Neuropsychological assessment

The compilation of clinical questionnaires and validated cognitive tests is summarized in Supplementary Table 1. Early post-concussive symptoms were assessed using the well-established Rivermead Post Concussion Symptoms Questionnaire (RPQ) (King et al., 1995). A German in-house translation of the RPQ was produced. This symptom checklist was selected as a core measure based on its capacity to detect somatic, cognitive, and emotional changes in patients with mTBI. The RPQ investigates the severity of 16 commonly reported post-traumatic symptoms, grading each item from 0 (not experienced) to 4 (severe problem). Patients rated perceived changes in performance occurred over the past 24 h, and HC were asked to rate the current severity of each symptom in comparison with their usual levels of functioning. The total RPQ score was calculated by summing all 16 items with a score of 2 or greater, as a score of 1 indicates that a symptom is not more problematic than before the trauma. In addition, two scales were used to assess emotional symptomatology: the German version (Hautzinger et al., 2006) of the Beck Depression Inventory 2nd edition BDI-II (Beck et al., 1996) was selected to control for manifestations of depression; and the German version (Margraf and Ehlers, 2007) of the Beck Anxiety Inventory BAI (Beck and Steer, 1993) was chosen to evaluate anxiety symptoms in response to mTBI. Further details regarding the neuropsychological test battery applied in the present study are provided in the Supplementary Materials online.

Magnetic resonance imaging data acquisition, preprocessing, and construction of the network

Data acquisition, preprocessing of DTI and SBM data and the construction of the structural networks are described in detail in the Supplementary Materials. Briefly, a 3 T scanner was used to acquire DTI, T1-weighted, and conventional MRI scans.

Presence of intracranial injury was determined by the evaluation of conventional MR scans (T1, T2, FLAIR, and SWI) and examined by the same radiologist (E.J.U.) not blinded to mTBI-diagnosis. In cases where any salient features were noted, the subjects were excluded from the sample.

Preprocessing of the diffusion-weighted MRI data is described in the Supplementary Materials and was performed with FSL tools (FMRIB software library; version 5.0.6; http://www.fmrib.ox.ac.uk/fsl/) (Smith et al., 2004) such as the FDT (FMRIB diffusion toolbox; version 3.0) (Behrens et al., 2003). For deterministic fiber tractography, we used the Diffusion Toolkit (DTK, version 0.6.2.1) and TrackVis software (version 0.5.2.1; http://trackvis.org/) (Park et al., 2009).

Cortical surface reconstruction is described in the Supplementary Materials and was performed with the FreeSurfer image analysis suite (version 5.3.0), which is documented and freely available online (http://surfer.nmr.mgh.harvard.edu/). The technical details of these procedures are described in prior publications (Dale et al., 1999; Fischl et al., 1999a,b, 2001, 2002, 2004a,b; Fischl and Dale, 2000). Reconstructed surface models were visually checked for accuracy by an experienced imaging analyst (J.H.) and no manual interventions were necessary.

Structural connectome

First, the structural connectivity of a single subnetwork was inversely related to the total RPQ score of the patients (mean r = −0.31, p = 0.045 corrected, primary threshold t = 1.80). The extent of this altered subnetwork comprised 59 inter-regional connections distributed over 50 nodes (Figure ​(Figure2A,2A, Supplementary video 1). The altered set of connections encompassed bilateral frontal, parietal, temporal, and subcortical regions such as hippocampus, thalamus and caudate nucleus. Connections going to or from the frontal cortex showed the highest concentration, i.e., 31 out of 59 connections (Supplementary Table 2). Further exploration of the dysconnected subnetwork revealed that the most affected connections were intra-hemispheric, with only 6 connections being inter-hemispheric. The nodes with the most connections related to severity of post-injury symptoms were located in the left insula (8 connections), left anterior cingulate (6 connections), left supplementary motor area (6 connections), and right insula (7 connections). Next, the same correlation between total RPQ and structural connectivity but based only on the 30 more symptomatic subjects was performed. Similar as for the entire patient group, the association was negative, but did not reach statistical significance (p = 0.180).

