Differential mechanisms of posterior cingulate cortex downregulation and symptom decreases in posttraumatic stress disorder and healthy individuals using real‐time fMRI neurofeedback

1 INTRODUCTION

The advent of functional magnetic resonance imaging (fMRI) has led to unprecedented insights into understanding the neurobiology of posttraumatic stress disorder (PTSD). It has been well documented that PTSD is associated with multiple functional disruptions in the brain that appear to underscore unique symptom presentations of the disorder (Fenster et al., 2018). Real-time fMRI neurofeedback (rt-fMRI-NFB) allows for such neural disruptions to be noninvasively regulated; as such rt-fMRI-NFB has been implemented in a broad range of prevalent psychiatric conditions (Linden et al., 2012; Kirsch et al., 2013; Li et al., 2013; Schoenberg and David, 2014; Paret et al., 2016a, 2019; Young et al., 2017; Mehler et al., 2018), including PTSD (Gerin et al., 2016; Nicholson et al., 2016a, 2018; Zotev et al., 2018; Zweerings et al., 2018; Chiba et al., 2019; Misaki et al., 2019; Weaver et al., 2020). Neurobiologically informed treatment interventions are particularly in demand for PTSD as suboptimal response rates to psychotherapy and pharmacological interventions have been reported (Bradley et al., 2005; Stein et al., 2006; Ravindran and Stein, 2009; Haagen et al., 2015; Krystal et al., 2017), where dropout rates remain high, particularly during trauma-focused interventions (Bisson et al., 2013; Goetter et al., 2015; Kehle-Forbes et al., 2016; Lewis et al., 2020).

In response to this demand, emerging scientific evidence suggests that directly regulating specific brain areas associated with PTSD symptomatology may be a viable treatment option for those affected by this illness (Reiter et al., 2016; Van der Kolk et al., 2016; Panisch & Hai, 2018; Chiba et al., 2019; Nicholson et al., 2020b, 2020c; Rogel et al., 2020). It has been hypothesized that normalizing the neural circuitry within large scale intrinsic connectivity networks (ICNs) is an essential treatment avenue for reducing PTSD symptoms (Lanius et al., 2015; Koek et al., 2019; Szeszko & Yehuda, 2019; Nicholson et al., 2020a, 2020b; Sheynin et al., 2020). Default mode network (DMN) functional disruptions among individuals with PTSD are thought to be related to traumatic/negative autobiographical memories, distorted and dysregulated self-referential processing, and alterations in social cognition (Bluhm et al., 2009; Daniels et al., 2010; Lanius et al., 2015; Tursich et al., 2015; Akiki et al., 2017; Fenster et al., 2018; Hinojosa et al., 2019; Frewen et al., 2020; Terpou et al., 2020). Indeed, to suffer from PTSD can be described as living with a disrupted self-narrative (Gerge, 2020; Lanius et al., 2020), where among individuals with PTSD, especially with early childhood maltreatment, there typically exists a highly rudimentary or shattered sense-of-self (Lanius et al., 2020).

