Enhanced inter-regional coupling of neural responses and repetition suppression provide separate contributions to long-term behavioral priming

Abstract

Stimulus identification commonly improves with repetition over long delays ("repetition priming"), whereas neural activity commonly decreases ("repetition suppression"). Multiple models have been proposed to explain this brain-behavior relationship, predicting alterations in functional and/or effective connectivity (Synchrony and Predictive Coding models), in the latency of neural responses (Facilitation model), and in the relative similarity of neural representations (Sharpening model). Here, we test these predictions with fMRI during overt and covert naming of repeated and novel objects. While we find partial support for predictions of the Facilitation and Sharpening models in the left fusiform gyrus and left frontal cortex, the data were most consistent with the Synchrony model, with increased coupling between right temporoparietal and anterior cingulate cortex for repeated objects that correlated with priming magnitude across participants. Increased coupling and repetition suppression varied independently, each explaining unique variance in priming and requiring modifications of all current models.

Trial registration: ClinicalTrials.gov NCT00001360.

Conflict of interest statement

The authors declare no competing interests.

Figures

Fig. 1. Neural models of repetition priming.

Four prominent models of repetition priming and repetition suppression are considered. The Synchrony model (upper left) holds that neural activity becomes more synchronized with repetition, permitting more coordinated propagation of activity at lower overall activity levels. The Predictive Coding model (upper right) holds that top-down causal influences are more strongly negative, leading to repetition suppression in the receiving region, along with a gain enhancement that leads to more rapid onset and offset of responses. The Facilitation model (lower left) claims that neural activity onset and offset are advanced in time, with earlier peak responses and a reduction in overall activity. The Sharpening model (lower right) holds that weakly tuned, poorly responsive cells are the ones driving repetition suppression, reducing downstream support for competing stimulus identities and speeding downstream stimulus-selective responses. Figure reproduced with permission from Gotts et al..

Fig. 2. Overt and covert picture naming tasks.

Participants were instructed to name pictures out loud (Overt Naming) or silently to themselves, pressing a button to mark the naming time (Covert Naming). Trials were structured in a slow event-related design in order to separate the peak responses of individual trials, and jitter of 8–14 s occurred from trial onset to trial onset. Name responses were marked for correctness and the onset time of the voice/button response was recorded for each trial. For analyses of local activity, an empirical hemodynamic response function was estimated across trials with a separate regressor at each timepoint following stimulus onset. For connectivity analyses, the peak BOLD response was calculated for each trial and in each fMRI voxel by averaging the 2 timepoints (4–6 and 6–8 s) adjacent to the expected peak of the hemodynamic response function, with a single peak value saved for each trial in order to minimize the contribution of the temporal contour of the evoked responses from functional and effective connectivity estimates.

Fig. 3. Regions engaged in object naming and showing repetition effects.

a Above-baseline responses to Overt and Covert Naming tasks are shown at two levels of significance, one at a minimum level of significance (pooling the tasks, P < 0.05, FDR q < 0.05; shown in orange) and another at a more restrictive level of significance (P < 0.0001, FDR q < 0.00016 in each task individually) for which responses can be said to replicate across tasks (shown in red). b Repetition effects, either repetition suppression (blue colors) or repetition enhancement (orange), are shown at two levels of significance, one at a minimum level of significance (pooling the tasks, P < 0.05, FDR q < 0.05; suppression shown in light blue, enhancement shown in orange) and another at a more restrictive level of significance (P < 0.00001, FDR q < 0.00006 in each task individually). Repetition effects were masked by the less restrictive threshold in (a), as all theories being evaluated make claims about regions engaged at above-baseline levels during the tasks. Statistics in (a) and (b) are based on N = 32 independent participant datasets for Overt Naming and N = 28 independent datasets for Covert Naming. c Repetition suppression was highly correlated (Pearson) with priming magnitude in left frontal cortex in terms of effect size (Cohen’s d) across N = 60 independent participants, combining both Overt and Covert Naming (corrected for multiple comparisons by both Bonferroni and FDR). The left frontal region is the same as the dark blue region showing Repetition Suppression (RS) at the more stringent threshold in (b).

