The measurement of cerebral perfusion [volume of blood delivered to a volume of tissue per unit time, referred to as the cerebral blood flow (CBF)] is a growing method for studying neural processes. O¹⁵-H₂O PET is the standard for quantifying CBF; however, this imaging requires the injection of a radioactive tracer. Alternatively, the in vivo quantification of CBF can be performed non-invasively using MRI through an arterial spin labeling (ASL) pulse sequence (Weber et al., 2013; Grade et al., 2015). ASL uses a simple modification to the standard MRI acquisitions to turn the blood in the neck into an MR tracer. This is completed by labeling blood in a slab inferior to the imaging field of view using magnetic inversion. This inversion will decrease the measured signal, and, thus, CBF can be extracted by comparing the labeled image with a control (unlabeled) image. ASL imaging is clinically used to identify the early pathophysiological changes in Alzheimer's disease (Du et al., 2006; Noguchi et al., 2008; Xu et al., 2010) and other disorders such as dementia (Xu et al., 2010; Borogovac and Asllani, 2012; Weber et al., 2013). In comparison to signals based on blood oxygen, CBF has better reliability and intersubject variability (Weber et al., 2013). Furthermore, CBF is directly responsible for the delivery of glucose and oxygen. Both oxygen and glucose are necessary to maintain ATP production and needs to be replenished to support the continued neural activity. Although CBF is not a direct measure of neural activity, it is a tightly coupled correlate: CBF changes with neural activity which occurs with a changing metabolism (i.e., resting activity) or during task activation (Borogovac and Asllani, 2012).

Behaviorally, our group has observed various effects from anodal prefrontal tDCS. Increased information throughput (Nelson et al., 2019) and multitasking throughput capacity (Nelson et al., 2016) during the multi-attribute task battery were observed in groups receiving 2-mA compared with those of sham stimulation. Improvements in the target detection were observed during an air traffic controller task in subjects receiving 2-mA compared to those in sham stimulation (Nelson et al., 2014). A similar improvement in the target detection was observed in a vigilance task from a group receiving a 2-mA stimulation compared to those receiving a lower-amplitude stimulation (0.5, 1, and 1.5 mA) as well as sham stimulation (McKinley et al., 2017a). McIntire et al. (2017a,b) observed attentional decrements due to sleep deprivation stress in a sham stimulation group; however, 6 h of improved attentional accuracy and reaction time following a single application of 2-mA stimulation was reported. In addition, self-reports from the 2-mA group revealed more vigor, less fatigue, and reduced boredom than those from the sham group. These effects were found to be reliable and repeated in a duplicated study on a new subject sample (McIntire et al., 2019). A final study observed a decreased sleep time without negative effects on mood or sleep quality following a single session of a 2-mA stimulation compared to the sham group (McIntire et al., 2020).

The study of the resting CBF in anodal prefrontal tDCS may provide critical, novel insights that could help elucidate the mechanisms of the anodal prefrontal tDCS resulting in these various behavioral effects. For instance, increased neural activity associated with anodal tDCS would increase the resting metabolism and, thus, would enhance CBF (Gsell et al., 2000; Nielsen and Lauritzen, 2001; Sheth et al., 2004). Few previous studies have used ASL to assess such neural effects of tDCS. In one study, increased regional CBF within and between subjects was found underneath the site of anodal tDCS, with transfer effects observed in brain regions functionally connected to the stimulation site following a single stimulation of 0.8- to 2-mA (Zheng et al., 2011). In another study where 1-mA anodal and cathodal stimulations were provided to the prefrontal cortex 1 week apart, immediate and lasting changes in CBF were found to be associated with the anodal left prefrontal tDCS (Stagg et al., 2013). Despite observing increased CBF in regions anatomically close to the dorsolateral prefrontal cortex, a widespread CBF was observed after both anodal and cathodal tDCS. Through the comparison of multiple levels of stimulation across concurrent days with that of the sham stimulation, we sought to identify tolerance or cumulative effects of tDCS on the resting CBF.

The previous research from our group, in between-subject experiments, has used Cohen's d in power analysis to help determine the sample size. Cohen's d of 0.8 or larger is considered a large effect. Using a two-sample t-test with a power of 0.8, an alpha error of 0.05, and a Cohen's d value of 0.8 results in 26 subjects per group. In the current study, there are three groups. Due to constraints of time and funding, and a plan to run follow-up studies, it was decided to use 20 subjects per group.

