Time-dependent differences in cortical measures and their associations with behavioral measures following mild traumatic brain injury

2.1 Participants

A total of 87 adults, recruited from the general population within the greater metropolitan area of Boston, MA and New England, participated in this study. Thirty-three participants were included as healthy controls (HCs, mean age = 24.52 ± 3.0 years, 19 female) and 54 participants with a recent mTBI were included in the mTBI group (mean age = 22.40 ± 4.6 years, 33 female, time since injury between 0 and 18 months, mean = 5.73 ± 3.9 months). Any participant from the HC group or mTBI group with any history of drug or alcohol abuse or current use of illicit substances was excluded. Current alcohol use was required to be lower than the Center for Disease Control criteria for excessive alcohol use (www.cic.gov/alcohol). All HCs were recruited as part of a separate study (although no data from these subjects regarding cortical thickness, volume, or surface area have been published previously) but with the same scanning parameters and on the same scanner as the mTBI group. Neuropsychological testing was completed at the Social Cognitive and Affective Neuroscience laboratory located at McLean Hospital. All participants underwent high-resolution anatomical brain imaging using a Siemens Tim Trio 3T scanner (Erlangen, Germany) located at the McLean Hospital Imaging Center.

2.2 Data acquisition

2.3 Data analysis

2.3.1 Identification of affected brain areas following mTBI

The “recon-all” pipeline in FreeSurfer (version 6.0) (https://surfer.nmr.mgh.harvard.edu/fswiki) was used to process anatomical images for all the participants (HCs and individuals with mTBI). Processing involved motion-correction, brain extraction (i.e., removal of skull, skin, neck, and eye-balls), automated transformation to the Talairach co-ordinate system, intensity correction, volumetric segmentation, and smoothing using a 15 mm full-width at half-maximum (FWHM) Gaussian kernel. For each HC and mTBI participant, we visually inspected raw T1-weighted image data to determine any possible imaging artifacts, which could affect FreeSurfer's segmentation accuracy. Accuracy of the FreeSurfer generated skull-stripped brain masks and brain surfaces (pial and white) were visually inspected for all the participants from the HC and mTBI groups. The measures of CT, CV, and CSA were calculated separately for the left and the right hemispheres for each participant. CT is defined as the mean distance from the white–grey matter interface to the nearest point on the pial surface (grey matter–CSF interface) and from that point on the pial surface back to grey/white matter interface (Fischl and Dale, 2000), CV is defined as the amount of grey matter that lies between the white–grey matter interface and pial matter (Winkler et al., 2010), and CSA is the sum of the areas of the triangles making up the surface model and is defined as the extent of the two-dimensional surface enclosed by the outer layer of the cerebral cortex (http://cna.hanyang.ac.kr/research/research02.htm) (Fischl, Sereno, & Dale, 1999) (Figure 1). For vertex-by-vertex general linear model (GLM) estimation across the left and the right cortical surface, CT, CV, and CSA were used as individual dependent variables. This method was used to generate statistical parametric maps to identify the brain areas, which showed significantly different CT, CV, or CSA in those with a mTBI (TP1, TP2, or TP3) compared to HCs and within three TPs (TP1 versus TP2, TP1 versus TP3, and TP2 versus TP3). These statistical maps display the distribution of p values. Effects of “age” (demeaned) and “gender” were regressed out when performing group analyses. As the group-wise sample size in our study is small and the differences in brain structure between HCs and mTBI groups and within the mTBI groups are not expected to be localized finely, we selected a moderately larger smoothing kernel size of 15 mm. Moreover, unlike volume-based analysis, larger smoothing kernel size in surface-based analysis never extends into bone/air/white matter. Furthermore, we used a cluster forming threshold (CFT) of p < 0.01. Multiple comparisons were corrected at a clusterwise statistical threshold (CWP) of p < 0.05 using Monte-Carlo simulations.

