this idea, and it was enthusiastically adopted as an unquestionable revelation. It is very hard nowadays for people to understand what the atmosphere was like then. This was the Revelation. It wasn’t doubted. But it was quite obvious to me it was wrong. It was obvious to me the moment I came across it, although initially, if your entire environment, and all the bright people in it, hold something to be true, you assume you must be wrong, not understanding it properly, and they must be right. And so I explored it further and finally came to the conclusion that I did understand it right, and it was rubbish, which indeed it is. .

So although we can certainly accept the importance of tokens for communication derived from Wittgenstein, maybe Plato deserves a second look. Let’s conduct a thought experiment, the Martian test. Imagine we give a Martian some of Cajal’s Golgi-stained material. I strongly suspect that she would come up with many of the same cell types as the great man himself. If that is true, it suggests to me that there is merit in both approaches. But to me as a scientist it also suggests that we should be looking for principled statistical approaches to define regularities that are organising principles of cell type. If we can do so, we have the chance not only to improve scientific communication but also to identify new, experimentally addressable features of circuit organisation.