30
$\begingroup$

As far as I understand, the basics of neurogenesis (abstracted down to the level that makes sense to a computer scientist) is as follows:

  1. Neural progenitor cells differentiate into new neurons that have zero (or very few) synaptic connections, but are sensitive to the local chemistry.
  2. (Optional) Sometimes (such as in adult neurogenesis in the olfactory bulb) these immature neurons-to-be are produced far away from where they are needed and follow standard pathways to migrate to the correct brain region
  3. The immature neuron extends dendrites towards upstream neurons and starts to develop an axon
  4. The immature neuron extends axon towards downstream neurons, and
  5. The neuron matures and becomes indistinguishable from the network it joined.

I've eliminated many of the biological details, since I want to just capture the details (more info). The part that seems to be not well described, and my question, is:

  • How do neurons select where to make their initial dendrite and axon connections?
  • How/when does a given network notify the progenitor cells that they should differentiate?

Partial results

Cecchi et al. (2001) proposed a model where the new neurons produce initial connections randomly, and then the only the ones that contribute to the function of the network end up firing frequently and surviving, and the rest die. This inspired by the observation of high death rate among immature neurons. However, there is not biological evidence to support that the initial connections are in fact random, and I hope that there has been more evidence gathered since the early work in 2001.

Notes

  • This question is motivated by the search for computational models with a biologically reasonable account of neurogenesis.

  • I am looking for some sort of abstract local explanation to my questions at-least at the rule-of-thumb level. For examples, if I was asking about plasticity of established connections (and not neurogenesis in particular), then Hebb's "neurons that fire together wire together" would be a sufficient answer, although STDP would be a better one. However, I do not need all the biological details, just the high level rule if one is known.

References

  • Cecchi GA, Retreanu LT, Alvarez-Buylla A, Magnasco MO – “Unsupervised Learning and Adaptation in a Model of Adult Neurogenesis” Journal of Computational Neuroscience; 11:175-182; 2001 [preprint]
$\endgroup$

2 Answers 2

22
+300
$\begingroup$

There is a huge body of literature on axon growth cone guidance which will give you some insights into how the biology works. Unfortunately, incorporating it all into a model is probably going to make it unwieldy unless your express purpose is to model the physiology, which doesn't seem like the case.

Here are some references:

Hong K, Nishiyama M. (2010). From guidance signals to movement: signaling molecules governing growth cone turning. Neuroscientist, 16(1),65-78.

This is pertinent because it explicitly mentions adult neurogenesis, as much of the subject is devoted to the developing nervous system in models like the developing chicken.

Kolodkin AL, Tessier-Lavigne M (2011). Mechanisms and molecules of neuronal wiring: a primer. Cold Spring Harb Perspect Biol, 3(6), 1-14 Free PDF

To call Marc Tessier-Lavigne a leader in the field of neuronal growth would be an understatement. This article also covers some of the cues of synaptogenesis as well.

Simpson HD, Mortimer D, Goodhill GJ (2009).Theoretical models of neural circuit development. Curr Top Dev Biol, 87, 1-51.

Regrettably, I do not have access to this article, but it appears to be more along the lines of computationally realistic representations. It does state that their models for synaptic strengthening are based on Hebbian Learning, so that's at least in line with what you presumed that you needed.

In searching for more computational models, ephrins and integrins are two of the cell surface agents that are heavily involved in the process, so any abstractions of those would make for a good model.

$\endgroup$
0
7
$\begingroup$

For the dentate gyrus, which is probably more closely analogous to a feedforward hidden layer in a memory network, here are some answers:

  1. Axon and dendrite connectivity is essentially local and can probably be assumed to be initially random within that local region. That is, a neuron integrating into the DG at the midpoint (along the long hippocampal axis) will have its dendrite arborize in that same location, thus receiving cortex inputs from the region of the cortex that topographically projects to that area. Likewise, its axon will target CA3 neurons at the midpoint of the hippocampus. For this reason, the general connection statistics of a newborn neuron will ultimately be similar to a neighboring neuron; though the (likely) competition between young and old synapses and the high synaptic plasticity of young neurons will distinguish the two neurons computationally.

  2. In the DG, the rate of neurogenesis is regulated by some external conditions, such as running and (likely) affect/mood. These can probably be thought of as predictive indicators of future need of neurons. Network activity is probably more critical for the survival of neurons.

  3. The OB is most likely at least somewhat different on both these points. One should take into account that newborn OB neurons are not "typical" neurons (in the abstract NN sense) by any means; OB-GCs rely more on dendro-dendritic communication (no axon) and are inhibitory.

$\endgroup$
3
  • 2
    $\begingroup$ Do you have the references for this information? I'd like to look into this further. $\endgroup$ Feb 17, 2012 at 12:24
  • $\begingroup$ Great answer, though as jonsca said, references would be ideal for support/further reading $\endgroup$
    – Ben Brocka
    Feb 23, 2012 at 20:07
  • 1
    $\begingroup$ I don't understand your first point. You start with "it is random in the local region" but end with "the general connection statistics of a newborn neuron will ultimately be similar to a neighboring neuron"? Which is it? Random connections and preferential attachment tend to produce very different network structures. I also agree with @ChuckSherrington and would prefer if you included some references for people like me that are not familiar with neuroscience. $\endgroup$ Apr 29, 2012 at 5:03

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.