3.1 Structural measures for HCs versus three mTBI groups

CT: Compared to HCs, 2 clusters—the right insula and the right superior temporal gyrus (STG)—in the mTBI group at TP1 showed significantly greater CT (Figure 2a–d). Compared to HCs, 4 clusters—the left rostral middle frontal gyrus (RMFG), the left supramarginal gyrus (SMG), the right lateral orbitofrontal cortex (LOFC), and the right postcentral gyrus (PostCG)—in the mTBI group at TP2 showed significantly greater CT (Figure 2e–h). These findings are summarized in Table 2. There were no significant difference in CT between HCs and mTBI group at TP3. Within the three TPs also, we did not find significant differences in CT, that is, for TP1 versus TP2, for TP2 versus TP3, or for TP1 versus TP3.

CV: We did not find significant differences in CV when compared between HCs and any of the three mTBI groups and within three mTBI groups.

CSA: Compared to HCs, 3 clusters—the right PostCG, the right inferior temporal cortex, and the right superior frontal cortex—in the mTBI group at TP2 showed significantly lower CSA (Figure 3a–d). There were no significant difference in CSA between HCs and mTBI groups at TP1 or TP3. These findings are summarized in Table 3. Within the three-mTBI groups, 3 clusters—the left STC, the left PostCG, and the right isthmus of cingulate gyrus—in the mTBI group at TP3 showed significantly higher CSA compared to TP2 (Figure 3e–h). These findings are summarized in Table 4.

3.2 Correlation analysis between structural measures, RBANS ATT, and sleep measures

For two ROIs—the right postcentral gyrus (R. PostCG) and the left rostral middle frontal gyrus (L. RMFG)—there were negative correlations between CT and the RBANS ATT index within TP2 (R. PostCG: r = −.44, p = .03 and L. RMFG: r = −.41, p = .05) (Figure 4a,b). However, the mTBI groups were not significantly different on RBANS ATT (F(2,53) = 1.47, p = .24, one-way ANOVA), ESS (F(2,53) = 0.41, = 0.66, one-way ANOVA) or PSQI (F(2,53) = 0.66, p = .52, one-way ANOVA) (Table 1).

4.1 Time-dependent cortical differences following mTBI

CT: In general, previous studies on CT following an mTBI reported thinning and thickening of the cortex (Govindarajan et al., 2016; Wang et al., 2015). It has been suggested that cortical differences might depend on several factors, including the time since injury, symptom severity, regional microedema, localized microhemorrhages, and cytotoxic edema (Lewen, Fredriksson, Li, Olsson, & Hillered, 1999; Wang et al., 2015). It was also suggested that due to subsequent cortical thinning after several weeks, differences in cortical thickness were undetectable at later time points (Govindarajan et al., 2016; Lewen et al., 1999; Tate et al., 2014). However, in this study, compared to HCs, we reported thickening within the right insula and the right STG among mTBI individuals who were between 0 and 3 months of injury and within the left RMFG, left SMG, right LOFC, and right PostCG among mTBI individuals who were between 3 and 6 months of injury.

Recently, brain regions including the insula, STC, and PostCG have been shown to display greater neural activation among individuals with mTBI relative to controls (Dretsch et al., 2017). In that study, it was proposed that several psychological health symptoms such as depression and attentional bias toward negatively valenced stimuli could be responsible for the neural hyperacivation within several regions of interest in the mTBI group. However, the validity of similar mechanisms resulting in cortical thickening within these regions following an mTBI still needs to be confirmed. In a separate study, higher numbers of mTBIs were also associated with reduced CT within the bilateral insula and right middle temporal gyrus (List, Ott, Bukowski, Lindenberg, & Floel, 2015). In that study, it was hypothesized that recurrent mTBIs may induce distinct alterations, especially thinning of the cortex. Consistent with our findings, it was proposed that cortical alterations from the acute phase following an mTBI may normalize in the chronic phase. Moreover, cortical thickening within the right RMFG was reported immediately following an mTBI (Wang et al., 2015). At 3 months post-mTBI, no more cortical thinning was observed in the supramarginal gyrus (Govindarajan et al., 2016). However, we observed thickening of the supramarginal gyrus at 3 months post-mTBI. During the first year after mTBI, changes in CT indicated thickening of the prefrontal cortex, including orbitofrontal cortex in mTBI patients (Dall'Acqua et al., 2017; Wilde et al., 2012). Cortical thickening during initial scans following an mTBI and cortical thinning in later scans may reflect progressive normalization of CT, that is, physical recovery from brain lesions (Lewen et al., 1999; Wang et al., 2015). In addition, the brain areas, such as RMFG, which are more susceptible to direct impacts following a frontal–rear axis head injury, may result in the release of excitotoxins from damaged tissues causing inflammatory reactions, including microedema (Barkhoudarian, Hovda, & Giza, 2011; Lillie, Urban, Lynch, Whitlow, & Stitzel, 2013; Patterson and Holahan, 2012; Urban et al., 2012). These inflammatory reactions have been reported to elevate fractional anisotropy, thicken the cortical regions initially but cause cortical thinning over time with the reduction of microedema (Lewen et al., 1999; Ling et al., 2013).

