Both the LL and ANP theories rely on certain common assumptions, including the concept of pooling the activities of neurons. Under LL theory, neurons are pooled into groups or “types”, in which there are as many different types of cells as there are taste qualities. Here, each type is composed of neurons that share a common optimal stimulus and it is the quality of only this stimulus that a type encodes. On the other hand, ANP theory posits that taste qualities are encoded by the pooled response of many like-tuned and heterogeneously-tuned neurons, which gives rise to a pattern of activation across the population. Curiously, ANP theory also relies on the concept of neuron types. For ANP theory to work, the pool of neurons that generates the patterns must be composed of cells with different tuning characteristics: There must be a mix of different types of cells to get unique patterns for different stimuli. Indeed, there are variants of ANP theory in taste that define explicit roles for groups of differently tuned neurons in generating distinct ANPs (Smith et al. Reference Smith, St John and Boughter2000).

But the way that neural groups and pools are typically defined and/or interpreted in the taste literature is subject to caveats that are sometimes overlooked or not explicitly acknowledged. One issue stems from the fact that neural data sets in taste are usually composed of many neurons sampled from many animals. The taste response characteristics of central gustatory neurons are influenced by a number of factors and neural responding could vary widely among animals as a function of variables that are not usually controlled, such as satiety state (Giza et al. Reference Giza, Scott and Vanderweele1992). Such variation could induce vast differences in the sensitivities of neurons that are not exactly a result of strict differences in the efficacy of sensory input. With just this in mind, it becomes questionable as to whether or not sets of neurons generated from many-animal data sets indeed reflect accurately those that would be observed across many neurons in an individual animal. Newer recording and optical technologies may prove fruitful in monitoring the activities of many neurons simultaneously.

Spatial codes that require cells to be segregated into defined types must assume that all neurons composing a type serve the same function and that the response properties of cells within a group are orthogonal. These assumptions refer to the independence of responses that is implied to exist among cells that are of a common type. These assumptions, however, cannot be easily validated using traditional taste data sets. There is presently only very limited means of relating the tuning properties of taste neurons to their function, such as, for example, whether they contribute to taste perceptual identifications or oromotor responding. What is more, there is not a method to determine independence among the activities of taste neurons based on their stimulus-response characteristics. For example, two cells with common tuning that would be grouped together and assumed independent under a classic typing scheme could actually be synaptically related, where one cells serves as a driver for the other. Here, there is not independence in the firings of these cells but statistical dependence, whereby a spike in the presynaptic cell increases the probability of firing in the follower neuron. To resolve response relationships among cells would depend on knowing the architecture of the neural circuit into which the cells are embedded, the exact locale of each neuron in the circuit, and the patterns of connectivity among cells. Data on these topics for taste nuclei are scant at best.

Models of coding must also account for the variability in responding that is inherent to taste neurons. A recent study revealed that in many gustatory neurons the response to a given stimulus can vary widely from trial to trial (Di Lorenzo & Victor Reference Di Lorenzo and Victor2003). For some cells, within-neuron response variability impacted the definition of their best stimulus, where variability in responding rendered a changeable best stimulus over trial blocks. This observation poses a clear challenge to a LL mode of operation. It also could be a potential issue for discriminating certain stimuli under an ANP code, particularly stimuli that are perceptually distinct but generate ANPs that differ only mildly. Tastants classified as bitter or sour are examples of such stimuli. Studies of central taste neurons in rodents have shown that ANPs evoked by these stimuli, as based on single-trial data, are strongly correlated (Lemon & Smith Reference Lemon and Smith2005). Factoring in a certain element of across-trial response variability among neurons renders a variance about the neural mass calculated from a stimulus pattern: ANPs to certain bitter and sour tastants could actually at times be more similar than the strong correlation already revealed by the single-trial data. This becomes paradoxical when viewed against behavioral data showing that rats can readily discriminate between the tastes of sour and bitter tastants (Grobe & Spector Reference Grobe and Spector2006). Part of this conundrum lies in that it is not exactly clear how much of a difference between ANPs would be sufficient to compute perceptual taste discriminations.

Considerations for further development of spatial codes in taste might include quantifying neuronal responses along behaviorally relevant stimulus-response windows. For example, perceptual identifications of taste stimuli can happen in less than 1 second as indexed in rat behavioral experiments (Halpern & Tapper Reference Halpern and Tapper1971), although many neurophysiological studies have quantified taste responses over 5- or 10-second periods. What is more, in some taste neurons there is a particular envelope to a stimulus response, wherein the magnitude of firing dynamically changes over the response time course in a stimulus-specific way (Katz et al. Reference Katz, Simon and Nicolelis2001). Such neuronal dynamics would suggest that the overall spatial neural signal, whatever its format, possesses a time-dependent plasticity that could be captured only by considering time as a factor in the code. These time dependencies in spatial activity might reflect parsing by the nervous system of different features about taste stimuli as related to ongoing streams of perceptual and behavioral processing (Katz et al. Reference Katz, Simon and Nicolelis2001). These are only a few issues associated with models of spatial coding in taste.