Frontal networks for learning and executing arbitrary stimulus-response associations

Abstract

Flexible rule learning, a behavior with obvious adaptive value, is known to depend on an intact prefrontal cortex (PFC). One simple, yet powerful, form of such learning consists of forming arbitrary stimulus-response (S-R) associations. A variety of evidence from monkey and human studies suggests that the PFC plays an important role in both forming new S-R associations and in using learned rules to select the contextually appropriate response to a particular stimulus cue. Although monkey lesion studies more strongly implicate the ventrolateral PFC (vlPFC) in S-R learning, clinical data and neurophysiology studies have implicated both the vlPFC and the dorsolateral region (dlPFC) in associative rule learning. Previous human imaging studies of S-R learning tasks, however, have not demonstrated involvement of the dlPFC. This may be because of the design of previous imaging studies, which used few stimuli and used explicitly stated one-to-one S-R mapping rules that were usually practiced before scanning. Humans learn these rules very quickly, limiting the ability of imaging techniques to capture activity related to rule acquisition. To address these issues, we performed functional magnetic resonance imaging while subjects learned by trial and error to associate sets of abstract visual stimuli with arbitrary manual responses. Successful learning of this task required discernment of a categorical type of S-R rule in a block design expected to yield sustained rule representation. Our results show that distinct components of the dorsolateral, ventrolateral, and anterior PFC, lateral premotor cortex, supplementary motor area, and the striatum are involved in learning versus executing categorical S-R rules.

Figure 1.

Behavioral task and stimulus sets. A, Diagrammatic view of the trial structure, with block and run organization. B, Graphic representation of stimulus set construction, showing two example sets. Random numbers are generated to create unique 8 × 8 matrices (left), and each stimulus is created by filling each square in the matrix with the appropriate color from a given color map. The 10-color maps each used the same 10 colors, but in a different order (center). Two example stimulus sets (set A and set B) constructed via this method are shown on the right. Each block (excepting the NR condition) consisted of a mixture of two such sets. See Materials and Methods for more details.

Figure 2.

Behavioral results: accuracy over time. Data are plotted as a percentage of correct responses for each block type, with data collapsed into three bins of two runs each (early, mid, and late). The data represent all sessions (n = 14). Plots show the mean ± SEM for each condition. ○, fam; ♦, nov;, NR.

Figure 3.

ROI analysis. ROI delimited statistical parametric t maps contrasting nov versus fam activity. A, C, nov (A)- and fam (C)-selective activity of representative single subjects overlaid onto a standardized anatomical image N>F, nov>fam; F>N, fam>nov; L, left; R, right. B, D, Mean parameter estimates derived for the four ROIs depicted in A and C are shown for nov, fam, and NR blocks. Paired t tests contrasted the nov and NR conditions (n = 14; ^*p < 0.05; ^**p < 0.001).

Figure 4.

Decay in rule-learning activity. Rule-learning-selective activation sites in the MFG and SMA show a decline in BOLD response magnitudes across the scan session only in the nov condition, reflecting the time course of rule acquisition. nov>fam activation sites in the PCG and striatum do not show this effect. Paired t tests contrasted activity in the early and late runs (n = 14; ^*p < 0.02; ^***p < 0.004).

Figure 5.

Learning-correlated BOLD signal. A, Block-by-block correlation of the estimated response magnitudes with subject accuracy within nov-selective regions for a single subject. B, Correlation values, like those shown in A, from all subjects regressed against average accuracy for the nov condition, an index of learning in this task. C, Data as in B median split according to S-R rule-learning ability. ^*p < 0.02; ^**p < 0.002. D, Correlation between blockwise response strength and accuracy for each ROI and each condition with subjects median split by rule-learning performance.

Figure 6.

Plots of activity correlations in the fROIs with respect to one another. Activations are shown in terms of estimated BOLD response magnitudes (arbitrary units), and the correlation coefficient is shown. A, Two representative subjects, one from each group (high performer vs low performer). B, Summary of interregional ROI correlation data from all subjects, median split by performance across all task blocks. C, The degree of correlation between signal strength in the SMA and MFG ROIs was highly predictive of a subject's overall task performance (^*p < 0.02).

Figure 7.

Novel rule versus familiar rule activity. Statistical parametric t maps contrasting activity in nov rule blocks and fam rule blocks, overlaid on a standard T1-weighted anatomical image. F Oper, Frontal operculum.