The posterior cingulate cortex (PCC) is the major hub of the posterior default mode network (DMN) (Greicius et al., 2003; Buckner et al., 2008; Spreng et al., 2008; Qin and Northoff, 2011; Akiki et al., 2018). The PCC and the DMN are highly associated with PTSD symptoms, and display disrupted functional connectivity both at rest (Bluhm et al., 2009; Sripada et al., 2012; Chen & Etkin, 2013; Tursich et al., 2015; Yehuda et al., 2015; Lanius et al., 2015; Koch et al., 2016; Akiki et al., 2017, 2018; Barredo et al., 2018; Hinojosa et al., 2019; Nicholson et al., 2020a) and during executive functioning tasks in PTSD (Daniels et al., 2010; Melara et al., 2018). During rest, it has been shown previously using graph theoretical analyses that connectivity within the posterior community of the DMN involving the PCC and precuneus may be increased, relative to decreased connectivity within the anterior community of the DMN involving the medial prefrontal cortex (mPFC) (Shang et al., 2014; Kennis et al., 2016; Akiki et al., 2018; Holmes et al., 2018). Additionally, studies exploring seed-based functional connectivity patterns within the DMN at rest have revealed decreased coupling between the PCC, vmPFC, and other DMN structures, which together have been associated with PTSD symptoms (Bluhm et al., 2009; Qin et al., 2012; Sripada et al., 2012; Koch et al., 2016; Miller et al., 2017; DiGangi et al., 2016). During working memory tasks that require executive functioning, enhanced connectivity of the PCC with other DMN areas has also been reported among individuals with PTSD as compared to increased central executive network (CEN) and salience network (SN) connectivity among healthy individuals (Daniels et al., 2010). With respect to executive functioning tasks in PTSD, suboptimal downregulation of DMN regions may underscore difficulties in disengaging from internally focused self-referential processing in order to attend to external cognitive demands (Aupperle et al., 2016). Notably, the DMN also exhibits altered activation patterns during threatful- and trauma-related conditions in PTSD. Indeed, a recent meta-analysis has shown that both reexperiencing and retrieval of trauma-related autobiographical memories are associated with enhanced activation within the PCC and other DMN regions among individuals with PTSD as compared to healthy controls (Thome et al., 2019). Meta-analytic results reported elsewhere also suggest that traumatic imagery tasks uniquely induce activation in the PCC, with coactivation of the precuneus, relative to healthy controls (Ramage et al., 2013). Similarly, the presentation of trauma-versus-neutral words has been shown to increase activation in the PCC, the mPFC, the midbrain, and the bed-nucleus of the stria terminalis (BNST), with concomitant decreases in activation within dlPFC emotion regulation areas in PTSD as compared to healthy controls (Awasthi et al., 2020). This is supported by years of experimental work in the field linking these neural correlates with PTSD symptoms during both script-driven imagery and the recall of trauma-related autobiographical memories in PTSD (Hopper et al., 2007a; Lanius et al., 2007; Frewen et al., 2011; Mickleborough et al., 2011; Ramage et al., 2013; Liberzon & Abelson, 2016; Fenster et al., 2018; Thome et al., 2019). As such, during trauma-related stimulus exposure, it has been suggested that enhanced DMN recruitment in PTSD may coincide with self-related processes that are seemingly fused with experiences of trauma, indeed reflecting the self-coupled nature of the disorder (Terpou et al., 2019; Lanius et al., 2020). Furthermore, the PCC has been shown to be hyperactive in PTSD during emotion-processing tasks in comparison to healthy individuals, where critically, longitudinal improvements in PTSD symptoms in response to trauma-focused cognitive behavioral therapy (CBT) have been found to be associated with decreased PCC activation in youth with PTSD (Garrett et al., 2019). Taken together, regulating the PCC and the DMN may represent a critical avenue to explore with respect to neurobiologically informed treatment interventions for PTSD (Lanius et al., 2015; Akiki et al., 2018; Nicholson et al., 2020c).

In support of this, previous studies in PTSD using electroencephalography neurofeedback (EEG-NFB), including a randomized controlled trial by our group (Nicholson et al., 2020b), have examined the regulation of brain oscillations tied to the PCC and DMN (Kluetsch et al., 2014; Nicholson et al., 2016b). Notably, one session of EEG-NFB has been shown to lead to acute decreases in arousal symptoms among individuals with PTSD, which has been associated with a normalization of both DMN and amygdala resting-state functional connectivity (Kluetsch et al., 2014; Nicholson et al., 2016b). In these aforementioned EEG-NFB studies, the target of NFB was the desynchronization of alpha rhythms over the PCC. Alpha oscillations are correlated with DMN activation (Mantini et al., 2007; Jann et al., 2009; Clancy et al., 2020), where alpha-rhythm reductions are commonly observed during the resting-state in PTSD over the main hubs of the DMN (PCC and mPFC) (Clancy et al., 2020), hypothesized to be related to chronic hyperarousal (Ros et al., 2014; Liberzon & Abelson, 2016; Abdallah et al., 2017; Clancy et al., 2017, 2020; Sitaram et al., 2017; Nicholson et al., 2020c). Additionally, during a 20-week randomized controlled trial of alpha-desynchronizing EEG-NFB in PTSD (Nicholson et al., 2020b), individuals in the experimental group demonstrated significantly reduced PTSD severity scores post-NFB and at the 3-month follow-up, which was associated with a shift towards normalization of DMN resting-state functional connectivity. Specifically, PTSD patients in the experimental group were found to display decreased PCC connectivity with the anterior DMN after NFB treatment (Nicholson et al., 2020b). It was hypothesized that this may reflect normalized connectivity within over utilized posterior DMN communities consisting of the PCC and precuneus (Akiki et al., 2018; Holmes et al., 2018) after NFB treatment (Nicholson et al., 2020b). Notably, PTSD remission rates as well as decreases in PTSD severity scores in the experimental group were comparable to that of current gold-standard treatments for PTSD (Nicholson et al., 2020b). Collectively, preliminary results from our previous alpha-desynchronizing EEG-NFB studies suggest that feedback signals tied to the DMN, and more specifically the PCC, may represent a viable target for NFB treatment in PTSD. Critically, in comparison to EEG-NFB, rt-fMRI-NFB allows for increased spatial specificity with respect to precisely targeting areas in the brain and provides increased spatial resolution for examining mechanistic evidence associated with regulation.