Open in a separate window

Figure 2

Inverse correlation between structural connectivity and Rivermead Post-Concussion Symptoms Questionnaire scores of the patients and group comparison. The 3D visualization of the network (Xia et al., 2013) shows a lateral section of the left and the right hemispheres (upper panels), horizontal and coronal sections (lower left: superior perspective, lower right: anterior perspective). The violet-colored dots correspond to the 90 cortical and sub-cortical automated anatomical labeling regions. The violet lines represent the supra-threshold connections of the subnetwork. The thickness of the lines represents the absolute connectivity strength of each connection. (A) The connectivity strength of one subnetwork is inversely correlated with the total RPQ score of the patients (n = 51, mean r = −0.31, p = 0.045 corrected). (B) Reduced structural connectivity in the subgroup of 30 mTBI patients compared to 30 matched controls (Cohen's d = 0.95, p = 0.027).

Finally, no correlation between connectivity strength and total RPQ score was observed in the HC group (n = 53), although some HC also indicated post-concussion-like complaints (Figure ​(Figure11).

Second, the results of the group comparison revealed no significant differences when considering the entire sample (n = 51 patients/53 HC, p = 0.626), whereas investigation of the more severe subgroup revealed a single dysconnected subnetwork with reduced connectivity strength in the 30 patients compared to their 30 HC (Cohen's d = 0.95, p = 0.027, Figure ​Figure2B,2B, Supplementary Table 3). The dysconnected subnetwork was composed of 57 connections distributed over 48 nodes; the primary threshold was set to t = 1.81. A closer examination of the subnetwork pattern revealed the presence of 11 connections identical to those resulting in the correlation analysis. No subnetwork showed greater structural connectivity in patients than HC.

Surface-based morphometry

First, correlation analysis conducted within the mTBI group revealed an inverse association between total RPQ score and cortical surface area in five clusters. Bilateral clusters in the frontal and parietal cortices and one cluster in the right inferior temporal gyrus (ITG) were found to be negatively correlated with early mTBI-related symptoms, showing that patients reporting more severe symptoms had lower cortical surface area (Figure ​(Figure3,3, Table ​Table2,2, Supplementary Video 2A for the left and Supplementary Video 2B for the right hemisphere). In the left hemisphere, a corresponding temporal cluster with reduced surface area was also detected, which did not, however, reach statistical significance after correction for multiple comparisons (r = −0.11, df = 48, cluster-wise corrected p-value, CWP = 0.22). The largest cluster was identified in the frontal region of the left hemisphere. The clusters with the strongest associations between RPQ and area were located bilaterally in the frontal cortex and in the right frontoparietal cortex. The RPQ score did not have any association with cortical thickness, but correlation with cortical volume revealed the same clusters as in the analysis of cortical surface area (with the exception of one cluster in the right precuneus instead of the right frontal cortex). No significant correlation between RPQ score and any of the surface-based measures was found in the HC. Driven by the NBS finding, post hoc ROI analyses focusing on the insula were performed, revealing negative correlations with symptoms level in both cortical surface areas (left, r = −0.270, p = 0.029; right, r = −0.212, p = 0.070; uncorrected). Similar correlations were found with cortical volume.

Open in a separate window

Figure 3

Inverse correlation between cortical surface area and Rivermead Post-Concussion Symptoms Questionnaire scores in the patient group (n = 51). Panels of the top row show the lateral hemispheres (A = left, B = right), whereas the second row shows the medial hemispheres. The third and bottom rows represent the superior and inferior views, respectively. Bilateral frontal clusters (yellow) encompass the lateral and medial prefrontal cortex, the orbitofrontal cortex and the anterior cingulate cortex. The parietal cluster (blue) in the left hemisphere includes the postcentral gyrus (PoCG) together with parts of the central sulcus, while the frontoparietal cluster (blue) in the right hemisphere stretched over the PoCG and the precentral gyrus. The red cluster in the right hemisphere comprises the inferior temporal gyrus. Only clusters exceeding a cluster-wise corrected probability of p < 0.05 are shown.

