Abstract
BACKGROUND:
Traumatic brain injury (TBI) often results in chronic impairments to cognitive function, and these may be related to disrupted functional connectivity (FC) of the brain at rest.
OBJECTIVE:
To investigate changes in default mode network (DMN) FC in adults with chronic TBI following 40 hours of auditory processing speed training.
METHODS:
Eleven adults with chronic TBI underwent 40-hours of auditory processing speed training over 13-weeks and seven adults with chronic TBI were assigned to a non-intervention control group. For all participants, resting-state FC and cognitive and self-reported function were measured at baseline and at a follow-up visit 13-weeks later.
RESULTS:
No significant group differences in cognitive function or resting-state FC were observed at baseline. Following training, the intervention group demonstrated objective and subjective improvements on cognitive measures with moderate-to-large effect sizes. Repeated measures ANCOVAs revealed significant (p < 0.001) group×time interactions, suggesting training-related changes in DMN FC, and semipartial correlations demonstrated that these were associated with changes in cognitive functioning.
CONCLUSIONS:
Changes in the FC between the DMN and other resting-state networks involved in the maintenance and manipulation of internal information, attention, and sensorimotor functioning may be facilitated through consistent participation in plasticity-based auditory processing speed training in adults with chronic TBI.
Keywords
Introduction
Approximately 2% of the US population is currently living with long-term disability related to traumatic brain injury (TBI; Faul & Coronado, 2015). Among the most commonly reported cognitive deficits to chronically persist after TBI are impairments in processing speed and attention (Dikmen et al., 2009). A primary cause of cognitive dysfunction in TBI is traumatic axonal injury (TAI) from deformation injury to white matter fibers, resulting in diffuse damage to the brain’s structural connectivity (Bigler & Maxwell, 2012).
Functional connectivity (FC) refers to the temporal coherence or correlation between the activity recorded in discrete anatomical brain regions simultaneously (Broyd et al., 2009). FC is measurable through spontaneous blood oxygenation level-dependent (BOLD) signal fluctuations while the brain is at rest through resting state functional MRI (rsfMRI). Typically, resting state networks (RSNs) show suppressed activation during goal-directed tasks and increased activation when the brain is at rest. The most commonly studied RSN is the default mode network (DMN), which is largely involved in introspection and self-referential processes; the core regions of the DMN include the medial prefrontal cortex (mPFC), the posterior cingulate cortex (PCC), and the inferior parietal lobules (IPL; Buckner et al., 2008). These and other regions of the DMN are densely interconnected with other areas throughout the brain, forming a large part of the brain’s core structural network (Horn et al., 2014).
Previous rsfMRI investigations have further described the DMN as a counterpoint to task-positive networks (TPNs), including the salience (SN), central executive (CEN), and dorsal attention (DAN) networks, which demonstrate increased activity during tasks that involve externally directed cognitive control (Kelly et al., 2008; Uddin et al., 2009). DMN activity is proportionately decreased during such tasks; thus, DMN activity is considered to be negatively correlated, or anticorrelated with TPN activity ( & Raichle, 2007; Fox et al., 2005). The anticorrelated patterns inherent between the DMN and TPN are indicative of temporal modulation in the opposite direction, reflecting a fundamental property of intrinsic functional organization (Fox et al., 2005) and demonstrating a task-related dichotomy between internally and externally oriented attentional control (Corbetta & Shulman, 2002; Uddin et al., 2009); this suggests that anticorrelated networks play a role in the segregation of competitive neuronal processes, which is crucial for attention and the efficient processing of information.
Hypersynchronous neuronal signaling between the DMN and TPN regions has been demonstrated in chronic TBI and is suggested to reflect a trade-off between metabolic cost and communication efficiency while networks undergo reorganization after TAI (Hillary & Grafman, 2017; Roy et al., 2017). While hyperconnectivity is considered to be an adaptive response in the months following TAI, long-term increases in FC reflect disorganization and overcompensation, leading to disadvantageous functional outcomes (Hillary & Grafman, 2017). The chronic hyperconnectivity of frontal, temporal, and parietal regions reduces the inherent, anticorrelated FC between the DMN and TPN, thus preventing their role in the inhibition of competitive cognitive processes and ultimately resulting in impaired attentional capacity and inefficient information processing.
