Although the C. elegans connectome has been a boon for scientists who study this nematode's behavior, the past two decades of research have also underscored the staggering intricacy of even a relatively small nervous system. "When you move onto behaviors that are more complex than a quick reflex, you're dealing with especially complicated pathways that are not immediately interpretable because they are not simple circuits—they are networks," explains Scott Emmons of Albert Einstein College of Medicine.
This past summer, in an attempt to confront some of that complexity, Emmons and his colleagues published a connectome of the male nematode's tail, which contains most of the 81 extra neurons that distinguish it from the hermaphrodite (giving the male a total of 383 neurons). It took Emmons and his team about three years to complete and publish the partial connectome: he used more or less the same techniques that Brenner relied on in the 1970s, albeit with faster computers, more powerful microscopes and digital cameras. The male C. elegans connectome also features a crucial piece of information missing from the original draft of its counterpart: synaptic weights. The many connections between neurons are not equal in strength—the more two neurons communicate, the stronger their link becomes and the more likely one is to fire when the other fires. Neurons may also be genetically programmed to form stronger connections with certain partners as the nervous system develops.
Analyzing synaptic weights in the male nematode's connectome has already given Emmons some ideas about neural development. Some neuroscientists have proposed that genes tightly regulate the strongest connections between neurons in C. elegans, whereas weaker connections are more or less accidents—neurons hooking up with whomever they bump into. Emmons's preliminary analysis shows that homologous pairs of neurons on either side of the nematode's body form highly similar strong and weak connections, suggesting that even the weak connections are not entirely random.
Dynamic networks
Synaptic weights are just one of the many layers of information missing from typical connectomes. To understand how neural circuits work, one also needs to know whether the relevant neurons are excitatory—increasing the likelihood that linked cells fire—or inhibitory, muffling their partners instead. Further complicating things, neglected neurons shrivel in the developing brain and new neurons sprout to replace them; in the adult brain, neurons change the strength of their connections with one another daily—such flexibility is essential for learning and memory. Yet another level of complexity involves neuromodulators: certain kinds of neurotransmitters and other small molecules that linger in the fluid surrounding neurons, changing how neurons behave in ways we do not yet fully understand. A prediction about how information will flow through a particular circuit based on a wiring diagram and synaptic weights might be completely wrong if one does not know which neuromodulators are hanging around at any given time.
A good example of how a static connectome fails to capture the dynamics of living neural networks comes from research on the stomatogastric ganglion (STG), a pair of neural circuits in crustaceans—including crayfish, crabs, lobsters and shrimp—that generate rhythmic behavior in response to food. One subcircuit repeatedly constricts and dilates the pyloric region of the stomach, the foyer to the small intestines. Another subcircuit pulsates the gastric mill, a muscular pouch lined with chitinous teeth that help break down food. Mapping all the connections between the 30 neurons in the crustacean STG was an important first step toward understanding how the STG controlled the crustacean digestive system. But it was by no means sufficient. Eve Marder of Brandeis University and others have shown that the neurons in the stomatogastric ganglion do not always use the same unchanging set of connections to communicate with one another. In the presence of certain neuromodulators, a neuron that contributes to the pyloric subcircuit might switch teams, joining the gastric mill subcircuit instead by changing the tempo at which it fires.
Because any brain or nervous system is so much more complex than what a connectome by itself represents, Movshon is certainly not alone in thinking that researchers' limited resources are better devoted to other areas of neuroscience. "I'm all in favor of Seung and others," Bargmann says, "but I don't think we should have a Manhattan Project for the connectome with such a huge amount of resources. We are not quite good enough at reading them. It wasn't like the human genome project, where we knew how to sequence DNA and said, 'Yeah, let's go for it!' Scaling up connectomes is a different issue."
Oliver Hobert of Columbia, another longtime C. elegans researcher, agrees that connectomics only scratches the surface. "It's like a road map that tells you where cars can drive, but does not tell you when or where cars are actually driving," he says. "Still, connectomics of C. elegans has given us wonderful testable hypotheses in terms of how neural circuits work. What we have learned from C. elegans diagrams are not just specific worm behaviors—they are logical principles common to much of biology."
