The present study is a subanalysis of the ESCAPE (Effect of SwitChing AtriPla to Eviplera on neurocognitive and emotional functioning) trial, which was conducted at two major HIV treatment centers in the Netherlands (OLVG (Amsterdam) and Universitair Medisch Centrum Utrecht (Utrecht)) from 2015 until its completion in 2017 [18]. Strict inclusion and exclusion criteria were chosen to ensure a homogenous study population as PLWH exhibit greater variability with respect to fMRI measurements, and fMRI results can be readily influenced by confounding factors [30, 31]. To summarize, asymptomatic male PLWH aged 25–50 years stable on FTC/TDF/EFV for over 6 months were included. Prospective participants were excluded in case of an active psychiatric or neurological disorder, an active or past central nervous system infection, or a history or evidence of alcohol or drug abuse as assessed by the Drug Abuse Screening Test [32]. During the trial, participants with a viral load (VL) of greater than 200 copies/mL were excluded from analysis, as we judged this might interfere with fMRI results. For the full list of inclusion and exclusion criteria, see the published study [18].

The trial was reviewed and approved by the Medical Research Ethics Committee of the UMC Utrecht, performed in accordance with the Declaration of Helsinki and registered at Clinicaltrials.gov [NCT02308332]. Findings were reported in accordance with the CONSORT guideline [33]. The trial was funded by Gilead Sciences. The funder had no role in trial design, data collection or analysis, or in the preparation of the manuscript. Data were collected by the investigators with the use of case report forms. All participants provided written informed consent. The data and corresponding analysis code that support the findings of this study are available upon reasonable request from the corresponding author. The data are not publicly available because of privacy and ethical restrictions.

Participants on FTC/TDF/EFV were randomly assigned in a 2:1 ratio, using computer-generated block randomization with a variable block size (range 3–9), to switch to FTC/TDF/RPV or to continue taking FTC/TDF/EFV. A study nurse, not involved in the study, generated the random assignment sequence and allocated participants. FTC/TDF/RPV was chosen for the switch group as it is a single-tablet regimen comprising the same backbone and a similar NNRTI anchor drug as FTC/TDF/EFV. Head-to-head comparison between RPV and EFV in the ECHO and THRIVE trials showed significantly fewer neuropsychiatric side effects, though they still were present in the RPV study groups [34]. Participants were instructed to take one tablet daily and, in the case of FTC/TDF/RPV, with a significant amount of food. The NPA was performed by neuropsychologists who were unaware of the assigned treatment. Researchers performing the fMRI scan and participants were not blinded, as we believed that their knowledge of the treatment would not affect our objective outcome of fMRI brain activity.

All participants had fMRI scans at baseline and after 12 weeks. The MRI scans were reviewed by a radiologist for intracranial pathology. Cognition was examined by way of NPA and it was ascertained whether the distribution of potentially confounding asymptomatic neurocognitive impairment, as defined by the Frascati criteria, was comparable between groups [3]. Routine blood samples were obtained to assess laboratory abnormalities and confirm virologic suppression. Participants switching to FTC/TDF/RPV had two additional routine outpatient visits after 2 and 4 weeks to monitor for side effects and obtain blood samples. Lastly, participants completed multiple questionnaires at baseline and week 12, including the Hospital Anxiety and Depression Scale (HADS) and Patient Reported Outcome Measurement Information System (PROMIS) questionnaires testing depression, anxiety, and sleep disorders. The HADS questionnaire consisted of a 7-item scale with a maximum of 21 points, with score of 11 points or more indicating a probable mood disorder. The raw PROMIS questionnaire scores for depression, anxiety, and sleep disorders were transformed into T-scores with a mean of 50 and a standard deviation of 10. For full information on these and other study questionnaires used, see the published study [18].

The NPA consisted of 16 subtests and tested for seven cognitive domains [18]. The tests were specifically selected to detect minimal changes in neurocognitive performance, as our study population was asymptomatic. For attention and processing speed, the Letter-Number-Sequencing WAIS-IV NL, Paced Auditory Serial Addition Test, Digit Symbol WAIS-IV NL, Symbol Search WAIS-IV NL, and Trailmaking Test part A were used [35,36,37].