Recently, the application of rt-fMRI-NFB in PTSD has expanded greatly, where previous studies have largely focused on the regulation of the amygdala (Gerin et al., 2016; Nicholson et al., 2016a, 2018; Misaki et al., 2018b, 2018a, 2019; Zotev et al., 2018; Chiba et al., 2019), a limbic region associated with emotion reactivity and highly implicated in PTSD symptoms (Fenster et al., 2018). Nicholson et al. (2016a) found that downregulating the amygdala in PTSD during trauma triggers increased activity and connectivity of the dlPFC and vlPFC involved in emotion regulation and executive functioning, findings supported by other rt-fMRI-NFB groups (Misaki et al., 2018b; Zotev et al., 2018). With regard to ICNs, Nicholson et al. (2018) also found that downregulating the amygdala with rt-fMRI-NFB led to increased recruitment of the CEN and stabilized DMN recruitment over NFB training. This represents a critical finding as individuals with PTSD have been shown to maladaptively recruit the DMN instead of the CEN during tasks that require cognitive control (Daniels et al., 2010). Of importance, Zotev et al. (2018) also showed in a randomized controlled study that amygdala regulation using rt-fMRI-NFB leads to significantly reduced PTSD severity scores, including significant reductions on avoidance, hyperarousal, and depressive symptoms. Extending the amygdala rt-fMRI-NFB literature, machine learning classifiers have also been utilized to improve performance on emotional conflict tasks by differentially selecting for brain states associated with targets as compared to trauma distractors (Weaver et al., 2020). Additionally, upregulating anterior cingulate cortex (ACC) activity has also been utilized in PTSD as a means to improve implicit emotion regulation capacities (Zweerings et al., 2018). Taken together, these results suggest that regulating specific brain areas tied to the manifestation of PTSD symptoms (e.g., the PCC of the DMN) may result in clinically meaningful changes, where additional studies are urgently needed to explore novel neurofeedback targets in PTSD.

1.1 Current study

Here, we utilized rt-fMRI-NFB to train PCC downregulation during emotion induction paradigms (presentation of trauma-related/distressing words) among individuals with PTSD and healthy controls. The rationale of the current study to downregulate the PCC was threefold: (1) the PCC is highly associated with PTSD symptomatology which together with other DMN areas, displays hyperactivity when trauma memories become activated (Frewen et al., 2020; Thome et al., 2019); (2) regulating neural signals related to the PCC/DMN using EEG-NFB has shown promising preliminary evidence in a randomized controlled trial (Nicholson et al., 2020b); and (3) the feasibility of downregulating amygdala activation using rt-fMRI-NFB in patients with PTSD has been demonstrated, which resulted in a shift toward normalization of DMN connectivity and reduced PTSD severity scores (Nicholson et al., 2016a, 2018; Zotev et al., 2018). Given the dynamic interplay between intrinsic brain networks (Menon, 2011), we hypothesized that PCC downregulation would lead to concomitant alterations in activation among regions within the DMN (e.g., mPFC), SN, and CEN (e.g., dlPFC involved in emotion regulation). We further predicted that NFB training would lead to decreased state PTSD/emotional symptoms. Moreover, given the well-documented differences between PTSD and healthy controls with respect to DMN recruitment during both emotion induction paradigms and executive functioning tasks, we hypothesized unique neural mechanisms associated with regulation (i.e., psychopathological specificity) and predicted that machine learning models would be able to accurately classify PTSD versus healthy controls during NFB training.

