The pyramidal neuron is the principal cell type in the mammalian forebrain, but its function remains poorly understood. Using a detailed compartmental model of a hippocampal CA1 pyramidal cell, we recorded responses to complex stimuli consisting of dozens of high-frequency activated synapses distributed throughout the apical dendrites. We found the cell's firing rate could be predicted by a simple formula that maps the physical components of the cell onto those of an abstract two-layer "neural network." In the first layer, synaptic inputs drive independent sigmoidal subunits corresponding to the cell's several dozen long, thin terminal dendrites. The subunit outputs are then summed within the main trunk and cell body prior to final thresholding. We conclude that insofar as the neural code is mediated by average firing rate, a two-layer neural network may provide a useful abstraction for the computing function of the individual pyramidal neuron.

No abstract

We consider the combined effects of active dendrites and structural plasticity on the storage capacity of neural tissue. We compare capacity for two different modes of dendritic integration: (1) linear, where synaptic inputs are summed across the entire dendritic arbor, and (2) nonlinear, where each dendritic compartment functions as a separately thresholded neuron-like summing unit. We calculate much larger storage capacities for cells with nonlinear subunits and show that this capacity is accessible to a structural learning rule that combines random synapse formation with activity-dependent stabilization/elimination. In a departure from the common view that memories are encoded in the overall connection strengths between neurons, our results suggest that long-term information storage in neural tissue could reside primarily in the selective addressing of synaptic contacts onto dendritic subunits.

The mechanisms that determine how information is allocated to specific regions and cells in the brain are fundamentally important for memory capacity, storage and retrieval, but are poorly understood. Here, we manipulated CREB in a subset of lateral amygdala (LA) neurons with a modified Herpes Simplex Virus (HSV), and reversibly inactivated transfected neurons with the Drosophila allatostatin G-protein-coupled receptor (AlstR)/ligand system. We found that inactivation of the HSV-CREB subpopulation of neurons with allatostatin (AL) during training disrupted memory for tone conditioning, while inactivation of a similar proportion of HSV-LacZ control neurons did not. Whole-cell recordings of fluorescently tagged HSV-CREB neurons revealed that neurons with higher CREB levels are more excitable than neighboring neurons, and show larger synaptic efficacy changes following conditioning. Our findings demonstrate that CREB modulates the allocation of fear memory to specific cells in lateral amygdala, and suggest that neuronal excitability plays a key role in this process.

The rules of synaptic integration in pyramidal cells remain obscure, in part due to conflicting interpretations of existing experimental data. To clarify issues, we developed a CA1 pyramidal cell model calibrated with a broad spectrum of in vitro data. Using simultaneous dendritic and somatic recordings and combining results for two different response measures (peak versus mean EPSP), two different stimulus formats (single shock versus 50 Hz trains), and two different spatial integration conditions (within versus between-branch summation), we found that the cell's subthreshold responses to paired inputs are best described as a sum of nonlinear subunit responses, where the subunits correspond to different dendritic branches. In addition to suggesting a new type of experiment and providing testable predictions, our model shows how conclusions regarding synaptic arithmetic can be influenced by an array of seemingly innocuous experimental design choices.