*Editor's Note: The author is a member of NeuWrite, a workshop of scientists and writers that organized the debate at Columbia.
See what we're tweeting about
15 Comments
Add CommentTypical, this new thing hasn't created any big discoveries therefore it is useless. People speaking in absolutes about complex subjects. A circuit diagram of a computer doesn't tell me what a specific program will do but saying the circuit diagram is useless as a result is idiotic. It is also one of those situations that you don't understand how relevant it is until you have it and start trying to apply it. Even if you discovered that the connectome is less important than say spike timing at least you now know something about organisation. It is similar to genetics and how our understanding of the genome revealed how important the non-coding regions are.
Reply | Report Abuse | Link to thisThanks for speaking up. This is a drive-by review of the literature, with no particular conclusions, passing as a state of affairs. I give the author opportunity to respond.
Reply | Report Abuse | Link to thisI've been following this matter for years. I'm also very interested in brain simulation.
Reply | Report Abuse | Link to thisBlue Brain Project conducted by Henry Markram is attempting to do it. Cognitive Computing leaded by ibmer Dharmendra Modha is also trying it.
However, nobody (as far as I know) has ever try to simulate C. elegans nervous system. ¿Why?
To my understanding there must be enormous advantages in achieving it. First in test computer simulations. Second in understanding C. elegans behavior.
Thanks to put light on this.
Worth it? I have no idea.
Reply | Report Abuse | Link to thisThere are limited research dollars so each discipline needs to decide how they get the most value out of those funds.
It's like space exploration. 10 billion dollars...do you spend it on a,b or c? They may all have benefits but funding for one mission means less for another.
Antonio Orbe -- that is just what we are trying to do at the OpenWorm project mentioned above. Please see openworm.org for more details. Thanks for your interest!
Reply | Report Abuse | Link to thisThe discussion which the debaters had in this situation was not nearly so black and white as "should we build a connectome or not?" Rather, the implicit debate was over whether it is useful to have such a highly detailed connectivity map as a "connectome" implies.
Reply | Report Abuse | Link to thisWhen this frame is taken, Movshon's argument clearly wins; which is to say that, as the author of this article alluded to, a static, highly detailed connectome will very quickly lose value as much of that detail changes (as would be seen in a mouse brain). Indeed, the C. Elegans "brain" is probably not a good analogy specifically because it is organized in a highly static fashion (connections are virtually identical across two different animals), which makes this issue of dynamicism much less problematic. In contrast, when trying to understand a more plastic mammalian brain, a more general, statistically reliable (but not perfectly detailed) connection map, as current tract-tracing techniques have sought to build in non-human mammals, suffers less from loss of detail due to circuit plasticity because it doesn't map out many of the connections undergoing dynamic restructing.
It is this argument, comparing a "full", highly detailed connectome to a partial connection map which is what we should actually be framing this debate about. And, when we look at the topic from this perspective, I think it is clear that Seung's argument for a detailed connectome map is flawed.
Finding software in worms
Reply | Report Abuse | Link to thisSoftware means information in memory. Any system that can learn must have software. That’s because software only exists in memory, and learning always means to modify that software. Now, in simple systems, there might not be a box labeled “memory system”, but if the system is learning you know a memory must be emerging. In any computing system with memory (almost all have it) you have to ask, what is the software system doing? This is important because in most systems it’s the software that actually solves the highest level problems and determines complex behavior.
In simple systems like a flatworm most the evolutionary design pressure will be on neuron hardware because that’s most of what there is to the system. As species complexity ramps up, there should be a simplification of their operation, because the design focus moves to a higher level of abstraction (like making big memories – to hold a bigger software system). In other words, it’s a lot more valuable to the organism to make a big memory system than to add more neuron features.