This study reports the findings from 47 healthy volunteers (mean age = 27.9 ± 4.85, 9 women). In total, we recruited 77 healthy, active-duty, Air Force military members aged 18–42 that did not meet our exclusion criteria (see Supplementary Table 1 for a full list of exclusion criteria). Participants were recruited from Wright Patterson Air Force Base, Ohio, and were randomly assigned to one of our three experimental groups. Withdrawals and disqualifications (detailed below) during the experimental progress resulted in exceeding our planned recruitment of 60 subjects. Additional disqualifications were made during our data analysis due to data issues, resulting in 47 participants being included in this report.

Written informed consent was obtained from each participant prior to any experimental procedures. At the time of consent (~1–2 days prior to the first experimental session), participants were randomly assigned to one of the three groups, received written instructions, and practiced tasks including 5-min of training on the Mackworth Clock test (McIntire et al., 2017b; McKinley, 2018). The experimental protocol was approved by the Air Force Research Laboratory Institutional Review Board at Wright-Patterson Air Force Base under Protocol # FWR20130126H. Participants eligible for compensation (i.e., if participation occurred in an off-duty status) received equal remuneration.

Of the 77 participants recruited and consented, the reported cohort was reduced due to medical disqualification (n = 1), withdrawal prior to MRI procedures (n = 6; e.g., family issues, being uncomfortable with MRI procedures once seen in person, or illness), self-withdrawal due to illness/family issues or being uncomfortable with MRI procedures (e.g., noise; n = 6), incomplete data collection due to MRI scheduling conflicts (n = 1), or being medically disqualified due to incidental findings during the initial MRI scan (n = 1). Additionally, participants were removed due to missing or corrupted data (n = 13) and bad registration between ASL and anatomical images (n = 2). Data from the remaining 47 participants were evaluated.

This study was a parallel-group sham-controlled design with two active tDCS conditions (1- and 2-mA for 30 min) and one sham condition (30 s of 2-mA followed by 29.5 min of no stimulation). Each participant completed three experimental sessions, with each session separated by ~24 h. The procedures at each session were identical – first, an initial MRI was performed followed by tDCS executed outside of the MRI, and finally, a second MRI with identical procedures similar to the first MRI (see Figure 1).

The sessions were conducted in the evenings to not only reduce the work-related conflicts but also conform to the availability of MRI. For most of the sessions, two participants were grouped on the same days with staggered start times (see Table 1). We attempted to hold start times consistent within the participants across the three sessions; however, the variability in MRI availability and participant delays could not be fully accounted.

Each of the three groups received the same instructions and performed the same tasks with the exception of stimulation. In the two experimental groups, 1-mA (ACT_1mA, n = 15, mean age = 26.93 ± 3.53, 2 women) or 2-mA (ACT_2mA, n = 17, mean age = 28.61 ± 5.79, 4 women), stimulation was provided for 30 min, while in the control group (CON, n = 15, mean age = 28.14 ± 5.08, 3 women), sham stimulation consisting of 2-mA was applied for 30 s followed by 29.5 min of no stimulation. The study was a single-blinded study – the participants, not the experimenters, were uninformed of the validity and intensity of the simulation. Since handedness was not controlled, self-reported handedness was queried. The CON group consisted of one left-handed participant, ACT_1mA had none, and ACT_2mA had seven.

Cerebral blood flow maps (see Supplementary Figure 1) were computed and quantified from the automated functions in the GE reconstruction software. First, the three tagged and three control volumes were averaged in place (without motion correction). Then, difference images were calculated for all participants by subtracting the average tagged volume from the average control volume. Finally, quantitative CBF maps (see Supplementary Figure 1 for example of raw CBF maps) were generated from the difference images, the associated proton density (PD)-weighted volumes, and the standard single-compartment model (Alsop and Detre, 1996; Mutsaerts et al., 2014; Alsop et al., 2015) using the formula:

where CBF is calculated in ml/100g/min. In this equation, T_1b is the T1 of blood and is assumed to be 1.6 s at 3 T. The partial saturation of the reference image (PD) is corrected using a T_1t of 1.2 s (typical of gray matter). The saturation time, ST, is set to 2 s, and the partition coefficient λ is set to a whole brain average of 0.9. The efficiency, ε, is the overall efficiency (0.6), a combination of both inversion efficiency (0.8) and background suppression efficiency (0.75). The PLD used for the ASL protocol was 2,025 ms, and the labeling duration, LT, was set to 1.5 s in the current version. PW is the perfusion weighted or the raw difference, and SF_PW is the scaling factor of the PW sequence. The number of excitations for PW images, NEX_PW, was set to 3.

