Dendrites are simply a part of a neurone; the long and intricate tree like structure that receive thousands of signals. They look spectacular, and have fascinated neuroscientists since their discovery.
But what do they do and why have dendrites.1 Do they simply efficiently add physical space for all the inputs, or do they amplify or alter the input signals that then switch the neuronal somata on, or are they themselves in fact a switch – a transistor and if so for what?
Dendrites add space for the 10’s of thousands of input connections each neurone receives. But these large dendritic trees also separate the connections in space and this may be a functional segregation. Many neurones receive inputs from one brain region onto one dendritic area. For instance, granule cell in the dentate gyrus receive input from the lateral entorhinal cortex on their distal dendrites and from the medial entorhinal cortex on their more proximal dendrites. Interestingly, these different input regions carry different types of information – the medial entorhinal cortex transmits spatial information while the lateral entorhinal cortex transmits contextual information. The spatial organisation of input onto dendrites may be even more finely tuned with each branch of the dendritic tree receiving functionally distinct input and this spatial compartmentalisation is an important ability of dendrites.
Form affects function and of course, where inputs are placed on dendrites affects their function in neuronal computation. The traditional view of neuronal computation is that inputs from the dendrites travel to the soma where they are integrated together, and if the inputs are sufficient the axon generates an action potential output. The further the inputs are from the soma, the less effective the inputs are – simply think of water travelling along a long leaky hose. So then the question becomes are dendrites actually a bug or a feature. If dendrites add space so as to add more inputs, but the distal inputs have no voice at the soma, what is the advantage to neuronal computation? Many studies have now shown that dendrites do not simply act as passive ‘leaky’ cables, but contain many active properties that not only compensate for the location of the inputs but change the input signal.
Dendrites are not just passive communicators. Active signals such as dendritic spikes (d-spikes) change the input signal. In the traditional view above, all input on the dendrite is linearly summed. A computational model of a neurone without dendrites (known as a point neurone model) successfully recapitulates this integrate and fire behaviour. But what if dendrites were capable of integrating input non-linearly and generating d-spikes. The addition of active properties to dendrites means dendrites are capable of supra-linear integration, and so clustered input generates larger responses and more likely to activate the neurone. Theoretically, this addition of dendrites with active properties is though to allow the combining of features into a discrete object (eg. red colour and apple shape into an apple).2 And when features are connected on different dendrites the neuron can perform an exclusive-or (XOR) computation (ie the neurone will be active for a yellow banana or a red apple but not the combination of red colour and banana shape). These active properties that allow the dendrites to generate supralinear integration vastly increase the computational power of a single neurone increasing the number of different combinations a neuron can differentiate.3 The computational work suggests that dendritic properties of the sub-compartment determine how the inputs combine and if the feature is encoded in the memory engram.
Dendrites and behaviour. Much recent work has investigated if dendritic processing actually contributes to behaviour. Non-linear dendritic processing has been observed in various computations in the behaving animal (orientation tuning of V1 neurones).4 The ability of animals to navigate space necessitates combining contextual and spatial features. The combination of specific contextual and spatial features create ‘place’ and are coded for by place cells. A recent study measured the electrical activity of CA1 neurones as an animal walked along a linear track. Different cells were active at different places along the track. Dendritic spikes were correlated with this place field firing and strengthened coding of this specific place. Dendritic spikes were also seen in previously quiescent cells and induced place field firing, triggering these cells to now code for that place. Thus dendritic activity is thought to be important for combing contextual and spatial information to code for place.
Dendritic spikes were also correlated with the ability of an animal to perceive touch. Animals’ whiskers were moved and when moved large enough the animal perceived the touch and licked a spout to received a water reward. Calcium imaging was used to assess activity in individual neurones and their dendrites during the behaviour. Dendritic Ca2+ events were correlated with the animals perceiving the touch and licking (a hit). When the animals’ whiskers were touched but they did not respond by licking (a miss), the same dendrites were not active.
Not only are dendrites quite beautiful structures but their properties add to the computational ability of neurones. We are now beginning to see how the computational advatages of dendrites are used in behaviour.