Synaptic integration — the summation of all the inputs that a neuron receives — is a challenging problem in cellular neurobiology, owing to the unanticipated complexity of the active properties of the dendritic tree, which depend on the presence of many types of voltage-gated ion channels. One consequence of this complexity is the existence of contradictory data that point to different rules of synaptic summation. Two recent papers in Neuron have tried to go beyond these contradictions by proposing a new model of CA1 pyramidal neurons, which might help to clarify how integration takes place.

The model that Poirazi et al. propose is broader than previous examples in that it combines results from different response measures (mean versus peak amplitude of synaptic potentials), stimulus formats (single stimuli versus trains) and conditions of spatial integration (within versus between dendritic branches). Using this model, they found that, below the threshold for action potential generation, inputs sum non-linearly within higher-order branches and linearly between branches. This result has profound implications for the design of future experiments and for the interpretation of previous data.

But what about stimuli that elicit action potentials when integrated? In the second paper, Poirazi et al. argue that we can think about a CA1 pyramidal neuron as a two-layer 'neural network'. In the first layer, higher-order branches act as independent subunits that add inputs non-linearly. In the second layer, the dendritic trunk and the soma linearly sum the output of the subunits and compare the result to the threshold for action potential firing.

Although the model does not address other aspects of synaptic integration, such as temporal summation, and might not apply to other classes of neurons, the work of Poirazi et al. constitutes a useful heuristic tool that makes specific predictions, which can now begin to be empirically tested.