The CBF maps from each session were exported from the MRI scanner and processed using the FMRIB Software Library (FSL; Smith et al., 2004; Woolrich et al., 2009) on a 128-core Rocks Cluster Distribution (www.rocksclusters.org) high-performance computing system capable of running 256 threads in parallel. Then, the high-resolution structural image of an individual was registered to the MNI-152 T1-weighted 2 mm template provided in FSL (Collins et al., 1995; Mazziotta et al., 2001) using a 12-parameter model (Jenkinson and Smith, 2001; Jenkinson et al., 2002; see Supplementary Figure 2A). Next, the raw PW images were registered to the high-resolution structural image by estimating the motion from a boundary-based registration method, which includes a field-map-based distortion correction (Greve and Fischl, 2009) (see Supplementary Figure 2B). In order to co-register all volumes, the CBF maps were converted to the standard space using the transforms responsible for morphing the PD-weighted image of each data set to the structural image and the structural image to the template (see Supplementary Figure 2C).

Voxelwise non-parametric analyses were performed using the conditional Monte Carlo permutation testing based on the method of Freeman and Lane (1983) implemented in randomise of FSL (Anderson and Robinson, 2001; Winkler et al., 2014). Due to the mixed effect of our design and how the data would need to be permuted, we were not able to perform proper between-group repeated measures ANOVAs. However, we implemented the following steps to evaluate run x group interaction effects. First, we subtracted each post-stimulation CBF map (in the standard space) from the corresponding baseline (session 1 pre-stimulation). Next, unpaired t-tests were performed to evaluate the between-group differences in the change from baseline per post-stimulation measurement (sessions 1–3 post-stimulation). These tests were conducted to compare ACT_1mA and ACT_2mA groups with the CON group separately. This resulted in a total of six analyses, analogous to the post-hoc pairwise testing that would be conducted to interpret a significant interaction effect. Null t distributions for the contrast representative of the between-group difference were derived by performing 2,000,000 random permutations of the data. Each permutation was created by exchanging the assigned group (Nichols and Holmes, 2003). A final t statistic was computed for each voxel by testing the unshuffled, real arrangement against the permutation distribution.

Additionally, voxelwise non-parametric, within-group one-way ANOVAs were performed on the session 1 pre-stimulation, session 1 post-stimulation, session 2 post-stimulation, and session 3 post-stimulation co-registered resting CBF maps. This analysis was also conducted using randomise of FSL. Null distributions for contrasts representative of the main effect of session (sessions 1–3 post-stimulation subtracted from the baseline) were derived by performing 2,000,000 random permutations of the data. Each permutation was created by exchanging the assigned session while maintaining subject membership. Then, an F test compared the means from each session (sessions 1–3 post-stimulation subtracted from the baseline). Pairwise comparisons were executed during the completion of the one-way ANOVAs. The results of the pairwise comparisons were further corrected for multiple comparisons to account for false positives due to the multiple comparisons (Worsley, 2001). This method considered adjacent voxels with a t statistic of 2.3 or greater to be a cluster. The significance of each cluster was estimated and compared to a threshold of p < 0.05 using the Gaussian random field theory-based maximum height thresholding and a one-tailed t-test. The significance of voxels that either did not pass the significance level threshold or do not belong to a cluster was set to zero.

