The same analysis was conducted on the output side of the ventral pathway (Figure 5B). The activation patterns were measured at the third time tick of speaking/naming (see Supplemental Experimental Procedures). Again there was a gradual shift in the type of similarity structure encoded across the

successive layers Doxorubicin (vATL to the insula-motor output layer: left to right in Figure 5B), becoming increasingly sensitive to phonological similarity and less so to semantic similarity. While the pattern of behavioral dissociations varies according to the location of brain damage, the qualitative nature of impairments can also change. For example, recent voxel-symptom lesion mapping studies have demonstrated significant variation in the rate of semantic speaking/naming errors according to the location of stroke-related damage, peaking in the aSTG (Schwartz et al., 2009). The rate of semantic errors produced by the model was compared after various levels of damage to each of its internal layers (to permit a fair comparison, the level of damage for each was titrated to equate overall speaking/naming accuracy, and lesioning was repeated ten times with different random seeds to avoid idiosyncratic results [Figure 6A]; see Supplemental Selleck trans-isomer Experimental Procedures for further methodological details). Figure 6B (bottom) shows

the rate of semantic errors as a function of the location and degree of damage. A 4 (place of damage) × 4 (severity level) ANOVA revealed a significant interaction (F(9, 81) = 19.27, p < 0.001). Increasing damage in MTMR9 aSTG significantly augmented the rate of errors (F(3, 27) = 33.26, p < 0.001). This pattern, albeit less pronounced, was also found for mSTG (F(3, 27) = 40.773, p < 0.001), but not for iSMG or opercularis-triangularis (Fs < 1). In parallel to the patient data, these simulations revealed that the rate of semantic errors was most pronounced after aSTG simulated damage (aSTG versus mSTG: F(1, 9) = 10.82, p = 0.009). Importantly, like the original patient study, these simulation

analysis outcomes remained, even after the comprehension accuracy was controlled (ANCOVA). To explain their results, Schwartz et al. (2009) proposed that aSTG mediates lexical access for speech production. This raises a conundrum, however, in that the same region is associated with both verbal and nonverbal auditory comprehension in patient, rTMS, and functional neuroimaging literature (Patterson et al., 2007, Pobric et al., 2007, Scott et al., 2000 and Visser and Lambon Ralph, 2011). Using the regression-based method (see above), we probed the functioning of the aSTG layer of the model across tasks (comparing the aSTG-associated β values [highlighted with a light-gray box] in Figure 5A [comprehension] versus Figure 5B [speaking/naming]). Given the structure of the model and the lesion simulations summarized above, it is clear that the region is critical both in auditory comprehension and speaking/naming.