Evoked brain responses are generated by feedback loops

Abstract

Neuronal responses to stimuli, measured electrophysiologically, unfold over several hundred milliseconds. Typically, they show characteristic waveforms with early and late components. It is thought that early or exogenous components reflect a perturbation of neuronal dynamics by sensory input bottom-up processing. Conversely, later, endogenous components have been ascribed to recurrent dynamics among hierarchically disposed cortical processing levels, top-down effects. Here, we show that evoked brain responses are generated by recurrent dynamics in cortical networks, and late components of event-related responses are mediated by backward connections. This evidence is furnished by dynamic causal modeling of mismatch responses, elicited in an oddball paradigm. We used the evidence for models with and without backward connections to assess their likelihood as a function of peristimulus time and show that backward connections are necessary to explain late components. Furthermore, we were able to quantify the contribution of backward connections to evoked responses and to source activity, again as a function of peristimulus time. These results link a generic feature of brain responses to changes in the sensorium and a key architectural component of functional anatomy; namely, backward connections are necessary for recurrent interactions among levels of cortical hierarchies. This is the theoretical cornerstone of most modern theories of perceptual inference and learning.

Fig. 1.

Model specification. The sources comprising the network are connected with backward (gray) and/or forward (dark gray) connections as shown. A1, primary auditory cortex; STG, superior temporal gyrus; IFG, inferior temporal gyrus. Two different models were tested within the same architecture, with and without backward connections (A and B, respectively). (C) Sources of activity, modeled as equivalent dipoles (estimated posterior moments and locations), are superimposed in an MRI of a standard brain in MNI space and their prior mean locations are: lA1 [−42,−22, 7], rA1 [46, −14, 8], lSTG [−61, −32, 8], rSTG [59, −25, 8], rIFG [46, 20, 8], in millimeters.

Fig. 2.

Bayesian model comparison among DCMs of grand mean ERPs. (A) Grand mean ERP responses, i.e., averaged over all subjects, to the deviant tone overlaid on a whole-scalp map of 128 EEG electrodes. (B) Overlapped ERP responses to deviant tones from all 128 sensors over the peristimulus interval [0, 400] (in milliseconds). (C) Differences in negative free-energy or log-evidence comparing the model with backward connections (FB) against the model without (F). The gray patch indicates the interval chosen to model the ERPs for each individual subject (see Fig. 3).

Fig. 3.

Bayesian model comparison across subjects. (A) Comparison of the model with backward connections (FB) against the model without (F), across all subjects over the peristimulus interval 180–260 ms. The dots correspond to differences in log-evidence for 11 subjects over time. The solid line shows the average log-evidence differences over subjects [this is proportional to the log-group Bayes factor (Bf) or to the differences in the free energy of the two models (ΔF); see Materials and Methods for details]. The points outside the gray zone imply very strong inference (≥99% confidence that one model is more likely), i.e., model FB supervenes over F for positive points and the converse for negative points. (B) Histogram showing the number of subjects in each of seven levels of inference on models with and without backward connections across the peristimulus interval 180–260 ms.

Fig. 4.

Contribution of extrinsic coupling to source activity. The graphs show predicted (solid) and observed (broken) responses in measurement space for the first spatial mode, which was obtained after projection of the scalp data onto eight spatial modes, for (A) FB model and (B) F model. The first mode accounts for the greatest amount of observed variance. The improvement of model fit due to backward connections for later components is evident. Predicted responses at each source (solid line) and changes in activity with respect to a unit change in forward (dotted line) and backward connection (dash-dotted line) for (C) right IFG, (D) right STG, and (E) right A1. The gray bar covers the same period of peristimulus time as in Fig. 2.