I have been reading about neuron creation, guidance cues and all sorts of highly complex mechanisms used to allow one neuron axon to extend or connect - but to what end?

Why does one neuron end up connecting to another? (instead of all the others)

  • $\begingroup$ Are you asking for a certain neurobiological mechanism or making a philosophical question implying some sort of decision making? $\endgroup$ – Lazaros Mitskopoulos Sep 19 '14 at 11:35
  • 1
    $\begingroup$ I'm after the effect of the biological mechanism - i.e. I know there are factors inside and outside the neuron which the mechanism consists of - but I want to know what the effect is. E.g. Complex mechanism inside a neuron cause an action potential but the overall effect is an addition - "neural integration". I want to know the effect of neuron guidance/axon growth etc for a single neuron. $\endgroup$ – James Sep 19 '14 at 11:42

Donald Hebb's postulate only applies when two neurons are already connected. It seems you are asking more specifically, 'when two neurons are not already connected and they want to connect, how and why?' Correct?

In this case, we do not know. StrangeLoop mentions it is due to the location of the neurons and the spreading of activation. Yes these might be the case but evidence is extremely limited.

It may be the case that correlated activation of two neurons instigates the formation of a synapse between them. It may also be the case that synapses form completely randomly and plasticity takes advantage of any new synapses (don't forget anti-hebbian). It may also be a combination of both. We do not know.

We further do not know concretely what happens in the reverse; for example, what causes a synapse to be eliminated, is it activity driven or a stochastic process?

Some recent evidence showed that spine generation and elimination are related to 'learning' and 'unlearning'. This suggests that activity driven plasticity can form/eliminate synapses, but my previous reference shows strong evidence that formation/elimination are stochastic -- hence we do not know!

Another thing which is not clear from your question, as James mentions, is at what age you are considering. Early development of synapses is more well understood and it may be the case that during early development a group of mechanisms determine why/where/how a synapse forms/eliminates and during adult another set of mechanisms are in charge.

For example, in the early post-natal visual system, it is found that cells have responses with similar properties as adult; however, later on in development visual experience is required to maintain these properties. Furthermore, these changes must occur within a critical period for activity-dependent plasticity.


Donald Hebb originally formulated what would later come to be known as spike-time dependent plasticity by famously stating "neurons that fire together wire together". In actuality, the firing has to be sequential (not simultaneous) and causal: if a neuron A fires and causes B to consequently fire, the synaptic strength between them increases. This is how the brain implements learning and memory. [1] This principle fits with the idea that the brain essentially learns causal relationships that exist in the world, as formalized in the Bayesian brain hypothesis [2].

[1] Markram et al., "A history of spike-timing-dependent plasticity": http://infoscience.epfl.ch/record/168812/files/fnsyn-03-00004.pdf

[2] Clark et al., "Whatever next? Predictive brains, situated agents, and the future of cognitive science": http://www.fil.ion.ucl.ac.uk/~karl/Whatever%20next.pdf

  • 1
    $\begingroup$ This is a very helpful answer, thankyou. Do you think the statement "neurons that fire together wire together" could help explain why two separate neurons have a third come in and connect them? I suspect that's not what the statement meant, if not, how does this logic encourage a neuron to connect to another specific neuron do you think? $\endgroup$ – James Sep 19 '14 at 12:02
  • $\begingroup$ You're welcome. I really recommend reading some of the papers contained the second article, by the way, it gives a nice mix of psychological/neuroscientific/philosophical positions on the "predictive brain". As to your follow-up question: if you imagine that a certain neuron represents an event/thing, that something can have many causes! If the environment repeatedly tells the brain (by causing the respective activations) that patterns A, B, and C tend to cause D, let's say, ALL of these connections will be enhanced. $\endgroup$ – user6682 Sep 19 '14 at 12:10
  • 1
    $\begingroup$ That second link does look very interesting, I'll have a read - thanks. I understand the point about connection strengthening, my question is more, given a collection of neurons already connected, say A and B feed into C, where does a new neuron D decide to join up? $\endgroup$ – James Sep 19 '14 at 12:17
  • 5
    $\begingroup$ @James There's a lot of movement of axons during early development that is driven by gradients of different growth factors that guide the growth cones of the nascent axons, so while StrangeLoop's answer is partly correct, it's ignoring the embryology a bit. $\endgroup$ – Chuck Sherrington Sep 19 '14 at 18:03
  • 1
    $\begingroup$ "The main thing that changes is how D connects to the others, and that changes based on the direction in which activation has previously spread." @StrangeLoop What is the evidence for this? I suspect this is conjecture. James, as Chuck mentions, early development, both pre- and postnatal, are driven by growth factors. This sets out an initial topology. It is observed that synapses can then form between adjacent dendrite and axon; what causes this is not fully understood (see references in my answer). $\endgroup$ – James Oct 2 '14 at 23:56

From a developmental perspective, this question is partially addressed in the pages of Mechanisms for Connectome Development by Marcus Kaiser. Spatial considerations are an important constraint on the probability of a connection between pairs of neurons in two ways:

  1. The first factor is the neuronal density which he explains very clearly:

...as a higher number of neurons per volume make it more likely to hit a potential target neuron. Say that for each volume element (e.g., 1 μm3) there is a probability $p$ that the space contains a neuron and a probability $q = 1 − p$ that the space is empty. Hitting another neuron three volume elements away from the starting point given a (certain) direction means two times passing through empty volume elements and one time (the last growth step) entering a volume element that contains a neuron (Figure 2B): the probability to hit another neuron after n steps or passed volume elements is $P(X = n) = qn−1 > \cdot p$. The exponential decay of the probability to encounter another neuron as a function of distance between two neurons (or $n$ steps) means that it is more likely to hit a nearby neuron than hitting a neuron that is far away after failing to encounter many nearby neurons. Imagine going through a crowded room in a straight line. You are more likely to bump into someone early on than only passing through an empty space and reaching a person who is farther away. Therefore, basic considerations for axon growth can already account for the observed exponential decay of connection probability with distance.

  1. The degree of curvature during axonal growth is another consideration as a random walk in space would actually increase the probability of revisiting spaces that don't contain a neuron. But, there may be reasons for changing course:

While axons normally grow in a straight line, their target neurons might often be off-course leading to the need to change the growth trajectory. Causes for changing direction could be physical obstacles that either block the path or lead to adhesion of the axon. Alternatively, concentration gradients can lead an axon towards a target. In this way, axonal growth cones detect the concentration of a molecule and follow a gradient of increasing concentration towards the source of that molecule [22]. However, the concentration of a molecule decays exponentially with the square of the distance between source and growth cone. Given the diffusion constant in vivo [23], the estimated maximum distance across which a growth cone can detect molecules emitted from a target neuron is 1 cm [24].

And what about axons that are more than a centimeter in length? Dr. Kaiser addresses this case as well. In fact, the whole paper is an enjoyable read and covers many related questions.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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