Injury-induced alterations to functional and structural connectivity contribute to the cognitive impairments typically seen in TBI; however, changes to the brain after an injury also reflect the brain’s plasticity and capacity for recovery (Castellanos et al., 2011; Hillary et al., 2015). Organized patterns of anticorrelated activity form between two networks as the result of repetitive activation of one network with simultaneous deactivation of the anticorrelated network throughout training (Bonavita et al., 2015; Jolles et al., 2013; Lewis et al., 2009; Voss et al., 2010). By implementing treatment strategies that enhance neuronal synchrony within regions comprising coherent and distinct networks in the healthy brain, it is possible to accelerate the recovery of FC in damaged networks (Fox & Raichle, 2007; Honey et al., 2009).
The purpose of the present study is to explore the effects of plasticity-based auditory processing speed training on resting state FC of the DMN in adults with chronic TBI. We hypothesized that atypical patterns of FC would be present in our sample of adults with chronic TBI at baseline, consistent with the results of similar studies (Hillary et al., 2011; Sharp et al., 2011; Sours et al., 2015; Venkatesan et al., 2015). It was also hypothesized that consistent participation in Hebbian plasticity-based auditory processing speed training would repeatedly activate TPN regions, while simultaneously suppressing activity in regions of the DMN. Specifically, relative to a non-intervention TBI control group, those in the intervention group would demonstrate greater distinctions in coherent network FC and increased anticorrelated activity between the DMN and TPN. The relationship between changes in FC and performance on cognitive tasks was also explored.
Method
Participants
Eighteen adults with chronic mild to severe TBI were recruited from outpatient brain injury rehabilitation centers and clinics in the New York metropolitan area. Demographic and injury characteristics between the experimental (n = 11) and control (n = 7) groups can be found in Table 1, and patient-specific demographic and injury characteristics are reported in Table 2. Inclusion criteria required participants to be between the ages of 18–70 years and to have sustained at least one brain injury no less than 12 months prior to participation. Participants were recruited at least one year post-injury to rule out spontaneous recovery as the source of functional gains that occur during enrollment in the study. Exclusion criteria included a history of other neurological disorder(s), diagnosis of bipolar disorder or schizophrenia, current use of benzodiazepine, anti-convulsant, or neuroleptic medications, and/or current drug or alcohol use disorder. Injury severity was classified according to initial Glasgow Coma Scale (Teasdale & Jennett, 1974) score, duration of loss of consciousness, and duration of posttraumatic amnesia in accordance with the American Congress of Rehabilitation Medicine (Kay et al., 1993) and Mayo Classification System for TBI Severity (Malec et al., 2007). Twenty-one adults with TBI were initially included in the present analysis, however two participants in the control group and one participant in the intervention group were excluded due to excessive head motion, later diagnosis of amyotrophic lateral sclerosis, and pre- and post-imaging occurring on different scanner models, respectively.
Sample demographic and injury characteristics
Sample demographic and injury characteristics
MVA = motorized vehicle accident. TSI = time since injury (months). WTAR = Wechsler Test of Adult Reading, values represented are age and education corrected standard score (SS, M = 100, SD = 15).
Patient-specific injury and demographic characteristics
TSI = time since injury. BL = baseline. FU = follow-up. FSIQ = estimated premorbid full scale IQ based on the Wechsler Test of Adult Reading (WTAR; Wechsler, 2001). M = male. F = female. MVA = motorized vehicle accident (Pass = passenger, Ped = pedestrian).
This study is registered at Clinicaltrials.gov through the US National Institute of Health/National Library of Medicine (NCT02507271). Approval was obtained from the New York University School of Medicine Institutional Review Board prior to beginning the study. All participants were screened for eligibility before enrollment and provided written informed consent prior to participation. A quasi-randomized technique was used for group assignment, where participants were offered 30-day unlimited MTA metro cards as an incentive to undergo the training. In addition to compensation for the cost of travel to and from the laboratory for the training sessions, the unlimited-use metro cards could be applied toward any other use of public transportation during the enrollment period. Participants who accepted the incentive were assigned to the intervention group and matched on age, sex, education, and injury severity with a participant assigned to the non-intervention, control group. It is of note that the sample used in the present study is a subset of the larger, matched sample, where only those who underwent baseline and follow-up neuroimaging are included in the present analysis.
Baseline and follow-up visits consisted of structural and functional neuroimaging and a short neuropsychological battery. During the rsfMRI scans, participants were instructed to remain awake with their eyes open. Over the course of 13 weeks following the baseline assessment and neuroimaging scan, intervention participants completed 40 one-hour training sessions. Over the same period of time, control participants were contacted weekly by telephone and asked about the kinds of cognitively stimulating activities they had engaged in over the previous week. The follow-up assessment and neuroimaging took place within two weeks after the 13-week intervention period. Participants were provided $150.00 upon completion of the baseline visit and again upon completion of the follow-up visit. A flow diagram demonstrating the participation timeline is provided in Fig. 1.