Participants performed the SSAT, a task based on the theory by Logan and Cowan [27, 38]. They postulated that a response, either starting or stopping, is the result of a race between the “GO” and “STOP” brain processes. If the STOP process is finished before the GO process reaches the execution threshold, the GO response is stopped.

The task and experimental procedures are the same as previously described by Zandbelt and Vink [27]. The experiment was performed using Presentation® software (Version 14.6, www.neurobs.com). In short, participants were presented with three background lines (Fig. 1). On each trial, a bar moved at a constant speed from the bottom towards the top bar, reaching the middle line in 800 ms. On GO trials, participants were asked to stop the bar as close as possible to the middle line, by pressing a button. If the bar passed the top line after 1000 ms, the GO trial was considered a failure. STOP trials were identical to GO trials, except that the bar stopped moving automatically before the middle bar, indicating a STOP signal. Participants were then required to withhold the button press (i.e., reactive response inhibition). To measure proactive response inhibition, the probability that a STOP signal would appear was manipulated across trials and could be anticipated on the basis of the color of the middle line. There were five STOP signal probability levels: 0% (green), 17% (yellow), 20% (amber), 25% (orange), and 33% (red). The interval between start of a trial and the STOP signal, the stop signal delay (SSD), was initially 550 ms and varied for each STOP signal according to the participant’s performance. In case of a successful STOP trial, the trial difficulty was increased as the SSD was raised by 25 ms. If the STOP trial was unsuccessful, the SSD was reduced with the same time limit, ensuring an equal amount of successful and unsuccessful STOP trials. The intertrial interval was kept at 1000 ms. In total, 414 GO trials (0%, n = 234; 17%, n = 30; 20%, n = 48; 25%, n = 54; 33%, n = 48) and 60 STOP trials (17%, n = 6; 20%, n = 12; 25%, n = 18; 33%, n = 24) were presented in a single run in pseudorandom order.

Schematic representation of the Stop-Signal Anticipation task. Three horizontal lines were displayed during the task. A bar moved from the bottom line to the top in 1000 ms. At 800 ms the bar reached the middle colored line and had to be stopped (GO trials, A). In a small proportion of trials, the bar stopped moving on its own before reaching the middle colored line, requiring the stop response to be withheld (STOP trials, B). The color of the middle line indicated the stop signal probability (C) [27]

All participants received standardized training in task performance before scanning. They were instructed that the GO and STOP trials were equally important and that it would not always be possible to suppress a response when a STOP signal occurred. We informed them that a STOP signal would never occur on a trial with a green cue and that they were more likely in the context of, in consecutive order, yellow, amber, orange, and red cues. The total task duration was 16 m 36 s.

Motor execution was studied using the response time and accuracy of GO trials with no possibility of a STOP signal (0%). Reactive inhibition was analyzed using the stop signal reaction time (SSRT), which was computed according to the integration method and calculated across all STOP signal probability levels (17–33%) [38]. The SSRT reflects the latency of the inhibition process and better reactive inhibition is indicated by a smaller SSRT.

Proactive inhibition is the anticipation of stopping based on the STOP signal probability and was measured as the slope of the mean response time to increasing STOP signal probability (0–33%). In general, participants slow their response as the STOP probability increases, resulting in larger response times. When proactive inhibition is impaired, participants thus show a reduced effect of the STOP signal probability on their response times, reflected by a less steep slope [27]. Repeated measures analyses of variance (ANOVA) were conducted on the mean response times, response accuracy, and on the slope of the response time to stop signal probability, with the STOP signal probabilities, group (FTC/TDF/RPV versus FTC/TDF/EFV), and time (baseline versus 12 weeks) as factors.