2 METHODS 2.1 Participants

Our neuroimaging sample consisted of n = 30 individuals [n = 15 patients with a primary diagnosis of PTSD and n = 15 healthy participants (see Table 1 for demographic and clinical characteristics of the study sample)]. The sample size of this preliminary investigation was based on study feasibility during the time of recruitment. One participant in the PTSD group was excluded from the analyses since they reported having fallen asleep in the scanner during the transfer run, thus leaving the final sample size n = 14 in the PTSD group and n = 15 in the healthy control group. No individual had previously received NFB, and there was no sample overlap with our previous NFB investigations (Nicholson et al., 2016a, 2018). There were nonsignificant differences with respect to biological sex between the PTSD and healthy control groups. However, the mean age of participants in the PTSD group was significantly higher as compared to the healthy control group. Importantly, when age was included as a covariate within the analyses described below, our neural activation results were not significantly affected. Participants were recruited from 2017 to 2019 through referrals from family physicians, mental health professionals, psychology/psychiatric clinics, community programs for traumatic stress, and posters/advertisements within the London, Ontario community.

TABLE 1. Demographic and clinical information PTSD group Healthy control group N 14 15 Sex 6 females, 8 males 10 females, 5 males Years of age 49.50 (± 5.11) 37.73 (±12.86) CAPS-total 43.21 (±8.26) 0 (±0) BDI-total 32.14 (±12.55) 1.2 (±2.46) CTQ-total 61.50 (±25.84) 31.13 (±8.44) MDI-total 87.36 (±28.23) 43.2 (±4.36) DERS-total 107.64 (±24.84) 52.80 (±9.03) MDD Current = 9, past = 2 Current = 0, past = 0 Agoraphobia Current = 1, past = 0 Current = 0, past = 0 Panic disorder Current = 1, past = 0 Current = 0, past = 0 Somatization disorder Current = 3, past = 0 Current = 0, past = 0 Psychotropic medication 10 0 Note: Values in bracket indicate standard deviation. Abbreviations: PTSD = Posttraumatic Stress Disorder, CAPS = Clinician Administered PTSD Scale, BDI = Becks Depression Inventory, CTQ = Childhood Trauma Questionnaire (none or minimal childhood trauma = 25–36, moderate = 56–68, extreme trauma > 72), MDI = Multiscale Dissociation Inventory, DERS = Difficulty in Emotion Regulation Scale, MDD = Major Depressive Disorder.

The inclusion criteria for PTSD participants included a primary diagnosis of PTSD as determined using the Clinician-Administered PTSD Scale (CAPS-5) and the Structured Clinical Interview for DSM-5 (SCID) (First et al., 2002; Weathers et al., 2013). Patients with PTSD currently receiving psychotropic medication were on a stable dose for 1 month prior to their participation in the NFB study. Exclusion criteria for PTSD patients included alcohol or substance use disorder not in sustained full remission within the last 3 months prior to scanning and a lifetime diagnosis of bipolar or psychotic disorders. PTSD patients were also excluded from the study if they had prominent current suicidal ideation within the past 3 months or self-injurious behaviours in the last 3 months requiring medical attention. Exclusion criteria for the healthy control group included lifetime psychiatric illness and current use of any psychotropic medications. Exclusion criteria for all participants included past or current biofeedback treatment, noncompliance with 3 Tesla fMRI safety standards, significant untreated medical illness, pregnancy, a history of neurological or pervasive developmental disorders, and previous head injury with loss of consciousness. Please see the supplementary materials section (Table S1) for a detailed report on the history of trauma exposure in each group.

Participants completed a battery of assessments before the NFB experiment, which consisted of the Beck's Depression Inventory (BDI) (Beck et al., 1997), the Childhood Trauma Questionnaire (CTQ) (Bernstein et al., 2003), and the Multiscale Dissociation Inventory (MDI) (Briere, 2002). In addition, in order to assess state changes in emotion-related symptoms during NFB, participants completed the Response to Script Driven Imagery (RSDI) Scale (Hopper et al., 2007a) after each of the 4 fMRI runs, which consisted of the following symptom subscales: reliving, distress, physical reactions, dissociation, and numbing. All scanning took place at the Lawson Health Research Institute in London, Ontario, Canada. The study was approved by the Research Ethics Board at Western University, Canada, where participants gave written and informed consent and received financial compensation for participating in the study.