Fig. 4. Repetition priming and Primeability.

a When averaged across participants within each task, priming magnitudes specific to each of the 200 stimuli are highly reliable across task. For the Covert Naming task, these priming magnitudes were assessed in a post-fMRI session during which participants overtly named all stimuli (either pre-exposed as OLD or novel to the fMRI session as NEW). b When responses are averaged across both participants and tasks, a strong relationship was observed between average response time (RT) to an object when NEW and the subsequent priming magnitude observed. Items were subjected to a median split based on the RT when NEW (using the group-average normative data) to classify objects as either Strong Primeable (slow RT) or Weak Primeable (fast RT), permitting a within-participant measure of “Primeability”. Statistics in (a) and (b) are based on N = 200 independent stimuli. c When grouping objects by Primeability (either Strong or Weak), priming effect sizes for each participant were indeed greater for Strong compared to Weak Primeable objects in the Overt Naming task (left panel); this is expected since these responses contributed to the original calculation of Primeability. However, this same relationship was also found for the Covert Naming button-press response times during fMRI, which were not used in calculating Primeability (right panel). The middle horizontal line in each box plot represents the median (50th %ile), the horizontal lines just above and below the median represent the 25th and 75th %iles, the top and bottom horizontal lines represent the minimum and maximum values, and the boundaries of the horizontal notches inside the 25th and 75th %iles depict the 95% confidence limits of the median. Individual datapoints are plotted as open circles. Statistics in (c) are based on N = 32 independent participant datasets for Overt Naming and N = 28 independent datasets for Covert Naming. For related content in SI, see Supplementary Fig. 1.

Fig. 5. Across- and within-participant relationships between repetition suppression and priming.

a Correlations (Pearson) across participants between Repetition Supression (RS) magnitude (NEW–OLD) and priming effect size (Cohen’s d) reveal a significant relationship in left frontal cortex and non-significant trends in left and right fusiform regions (left-most plot same as Fig. 3c). b Separating trials into Strong and Weak Primeable conditions show that significant within-participant relationships between RS and priming are found in left frontal, left fusiform, and ACC regions (greater RS in the Strong Primeable condition by paired t-tests). The middle horizontal line in each box plot represents the median (50th %ile), the horizontal lines just below and above the median represent the 25th and 75th %iles, the bottom and top horizontal lines represent the minimum and maximum values, and the boundaries of the horizontal notches inside the 25th and 75th %iles depict the 95% confidence limits of the median. Individual datapoints are plotted as open circles. Statistics (paired t-tests) are based on N = 60 independent participants (combining Overt and Covert Naming), and multiple comparisons were corrected by FDR (q < 0.05).

Fig. 6. Functional connectivity shows interaction between Primeability and Repetition.

a A right temporoparietal (R TP) region exhibited an interaction in whole-brain connectedness between Repetition and Primeability, similar to that seen in the behavioral priming results (P < 0.001, corrected to P < 0.025). When used as a seed, R TP jointly exhibited this interaction with the anterior cingulate (ACC), right putamen, right STG, and the right fusiform gyrus. b Regions showing a Repetition × Primeability interaction were combined with regions showing repetition suppression in both Overt and Covert Naming Tasks. c Region-by-region functional connectivity interactions of Repetition × Primeability are shown for all 7 regions (left panel). FDR-corrected effects are indicated by black boxes (P < 0.01, q < 0.05). Region-by-region comparisons of OLD versus NEW functional connectivity are shown for Strong and Weak Primeable conditions separately in the right panels. Warm colors (red) indicate OLD > NEW and cool colors (blue) indicate NEW > OLD. FDR-corrected effects (P < 0.0286, q < 0.05) were calculated among all region-by-region combinations showing a significant Repetition × Primeability interaction, indicated by black squares. Statistics are based on N = 60 independent participants, organized into a factorial design with Task (Overt, Covert Naming) as a between-participant variable. For related content in SI, see Supplementary Fig. 2 and Supplementary Table 1.

