Clinical examination at admission revealed a score of 8 on the Glasgow Coma Scale, and a score of 11 on the FOUR scale. The presence of an oral automatism reflex. Muscle strength could not be reliably determined due to limited contact with the patient. Passive movements in the extremities are limited due to increased muscle tone: flexor set of hands and feet. Muscle tone is elevated in the spastic type: in the right hand in the hand 4 points, proximally 3 points; in the left hand in the hand 4 points, proximally 2 points; in the right leg 2 points, in the left leg 3 points on the modified Ashworth spasticity scale. There are pathologic Babinski reflexes on 2 sides. No control of pelvic organ functions.

The patient underwent trial spinal cord stimulation from April 7 to 14, 2020. To monitor treatment effects, repeat clinical and instrumental examinations were performed. Neurological assessment and fMRI were conducted no more than 24 h before and after the stimulation period. Pre- and post-study neurological evaluation included an assessment of consciousness using the JFK Coma Recovery Scale-Revised (CRS-R) (Giacino et al., 2004) since CRS-R is established as the most reliable tool for chronic DOC assessment (Seel et al., 2010). Validation of the Russian adaptation of the Coma Recovery Scale-Revised was conducted in Research Center of Neurology (Moscow) from October 2016 until April 2017, registered at clinicaltrials.gov (Mochalova et al., 2018).

We assessed the level of spasticity using the Ashworth scale (AS). It provides a semiquantitative measure of the resistance to passive movement but has limited inter rater reliability. The AS is simple, requires no instrumentation and is easy and quick to carry out, and has been used in many studies (Bohannon and Smith, 1987). The CRS-R and the spasticity assessments were performed two times: on the day of pre- and postsurgical fMRI by two independent neurologists.

Neurological examinations were conducted by two independent neurologists from the Center, with no financial conflicts of interest. Both specialists received comprehensive training, including: (1) general intensive care unit protocols and (2) specific certification in administering the Russian-adapted version of the Coma Recovery Scale-Revised (CRS-R).

According to the clinical manifestations the aim of achieving a certain cervical level was the objective that guided the installation of the electrodes. A trial epidural electrode was implanted in the C3 to C5 area under general anesthesia on April 7, 2020.

The choice of placement area was determined based on existing evidence that this leads location is reported in the literature to be effective for the treatment of spasticity of various etiologies (Waltz et al., 1987; Waltz et al., 1981) and, given its proximity to the central nervous system and the ascending activating reticular system in particular, is most commonly used in DoC patients to enhance consciousness (Xia et al., 2018).

The patient was positioned on the left side and the head was secured using a Mayfield cranial stabilization system. The procedure included parasagittal transcutaneous puncture of the epidural space in the Th1-Th2 area with a Tuohy needle and insertion of the electrode in C3-C5 under intraoperative X-ray monitoring. An eight-pin monoaxial electrode (Octrode kit, St. Jude Medical, USA) was used. A test pulse generator (St. Jude Medical, USA: model 3599)¹ was then connected to the distal part of the electrode to check impedance. The peripheral part of the electrode was fixed on the skin. The entire procedure took 15 min and was performed by a qualified neurosurgeon. A CT scan of the neck was performed after surgery to confirm the position of the electrode and rule out surgical complications (Figure 1).

The stimulator was connected on the 1st day after surgery. Stimulation options were isolated tonic mode, “Burst”, or combinations thereof. The decision to choose a mode was made based on the presence of clinical effect. With tonic mode the pulse width was set to 200 μs, and with “Burst” mode the pulse width was set to 1000 μs (5 pulses per burst). The stimulation frequency was 40 Hz for both modes. For testing the initial current was set at 1 mA, then it was gradually increased to the value at which the limb tremor appeared. Then the current was decreased by 1 mA. This was the value of the current at which the stimulation was performed. On April 14, 2020, following completion of six neuromodulation cycles, the trial stimulation period was concluded and the electrode array was explanted.