MRI scans were acquired using a 3.0 T Philips Achieva MRI scanner (Philips Medical Systems, Best, the Netherlands) in the UMC Utrecht. An eight-channel sensitivity-encoding (SENSE) parallel-imaging head coil was used to acquire the images. Head motion was restricted using a vacuum cushion and foam wedges. Whole-brain T2-weighted echo planar images with BOLD contrast, oriented in a transverse plane tilted 20° over the left–right axis, were acquired in a single run (622 volumes; 30 slices per volume; repetition time 1600 ms; echo time 23.5 ms; field of view 256 × 208 mm × 256 mm; flip angle 72.5°; 64 × 64 matrix; 4 × 4 mm in-plane resolution; 4 mm slice thickness SENSE factor 2.4 (anterior–posterior)). We discarded the first six images to allow for T1 equilibration effects. A whole-brain three-dimensional fast field echo T1-weighted scan (185 slices; repetition time 8.4 ms; echo time 3.8 ms; flip angle 8°; field of view 288 × 252 × 185 mm; voxel size 1 mm isotropic) was acquired for within-subject registration purposes.

Image data were analyzed with SPM (https://www.fil.ion.ucl.ac.uk/spm/). Pre-processing and first-level statistical analyses were performed as described previously [27]. In short, pre-processing included correction for differences in slice timing, realignment to correct for head motion, spatial normalization according to the Montreal Neurological Institute template brain and spatial (8 × 8 × 8mm) smoothing to account for interindividual differences in neuroanatomy. Head motion parameters were analyzed to ensure that the maximum motion did not exceed a pre-defined threshold (scan-to-scan > 3 mm) [39]. If this threshold was exceeded, the MRI scan was considered of insufficient quality and the participant was excluded from the analysis.

Each participant’s pre-processed fMRI data were high-pass filtered (cutoff 128 Hz) to remove low-frequency drifts and were modelled voxel-wise using a general linear model. The following events were included as regressors: timed GO trials with STOP signal probability above 0%, successful STOP signal trials, and unsuccessful STOP signal trials. For the GO trials with a STOP signal probability above 0%, we included a parametric regressor modelling the STOP signal probability level and variation in response time. In addition, GO trials with 0% STOP signal probability and activity were also modelled. We computed two contrast images for each participant: activation during successful STOP trials versus unsuccessful STOP trials (to assess reactive inhibition) and the parametric effect of STOP signal probability on GO trial activation (to assess proactive inhibition).

Differences in activation between groups were assessed in pre-defined regions of interest (ROIs), using mask-based activation maps acquired in a previous experiment in healthy controls performing the same task (Fig. 2) [27]. These 17 ROIs were defined using a cluster-level threshold (cluster-defining threshold of p < 0.001, cluster probability of p < 0.05, family-wise error corrected for multiple comparisons). Mean activation levels during reactive and proactive inhibition were calculated over the ROIs as defined by the a priori masks. For both reactive and proactive inhibition, the correlation between the change in BOLD fMRI activation levels and change in processing speed and attention Z-scores was examined in the 17 ROIs in both groups using both the Pearson and Spearman correlation coefficient. These coefficients were then compared to examine a differential pattern after switching from FTC/TDF/EFV to FTC/TDF/RPV versus continuing with FTC/TDF/EFV. Using a multivariate analysis of variance (MANOVA) and using Pillai’s trace, groupwise differences in the overall change in BOLD fMRI activation between baseline and 12 weeks after therapy switch in the 17 regions were also assessed. A two-sided alpha level of 0.05 was used and statistical analyses were conducted using SPSS version 25.0 (IBM Corp. Armonk. NY).

Regions used to assess activation levels related to reactive and proactive inhibition after discontinuing EFV. Regions were (1) right striatum; (2) right inferior frontal cortex ventral; (3) left middle frontal gyrus; (4) left temporoparietal junction; (5) left superior parietal gyrus; (6) right superior parietal gyrus; (7) right temporoparietal junction; (8) left precuneus; (9) anterior cingulate gyrus; (10) right superior frontal gyrus; (11) left superior frontal gyrus; (12) left inferior frontal gyrus; (13) right anterior insula; (14) right inferior frontal cortex dorsal; (15) right caudate; (16) left subthalamic nucleus; (17) right subthalamic nucleus