Participants

Our sample included 27 healthy controls (HC) and 22 patients with schizophrenia, of whom 13 were first episodes (FE). Patients were diagnosed by one of the experienced psychiatrists in the group (VM) according to the criteria of the Diagnostic and Statistical Manual of Mental Disorders 5th edition, considering current mental state, clinical records, and relatives’ information. Exclusion criteria included (a) intelligence quotient under 70; (b) present or past substance dependence (excluding caffeine and nicotine); (c) head trauma with loss of consciousness; (d) neurological or mental diagnosis different to schizophrenia (patients); (e) any current neurological or psychiatric diagnosis (controls); (f) receiving any other treatment affecting central nervous system; and (g) not being safe to undergo TMS. All participants provided informed written consent after full written information before inclusion. The local ethics committee of the Clinical University Hospital of Valladolid endorsed the study (PI 22–263). This work complies with the ethical standards of the Helsinki Declaration of 1975, as revised in 2008.

Transcranial magnetic stimulation

TMS stimulation was performed using a MagPro X100 stimulator (MagVenture, Denmark) and a figure-of-8 coil. Participants sat comfortably and were instructed to look directly ahead with their eyes open. An EEG cap was fitted to their head and electrodes were placed over the right abductor pollicis brevis (APB) muscle for electromyographic recordings. The resting motor threshold (RMT) was determined over the motor cortical region following the relative frequency method [28], defined as the minimum intensity required to elicit a motor evoked potential (MEP) of > 50 µV peak-to-peak amplitude in at least five of ten subsequent trials. The optimal coil location to determine the RMT was identified as the position that consistently elicited the largest MEPs in the right APB muscle by slightly suprathreshold single-pulse TMS. Afterward, 75 monophasic TMS single pulses at an intensity of 120% RMT were applied over the left DLPFC with randomized jittered inter-stimulus interval from 5 to 7 s to reduce anticipation of the next trial. The coil was positioned in the middle of a line between the F3 and F5 electrodes with a 45º rotation relative to the midline, producing a posterior–anterior current flow in the underlying cortex. This position provides the most accurate estimation of left DLPFC (border of BA9 and BA46) in the absence of neuronavigational equipment [29,30,31]. To assess potential auditory-evoked potentials that could confound genuine TMS cortical reactivity findings, 40 participants (of them 20 HC) received sham TMS pulses. The sham condition was conducted by placing the coil perpendicular to the left DLPFC.

Auditory oddball task

During the same session, participants performed a 3-condition auditory oddball task in which 600 stimuli were randomly presented: target (500 Hz tone, probability of 0.2), distractor (1000 Hz tone, probability of 0.2), and standard (2000 Hz tone, probability of 0.6). Each tone lasted 50 ms, with a rise and fall time of 5 ms and an intensity of 90 decibels. The inter-stimulus interval between tones randomly jittered between 1.16 and 1.44 s. Participants were asked to keep their eyes closed and to press the mouse button upon hearing target tones. Target tones were considered attended when followed by a button press. Only attended target tones were considered for further analysis.

EEG data acquisition

EEG activity was collected using a 64-channel system [Brain Vision (Brain Products GmbH)] following the international 10–10 system. Impedance for all electrodes was lowered to ≤ 5 kΩ. The channels were referenced over Cz during acquisition and re-referenced offline to the averaged activity of all sensors [11, 32]. During the auditory oddball task, the sampling rate was 500 Hz. TMS–EEG data were recorded with a sampling rate of 25 kHz.

EEG data pre-processing

After recording EEG activity during the auditory oddball task, the following three-step artifact rejection algorithm was applied to minimize electrooculographic and electromyographic contamination [12]: (i) an independent component analysis (ICA) was performed to discard noisy ICA components; (II) the signals were divided after ICA reconstruction into trials of 1 s (from 300 ms prior to the stimulus onset to 700 ms after); and (iii) the trials with amplitudes that exceeded an adaptative statistical-based threshold were automatically rejected [33]. The signals were band-pass filtered between 1 and 70 Hz, and a 50-Hz notch filter was applied to remove the power line artifact.