Concerns regarding the potential bias in the data due to the within-subject variability in scanning start time arose during the analysis (see Supplementary Table 2 for a complete list of the scan initiation times). To address this concern and the potential impact of the session time on any one group in particular, we evaluated the scan initiation times between groups by the session and scan. First, we extracted the scan initiation times from the log files. Next, we performed between-group one-way ANOVAs separately for the pre- and post-stimulation scan initiation times for each session. We did not correct for the family-wise error to be more sensitive to timing differences between groups. Our analyses did not find any significant variability (p > 0.05; Table 2) between groups for the scan start time for any of the scan sessions (Figure 2). There was also concern raised regarding the variable durations between the stimulation and the post-stimulation ASL that may disproportionally affect the results. Unfortunately, the stimulation time was not recorded in reference to the scan time. Therefore, we used the end time for the pre-stimulation task scan as the best approximation since the setup time for tDCS and post-task scanning was fairly consistent and much less variable than the total scan times. The duration between the pre-stimulation task end time and the post-stimulation scan start time was computed, and one-way ANOVAs were conducted to compare the groups across each session. Again, we did not correct for a family-wise error. The results did not reveal any significant variability (p > 0.05) in duration between the pre- and post-stimulation scans (Figure 3).

Interaction effects between the run and stimulation were assessed via the unpaired t-tests following a subtraction of the post-stimulation measurement from the pre-stimulation measurement from session 1 (i.e., baseline). The comparison of resting CBF from baseline and post-stimulation between CON and ACT_1mA groups resulted in significant variations across the three sessions (Figure 4). A similar trend was observed between CON and ACT_2mA (Figure 5). Frontal areas in these analyses appeared with an increasing statistical significance and extent, with stronger effects observed in the comparison between CON and ACT_2mA. These consisted of the bilateral superior frontal gyrus (SFG) and the right middle frontal and inferior frontal gyri (MFG and IFG, respectively). Posterior regions such as the right superior parietal lobule, the inferior parietal lobule, the middle temporal gyrus, and the precuneus demonstrated a trend with a decreasing statistical significance and extent.

These results are difficult to interpret alone and could represent either a greater decrease from baseline or a smaller increase from baseline between the CON group and the active group. Therefore, the within-group analyses were conducted to provide a clearer understanding of these effects. For the CON and ACT_1mA groups, repeated-measure one-way ANOVAs revealed significant differences in CBF across four different measurements (baseline and sessions 1–3 post-stimulation; see Figure 6). Regions identified in these analyses include the LC, superior temporal gyrus (STG), inferior temporal gyrus (ITG), supramarginal gyrus (SMG), and SFG. There were no significant findings for the main effect of the session in the ACT_2mA group. However, pairwise comparisons were further conducted to evaluate CBF each post-stimulation measure in comparison to the baseline for each group. The results summarized from these analyses identifying a decreased CBF from baseline are shown in Table 3 and the increased CBF from baseline are shown in Table 4.

Compared to the baseline, decreases in CBF were observed in all three post-stimulation sessions for the CON group (Figure 7; Supplementary Figure 3). The amount CBF was lowered in comparison to that of session 1 pre-stimulation increased in statistical reliability, extent, and magnitude by session 3 post-stimulation. Of note, decreased CBF was observed consistently in the bilateral STG and pre-central gyrus. Decreases in CBF in the ITG and SMG were observed in the post-stimulation measures for sessions 1 and 3. Additionally, decreased CBF in the LC was observed in the post-stimulation measures for sessions 2 and 3.

Few significant findings were present in the ACT_1mA group at session 1 post-stimulation but widespread decreases in CBF were observed at session 2 post-stimulation (Figure 8, top and middle rows; Supplementary Figure 4). These decreases share similarities with that observed in post-stimulation measures from sessions 1 and 3 in the sham group including the SMG and ITG. A decreased CBF diminished by session 3 with only a few small clusters remaining (Figure 8, bottom row). The diminished CBF at session 3 was accompanied by an increased CBF appearing bilaterally in the SFG and the anterior cingulate cortex (ACC).

In contrast with the CON and ACT_1mA groups, no significant changes in CBF were observed between baseline and the session 1 post-stimulation measurement for the ACT_2mA group. Less defined changes in CBF were observed in the session 2 post-stimulation scan (Figure 9, top row; Supplementary Figure 5). The largest clusters of decreased CBF were observed in the left STG and the left pre-central gyrus; these areas were also observed to have a decreased CBF in the CON group across all three comparisons. For the ACT_2mA group, more defined increases were observed by the third post-stimulation session (Figure 9, bottom row). Clusters were observed in the LC, the left hippocampus, the bilateral fusiform gyrus, the left thalamus, the right insula, the right IFG, and the left SFG.