Flow diagram of the timeline of participation in the study. Baseline and follow-up evaluations include the cognitive assessments and neuroimaging scans.
The cognitive intervention used in this study was the Brain Fitness Program (BFP) version 2.0.1b (Posit Science, San Francisco, CA). This computerized training program is designed to improve speed of auditory information processing and is effective for the long-term maintenance of enhanced cognitive functioning and structural integrity in healthy older adults (Anderson et al., 2014; Mahncke et al., 2006; Smith et al., 2009; Styliadis et al., 2015). The BFP is comprised of six 15–minute modules that involve discrimination, recognition, and syllable pair matching, along with the reconstruction of sequences of verbal instructions and the identification of story details presented verbally (Mahncke et al., 2006; Smith et al., 2009). These exercises become progressively more difficult through the systematic reduction of the inter-stimulus interval or stimulus duration. The exercises automatically adjusted to user performance to sustain an 80% correct response rate, maintaining a mismatch between user ability and task demand (Zelinski et al., 2011). As the intervention involves both visual and auditory modalities, participants were provided with noise-canceling headphones to wear during the training.
Neuropsychological testing
The cognitive battery included measures of processing speed and attention, including the Paced Auditory Serial Addition Task (PASAT; Gronwall, 1977), Trail Making Test, parts A and B (TMTA, TMTB; Reitan & Wolfson, 1993) and the Understanding Directions subtest from the Woodcock-Johnson III Tests of Cognition battery (WJUD; Woodcock et al., 2001). Additionally, self-reported responses were collected from the Beck Depression Inventory (BDI-II; Beck & Steer, 1993b), Beck Anxiety Inventory (BAI; Beck & Steer, 1993a), and the cognitive (CSR-C) and total (CSR-T) scales of the Cognitive Self-Report Questionnaire (Spina et al., 2006). Raw scores from all cognitive tests were converted to demographically-corrected standardized scores using normative data provided by test publishers for the PASAT, CSR-C, CSR-T) or Heaton et al. (2004) for the TMTA and TMTB.
Neuroimaging procedure
Image acquisition
All MRI data were acquired at the Center for Biomedical Imaging of New York University’s Langone Medical Center on a 3T Siemens Tim Trio or Prisma Fit whole-body scanner using a 12-channel head coil. Structural images were acquired on the sagittal plane using a 3D T1-weighted MP-RAGE sequence via interleaved sampling with 192 slices and an isotropic 1.0 mm voxel resolution (TR/TE/TI =2300/2.98/900 ms, FOV = 240×256 mm, matrix size = 256×240 mm, flip angle = 9°). The 2D resting-state functional MR images consisted of 250 volumes acquired on the transverse plane via a T2*-weighted single-shot EPI pulse sequence with a 3.44×3.44×3.40 mm voxel resolution (TR/TE =2000/25 ms, FOV = 220×220 mm, matrix size =64×64 mm, flip angle = 90°). Participants were scanned on the same scanner at baseline and follow-up.
Image preprocessing
Advanced Normalization Tools (ANTs, version 3.0; Avants et al., 2009) was used for the preprocessing of structural imaging data. Structural data were AC-PC aligned before undergoing a N4 bias field correction. Each image was then registered to standard space using the ICBM 2009c nonlinear symmetric 1×1×1 mm MNI152 template (Fonov et al., 2009) before undergoing affine registration to create a target template space. The data was then diffeomorphically transformed and combined into a custom, whole-sample model template, which was skull-stripped using the priors from the MNI152 OASIS-30 dataset (Klein, 2016). This model template was used as the standard anatomical space for the functional imaging analyses.
Toolkits from the Analysis of Functional NeuroImages (AFNI, version 20.2.16; Cox, 1996) software were used for preprocessing of all functional imaging data. The data were initially despiked, slice-time corrected, and motion corrected via rigid body correction before undergoing volume registration. The preprocessed structural datasets (in subject space) were initially segmented into CSF, white matter, and gray matter (Vovk et al., 2011). CSF and white matter segmentations were resampled to the functional grid, and the average time course was extracted from each.