Remember that the software design doesn't come from biology. Software systems in computers don’t come from electronics either. Software is an independent system made of information (typically piped in from the outside), not particles. The genome is another kind of software system, but it’s not really made of DNA. Software is the contents of the memory, not the hardware memory system itself.
Understanding minds and genomes means understanding software systems.
Folks, you are really doing it. Not only neurons but the complete set of cells. Including interaction with the environment. I think you are in the right direction. Actually the most promising project I can imagine. I'm following you. Good luck.
Reply | Report Abuse | Link to thisi vote yes if for no other reason, it can lead to perhaps some understanding of the people who watch the fox news channel.
Reply | Report Abuse | Link to this@jsweck, biological nervous systems are not von Neumann machines and neurons aren't Turing machines. It is a completely different computational paradigm.
Reply | Report Abuse | Link to this"The genome is another kind of software system, but it’s not really made of DNA. Software is the contents of the memory, not the hardware memory system itself." not really sure what you mean by that but the genome is encoded DNA and RNA. Again, you seem to be applying the model of a PC to DNA which is not how DNA works though there are some similarities to Turing machines. Realize that biology is electro-chemical so you really have a hard time differentiating between hardware and software, which is partly why they refer to it as single entity when they call it wetware.
Modeling an entire nervous system, or organism for that matter, would be useful - if done in timeslices a few nanoseconds apart.
Reply | Report Abuse | Link to thisToo bad the tools necessary to perform this task are too costly for a single scientist to afford. This may not be the case many years hence.
Cutting edge science being affordable only via government grant is necessarily shackled and hindered.
C. Elegans is an animal with a complete nervous system and I'm gonna go out on a limb and suggest that there's something it's like to be one of these animals. I mean that it has experience, perhaps including touch/pressure, temperature and chemical discrimination; and that these experiences might involve direct sensation or maybe some kind of emotional feeling. Obviously (?) any such experice would be minimal, and probably wouldn't include any sense of self but in the search for a solution to the hard problem I'd sure like to see the researchers root around in the molecules and proteins that form this animals neurons to look for something that relates the operation of the non-motor neurons to the expression of any kind of experience.
Reply | Report Abuse | Link to thisHi RSchmidt,
Reply | Report Abuse | Link to thisSoftware means information in memory, and that’s it. This doesn’t imply any particular computational paradigm. Any computation (or no computation) can be used to stir the software pot. Software design cares about memory, not so much computation.
Wherever you have a hardware memory system, you must have software to fill it. Software is not some sort of option in computational systems – it must exist. Any system of any complexity has software running at its core, even if the people working with it don’t call it that. When stored in memory, here are some synonyms for the software: state, configuration, data, experience, etc.
I’m not applying a PC model to DNA, I am applying my definition of software, and so I’m able to differentiate the hardware of the DNA memory system from the software of the genome. The genome is information in memory, which means software. The DNA is part of the chemical hardware.
Hardware and software are never a grey area. In any computational system they both must exist separately and independently, like oil and water. The hardware/software boundary is why there is a “nature/nurture” boundary in all computational systems. All software is made of information – not chemicals. With software you create your own completely decoupled and novel structures – in effect your own universe. This is why you can create artificial universes with it.
I hope I’ve answered your questions.
The first known work by Sigmund Freud was a neurologic study of the movement of worms. When you start following a path, you never know where the first step will finally lead you.
Reply | Report Abuse | Link to thisSoftware with no computation
Reply | Report Abuse | Link to thisLet me give an example of software without computation: a book. A book has a hierarchy of hardware that makes it a kind of memory system. A book with the text removed is empty and contains no software, but keeps all of its hardware memory system. The author’s story is normally the content of that memory system. Any memory content is software. So stories in books are software entities that don’t change, and “no change” means “no computation”. Informationally, stories are like genomes, in that they are fixed blocks of information in memory. They simply have different memory system hardware containers.
You can see the separation of hardware and software simply by observing how easy it is to cross the information streams to utterly different hardware - making for example a book filled with the genome, or a DNA strand filled with your favorite story. The software doesn't care because it’s made of information.
Bye.