2.2 Neurofeedback paradigm

We implemented an experimental protocol and paradigm that was identical to our previous NFB investigations (Nicholson et al., 2016a, 2018); however, we trained individuals to downregulate the PCC as opposed to the amygdala (Figure 1). Participants were instructed that they would be “regulating an area of the brain related to emotional experience,” that is, to decrease activation within the PCC. In order to elicit unbiased and personalized regulatory strategies, specific instruction on how to regulate the brain region-of-interest was not provided (Paret et al., 2014, 2016a; Nicholson et al., 2016a, 2018; Zaehringer et al., 2019). During training trials, feedback of PCC activation was displayed to participants in the form of two identical thermometers on the left and right side of a screen projected inside the scanner. The bars on the thermometer increased/decreased as BOLD signal in the PCC target fluctuated, where an orange line on the thermometer indicated baseline PCC activation (Figure 1).

image

Schematic of the real-time fMRI neurofeedback set-up. Brain activity in the neurofeedback target region (posterior cingulate cortex) was processed in real-time and presented to participants in the fMRI scanner as thermometers that increased or decreased as activation fluctuated. Participants completed three neurofeedback training runs and a transfer run without neurofeedback signal. Figure created with BioRender.com.

Our neurofeedback protocol consisted of three conditions: (i) regulate, (ii) view, and (iii) neutral. During the regulate condition (Figure 2), individuals were asked to decrease activity in the brain target (decrease the bars on the thermometer corresponding to PCC activation) while viewing either personalized trauma-related words for the PTSD group or a matched stressful word for the healthy control group (Nicholson et al., 2016a, 2018). During the view condition, individuals were asked to respond naturally to their personalized trauma/stressful words while not attempting to regulate the target brain area. Neutral trials consisted of asking individuals to respond naturally to personalized neutral words for both groups. Personalized trauma/stressful words (n = 10) and neutral words (n = 10) were selected by participants with a trauma-informed clinician and matched on subjective units of distress to control for between subject/group variability. The personalized trauma words selected by participants with PTSD were related to individual experiences of trauma. Furthermore, personalized stressful words selected by healthy controls were related to the individual's most stressful life event. Stimuli were presented with Presentation software (Neurobehavioral Systems, Berkeley, CA). Participants were first provided with written instructions, followed by a single example trial within the scanner. Our experimental design then consisted of three consecutive neurofeedback training runs, which was followed by one transfer run in which individuals were presented with the same three conditions but without neurofeedback from the thermometer. Instructions were presented for 2 s before each condition; individual conditions lasted for 24 s and were followed by a 10 s implicit resting state where participants viewed a fixation cross (Figure 2). An experimental run lasted about 9 min and consisted of 15 trials (5 of each condition, counterbalanced and separated by an intertrial fixation cross) (Nicholson et al., 2016a, 2018).

image

Neurofeedback experimental procedure for the regulate condition. The same timing was utilized for (i) view conditions in which participants viewed trauma-related/distressing words while not attempting to regulate and (ii) for neutral conditions in which participants viewed neutral words and did not attempt to regulate. A trial started with a 2 s instruction slide indicating trial type (i.e., regulate, view, neutral). In the following block, participants saw either a trauma-related/distressing word or a neutral word with a thermometer at both sides. The thermometer displayed the change in brain activation and was updated every 2 s.

One bar on the thermometer display corresponded to 0.2% signal change in the PCC, consisting of an upper activation range with a maximum of 2.8% signal change and a lower activation range with a maximum of 1.2% signal change (Paret et al., 2014, 2016b; Nicholson et al., 2018). Participants were instructed to visually focus on the word during its entire presentation and to view the two thermometers in their peripheral vision. Emotion-induction effects of personalized stimuli were confirmed both on the subjective experience level via inspection of RSDI scores and on the neurobiological level by contrasting view as compared to neutral conditions (see results section below). Participants were also informed of the temporal delay that would occur during neurofeedback, corresponding to both the BOLD signal delay and real-time processing of this neural activation. Finally, when a neurofeedback run was completed, individuals were asked to rate their perceived ability to regulate the target brain area. Specifically, we asked participants to rate the extent to which they were able to gain control over the neurofeedback signal, which ranged from 0 (not at all) to 6 (a great deal).