The structural datasets were co-registered to functional space and resampled onto the functional grid. The resampled structural and preprocessed functional datasets were then registered to MNI space using the previously calculated deformation matrix for each participant in ANTs. These data were then skull-stripped in AFNI for use as the individual subject’s brain mask dataset. Sources of spurious variance, including the six parameters of head motion, which were obtained through rigid body correction, and the time course signals extracted from the white matter and CSF, were excluded from the functional data via multiple linear regression using AFNI’s 3dDeconvolve tool. The global brain signal was not regressed out of the data, as doing so may bias interregional correlations downward, artificially introducing anticorrelations between brain regions where no significant correlations would otherwise exist (Murphy et al., 2009; Saad et al., 2012). A temporal bandpass filter of 0.008 to 0.09 Hz was applied to the data, and the subject’s skull-stripped brain mask was applied to exclude any unwanted voxel noise from the regression. The output of the regression was smoothed with a 6-mm Gaussian blur.
Functional connectivity analysis
A modified seed-based correlational approach (Uddin et al., 2009) was used for subject-level analysis of the functional data. Separate 8-mm (radius) seed regions were created for the mPFC (MNI xyz = –2, 46, –10), PCC (MNI xyz = –2, –50, 26), and bilateral IPLs (Right: MNI xyz = 50, –60, 32; Left: MNI xyz = –46, –66, 32) from the functional datasets, with coordinate locations determined by an automated probabilistic mapping approach through the online NeuroSynth framework (Yarkoni et al., 2011). Bivariate correlation coefficients (Pearson’s r) were computed between the BOLD time course fluctuations of each seed region and that of all other voxels in the brain, and a Fisher’s r-to-z transformation was applied to the correlational data to ensure that the coefficients were normally distributed. Prior to beginning group-level analyses, a binary, skull-stripped brain mask was created from the model template in MNI space and resampled to the functional grid; this brain mask was included in the following group-level analyses to negate any non-brain voxel activity from influencing the results.
Statistical analysis
All statistical analyses for demographic and cognitive data were completed using Stata 16.1. Homogeneity of variance between groups was confirmed via Levene’s test prior to each between-group comparison of continuous data. Between-group differences in demographic and injury characteristics were analyzed using an independent samples t-test for continuous variables and Pearson’s chi-squared or Kruskall Wallis tests for categorical variables. Welch’s t-tests were used to compare baseline performance on cognitive tests and self-reported measures of function between the two groups. Baseline to follow-up change scores were computed for the investigation of change in performance on cognitive and self-report measures over time by subtracting the baseline score from the follow-up scores. Welch’s t-test and Wilcoxon’s ranked-sums tests were used to compare change in performance on cognitive tests and self-report measures between the two groups, with Hedges’s g and Rosenthal’s r included as measures of effect size, respectively. For all analyses, statistical significance was thresholded at an α = 0.05, and effect size is interpreted according to the conventions outlined by Cohen (1988). To explore the relationship between changes in FC and changes cognitive function in the intervention group, mean z-transformed correlation coefficients were extracted from brain regions that showed a significant training-related effect, and semipartial correlations were used to evaluate the relationship between these changes in DMN FC and changes in performance on cognitive and self-report measures that were significantly different between the two groups; the effects of scanner model on DMN FC were controlled for. Due to the exploratory nature of the investigation into changes in the brain-behavior relationship following training, these results are interpreted according to effect size only.
Functional connectivity
Baseline correlational data for each seed region were subjected to a one-sample t-test to determine FC of the DMN at baseline. A 2×2 repeated measures ANCOVA was conducted to determine the effect of a group×time interaction on FC of each DMN seed region, with scanner model included as a covariate. A Monte Carlo simulation was used to determine the appropriate cluster-size threshold of within-brain voxels at a voxel-wise entry criterion of p < 0.001, where 3 contiguous voxels (using first-nearest neighbor clustering) represent a significant group-by-time interaction, thresholded at a two-tailed p < 0.001. Specific neuroanatomical locations of the MNI coordinates representing peak voxel activity within significant correlation clusters were determined using MRIcron (v. 1.0.20190902), which generates region names for specified MNI coordinate locations using the Automated Anatomical Labeling (Tzourio-Mazoyer et al., 2002) and Broadman area atlases (Brodmann, 2006).
Results
No significant differences were found for any demographic variable (Table 1), nor were any differences found on neuropsychological or self-reported measures between the two groups at baseline. Following training, group differences (Table 3) were observed on measures of visual and auditory attention, where performance was significantly increased in the Intervention group on TMTA (p = 0.001) and WJUD (p = 0.038), relative to the control group. Additionally, the Intervention group reported significantly fewer symptoms on the CSR-C (p = 0.003) and CSR-T (p = 0.025) following training, when compared to the control group (Table 3).