CV: CV is a composite of both CT and CSA, therefore, changes in CV could be due to changes in either CT or CSA, or both. Therefore, significant increase in CT and significant reduction in CSA or vice versa could be responsible for an unknown CV proportionality across the cortex or even the absence of differences in CV in the three mTBI groups, as observed in our study. Previously in a study on gene identification, it was reported that measures of grey matter volume are less sensitive than CT or CSA, where CT and CSA are also distinct from genetic origins (Winkler et al., 2010). In that study, there was no clear interpretation made from regional grey matter volume differences in terms of genetic influences. In the same study, it was also reported that since the variability in CSA was higher compared to CT, variability on CV might therefore be more associated with CSA as compared to CT. Our findings are partially consistent with these mechanisms as we also found more variability in CT measures as compared to CV and CSA. We acknowledge the fact that the preceding analogy is not ideal and is made between mTBI groups and a gene identification study but the geometrical relationships between these three cortical measures (CT, CV, and CSA) and relatively more dependence of CV on CSA compared to CT may partially explain the underlying mechanisms behind our findings.

CSA: We observed greater CSA at the later stages of mTBI (i.e., 6 and 18 months post-mTBI). Specifically, we found that there were many regions with significantly lower CSA at TP2 compared to HCs but greater CSA at TP3 compared to TP2. The observed differences in CSA contrast with prior findings, as decreases in CSA were previously reported to be one of the earlier existing and sensitive biomarkers for the quantification of brain damage following mTBI (Dall'Acqua et al., 2016). However, larger CSA was shown by others to be associated with complex brain interactions and better cognitive skills (Raznahan et al., 2011; Schnack et al., 2015). In humans, a larger proportion of CSA due to larger surface convolutions is attributed to an extended and dynamic network of brain projections (Hofman, 2014). This generation of an extended dynamic network may not be quick and immediate but might instead be a slow process, which could be the backbone for brain plasticity resulting in compensation of behavioral skills following an injury. Significant increases in CSA, regardless of increases in CT, have also been associated with an increase in radial column units during expansion of the neocortex in primate evolution (Rakic, 2009). Integration of these neocortical columns at higher levels of information processing sets the neural basis of multiple brain regions and their unique features to interact dynamically, which could result in greater synaptic plasticity (Budd and Kisvarday, 2012; Hofman, 2014). Moreover, at the chronic stage of mTBI (i.e., 6–18 months post-mTBI), increases in CV rather than CT and CSA individually, which can account for changes in both CT and CSA, could be an indication of an increase in the formation of dendrites resulting in modest remodeling of the cortex over time (Killgore et al., 2016). These improvements in functional and structural abnormalities could also be closely associated with beneficial neural reorganization of the affected brain hemisphere. Experience-based changes in brain structure over time, also known as experience-dependent neural plasticity, were also found to be beneficial for reducing behavioral and physical disorders (Kerr, Cheng, & Jones, 2011).

In sum, the differences in multiple structural measures following an mTBI might indicate that various brain systems change at different rates. Previous brain imaging studies on mild, moderate, and severe TBI showed that although TBI patients performed equally well as HCs, they recruited a larger number of brain areas, including the frontal and posterior cortices (da Costa et al., 2015; Turner and Levine, 2008). Larger recruitment of these areas during later stages of mTBI could be due to reduced involvement of damaged brain areas immediately after an mTBI or greater compensatory recruitment in more chronic stages. Diffusion imaging studies on TBI have also reported microstructural white-matter alterations that differ at various stages following injury, such as axonal swelling and/or an increase in glial cells (Pasternak et al., 2014), causing variations in CT, CV, or CSA across the recovery period following mTBI.

