Consequences of the burst duration code
Although the scope of the model was limited to the generation of a burst duration code at the level of a single granular cell, certain predictions can be made concerning propagation of this code to cells postsynaptic to the granular cell layer (medium ganglion, large fusiform, and large ganglion cells of the ELL), robustness of the code over a population of granular cells whose responses are subsequently pooled, and the influences of independent and correlated noise within this population. Furthermore, it is worthwhile to consider aspects of the granular cell burst that might provide additional information about stimulus intensity to the aforementioned postsynaptic cells, such as spike frequency and relative spike timing.
Generation of what appears to be a neural code for stimulus intensity in the granular cell layer, by the criteria of an external observer, does not guarantee that such a code is actually used by the fish. The biological relevance of apparent neural codes, first discussed extensively in the late 1960’s (Perkel and Bullock, 1968), remains a heated issue. While it may be quite difficult to completely describe all aspects of a neuron’s behaviors that serve a coding function in the organism, it is possible to reason about putative codes in the sense that physiology must provide a means for the code to be "read" at each neural relay as it propagates through the nervous system.
In order for the burst duration code to be succesfully "read" by postsynaptic cells, these cells must be sensitive to the full range of granular cell outputs. This could be limited even presynaptically by the transformation from granular cell spikes to amount of released neurotransmitter. High frequency presynaptic events can produce either facilitation or depression of neurotransmitter release, especially during a brief burst with particularly small interspike intervals (ISIs). Facilitation is enabled by residual Ca2+ in the presynaptic terminal that accumulates during the burst, which accelerates Ca2+ dependent fusion of neurotransmitter-containing vesicles to the cell membrane. Depression, on the other hand, is caused by depletion of the ready-releasable pool of vesicles following rapid, high frequency stimulation. It could be expected that, if the full granular cell burst duration code is biologically relevant at the postsynaptic cell, presynaptic facilitation predominates over depression during typical bursting events. If presynaptic depression played a major role in these bursts, the dynamic range of the code would be greatly reduced. Of course, these considerations are irrelevant for signal propagation via gap junction electrical synapses.
Postsynaptic pooling of granular cell responses within a spatially restricted area, despite potentially decreasing the spatial resolution of the code for local stimulus intensity at the skin, would increase the robustness of the burst duration code by eliminating independent noise created by the variability of single granular cell responses. As shown in model sensitivity tests, the transformation from input delay (i.e. primary afferent latency-to-fire) to spike number in the granular cell varies in its fidelity as parameters such as maximum ionic conductances and EPSP amplitudes and decay times are varied over a 50% range with respect to their putative baseline levels. Although for some parameters these perturbations produced nearly uniform burst durations over all input delays, an average of all burst duration vs. input delay curves over these parameter ranges would likely retain the desired coding properties.
Given that channel densities, channel gating kinetics, and input EPSP properties vary independently for each granular cell, response pooling via postsynaptic convergence or some form of ensemble representation would greatly decrease noise in the coding of stimulus intensity. Furthermore, it seems plausible that some mechanism of regulation might control the relative degree to which response pooling, which increases coding reliability, is balanced by use of single granular cell responses, which retains spatial resolution.
In any case, the use of a burst duration code for stimulus intensity at the granular cell layer is more robust than a binary single spike code. Single spikes are unreliable signals for several reasons: first, a single depolarizing event at the presynaptic terminal causes neurotransmitter release with a probability much less than 1 (Lisman, 1997). Second, any correlated noise among a population of cells coding with single spikes would shift all responses to either on or off. In other words, the burst duration code is desirable both because it is graded and relatively insensitive to correlated noise and because it operates within a range of spike numbers and frequency that is guaranteed to produce postsynaptic responses.