Spectral entropy

Spectral entropy modulation was calculated in the auditory oddball task and computed from the normalized continuous wavelet transform (CWT), which is a form of time–frequency representation of a signal that is conceptually related to the short-term Fourier transform [33]. The CWT allows for better detection of dynamic EEG components due to its balance between frequency and time resolution [33]. The time-dependent wavelet-based SE can be defined as follows:

$$SE\left(t\right)=-\frac}(M)}\cdot _WS\left(t,f\right)\dot }\left[WS\left(t,f\right)\right],$$

where SE is the spectral entropy (as a function of time) and WS is the normalized wavelet scalogram. Specifically, SE was computed in two windows: pre-stimulus (300 ms before stimulus to stimulus onset) and response (150 ms to 450 ms from the stimulus onset, centered around the P300 peak). Afterward, it was averaged in each of the two windows. As in our previous studies, SE modulation was calculated as the difference in SE between response and pre-stimulus windows (Gomez-Pilar et al., 2018b), providing a measure of the degree of the change of signal irregularity across time. Since a decrease in SE in the response window has been robustly observed in healthy controls, normal SE modulation is expected to be expressed in negative values [12, 13, 16]. Complete details of spectral entropy calculation are found in the Supplementary material.

TMS–EEG signal pre-processing

TMS–EEG signal pre-processing was performed using Fieldtrip [34] and MATLAB (R2021b; The Mathworks Inc., Natick, MA). Signals were epoched from − 1000 ms to 1000 ms relative to the TMS pulse. As the data samples where the TMS pulse appears are irretrievable, they were deleted (from − 1 ms to 10 ms related to TMS-pulse onset) and cubic interpolated [35]. To remove artifacts present in the signals, which encompassed TMS-induced, muscle, ocular, auditory, and noise-related artifacts, independent component analysis (ICA) was applied. The independent components (ICs) that represented the aforementioned artifacts were manually selected by three experts. The criteria to remove the ICs were based on their trial-averaged amplitude, spatial distribution, and activation and time–frequency maps [31, 35, 36]. Subsequently, bad channel interpolation and bad trial rejection were automatically performed. Finally, a baseline correction was applied using an interval of 800 ms before the TMS pulse onset. Data were resampled to 5 kHz and band-pass filtered between 0.5 Hz and 70 Hz.

TMS–EEG signal processing/LMFP-AUC computation

Artifact-free TMS–EEG data processing was performed in a region of interest (ROI) composed of the channels covering the DLPFC, i.e., Fp1, Af3, Af7, F1, F3, F5, F7, FC1, FC3, and FC5 [37]. To measure the activity induced by the TMS pulse on this ROI, the area under the curve (AUC) of the local mean field power (LMFP; in combination LMFP-AUC) was computed for each subject. First, the LMFP was calculated following the formula below:

$$}\left(t\right)= \sqrt_^\left(_\left(t\right)-_}}(t))}^\right]}},$$

where K is the number of channels, Vi (t) is the amplitude of the signal in channel i at instant t, and Vmean (t) is the mean amplitude of all channels of the ROI at instant t. Finally, the AUC was computed by integrating the LMFP signal from 30 to 250 ms after the TMS pulse.

The LMFP-AUC is a widely used neurophysiological measure that represents activity induced by TMS pulses across a specific subset of electrodes of interest [20]. Therefore, it might be interpreted as an index of the cortical reactivity of the area covered by those electrodes. Sham stimulation signals were pre-processed and processed analogously to the active stimulation signals.

Statistical analysis

Demographic characteristics were compared between healthy controls and patients using independent samples t test or Chi-square test wherever appropriate. Similarly, RMT was compared using independent samples t test to ensure that stimulation intensities did not differ between groups. Since SE modulation included many different potentially collinear variables, it was reduced to principal components using PCA, following our previous studies [16]. The number of factors retained was determined by scree plot examination. To obtain a stable solution, the PCA was carried out on a larger sample (n = 440) containing the participants of this study and reducing the number of electrodes to 32. The auditory oddball task was performed under the same conditions in that sample. Independent t tests were performed to compare SE modulation and LMFP-AUC values between groups and to compare active and sham stimulation signals.

The main hypothesis of the study was tested using Pearson correlation analyses between LMFP-AUC and SE modulation values, including patients and HC, and then repeating this analysis separately for each group. Finally, to rule out a major effect of treatment, correlation coefficients between LMFP-AUC and medication dose (based on chlorpromazine equivalents) were also calculated. Data analyses were performed using SPSS statistical software, version 23 for Windows (IBM).