Change in neuropsychological task performance and self-reported functioning from baseline to follow-up
Change in neuropsychological task performance and self-reported functioning from baseline to follow-up
PASAT = Paced Auditory Serial Addition Task (2-second trial). TMTA = Trail Making Test part A. TMTB = Trail Making Test part B. WJUD = Woodcock-Johnson III Cognition Battery, Understanding Directions subtest. BDI-II = Beck Depression Inventory –Second edition. BAI = Beck Anxiety Inventory. CSR-C = Cognitive Self Report Questionnaire, Cognitive subscale. CSR-T = Cognitive Self-Report Questionnaire, Total score. Welch’s Independent t-test was conducted on the TMTA, TMTB, WJUD, BDI-II, and BAI, with Hedges’s g as a measure of effect size, and Wilcoxon’s ranked-sums test was conducted on PASAT, CSR-C, and CSR-T due to normality violations, with Rosenthal’s r as a measure of effect size.
The one-sample t-test revealed significant (p < 0.001) correlations within the DMN and between the DMN and other functional networks for all participants at baseline (Table 4, Fig. 2). Positively correlated FC was observed within DMN regions (PCC – left supramarginal gyrus; mPFC – left IPL; right IPL – right medial orbitofrontal cortex, right angular gyrus, and left rectus gyrus; left IPL – left and right angular gyri and right precuneus), between the DMN and regions of the frontoparietal control network (FPN; PCC – left superior frontal gyrus and left supramarginal gyrus; mPFC – left inferior parietal lobe; right IPL – right middle frontal gyrus and right supramarginal gyrus; left IPL – left supramarginal gyrus and right middle frontal gyrus), and between the DMN and various TPNs, including the visual network (right IPL – right superior occipital gyrus and right lingual gyrus; left IPL – right middle occipital gyrus, bilateral calcarine cortex, and left superior occipital gyrus), DAN (right IPL – left cerebellum crus II and right middle temporal gyrus; left IPL – left superior parietal lobe), and sensorimotor network (SMN; right IPL – right precentral gyrus and right superior temporal gyrus; left IPL – right precentral gyrus).
Functional connectivity with the default mode network at baseline
Functional connectivity with the default mode network at baseline
BA = Brodmann’s area; k = cluster size/volume; PCC = posterior cingulate cortex; mPFC = medial prefrontal cortex; IPL = inferior parietal lobe; PCC seed = 8 mm radius (MNI xyz = –2, –50, 26); mPFC seed = 8 mm radius (MNI xyz = –2, 46, –10); Right IPL seed = 8 mm radius (MNI xyz = 50, –60, 32); Left IPL seed = 8 mm radius (MNI xyz = –46, –66, 32); x, y, and z refer to peak voxel coordinates in MNI152 space, following the LPI convention. All findings are thresholded at a clusterwise p < 0.001 following a Monte Carlo simulation (extent 3 contiguous voxels), and all between-group differences are significant at p < 0.001.
Whole sample default mode network (DMN) functional connectivity (FC) at baseline. The array of axial images (inferior-to-superior) demonstrates the patterns of FC with the DMN seed regions in the present sample (n = 18) at baseline. Stronger positively correlated FC with the DMN is indicated by red-to-yellow regions of activation, and stronger anticorrelated FC with the DMN is indicated by blue-to-green regions of activation. Images are overlaid on an averaged, whole-sample template brain standardized in MNI152 space, lateral-posterior-inferior (LPI) oriented, and represented in neurological view (left = left).
Anticorrelated FC was also observed within DMN regions (PCC – right parahippocampal gyrus and right ventral anterior cingulate cortex; mPFC – left parahippocampal gyrus and left ventral anterior cingulate cortex; left IPL – right hippocampus), between DMN and FPN regions (PCC – right superior, middle, and inferior frontal gyri, left middle temporal pole, and right supramarginal gyrus; mPFC – left middle frontal gyrus), and between the DMN and various TPNs, including the basal ganglia network (PCC – right basis pontis, bilateral pallidum, and left putamen; mPFC – left caudate nucleus), SN (PCC – left insula and bilateral dorsal anterior cingulate cortex; mPFC – right insula), ventral attention network (PCC – right pars triangularis and right pars orbitalis), SMN (PCC – right pre- and postcentral gyri; mPFC – right postcentral gyrus), CEN (right IPL – left thalamus), and visual network (PCC – left fusiform gyrus and right middle occipital gyrus).
