Study selection

We identified relevant studies examining longitudinal brain functional alterations following ketamine administration (i.e., post- versus pre-ketamine) in depressed patients. Twelve studies published before April 16, 2022, were initially selected from a comprehensive systematic review [Reference Medeiros, Matheson, Demo, Reid, Matheson and Twose12]. To search for additional qualified studies, we then performed a systematic literature search in PubMed and Web of Science to identify relevant studies published from April 16, 2022 to January 2, 2024. The following combination of search terms were used: (“ketamine”) AND (“MRI” OR “fMRI” OR “magnetic resonance imaging” OR “PET” OR “positron emission tomography” OR “SPECT” OR “single photon emission computed tomography” OR “ASL” OR “arterial spin labeling” OR “neuroimaging” OR “FC” OR “functional connectivity” OR “ReHo” OR “regional homogeneity” OR “ALFF” OR “fALFF” OR “amplitude of low frequency fluctuation*” OR “CBF” OR “blood flow” OR “glucose metabolism” OR “VBM” OR “voxel-based morphometry” OR “DBM” OR “deformation-based morphometry” OR “GMV” OR “gray matter” OR “gray matter” OR “EEG” OR “electroencephalogram” OR “MEG” or “magnetoencephalography”). A flow diagram of the study selection process is shown in Supplementary Figure S1. This protocol was registered on PROSPERO (https://www.crd.york.ac.uk/PROSPERO/; registration number: CRD42024565435). It is notable that our initial search terms included structural MRI, EEG, and MEG to ensure a comprehensive review. However, during the full-text screening, we found that the available structural and electrophysiological studies did not meet our inclusion criteria or lacked compatible coordinate data. Therefore, they were excluded from the final analysis. Because a single study may contain multiple contrasts (i.e., within-subject comparisons between pre- and post-ketamine in different brain functional measures), we focused our analyses on contrasts rather than studies. In cases where multiple publications or contrasts utilized the same group of participants, we assessed the neuroimaging measures used. If the measures reflect distinct brain functional properties and contain different neurophysiological information, we treated them as independent inputs. Coordinates of peak voxels of significant clusters reported in each contrast were extracted, with coordinates in Talairach space converted to Montreal Neurological Institute (MNI) space.

To ensure rigor, we designed a specific quality assessment framework based on the Newcastle-Ottawa Scale (NOS), adapted for this study’s context. We evaluated all included studies across four domains: 1) Participants & Selection, 2) Study Design & Treatment Protocol, 3) Imaging Quality (Modality Agnostic), and 4) Statistical Rigor. Based on these scores, studies were classified as: High Quality (14–18 points): no significant shortcomings in design, administration, imaging, or statistics; Moderate Quality (9–13 points): minor defects (e.g., single-blind design or imprecise scan timing) or limitations addressed in the discussion; Low Quality (0–8 points): major defects (e.g., no control, no correction, irregular administration). The detailed scoring criteria and results are provided in Tables S1 and S2 in the Supplementary Materials.

Brain functional connectome datasets

The resting-state fMRI datasets used for constructing the brain functional connectome included a healthy volunteer (HV) dataset and an MDD dataset. The HV dataset included 1113 healthy adults of Han Chinese descent (643 female, mean 32.66 ± 12.78 years, 1110 right-handed), who were enrolled from local universities and communities through poster advertisements. The exclusion criteria included neuropsychiatric or severe somatic disorders, a history of head injury with consciousness loss, MRI contraindications, or a family history of psychiatric diseases among first-degree relatives. The MDD dataset included 255 MDD individuals of Han Chinese (161 female, mean 40.99 ± 12.67 years, 252 right-handedness), who were recruited consecutively from Affiliated Psychological Hospital of Anhui Medical University. Following the International Classification of Diseases (ICD-10), two well-trained clinical psychiatrists confirmed the diagnosis of depression. The exclusion criteria included 1) the presence of other psychiatric disorders such as substance-induced (e.g., drug or alcohol) mood disorder, bipolar disorder, anxiety disorder, schizophrenia, substance abuse or addiction; 2) a history of severe physical or neurological disease; 3) a history of head injury with loss of consciousness; 4) contraindications for MRI such as pregnancy. This study was approved by the ethics committee of The First Affiliated Hospital of Anhui Medical University. Written informed consent was obtained from all participants after being given a complete description of the study. Demographic and clinical information of the two datasets is provided in Table S3 in the Supplementary Materials.

fMRI data acquisition and preprocessing

MRI data of the two datasets were collected on a 3.0-Tesla MR system (Discovery MR750w, General Electric, Milwaukee, WI, USA) with the same imaging protocol. During the scans, all participants were instructed to keep their eyes closed, relax but not fall asleep, think of nothing in particular, and move as little as possible. High-resolution 3D T1-weighted structural images were acquired by employing a brain volume (BRAVO) sequence with the following parameters: repetition time (TR) = 8.5 ms; echo time (TE) = 3.2 ms; inversion time (Tl) = 450 ms; flip angle (FA) = 12 °; field of view (FOV) = 256 mm × 256 mm; matrix = 256 × 256; slice thickness = 1 mm; no gap; 188 sagittal slices; acquisition time = 296 s. Resting-state blood-oxygen-level-dependent (BOLD) fMRI data were acquired using a gradient-echo single-shot echo planar imaging (GRE-SS-EPI) sequence with the following parameters: TR = 2000 ms; TE = 30 ms; FA = 90 °; FOV = 220 mm × 220 mm; matrix = 64 × 64; slice thickness = 3 mm; slice gap = 1 mm; 35 interleaved axial slices; 185 volumes; acquisition time = 370 s. Participants with poor image quality (e.g., visible artifacts) or incomplete brain coverage were excluded.