We used a cyclic stimulation mode of 15 min on and 5 min off in the daytime (from 8 a.m. to 8 p.m.) and turned the generator off at night to comply with the sleep– wake cycle. Contact 1 (the uppermost contact) or contact 2 of the electrodes were usually used as the stimulation contact. Stimulation parameters were selected according to previous data and the response of the patient. Stimulation contact and stimulation current were adjusted until the appearance of an expression of pain, consistent twitches of the extremities or significant alterations in respiration and heart rate, then, the stimulation current was turned down to an intensity at which no expression of pain, no consistent twitches of the extremities, and no significant alterations in respiration or heart rate occurred. Stimulation parameters were set at 2.5–3.5 mA with a frequency of 60 Hz and pulse width of 200 μs. Neuromodulation cycles were performed 4 times, with a break on weekends.

Rs-fMRI examinations were performed before and after surgical placement of the spinal cord stimulation system. Magnetic resonance imaging (MRI) was performed on a Magnetom Essenza 1.5 T Scanner (Siemens, Germany). To acquire anatomical images in the sagittal plane, the T1 MPRAGE sequences were used with the following parameters: TR = 1900 ms, TE = 3.4 ms, 174 slices, slice thickness = 1 mm, FoV = 250 mm, reconstruction matrix of 256 × 256, voxel size = 1 × 1 × 1 mm. Resting-state fMRI was performed according to the following protocol: standard two-dimensional echo-planar imaging, TR = 3000 ms, delay = 0 ms, TE = 45 ms, 35 slices (ascending interleaved order), slice thickness = 5 mm, FoV = 192 mm, matrix of 64 × 64, voxel size = 3 × 3 × 3 mm, 180 measurements (volumes) for each subject.

The fMRI data were processed on a PC HP Probook 430 G6 (no calculations were performed using a graphics processor) using MATLAB R2019b (MathWorks, Natick, MA, USA). Statistical Processing Package SPM12² was used for preprocessing. The first two functional volumes were excluded from the analysis; the remaining images were aligned in time (for resting-state fMRI was not carried out) as well as relative to the first volume for movement artifacts correction. Further coregistration of the average functional image with the structural image was performed. The normalization procedure in the MNI space was performed using the DARTEL (Diffeomorphic Anatomical Registration Through Exponentiated Lie Algebra) tool and the New Segment Tool. Finally, the image was smoothed using a Gaussian filter with a core size of 8 × 8 × 8 mm.

After preprocessing, the CONN software application³ was used to create functional connectivity matrices from the patient’s fMRI data before the procedure, after the procedure, and for the average matrix of the normal reference group. The following steps are described for working with SPM12 preprocessed data to construct functional connectivity matrices. Denoising was performed by removing the following confounders by linear regression: the blood-oxygen-level dependent (BOLD) signal from the white matter and CSF masks (5 principal components of each signal); scrubbing (the number of regressors corresponded to the number of identified invalid scans identified using the Art toolbox⁴); [motion regression (12 regressors: 6 motion parameters, 16 first-order temporal derivatives)]. In addition, manually created lesion mask in the MRIcron toolbox (Rorden and Brett, 2000) was included in the denoising step to regress out any lesion-related signal. The resulting signals were band-pass filtered in the range of 0.008–0.12 Hz. Using the atlas built into the program, we analyzed the ROI-to-ROI functional connectivity of 165 brain areas and networks, including network interactions. The BOLD-signal time course of every ROI was calculated. The Fisher-transformed correlation coefficients between all pairs of these signals were computed, comprising a symmetric connectivity matrix. Three matrices were constructed: the patient’s preoperative and postoperative connectivity matrix and the average matrix of the reference group of healthy controls. The averaged matrix for the healthy control group (group characterization is given in paragraph 3.0) was constructed from individual matrices (from the concatenated 3-dimensional matrix) using MATLAB functions (see Supplementary materials, section a-b). Then all the obtained matrices were transformed into the graphical output of matrices using MATLAB function (see Supplementary materials, section c).