The 2×2 repeated measures ANCOVA revealed significant (p < 0.001) group×time interactions for FC with the DMN (Table 5). Significant interactions for increased positively correlated FC (Fig. 3) were observed within DMN regions (PCC – right precuneus; mPFC – right precuneus; right IPL – left angular gyrus) and between the DMN and other networks, including the SMN (mPFC – right postcentral gyrus), the FPN (right IPL – right middle frontal gyrus; left IPL – right supramarginal gyrus, right middle frontal gyrus, and right superior frontal gyrus), the DAN (left IPL – right middle temporal gyrus), the language network (right IPL – right inferior temporal gyrus), and the visual network (left IPL – left lingual gyrus).
Changes in functional connectivity with the default mode network following training
Changes in functional connectivity with the default mode network following training
BA = Brodmann’s area; k = cluster size/volume; PCC = posterior cingulate cortex; mPFC = medial prefrontal cortex; IPL = inferior parietal lobe; PCC seed = 8 mm radius (MNI xyz = –2, –50, 26); mPFC seed = 8 mm radius (MNI xyz = –2, 46, –10); Right IPL seed = 8 mm radius (MNI xyz = 50, –60, 32); Left IPL seed = 8 mm radius (MNI xyz = –46, –66, 32); x, y, and z refer to peak voxel coordinates in MNI152 space, following the LPI convention; positive t-values indicate that the intervention group increased in positively correlated network connectivity from baseline to follow-up, relative to the control group; negative t-value indicates that the intervention group increased in anticorrelated network connectivity from baseline to follow-up, relative to the control group; all findings are thresholded at a clusterwise p < 0.001 following a Monte Carlo simulation (extent 3 contiguous voxels), and all between-group differences are significant at p < 0.001.
Increases in positively correlated functional connectivity (FC) with the DMN following training. Peak voxel activity of positively correlated regions with the DMN are shown on the axial plane. Images are overlaid on an averaged, whole-sample template brain standardized in MNI152 space, lateral-posterior-inferior (LPI) oriented, and represented in neurological view (left = left).
Significant interactions for increased anticorrelated FC (Table 5, Fig. 4) were observed between the DMN and regions of other networks, including the DAN (left IPL – left middle temporal gyrus, left cerebellum lobule VIII, and right cerebellum lobule VI), SMN (right IPL – left superior temporal gyrus and right superior temporal pole; left IPL – right superior temporal pole, right superior temporal gyrus), FPN (right IPL – right inferior frontal gyrus; left IPL – left inferior frontal gyrus and right cerebellum lobule IX), SN (PCC – right pre-supplementary motor area and right dorsal anterior cingulate cortex), visual network (left IPL – right fusiform gyrus), CEN (PCC – right medial frontal gyrus), language network (left IPL – left inferior temporal gyrus), and basal ganglia network (right IPL – right putamen).
Increases in anticorrelated functional connectivity (FC) with the DMN following training. Peak voxel activity of anticorrelated regions with the DMN are shown on the axial plane. Images are overlaid on an averaged, whole-sample template brain standardized in MNI152 space, lateral-posterior-inferior (LPI) oriented, and represented in neurological view (left = left).
Semipartial correlations revealed several meaningful relationships between training-related improvements in cognitive function and FC with the DMN within the intervention group, which are demonstrated by moderate-to-large effect sizes, suggesting that a meaningful proportion of variance in observed changes in cognitive function is accounted for by changes in FC with the DMN (Table 6, Fig. 5). In the intervention group, improved TMTA performance was associated with increased anticorrelated FC (more negative) of the PCC – right dorsal anterior cingulate gyrus and of the right IPL – right putamen and right inferior frontal gyrus. Improved WJUD performance was associated with increased positively correlated FC of the left IPL – right supramarginal gyrus, right middle frontal gyrus, left lingual gyrus, and right superior frontal gyrus and with increased anticorrelated FC of the left IPL – right superior temporal pole and right superior temporal gyrus. Decreased CSR-C score (suggesting subjective improvements in cognitive function) over time was associated with increased positively correlated FC of the mPFC – right postcentral gyrus and right precuneus. Finally, decreased CSR-T score (suggesting subjective improvements in overall function) over time was associated with increased positively correlated FC of the PCC – right precuneus and mPFC – right postcentral gyrus as well as increased anticorrelated FC of the PCC – right pre-supplementary motor area and right dorsal anterior cingulate gyrus and of the right IPL – right inferior frontal gyrus.
