08 March 2012

Overselling the connectome

In the last few years, there has been much discussion about the prospect of tracking the neural connections of mammalian, and particularly human, brains at very high levels of detail. Following the concept of a genome – every gene in an organism – a connectome is a map of every anatomical connection between every neuron in an organism.

One of the major proponents of this effort has been Sebastian Seung, who you can see giving a TED talk here. He sells the idea that understanding the human connectome will help us understand human identity. In his talk, Seung encourages his audience to say along with him. “I am my connectome.” He’s written a new book on this subject, Connectome: How the Brain’s Wiring Makes Us Who We Are.

This is an ambitious research project that will no doubt yield highly improved techniques to get anatomical information and analyze it. We will learn a lot from it.

And it will fail.

The allure, promise, and shortcomings of the connectome approach are yesterday’s news to neuroethologists. In the 1960s and 1970s, neuroethologists put in a lot of effort to crack partial connectomes of several species. These were usually referred to as “circuits” or “wiring diagrams.” (“Connectome” only appeared when neuroscientists got genome envy.) We made good progress on these. For example, we can explain escape behaviour in fishes and crayfish by the main synaptic connection between the critical neurons. That said, escape systems were chosen specifically because they were unusual behaviours. they are very sterotyped, very fast, and dedicated to one single task.

As other circuits were cracked, they revealed a much more subtle story.

ResearchBlogging.orgA new paper by Bargmann details the case histories of a few of the species that neuroethologists have basically cracked the circuit. And contrary to some expectations, getting the complete set of synaptic connections did not solve the problems of understanding behaviour. I’m very glad that Bargmann wrote this paper, because it saves me the trouble of writing a much longer blog post.

For example, the nematode worm Caenorhabditis elegans has 302 neurons, and all the connections between them are known (the first complete connectome in the animal kingdom). Bargmann writes:

At a more profound level, however, the wiring diagram was and remains difficult to read. The neurons are heavily connected with each other, perhaps even overconnected – it is possible to chart a path from virtually any neuron to any other neuron in three synapses. ... Circuit studies suggest a reason for this failure: there is no one way to read the wiring diagram.

One of the major lessons that emerged in the 1990s from the study of these small circuits where we knew all the synaptic connections was the importance of neuromodulation. Neurons’ functions were not set only by their anatomical connections. They were profoundly influenced by a cocktail of neuroactive chemicals that could change the physiological responses of neurons.

Bargmann breaks it down. First, she shows that only rarely can you link single neurons to single behaviours. Then, she shows how one behaviour can result from several circuits, and how one neuromodulator can influence several behaviours. She notes that given how neuromodulation has appeared pretty much in every nervous system where we’ve looked, there’s every reason to expect it’s going to be a major factor in determining human neural activity, and thus, human identity.

In his TED talk, Seung draw an extended metaphor that the connectome is like the bed of a river.

I would like to propose a metaphor for the relationship between neural activity and connectivity. Neural activity is constantly changing. It's like the water of the stream; it never sits still. The connections of the brain's neural network determines the pathways along which neural activity flows. And so the connectome is like bed of the stream; but the metaphor is richer than that, because it's true that the stream bed guides the flow of the water, but over long timescales, the water also reshapes the bed of the stream. And as I told you just now, neural activity can change the connectome. And if you'll allow me to ascend to metaphorical heights, I will remind you that neural activity is the physical basis – or so neuroscientists think – of thoughts, feelings and perceptions. And so we might even speak of the stream of consciousness. Neural activity is its water, and the connectome is its bed.

Knowing the bed of the river still doesn’t tell you everything. The same river bed can have a trickle one day, and a flash flood the next. Neuromodulation is a bit like a dam partway along the river. It can regulate whether you have a torrent or a trickle.

Bargmann and I agree that connectome projects are very useful. But they alone will not solve the question of human identity. But at least when they fail, they will fail in an interesting way.

Reference

Bargmann C. 2012. Beyond the connectome: How neuromodulators shape neural circuits. BioEssays: In press. DOI: 10.1002/bies.201100185

12 comments:

Bashir said...

I had a similar reaction upon seeing that book. I think that's just the pattern we are in. Every new neuroscience method is oversold for a bit when it first becomes popular.

Zen Faulkes said...

That's part of the puzzle for me. When there are people out promoting some new technique or project, do they believe their own hype? Or is it just a calculated marketing ploy?

The Naive Observer said...

I think if you had an honest conversation with any connectomics geek they would acknowledge that the connectomes doesn't hold THE ANSWER. They will, however, argue that to find that answer, we need the connectome.

For instance, your main argument against connectomics is the the fact that the function of a given circuit can be modulated by a slurry of neuromodulators. No one will refute that. But would you suggest that the answer lies solely in the neuromodulators? If that was the case, we should all go back to believing the erroneous notion of "chemical imbalances" and assume that applying an SSRI to the entire brain will cure depression, and not have different effects based on what region and what circuit the drug is acting on.