2.3 Real-time signal processing for neurofeedback

Anatomical scans were first imported into BrainVoyager (Brain Innovations, Maastrict, the Netherlands), skull-stripped, and then transformed into Talairach space. Normalization parameters were then loaded into TurboBrainVoyager (TBV) (Brain Innovations, Maastricht, the Netherlands). Motion correction features and spatial smoothing using a 4-mm full-width-half-maximum (FWHM) Gaussian kernel were implemented in TBV, and the initial 2 volumes of the functional scans were discarded before real-time processing. We defined the target PCC using a 6 mm sphere over the following coordinate (MNI: 0 -50 20) (Bluhm et al., 2009). We used the “best voxel selection” tool in TBV to calculate the BOLD signal amplitude in the PCC. This tool identifies the 33% most active voxels for the view > neutral contrast. Further details on dynamic ROI definitions can be found in our previous publications (Nicholson et al., 2016a, 2018). The first two trials of each neurofeedback run consisted of view and neutral conditions thereby allowing for initial selection of PCC voxels based on the view > neutral contrast, which was dynamically updated as voxels selection was refined along the course of training. For each trial, the mean of the last 4 data points before stimuli onset (during the implicit resting state) were selected as a baseline and indicated to participants as an orange line on the thermometer display. The signal was smoothed by calculating the mean of the current and the preceding 3 data points (Paret et al., 2014, 2016b; Nicholson et al., 2016a).

2.4 fMRI image acquisition and preprocessing

Neuroimaging was conducted using a 3 Tesla MRI Scanner at the Lawson Health Research Institute (Siemens Biograph mMR, Siemens Medical Solutions, Erlangen, Germany) with a 32-channel head coil, where during scanning participants’ heads were stabilized. Functional whole brain images of the BOLD contrast were acquired with a gradient echo T2*-weighted echo-planar-imaging sequence (TE = 30 ms, TR = 2 s, FOV = 192 × 192 mm, flip angle = 80°, inplane resolution = 3 × 3 mm). One volume comprised 36 ascending interleaved slices tilted −20° from AC-PC orientation with a thickness of 3 mm and slice gap of 1 mm. The experimental runs comprised 284 volumes each, where T1-weighted anatomical images were acquired with a Magnetization Prepared Rapid Acquisition Gradient Echo sequence (TE = 3.03 ms, TR = 2.3 s, 192 slices and FOV = 256 × 256 mm).

Preprocessing of the functional images was performed with SPM12 (Wellcome Department of Cognitive Neurology, London, UK) within MATLAB R2020a. Our standard preprocessing routine included discarding 4 initial volumes, slice time correction to the middle slice, reorientation to the AC-PC axis, spatial alignment to the mean image using a rigid body transformation, reslicing, and coregistration of the functional mean image to the subject's anatomical image. The coregistered images were segmented using the “New Segment” method implemented in SPM12. The functional images were normalized to MNI space (Montréal Neurological Institute) and were smoothed with a FWHM Gaussian kernel of 6 mm. Additional correction for motion was implemented using the ART software package (Gabrieli Lab, McGovern Institute for Brain Research, Cambridge, MA), which computes regressors that account for outlier volumes.

2.5 Statistical analyses 2.5.1 First-level analysis

We defined separate sessions for each neurofeedback training run and the transfer run, where all events (initial rest, instructions, fixation, and conditions) were modeled as blocks of brain activation and convolved with the hemodynamic response function. In the first level, functional data were also high-pass filtered and serial correlations were accounted for using an autoregressive model. Additionally, ART software regressors were included as nuisance variables to account for any additional movement and outlier artifacts. The three experimental conditions (regulate, view, and neutral) were modeled separately on the first level.

2.5.2 Second-level analyses

We first conducted a split-plot full factorial 2 (group) by 3 (condition) by 3 (NFB training run) ANOVA within SPM12 to investigate changes in whole-brain activation, inputting separate condition specific contrast images generated in the first level. As we were specifically interested in differential activation during the regulate and view conditions (Nicholson et al., 2016a; 2018), we examined follow-up comparisons focusing on between condition effects within group, as well as between groups comparing the PTSD and healthy control groups. We then examined the transfer run separately, where we conducted a 2 (group) by 3 (condition) ANOVA and subsequently examined aforementioned direct follow-up comparisons. All analyses were whole-brain corrected for multiple comparisons using a clusterwise false discovery rate (FDR) threshold at p < .05, k = 10, with an initial clustering defining threshold in SPM at p < .001, k = 10 (Eklund et al., 2016; Roiser et al., 2016).