Semipartial correlations between changes in cognitive function and default mode network functional connectivity in the intervention group
Semipartial correlations between changes in cognitive function and default mode network functional connectivity in the intervention group
Relationships with moderate-to-large effect sizes are indicated by bold text. Shared variance between scanner model and regions of increased functional connectivity with the default mode network is controlled for in all correlational models. Higher scores on TMTA and WJUD and lower scores on CSR-C and CSR-T are indicative of improved function over time. TMTA = Trail Making Test part A; WJUD = Woodcock-Johnson Tests of Cognition Understanding Directions subtest; CSR-C = Cognitive Self Report Questionnaire, Cognitive subscale; CSR-T = Cognitive Self-Report Questionnaire, Total score; PCC = posterior cingulate cortex; mPFC = medial prefrontal cortex; IPL = inferior parietal lobe.
Semipartial correlations between changes in cognitive function and default mode network (DMN) functional connectivity (FC) in the intervention group. The top panel (A) demonstrates associations between objective and subjective improvements in cognitive function and increases in positively correlated FC with the DMN following training. The bottom panel (B) demonstrates associations between objective and subjective improvements in cognitive function and increases in anticorrelated FC with the DMN following training. Only relationships with moderate-to-large effect sizes are shown and shared variance between DMN FC and scanner model is controlled for in all correlational models. IPL = intraparietal lobe. SMG = supramarginal gyrus. MFG = middle frontal gyrus. LingG = lingual gyrus. SFG = superior frontal gyrus. WJUD = Woodcock-Johnson III Tests of Cognition, Understanding Directions subtest. mPFC = medial prefrontal cortex. PostCG = postcentral gyrus. CSR-C = Cognitive Self Report Questionnaire, Cognitive subscale. PCC = posterior cingulate cortex. CSR-T = Cognitive Self-Report Questionnaire, Total score. dACC = dorsal anterior cingulate cortex. TMTA = Trail Making Test part A. IFG = inferior frontal gyrus. STP = superior temporal pole. STG = superior temporal gyrus. Pre-SMA = pre-supplementary motor area.
In addition to these relationships, several semipartial correlations demonstrated moderate-to-large effect sizes in the direction opposite to that which was expected in the intervention group (Table 6). Improved TMTA performance was associated with decreased positively correlated FC of the right IPL – left angular gyrus and right inferior temporal gyrus and of the left IPL – right middle frontal gyrus, left lingual gyrus, right middle temporal gyrus, and right superior frontal gyrus. Improved TMTA performance was also associated with decreased anticorrelated FC (more positive) of the left IPL – right fusiform gyrus, left cerebellum lobule VIII, left inferior temporal gyrus, and right superior temporal gyrus. Improved WJUD performance was associated with decreased positively correlated FC of the mPFC – right postcentral gyrus and right precuneus and decreased anticorrelated FC of the right IPL – right superior temporal pole and of the left IPL – right fusiform gyrus and left middle temporal gyrus. Fewer symptoms endorsed on the CSR-C was associated with decreased positively correlated FC of the left IPL – right supramarginal gyrus and right superior frontal gyrus, and decreased anticorrelated FC of the right IPL – right superior temporal pole and of the left IPL – left cerebellum lobule VIII, left inferior temporal gyrus, and right cerebellum lobule IX. Finally, fewer symptoms endorsed on the CSR-T was associated with decreased positively correlated FC of the left IPL – right superior frontal gyrus as well as decreased anticorrelated FC of the PCC – right medial frontal gyrus and of the left IPL – left inferior temporal gyrus and right superior temporal gyrus.
Cognitive dysfunction is commonly reported by those with TBI and is likely the result of TAI-induced reductions to the integrity of the structural networks from which FC emerges (Honey et al., 2009; Honey et al., 2010). Studies assessing changes in FC after TBI demonstrate alterations in resting-state activity at both acute and chronic stages of recovery (Hillary et al., 2011; Stevens et al., 2012). These alterations appear to result from the initial damage to the structural integrity of the network’s connections, followed by a plasticity-based compensatory response to the damage, resulting in hypersynchronous activity between intrinsically anticorrelated regions. This inherent response to injury may also reflect the brain’s capacity for network reorganization that may be facilitated by neurorehabilitation (Castellanos et al., 2011; Honey et al., 2010; Merzenich et al., 2014). To that end, the present study aimed to measure changes in DMN FC following consistent participation in plasticity-based auditory processing speed training and to relate these findings to changes in cognitive functioning.