Next, cross-sectional whole-brain comparison of area, thickness and volume revealed no significant difference between traumatic group and HC (p > 0.15). Post-hoc we then compared the mean values between the groups within ROIs, defined as the clusters resulting from the correlation with the RPQ across both groups. To limit the number of comparisons, group analyses focused on the six surface area clusters that exhibited significant negative associations with total RPQ score across both groups (Supplementary Figure 1, Supplementary Table 4). When considering the entire sample, significant group differences were identified in the left parietal (p = 0.009) and left temporal cluster (p = 0.014; Supplementary Figure 2). The 30 patients suffering from more severe symptoms showed significant reductions in surface area in five of the six clusters compared to the 30 HC (Figure ​(Figure44).

Open in a separate window

Figure 4

Group differences in cortical surface area between the subgroup of 30 patients suffering from more severe symptoms and the 30 matched HC. The same colors used in Figure 3 were assigned to the cortical clusters (the right precuneus is newly shown in green). Depending on the cluster, the percentage of surface area reduction in the patient group varied between 3.9 and 9.1%. [Error bar = ± standard error of mean SEM; the strength of the group difference (i.e., effect size) was assessed using Cohen's d score].

Convergence of the white and gray matter findings

Anatomically consistent structural changes were observed in bilateral frontal brain regions, i.e., in lateral (lPFC) and medial prefrontal cortices (mPFC), orbitofrontal cortex (OFC), and anterior cingulate cortex (ACC) as well as in the pre- (PrCG) and postcentral gyrus (PoCG) of the right hemisphere (Figures ​(Figures55,1A–F). In the mTBI group, positive correlations (p < 0.001) between total number of reconstructed streamlines within the dysconnected subnetwork (59 edges) and surface area of the five clusters (mean surface area per vertex in mm²) have been found in the left frontal (r = 0.532), left parietal (r = 0.519), right frontal (r = 0.448), right frontoparietal (r = 0.673), and in the right temporal cluster (r = 0.439). Area reduction in the right frontoparietal and left frontal cluster explained up to 45 and 28%, respectively, of the variance in connectivity strength of the dysconnected subnetwork (Figures ​(Figures55,2A,B).

Open in a separate window

Figure 5

Anatomical overlap of white and gray matter changes in the patients. The 3D visualization of the network (Xia et al., 2013) includes the lateral view (1A = left, 1B = right), the medial view (1C,1D), and the horizontal superior (1E) and inferior (1F) views. The nodes in yellow and blue represent the brain regions of patients in whom changes were observed in both the surface- and connectivity-based analyses. The scatterplots illustrate the best correlations between surface area values and the sum of the streamlines within the 59-edge network (2A,B).

Network-based connectivity

Decreased structural connectivity within one subnetwork was significantly associated with an increase in self-reported complaints in the mTBI group. Descriptively, most of these edges (53 of 59) were intrahemispheric and involved both short and long connections. Although the dysconnected fibers appeared widespread when inspected by the naked eye, the involvement of frontal regions was highly pronounced (31 of 59 altered connections possessed a node in the frontal cortex). The insula also seems to play an essential role in the altered circuit, since the strongest connections related to self-reported complaints were identified in the insula. The close links of the frontal cortex and insula to symptom severity were found bilaterally and will be discussed later.

To date, only one graph-theoretical study using DTI has investigated connectivity in mTBI, although in children (Yuan et al., 2015). In nice agreement with our findings, the study demonstrated a correlation between symptom severity and nodal degree of the superior and middle frontal gyrus in the mTBI group.

