The original conception of across-fiber patterns (AFPs) described by Erickson was based on Young's theory of color vision, which included the idea that the eye is provided with three distinct sets of broadly tuned receptors (Young Reference Young1802; Reference Young, Teevan and Birney1807). Although Erickson agreed that receptor or neuron types are similarly required for taste, he stated their existence has not been established, as they have not been formally defined. This failure to formally define taste cell types results in an AFP theory that is not testable, and therefore, AFPs suffer the same deficit Erickson ascribed to the concept of labeled lines / basic tastes (LLs). A serious consequence is that the LL and AFP models of taste coding are largely indistinguishable, creating confusion at both theoretical and empirical levels.

The relationship between the two models can be compared to alternating perceptions of a Necker cube, an analogy used famously by Dawkins (Reference Dawkins1976) to suggest that what appeared to be two different theories were actually only two different ways of looking at the same theory. For instance, LL and AFP information could be encoded within the same neural representation: the rich information contained in an AFP could be utilized by brain circuits that discriminate tastes, and in parallel, best-response neuron activity embedded within that AFP could be used to convey categorical information for circuits that make rapid yes/no decisions, such as an infant's acceptance of sweet tastes and rejection of bitter ones. In instances where virtually all of the AFP information is carried in the best-response neurons, the AFP and LL signaling are nearly identical – tuning this narrow may actually occur for sweet taste in primate primary taste fibers (Danilova et al. Reference Danilova, Danilov, Roberts, Tinti, Nofre and Hellekant2002). Similarly, Erickson suggested that neural approaches for yes/no circumstances could require only LLs in the form of an AFP reduced to a single neuron and that this possibility is consistent with the AFP model.

The lack of uniqueness of LLs and AFPs leaves open the possibility of interpreting any data as being for or against either model. This is illustrated here by reviewing the types of evidence that have typically been taken as support for the LL model, followed by a demonstration of how this evidence can be turned on its head to fit nicely into an AFP framework.

Groupings of taste stimuli into basic tastes using cluster analysis of neuron responses to multiple basic taste stimuli have been largely replicated across species and nervous system levels (e.g., Scott & Plata-Salaman Reference Scott and Plata-Salaman1999). Moreover, their validity has been strengthened by showing that conditioned taste aversions (CTAs) involving a basic taste generalize to taste stimuli in the same basic taste category, but not to those of other categories (e.g., Nowlis et al. Reference Nowlis, Frank and Pfaffmann1980). Erickson rightly argues that although these cluster analyses may separate neurons, they do not define types. Nevertheless, he suggests that if the neurons within one group responded similarly to manipulations of the taste system, and neurons of other groups responded differently, such evidence would favor neuron types (Scott & Giza Reference Scott and Giza2000).

Indeed, CTA, pharmacological, nutritional, and hormonal manipulations, especially of salt- and sweet-responsive neuron groups, including sodium deprivation, lingual amiloride or gymnemic acid application, deoxycorticosterone acetate (DOCA) priming followed by intracerebroventricular renin, and intravenous insulin or glucose, have provided such evidence (Chang & Scott Reference Chang and Scott1984; Giza & Scott Reference Giza and Scott1987a; Reference Giza and Scott1987b; Hellekant et al. Reference Hellekant, Ninomiya and Danilova1998; Jacobs et al. Reference Jacobs, Mark and Scott1988; McCaughey & Scott Reference McCaughey and Scott2000; Scott & Giza Reference Scott and Giza1990; Verhagen et al. Reference Verhagen, Giza and Scott2005). These findings are strengthened by functional-anatomical relationships of taste pathways that are consistent with LLs, including the innervation of single or small numbers of taste buds by primary taste neurons in some species, transynaptic neural tracing demonstrating differential projections of sweet and bitter receptors, differential localization of the activation of brainstem sensory neurons by bitter versus sweet or acidic stimuli, and unique regions of cortical activation associated with each basic taste (Accolla et al. Reference Accolla, Bathellier, Petersen and Carleton2007; Harrer & Travers Reference Harrer and Travers1996; Sugita & Shiba Reference Sugita and Shiba2005; Travers Reference Travers2002; Zaidi & Whitehead Reference Zaidi and Whitehead2006).

Mouse transgenics have also been used to selectively manipulate individual taste receptor cell types, and the results have been taken by some to be a hands-down win for LLs (Chandrashekar et al. Reference Chandrashekar, Hoon, Ryba and Zucker2006). Knockouts of sweet or bitter taste receptors and ablation of cells expressing sour receptors each resulted in loss of sensitivity to the respective basic taste without affecting responses to other basic tastes (Damak et al. Reference Damak, Rong, Yasumatsu, Kokrashvili, Varadarajan, Zou, Jiang, Ninomiya and Margolskee2003; Huang et al. Reference Huang, Chen, Hoon, Chandrashekar, Guo, Trankner, Ryba and Zuker2006; Zhao et al. Reference Zhao, Zhang, Hoon, Chandrashekar, Erlenbach, Ryba and Zuker2003). Consistent results have been obtained using genetic “rewiring” of taste pathways: Expression of a bitter receptor in taste cells normally expressing only sweet receptors resulted in “rewired” mutant mice, which readily ingested a bitter stimulus that was aversive to normal mice (Mueller et al. Reference Mueller, Hoon, Erlenbach, Chandrashekar, Zuker and Ryba2005).

Additionally, one of the biggest stumbling blocks for LLs has been data suggesting taste receptor cells are broadly tuned (e.g., Gilbertson et al. Reference Gilbertson, Boughter, Zhang and Smith2001). However, it has recently been shown that each taste receptor type occurs in a different population of taste receptor cells (Chandrashekar et al. Reference Chandrashekar, Hoon, Ryba and Zucker2006), and about 80% of Type II taste receptor cells, each of which contains only one type of taste receptor, respond to only one basic taste (Tomchik et al. Reference Tomchik, Berg, Kim, Chaudhari and Roper2007), suggesting that taste receptor typologies – potential substrates for LLs – exist.

In total then, the empirical data seem quite strong in favor of LLs. Why then hasn't the knockout punch to the AFP model been delivered? It could be argued that manipulation of a specific taste neuron, receptor, or receptor cell type would not only affect a LL, but could also have a specific effect on AFP coding. For example, in the genetically “rewired” mice described earlier, sweet-best neurons would become bitter-best neurons, and cells in the population that normally responded in any degree to sweet receptor activation would now respond in a similar degree to bitter receptor activation. Consequently, a bitter taste would now evoke the AFP normally activated by sweet taste. Therefore, as a result of the failure to develop formally sound definitions for AFP and LL models, this genetic manipulation, or, in fact, any selective manipulation, cannot distinguish these models.