Fig. 7. Effective connectivity increases between R TP cortex and ACC correlate with priming magnitude.

a Structural equation modeling (SEM) was used to estimate effective connectivity among the 7 regions. The optimal 10-parameter SEM model is shown, with arrows indicating causal directionality and connections exhibiting a significant Repetition × Primeability interaction (P < 0.0063, q < 0.05) shown with thick red arrows (non-significant interactions shown with black arrows). b Region-by-region comparisons of SEM parameters for OLD versus NEW objects are shown separately for the Strong and Weak Primeable conditions. For directionality, sending ROIs are listed along the x-axes and receiving ROIs are listed along the y-axes. FDR-corrected OLD/NEW comparisons (P < 0.0091, q < 0.05) were assessed among connections exhibiting a Repetition × Primeability interaction, indicated with black squares. c The connection from the R TP ROI to ACC ROI shows a Repetition × Primeability interaction that is consistent with the Synchrony model (with increased coupling for OLD objects in the Strong Primeable condition). The middle horizontal line in each box plot represents the median (50th %ile), the horizontal lines just below and above the median represent the 25th and 75th %iles, the bottom and top horizontal lines represent the minimum and maximum values, and the boundaries of the horizontal notches inside the 25th and 75th %iles depict the 95% confidence limits of the median. Individual datapoints are plotted as open circles. d The R TP to ACC connection further exhibited a correlation across participants with observed priming magnitude, assessed by effect size (Cohen’s d). This correlation appeared to be driven by the Strong Primeable condition (rightmost panel), with priming effect size in the Strong Primeable condition correlated with the difference between OLD and NEW SEM parameters in the Strong Primeable condition. Statistics are based on N = 60 independent participants, organized into a factorial design with Task (Overt, Covert Naming) as a between-participant variable (Primeability and Repetition are both within-participant variables). For related content in SI, see Supplementary Fig. 3.

Fig. 8. MVPA tests of the Sharpening model.

a Spatial correlations of peak responses on each individual trial were calculated across all trials per condition type using the full extent of the four Repetition Suppression (RS) clusters (see Supplementary Table 1), retaining the median inter-item correlation per participant. A significant Repetition × Primeability interaction was observed for the Left Frontal and ACC ROIs (corrected by FDR, q < 0.05). The middle horizontal line in each box plot represents the median (50th %ile), the horizontal lines just below and above the median represent the 25th and 75th %iles, the bottom and top horizontal lines represent the minimum and maximum values, and the boundaries of the horizontal notches inside the 25th and 75th %iles depict the 95% confidence limits of the median. Individual datapoints are plotted as open circles. b Strong covariation of the spatial correlations with average beta coefficients and estimated signal-to-noise ratio eliminated most of the differences between conditions after adjustment for these variables, leaving only a significant Repetition × Primeability interaction in the Left Frontal ROI. Bar plots are used along with standard error of the mean (SE) estimates, since these adjusted means and model-residual error are not defined on the individual participants and only on the full LME model. Statistics are based on N = 60 independent participants, organized into a factorial design with Task (Overt, Covert Naming) as a between-participant variable (Primeability and Repetition are within-participant variables). For related content in SI, see Supplementary Fig. 4.

Fig. 9. Assessing activity timing predictions of the Facilitation and Predictive Coding models.

A hemodynamic response function model (gamma variate) was fit to the beta coefficients at each timepoint (TR) for each participant and experimental condition, permitting estimates of the peak time (tpeak) (see graphic at the bottom). Group-average response functions are shown with dashed lines for each condition, and group-average estimates of peak times are shown with vertical dotted lines (mean data on the actual measured beta coefficients are shown with solid lines). Main effects of Repetition on peak time were observed in the Left Fusiform and Left Frontal ROIs (OLD peaks earlier than NEW peaks), but there were no significant interactions between Repetition and Primeability on peak time. Statistics are based on N = 60 independent participants, organized into a factorial design with Task (Overt, Covert Naming) as a between-participant variable. A color key for the experimental conditions is shown at the bottom right (OLD, Strong Primeable = red; OLD, Weak Primeable = orange; NEW, Strong Primeable = gray; NEW, Weak Primeable = black).