Summary
A Duke-led computational preprint compares cortical-column models to explain the short-latency D- and I-wave responses produced by cortical stimulation. Models of motor and auditory cortex reproduced both wave types, while the somatosensory model generated rhythms with mismatched timing.
Researchers at Duke University have used computational models of cortical circuits to investigate how transcranial magnetic stimulation (TMS) and intracortical microstimulation produce the brain’s short-latency D- and I-wave responses. The bioRxiv preprint, posted on September 16, 2026, compared models of several cortical regions and identified circuit mechanisms that could account for the timing and strength of these responses.
The work is a computational study rather than a human or animal intervention. Its models were designed to reproduce experimentally observed neural signals and to test how different populations of cortical neurons contribute to them.
A model of stimulation across cortical layers
TMS applied over the primary motor cortex, or M1, produces a sequence of high-frequency volleys that can be recorded in the corticospinal tract. The earliest volley is called the D-wave, followed by later I-waves. Similar responses have been observed after intracortical microstimulation in the motor cortex of non-human primates.
Directly recording the immediate cortical response is difficult because stimulation creates electrical artefacts. To examine the underlying circuitry, the researchers implemented four previously published types of cortical-column model: a single-compartment M1 model, a multi-compartment M1 model, and multi-compartment models of the primary auditory cortex (A1) and primary somatosensory cortex (S1).
The models represented the layered organisation and region-specific connectivity of each cortical area. The researchers simulated the direct effect of the stimulation-induced electric field by activating different proportions of pyramidal neurons across the cortical layers. They then constructed dose–response curves for pyramidal neurons in layer 5, a major output layer of the cortex.
This approach allowed the team to vary which neuronal populations were recruited and examine how that changed the simulated D- and I-wave pattern. The researchers also tested the models against the reported effects of pharmacological agents and performed virtual lesions by removing selected neural populations from the simulated circuits.
Motor and auditory models reproduced the wave pattern
Both the M1 and A1 models generated D- and I-waves. In these simulations, the size of the I-waves increased as more pyramidal neurons in layers 2/3 and 5 were recruited, linking the later volleys to activity distributed across multiple cortical layers.
The S1 model also produced rhythmic firing, but its timing did not match the experimentally observed I-wave pattern. That contrast gave the researchers a way to compare how different regional connectivity schemes shape the immediate response to stimulation rather than assuming that every cortical area reacts in the same way.
The simulated pharmacological effects were consistent with experimental observations of changes to I-waves. Virtual lesions of particular neural populations also produced circuit-level explanations for the first and later I-waves. The authors describe these results as plausible microcircuit mechanisms: the simulations indicate which populations could generate or shape the volleys, while the precise contribution of each population remains a model-derived interpretation.
The cross-region comparison provides a framework for studying why stimulation produces distinct short-latency responses and for refining computational strategies used to design stimulation paradigms. The immediate value is mechanistic: the models connect the externally applied electric field to activity in specific cortical layers and neuronal populations.
The study is a bioRxiv preprint, so its conclusions are presented before formal peer review. The evidence is also computational and depends on the assumptions, cellular properties and connectivity included in the models. It addresses the generation and interpretation of stimulation-evoked signals, not the clinical effectiveness of a treatment or the outcome of a human trial.