Subsequently, we performed whole-brain comparison over the entire sample that yielded no significant group differences. Conversely, a reduction of structural connectivity within a single subnetwork comprising by 57 connections was detected in the more severe patient subgroup compared to their matched HC. The focus on this subgroup may have enhanced the detection of decreased WM integrity, an effect that was attenuated when the 21 less impaired patients were included. The importance of classifying patients according to symptom severity has already been put forth in classical DTI studies of mTBI patients (Messé et al., 2011; Wäljas et al., 2014), and other studies have even indicated that the level of symptom severity is more critical than the experience of mTBI per se (Sterr et al., 2006; Ponsford et al., 2012). As reported above, only one study on children with acute mTBI has applied graph theoretical analysis to investigate structural connectivity. They also observed a significant reduction in network efficiency compared to controls. Groupwise differences using tract-based spatial statics (TBSS) were previous largely reported in multiple, widespread white matter regions where the classical diffusion metrics as FA and mean diffusivity were significantly different in acute-subacute mTBI patients compared to controls (Messé et al., 2011, 2012; Toth et al., 2013; Croall et al., 2014; Zhu et al., 2014). Studies combining DTI tractography with graph theory to investigate patients with moderate-to-severe TBI have reported altered network organization and lower efficiency in the chronic phase after injury (Caeyenberghs et al., 2012, 2014; Kim et al., 2014). Caeyenberghs and colleagues suggested the idea of TBI as a dysconnection syndrome from a network perspective. Moreover, since recently TBI is considered an example of damage in which large-scale network alterations produce clinical impairments (Sharp et al., 2014). The present findings may support the extension of this assumption to the mild end of TBI. To our knowledge, this is the first study that demonstrates direct evidence for a dysconnection syndrome in adults acutely following mTBI.

As frequently reported in the literature over the last years (Shenton et al., 2012), biomechanically induced DAI could play a major role in mTBI. Our results are compatible with DAI and indicate that axonal damage may be responsible for the reduced connectivity strength in large-scale structural subnetworks.

Surface-based morphometry

The surface-based morphometry analyses revealed that severity of complaints was inversely correlated with cortical surface area in mTBI patients. Area reductions associated with greater complaints were identified in five clusters located in left and right frontal cortices, left PoCG and right PoCG/PrCG, as well as in the right ITG. Similar results were detected for cortical volume, although we did not detect cortical thinning associated with clinical symptomatology. This implies that volume reduction was mainly attributable to area reduction, since volume is defined as the product of area and thickness.

Other mTBI studies have demonstrated a negative association between post-concussive symptoms and cortical morphology (Zhou et al., 2013; Albaugh et al., 2015). Anatomically strongly in line with our findings, these studies focused on the subacute-to-chronic phase following injury and only reported changes in volume or thickness. For example, a study of young athletes reported widespread cortical thinning in frontal and bilateral temporoparietal cortices associated with post-concussive symptoms (Albaugh et al., 2015).

Between-group analysis showed significant differences in two clusters (left parietal and left temporal), with patients displaying reduced area. When considering the subsample of 30 patients reporting moderate-to-severe symptoms and their 30 HC, five of six clusters demonstrated area reductions between the two groups. Hence, focusing on the subgroup of patients with more severe symptoms may have increased area diminutions compared with HC. Other studies have devoted attention to the initial post-injury periods, but found no group differences in volume or thickness (Ling et al., 2013; Toth et al., 2013; Narayana et al., 2015). As suggested from these examinations, measurable atrophy due to neuronal loss may occur later in the post-injury course of mTBI. In fact, as reported by Mayer et al. (2015), signs of cortical atrophy were demonstrated over a 4-month study interval in mTBI children compared to controls (Mayer et al., 2015). The location of the observed bilateral frontal cluster is in line with our results.

In addition, morphometric studies concentrating on chronic effects of mTBI have provided evidence for progressive morphological alterations and more accentuated damage of GM integrity (Tremblay et al., 2013; Zhou et al., 2013; Tate et al., 2014).

For instance, significant decreases in brain volume have been demonstrated after a single mTBI over the first 13 months in the anterior cingulate gyrus (bilateral), left cingulate gyrus isthmus, and right precuneus (Zhou et al., 2013). Briefly, all these studies detected thickness or volume loss long after injury, leading us to speculate that area reduction might be specific to the acute period and that changes in cortical thickness might be observed later in the course of mTBI. As pathological interpretation of area reduction remains unclear, we can only conjecture possible explanations for our results. Tissue deformation in the form of area contraction could be a direct result of different forces acting on the brain tissue including angular acceleration (Feng et al., 2010; Hansen et al., 2013). In conclusion, a simultaneous but separate evaluation of cortical surface area, thickness, and volume is important to provide more comprehensive information about the underlying pathogenic changes following mTBI. To date, we are not aware of any morphometric mTBI studies that have considered the examination of area as a distinct cortical feature.