Resting-state fMRI data were preprocessed using Statistical Parametric Mapping (SPM12, http://www.fil.ion.ucl.ac.uk/spm) and Data Processing & Analysis for Brain Imaging (DPABI, http://rfmri.org/dpabi) [Reference Yan, Wang, Zuo and Zang33]. The first 10 volumes for each participant were discarded to allow the signal to reach equilibrium and the participants to adapt to the scanning noise. The remaining volumes were corrected for the acquisition time delay between slices. Then, realignment was performed to correct the motion between time points. Head motion parameters were computed by estimating the translation in each direction and the angular rotation on each axis for each volume. All participants’ BOLD data were within the defined motion thresholds (i.e., translational or rotational motion parameters less than 2 mm or 2 ° for healthy subjects and less than 3 mm or 3 ° for MDD individuals). We also calculated frame-wise displacement (FD), which indexes the volume-to-volume changes in head position. Several nuisance covariates (the linear drift, the estimated motion parameters based on the Friston-24 model, the spike volumes with FD > 0.5 mm, the global signal, the white matter signal, and the cerebrospinal fluid signal) were regressed out from the data. Next, the datasets were band-pass filtered using a frequency range of 0.01–0.1 Hz. In the normalization step, individual structural images were first co-registered with the mean functional images; the transformed structural images were then segmented and normalized to MNI space using a high-level nonlinear warping algorithm, i.e., the diffeomorphic anatomical registration through exponentiated Lie algebra technique [Reference Ashburner34]. Then, each filtered functional volume was spatially normalized to MNI space using the deformation parameters estimated during the above step and resampled into a 3-mm isotropic voxel. Finally, all functional data were spatially smoothed with a Gaussian kernel of 6 × 6 × 6 mm³ full-width at half maximum.

Functional connectivity network mapping

The FCNM approach was used to construct ketamine-induced brain functional alteration networks based on the extracted coordinates of pre- to post-ketamine changes (increase and decrease separately) in brain function (Figure 1). First, 4-mm radius spheres centered at each coordinate of a contrast were created and merged together to yield a contrast-specific combined seed mask (henceforth referred to as the contrast seed). Second, in combination with the preprocessed resting-state fMRI data of the HV and MDD datasets, we calculated a contrast seed-to-whole brain functional connectivity (FC) map for each subject, by computing Pearson’s correlation coefficients between time courses of the contrast seed and each voxel within the whole brain, followed by Fisher’s Z-transformation to improve normality. Third, the subject-level FC maps were entered into a voxel-wise one-sample t test to identify brain areas functionally connected to each contrast seed. Only positive FC was considered because the biological relevance of negative FC remains controversial [Reference Murphy, Birn, Handwerker, Jones and Bandettini35, Reference Murphy and Fox36]. Fourth, the resultant group-level t maps were thresholded at P < 0.05 adjusted for multiple comparisons and then binarized. Note that a voxel-level family-wise error correction was used for the HV dataset, while a less stringent voxel-level false discovery rate correction was used for the MDD dataset due to its reduced sample size. Finally, the binarized maps were overlaid to produce 2 network probability maps, which were thresholded at 50% to yield ketamine-induced hyper-functional and hypo-functional networks, respectively.

Relation to canonical brain networks

For ease of interpretability, we investigated the spatial relations between the ketamine-induced brain functional alteration networks and 8 well-established canonical brain networks. The 7 cortical networks were defined as the visual, somatomotor, dorsal attention, ventral attention, limbic, frontoparietal, and default networks according to the Yeo et al. study [Reference Yeo, Krienen, Sepulcre, Sabuncu, Lashkari and Hollinshead37]. The Human Brainnetome Atlas [Reference Fan, Li, Zhuo, Zhang, Wang and Chen38] was adopted to define the subcortical network including the amygdala, hippocampus, basal ganglia, and thalamus. The proportion of overlapping voxels between each ketamine-induced brain functional alteration network and a canonical network to all voxels within the corresponding canonical network was calculated to assess their spatial relation. Recently, a network correspondence toolbox (NCT) has been developed to provide a convenient means for computing Dice coefficients with spin test permutations to determine the magnitude and statistical significance of correspondence between user-defined maps and existing atlas labels [Reference Kong, Spreng, Xue, Betzel, Cohen and Damoiseaux39]. Here, we used the NCT to test the statistical significance of the spatial correspondence between the ketamine-induced brain functional alteration networks and the canonical brain networks.

Statistical significance testing

To determine whether the ketamine-induced brain functional alteration networks were statistically different from those generated by chance, we repeated the above-described FCNM procedure 1000 times with random contrast seeds, which yielded 1000 null networks. To quantify the similarity between the actual network and the 1000 null networks, we estimated their spatial overlap by calculating a Dice coefficient, defined as 2 × (overlapping voxels)/(network 1 voxels) + (network 2 voxels). A higher Dice coefficient indicates more similar networks.

Validation analysis

Several validation analyses were performed to verify the robustness and consistency of our findings. To ensure that our results were independent of seed size, the FCNM procedure was repeated using 1- and 7-mm radius spheres. To examine the potential heterogeneity arising from different post-treatment scan timings, we conducted a sensitivity analysis by using the contrasts focusing on the acute phase of ketamine action (scanning within 48 hours post-administration). Finally, to investigate the potential influence of different imaging techniques, we conducted a validation analysis stratified by imaging modality.