Finally, we conducted linear regression analyses across all subjects, examining potential correlations between trait-based symptoms and whole-brain activation during view as compared to regulate conditions over NFB training runs. Here, we examined PTSD symptom severity scores (CAPS-5 total), difficulty in emotion regulation scores (DERS total), and depressive symptoms (BDI total).

2.5.3 Neurofeedback PCC downregulation analysis

In order to evaluate PCC downregulation (neurofeedback success), we extracted the event-related BOLD response (peristimulus time histogram) from the PCC target area during the regulate and the view conditions using rfxplot software (Gläscher, 2009), using the same sphere definition that was used to generate feedback for participants in the fMRI scanner. Here, we extracted the event-related BOLD response from individual peaks within the search volume, and these values were then passed to SPSS (v.26) for statistical analyses. Within rfxplot software, event-related BOLD responses display the average height of the BOLD responses within a defined volume and time window (Gläscher, 2009). Rfxplot shows the actual data and does not rely on first- or second-level models. Event-related BOLD responses are estimated by a condition-specific Finite Impulse Response (FIR) model (Gläscher, 2009). Here, the condition duration in which the BOLD response is expected to fluctuate is parcellated into temporal bins (TR = 2 s) starting at the onset of all trials belonging to a particular condition. The parameter estimate for each bin of the FIR model is identical to the mean BOLD response in that bin, thus creating an event-related BOLD time course for each subject. For the final display (Figure 3), rfxplot averages subject-specific event-related BOLD responses based on group.

image

(a) Event-related BOLD response in the NFB target area (PCC), during the three training runs in the PTSD and healthy control groups. The red lines indicate PCC activation during the regulate condition, where the goal was to decrease activation while viewing trauma/stressful words. The green lines indicate PCC activation during the view condition, where participants were not attempting to decrease activation while viewing trauma/stressful words. Here, PCC activation was significantly lower during regulate as compared to view conditions for NFB training runs 1–3, for both the PTSD and healthy control groups. (b) Event-related BOLD response in the target area (PCC) during the transfer run when neurofeedback was not provided. Taken together, this demonstrates that both groups were able to gain control over downregulating their PCC with similar success. The x-axis of the graphs indicate time over the 24 s conditions; the y-axis indicates the event-related BOLD response (peristimulus time histogram) in the target area. Shaded areas of red and green indicate standard error of the mean. Abbreviations: PCC = posterior cingulate cortex, NFB = neurofeedback.

For the PTSD and healthy control groups separately, we computed repeated measures 2 (condition) by 4 (NFB run) ANOVAs. Subsequently, we then conducted a priori defined paired sample t-tests, comparing the average BOLD response within the NFB target area between conditions for each NFB run within groups. We also conducted independent samples t-tests comparing the average BOLD response within the NFB target area during a single NFB run for a given condition between groups. Lastly, we conducted repeated measures one-way ANOVAs for the regulate condition for each group in order to examine potential learning effects across NFB training.

2.5.4 State changes in emotional experience over neurofeedback

We examined state changes in subjective response to traumatic/stressful stimuli over the NFB training experiment, as measured by RSDI subscales. As collected data were not normally distributed, we computed nonparametric Friedman's repeated measures ANOVAs for each group for each RSDI subscale. Here, we Bonferroni corrected our statistical threshold (p < .05/5 = .01) for nonparametric ANOVAs. Paired comparisons between time points were conducted using nonparametric tests for related samples (Wilcoxon signed-ranks test). We then compared state symptoms across NFB runs between groups with Mann–Whitney U tests.

2.5.5 Machine learning classification analysis

We examined the accuracy of machine learning algorithms in classifying PTSD patients as compared to healthy individuals based on whole brain activation during view as compared to regulate conditions across NFB training runs. Here, we implemented L1-Multiple Kernel Learning (MKL) Classification algorithms within PRoNTo toolbox (http://www.mlnl.cs.ucl.ac.uk/pronto/) (Schrouff et al.,

留言 (0)

沒有登入
gif