Similar to previous reports of abnormal connectivity in TBI (Buckner et al., 2008; Fransson & Marrelec, 2008), our sample demonstrated abnormal FC at baseline relative to that typically seen in healthy adults (Uddin et al., 2009). Overall, the DMN showed relatively poor intrinsic FC with other regions of the DMN in our TBI sample. Similar findings have been demonstrated in healthy aging and dementia (Andrews-Hanna et al., 2007; Damoiseaux et al., 2008), schizophrenia (Zhou et al., 2007), and autism (Kennedy & Courchesne, 2008), where decreased or abnormal intrinsic FC of the DMN was related to dysfunctional cognitive, emotional, and/or introspective processes, which are common consequences of TBI. Furthermore, the DMN demonstrated positive FC with several TPN regions.
Following training, the intervention group demonstrated increased FC within DMN regions, between the DMN and the FPN, an intrinsic network that has been shown to have positive FC with the DMN and various TPNs and is suggested to be involved in the active manipulation and maintenance of internal information (Doucet et al., 2011). Contrary to expectations, positive FC was also observed between the DMN and TPNs following training, including the SMN, DAN, and the visual network. The abnormal synchrony between the DMN and TPN may reflect a similar phenomenon as that which takes place during the acute stages of TBI, where hyperconnectivity is seen as an adaptive response that reflects a trade-off between metabolic cost and communication efficiency while networks undergo reorganization (Hillary & Grafman, 2017; Hillary et al., 2011; Roy et al., 2017). Additionally, we found increased positive FC between the right IPL of the DMN and the right inferior temporal gyrus, which has historically been considered to be part of the task-positive language network; however, recent studies suggest that the inferior temporal gyrus may play a role in internal discourse and other intrinsic functions that involve some level of language processing (Muller & Meyer, 2014). Following training, the intervention group also demonstrated increased anticorrelated FC with core DMN regions and the FPN, and most importantly, with various TPNs, including the SN, SMN, DAN, CEN, basal ganglia, visual, and language networks.
Several of the observed training-related changes in FC with the DMN were also found to be associated with training-related cognitive improvements in processing speed, attention, and self-reported cognitive and overall functioning (Fig. 5). Changes in FC and cognitive performance have previously been demonstrated in various clinical populations such as multiple sclerosis (Fuchs et al., 2020), dementia (Ochmann et al., 2017), and schizophrenia (Penadés et al., 2019). Furthermore, previous studies have shown that FC moderates cognitive performance (Han et al., 2020; Konstantinou et al., 2018; Penadés et al., 2019), which is also seen presently, suggesting neuronal efficiency post-cognitive remediation.
The present study is limited by the small, unequal sample sizes of the intervention and control groups, and the lack of a healthy comparison group. A future study with a larger sample size would allow for greater power as well as an investigation of the differential effects of demographic and injury factors. Despite these limitations, our results have demonstrated that it may be possible to facilitate changes in the FC of RSNs through auditory processing speed training in adults with chronic TBI. These changes may further be the result of training-induced increases in the structural integrity of the networks through Hebbian strengthening models of neuroplasticity (Hebb, 1949; Honey et al., 2010; Merzenich et al., 2014). Long-term changes in neural plasticity, measured through EEG recordings and rs-fMRI, following auditory processing speed training using Posit Science’s BFP have been demonstrated in previous studies of older adults (Anderson et al., 2014; Styliadis et al., 2015), but the present study is the first to do so in chronic TBI.
Conclusion
The present study represents an important extension of the previous literature within the realms of both TBI and cognitive rehabilitation. Overall, our results suggest that plasticity-based auditory processing speed training may improve recovery from TBI in adults in a chronic stage by facilitating changes in resting-state FC with the DMN. Further research is warranted to investigate the relationship between training-related changes in FC and cognitive functioning. Nevertheless, our findings demonstrate that the damaged brain is plastic and responsive to plasticity-based cognitive rehabilitation, thereby providing preliminary evidence for the effectiveness of neurorehabilitation strategies in influencing recovery from brain injury.
Footnotes
Acknowledgments
This study was supported by the New York University Research Challenge Fund (R8740). The authors would also like to express their gratitude to all participants that made this study possible.
Conflict of interest
The authors declare no financial disclosures or potential conflicts of interest regarding the research, authorship, and/or publication of this article.