4.2 Attention and sleep measures following mTBI

We observed that between 3 and 6 months post-mTBI, abnormally higher cortical thickening within the left RMFG and right PostCG compared to HCs, was negatively associated with performance on measures of attention. Previous work has shown that both RMFG and PostCG are reliably associated with attention capacities. For instance, it was reported that the posterior region within the rostral middle frontal cortex is activated by various cognitive tasks, including the ones designed to engage in internal monitoring of action, error and attention (Amodio and Frith, 2006). A positive correlation between thickness within the left posterior middle frontal cortex and performance on a dichotic listening task (a measure of executive attention) further suggested that the left middle frontal cortex is part of an executive attention network (Andersson, Ystad, Lundervold, & Lundervold, 2009). Furthermore, the PostCG or somatosensory cortex, which is the most anterior portion of the parietal lobe, is also one of the three major major sites (intraparietal, postcentral, and precentral) of activation for attention (Corbetta, 1998). A study on a group of right hemisphere stroke patients also suggested a vital role of the PostCG/somatosensory cortex in visuospatial attention (Balslev, Odoj, & Karnath, 2013). Thus, it is clear that the RMFG and PostCG play an important role in attention.

Interestingly, in this study, we did not observe significant differences in attention or sleep measures between any of the three mTBI groups. There was no significant association observed between increased CSA and improved attention abilities or sleep quality. One possible explanation may involve the construct of “cognitive reserve,” or the ability to maintain cognitive functioning in the presence of brain damage or degenerative process (Stern, 2009, 2012). Previously, it was found that increased cognitive reserve might play a protective role against obstructive sleep apnea syndrome (OSAS)-related cognitive decline, including intelligence and attention (Alchanatis et al., 2005). Given the fact that our data did not include specific cognitive measures relevant to a range of sleep disorders, it is beyond the scope of our study to directly confirm that cognitive reserve played a role in the nonsignificant differences in attention and sleep measures across the three time-points. Regarding the sleep measures we used, another possibility is that some of the specific features of sleep biology are not well captured by self-reported measures (Lim et al., 2016). Future research in these areas is therefore needed to investigate these intriguing possibilities further.

4.3 What are the benefits of using multiple structural measures?

In this study, we report the time-line of differences in multiple cortical measures, especially CT and CSA, following mTBI. In particular, we report that when CT was significantly different following mTBI, there were no differences observed in CSA, whereas when CT did not differ across time-points, CSA appeared to be higher, which could be due to greater cortical folding. Human brain development is associated with increased cortical folding, which leads to a progressively more convoluted brain structure and gyrification along spatial and temporal scales (Armstrong, Schleicher, Omran, Curtis, & Zilles, 1995; Richman, Stewart, Hutchinson, & Caviness, 1975). Compared to other species, the folds in the human brain are unique and are associated with specific behavioral skills (Gautam, Anstey, Wen, Sachdev, & Cherbuin, 2015; Gregory et al., 2016). Approximately one-third of the brain's cortical surface is visible, whereas two-thirds of the surface is hidden from view among its folds, leading to overall greater CSA and extra space for the accommodation of additional neurons (Toro, 2012). Cortical folding also shortens the distance of cortical connections by reducing the fiber length necessary between neural regions, resulting in reduced conduction delays across axons (Buzsaki, Logothetis, & Singer, 2013; Chklovskii, Mel, & Svoboda, 2004). Mathematically, there is an interdependent relationship between cortical folding and structural cortical measures. More specifically, it is suggested that the amount of cortical folding increases as CSA increases, where CT becomes an important factor to consider (Mota and Herculano-Houzel, 2015). These relationships suggest that more brain folds lead to more CSA, and thicker cortex could be responsible for restricted brain folds, and both, that is, brain folds or CSA and CT might have unique contribution toward stronger behavioral responses. Therefore, it becomes crucial to consider multiple cortical measures to better understand the time-dependent differences in brain structure following an mTBI or in general.