From the three acquired matrices, we derived additional matrices representing all brain areas (132 regions), cortical-cortical (91 regions) and subcortical interactions (15 regions) by selectively extracting specific subsets of data from the original matrices (see Supplementary materials, section d).

In the final data processing step, we computed the index of connectome intactness (ICI) for both pre- and post-intervention matrices, using the healthy control group as reference. ICI is a normalized measure comparing the similarity (or divergence) of an individual’s structural/functional connectome to a reference template derived from healthy controls. It reflects the degree to which network topology, edge weights, or nodal connectivity are preserved. As a similarity metric, we used the Pearson correlation coefficient. All steps for calculating ICI (matrix stretching and computing the Pearson correlation coefficient) were also performed in the MATLAB software environment using its standard functions (see Supplementary materials, section e). The description of the MATLAB code, including examples for each of the outlined steps, is provided sequentially in the Supplementary materials.

Overall, there was a slight decrease in spasticity according to both neurologists (Table 1). The level of consciousness according to the clinicians was unchanged, the patient understood simple instructions, but speech and purposeful behavior were absent. As a result of SCS therapy, some regression of pain syndrome was also recorded in the medical history. Neurological status without negative dynamics. The physicians assessing the patient before and after the procedure were blinded to her spinal cord stimulation treatment. While they were aware of a “spinal cord stimulation efficacy study,” they had no knowledge of whether she was in the active treatment or control group.

Figure 2 shows from bottom to top the functional connectivity matrices of the patient before and after spinal cord stimulation and the healthy control group. The warmer shades reflect the higher degree of connectivity of the brain regions. Visually, the “after stimulation” matrix is more similar to the healthy control matrix than to the “before stimulation” matrix, which means that the patient’s total neuronal activity is closer to that of the healthy brain (both at the level of cortical-cortical interactions (red box) and subcortical interactions (yellow box).

To mathematically assess the degree of reorganization of the patient’s brain activity after 6-day spinal cord stimulation, we used the index of connectome intactness (Finn et al., 2015) calculated relative to the normal group. When assessing the changes in functional brain interactions in general after stimulation, some increase in the connectome integrity index was noted. Further, we made an attempt to evaluate in more detail which interactions contribute most to the obtained result and calculated the index separately for cortical-cortical and subcortical interactions. In all cases, we observed an increase in the connectome integrity index (relative to the control group) both at the level of cortical and subcortical interactions, with the greatest changes in the set of subcortical functional connections (Table 2).

On the penultimate day of spinal cord stimulation (13.04.2020), the patient underwent successful decannulation following clinical evaluation. She was subsequently transferred to the rehabilitation department for enhanced rehabilitation therapy. On April 28, 2020 (14 days after electrode removal), the patient experienced cardiac and respiratory arrest secondary to acute pulmonary embolism. The observed complication originated from deep venous system pathology in the lower limbs. After 8 min of resuscitation, spontaneous circulation was restored; however, the patient sustained severe anoxic brain damage. Given this confounding complication, subsequent evaluation of delayed spinal cord stimulation effects on consciousness was deemed clinically irrelevant and thus discontinued. The patient was discharged on July 31, 2020, in critical condition, remaining in a vegetative state with only minimal flexion in response to painful stimuli preserved. Thus, despite initially demonstrating highly favorable prognosis following the spinal cord stimulation course, the patient subsequently experienced cardiorespiratory arrest resulting in anoxic brain injury, which substantially worsened her clinical outlook. Historical and current clinical data for this care episode, organized as a timeline, are presented in Figure 3.

We scanned healthy volunteers using the same protocol to construct an average matrix of functional connectivity and further compare patient data with this mean functional connectivity matrix. The group comprised 10 healthy volunteers (5 women, median age 33 years, age range 25–41 years). The processing pipeline of the data completely coincided with those in the patients. Healthy participants of the control group were instructed to remain calm, with their eyes closed, not falling asleep and not thinking about anything. To minimize head movements, foam pads were used to fix the head during MRI acquisition.