The reason why the neuroethological studies you alluded to earlier have been so successful is, as you say, because they studied simple, robust behaviors. And if I remember my undergraduate material correctly, sorting out the behaviors relied heavily on mapping local circuits. Unfortunately, the behaviors we want to study in higher organisms are much more complex, and so to understand them we need more complicated wiring diagrams of the local circuits in question. Perhaps the local circuit for some higher functions will actually be the entire brain.

Yes, connectomics will have to take into account neuromodulators. And plasticity. And intrinsic properties. It will also require new and very complicated software and equipment. I don't see any of these lofty requirements as reasons to poo-poo the field though.

miko said...

It really is similar to the ridiculous human genome project hype. We've mostly forgotten the similar grandstanding promises (and the people who made them) from that era.

There are serious people working in connectomics who have more realistic goals and expectations. But I think for a certain kind of scientist, getting a TED talk and a pop book is the actual goal.

Zen Faulkes said...

Naive Observer: I was talking with some of the researchers today at University of Texas San Antonio about this question, In particular, one knows quite a few of the connectome researchers well, and he was of the opinion that many of them did think that connectomes would provide The Answer.

So if people in that field do have more reasonable, nuanced views, it’s not coming across even to their own colleagues.

Bashir said...

The one connectome researcher I know seems pretty reasonable. Very excited about the method but doesn't seem to greatly oversell the potential for THE ANSWER.

In some ways you could perhaps compare this to when fMRI was very new. There seemed to be a lot of hype about how we'd be able to figure out all of human cognition in 10 years. I think things have fallen back to earth a bit into the "useful but no magic bullet" range. Perhaps as connectome research will age the same way.

The FPR said...

Thanks for the insights from neuroethology! I liked Seung's book and also, very much, Olaf Sporns' Networks of the Brain (MIT, 2010). Seung and Sporns take rather different approaches and perhaps reading both gave me a more balanced view than if I'd read Seung alone.

Connectivist said...

Zen: After thinking a bit harder about it, I will say that you are right. At SfN this year I spoke to a number of connectomicists(?) and for the life of me I could not get them to acknowledge that looking at presnaptic bouton size, PSD size or number of vesicles (info that is all available in the EM series) would greatly enhance the interpretation of the function of the circuitry they were looking at. For some, it seems to be all about the connections, and not yet about the nature of those connections. However, one of them did mention that even if assessing those parameters was a priority, there would be no time for it - they are already maxed out as far as data mining goes.

I think though, that if you take the parameters that I mentioned and assess them for a local circuit, you might move toward a more reasonable understanding of that circuit.

As far as more reasonable connectomics researchers go, I quite like Clay Reid's approach, which is to look at cortical circuitry and try to parse out the "rules" that establish that circuitry. Granted, the rules will likely be very complex.

As for Dr. Seung, I won't share my personal opinion, but I will play devil's advocate and ask, if any of you were offered a TED talk, would you turn it down? I did, however, see him promoting his book by wearing a T-shirt that said something like: "'The neuron is my second favorite cell...' page 32." Hows that for fuel to the fire ?

miko said...

I agree there is a divide in the connectomics field between those who have realistic goals and those who think it is THE ANSWER. And thinking about this, I realized that it breaks down fairly cleanly into the former being from biology and neuroscience and the latter being from physics, engineering and computer science.

Jayarava said...

It's kind of obvious from research on neural networks that we won't understand exactly how any given neural network achieves what it does. Simply knowing what's connected to what isn't enough - one has know the weights of the connections, and what mechanisms can alter weights and how, and under what circunstances... and all that. And even then it will be so complex that we won't be able to reproduce any given state.

But the insights gained from neural network research seem worth it - that is De Bono's background after all and I think he has some very interesting insights on the workings of the mind.

Research on this scale always has unlooked for peripheral benefits that could be on enormous value.

In order to make progress one must be single minded, and persevere despite difficulties. I think it's great that some people are aiming for The Answer even if that is quixotic. If only less people were "realistic" and struck out for The Answer we'd know a lot more about everything.

So to me your assessment seems kind of defeatist.

Justin Boland said...

"When there are people out promoting some new technique or project, do they believe their own hype? Or is it just a calculated marketing ploy?"

Necessary synthesis. Ideas need evangelists for maximum fertilization. Those evangelists are of course shrill and shallow, but: they have to be.

Really enjoyed Seung's book. At least he's not, you know....Kurzweil.

Stephen Paul King said...

"Circuit studies suggest a reason for this failure: there is no one way to read the wiring diagram."

Maybe it is not something to be "read" at all. Maybe it is not a circuits within which signals flow. That "flow" may be merely a function that is supporting higher levels of abstraction. Consider how we can build software that emulates hardware virtually and such supports even more software. If we only look at the hardware, will we have a clue as to what is "running on it"? We should be careful about trying to shove Nature's methods into the box of what we understand.