Convergence of the white and gray matter findings

Patients reporting higher levels of complaints within the first week following mTBI showed a concomitant reduction of structural connectivity and cortical surface area. No significant relationships between symptom severity and structural alterations could be found in the HC, suggesting that WM and GM anomalies observed in the patients reflect injury-specific alterations. Moreover, the inverse correlations between both structural MRI measures and symptom severity together with the observed anatomical overlap clearly emphasize the clinical significance of the present findings. Given the availability of only one scan, it remains unclear whether reduced WM integrity is the trigger for area reductions seen in anatomically connected areas or vice versa. The findings reported for the frontal regions are in agreement with many mTBI studies regarding structural and functional alterations discussed in a recent meta-analysis (Eierud et al., 2014). Furthermore, the frontal cortex is among the brain regions identified as being more biomechanically vulnerable to impact forces (McAllister and Stein, 2010; Zappala et al., 2012; Wright et al., 2013). Temporal cortices, corpus callosum and subcortical structures are also regions susceptible to injury in mTBI (Viano et al., 2009; Bigler and Maxwell, 2012). Moreover, the occurrence of DAI is frequently located within the frontal lobes, as revealed by various epidemiological studies (Niogi et al., 2008; Chatelin et al., 2011). The structural WM/GM changes in the right PoCG are difficult to explain. Perhaps the primary somatosensory cortex is associated with headaches frequently reported by mTBI patients or with the experience of pain (Gosselin et al., 2012). Nevertheless, another study also observed an inverse correlation between cortical thickness in bilateral temporoparietal regions and post-concussive symptoms in the subacute post-injury phase (Albaugh et al., 2015). Another finding of the present study is the involvement of the insula, which showed decreased connectivity within the dysconnected subnetwork and decreased cortical surface area using the ROI approach. The insula is an important hub for monitoring, switching, attention, mental flexibility, error processing, cognitive control and is a key region involved in all subjective feelings from the body including pain (Menon et al., 2001; Sridharan et al., 2008; Craig, 2009; Ham et al., 2014; Wiebking et al., 2014). Since individuals with mTBI complain for example of poor concentration and greater irritability, damage to the insula's GM and/or its connectivity might have the potential to explain such conditions. Recent studies with resting-state fMRI, arterial spin labeling MRI and magnetoencephalography have also revealed a responsibility of the insula in early stage post-mTBI (da Costa et al., 2015; Meier et al., 2015; Zhan et al., 2015).

Correlations between structural brain alterations and cognitive performance

The extent to which WM/GM alterations linked to symptom severity is associated with executive-attentive functions was explored in the patient group. Only area reduction in the bilateral frontal clusters was related to poor performance in the Go/Nogo task, a paradigm commonly employed to study selective attention and response inhibition. Performance in the Go/Nogo task has been found to be particularly dependent on the functioning of the ACC (Botvinick et al., 2004), dorsolateral PFC (MacDonald et al., 2000), right inferior and superior frontal gyrus (Aron et al., 2004; Floden and Stuss, 2006), OFC (Eagle et al., 2008), striatum (Vink et al., 2005), and on frontostriatal connectivity (Liston et al., 2006). The majority of these structures has been shown to be altered in the present study.

We thank all study participants, the MRI acquisition staff of the Department of Psychology, Division Neuropsychology, University of Zurich (Carina Klein, Volker Baur), the neurology team (Gabriella Künig, Celia Kako, Klaus Bathke, Judith Sartorius) and the neuropsychological support staff (Denise Long, Karoline Wurmitzer) of Bellikon Rehabilitation Clinic. We also thank Ludwig Coray from SUVA for his management support and Christian Mondadori from Bellikon Rehabilitation Clinic for substantive comments with respect to the manuscript.