Diffuse axonal injury (DAI) was initially defined as widespread damage to axons throughout the white matter, evoked by intense shear and strain forces resulting from rapid acceleration and deceleration of the brain with or without impact after traumatic brain injury (TBI) [1, 2]. More recently, traumatic axonal injury (TAI) has been suggested as a more appropriate term for describing axonal damage because it encompasses not only the primary axonal damage specifically caused by shear/strain injury, but also secondary alterations of white matter such as metabolic, hypoxic and microvascular damage or excitotoxicity [3, 4].

Although TAI has been described in neuropathological terms, magnetic resonance imaging (MRI) allows the detection of microhemorrhages and other indirect signs in regions commonly affected by this injury such as the subcortical white matter, the corpus callosum and the dorsolateral quadrant of the rostral brain-stem. It has recently been demonstrated that T2*-weighted MRI at high field strength is a useful tool for the identification of traumatic microbleeds even in the chronic stage of TBI [5]. However, diffusion tensor imaging (DTI) has been suggested as the best technique for the detection of subtle white matter changes [6] given that it can reveal significant abnormalities in white matter in patients with normal findings in conventional MRI [7, 8].

DTI is a non-invasive MRI technique that identifies the microscopic physical properties of tissues directly through the observation of translational molecular movement of water [9]. Water diffusion in cerebral white matter tends to be anisotropic, because the highly linear organization of white matter fibers restricts movement in other directions [10, 11]. Fractional anisotropy (FA), one of the main DTI-derived indices, provides information of the degree of directionality of water diffusion and on microstructural white matter changes. DTI has been shown to be an efficient technique for determining white-matter integrity in several pathologies [12]. It has also been proposed as the most feasible biomarker of TAI and one of the best indicators of TBI severity [13, 14]. Reductions in FA have been detected not only in moderate and severe TBI patients [15–18] but also in cases of mild TBI [19–24]. Moreover, DTI has proved to be an excellent tool for evaluating structural changes after TBI in longitudinal studies [16–18].

The advantages of DTI have resulted in a growing body of scientific evidence regarding the relationship between white matter damage and neuropsychological deficits in TBI. Studies conducted with pediatric samples have identified correlations between FA values and various cognitive functions, including cognitive processing speed and interference, executive functioning, IQ, verbal working memory, reading comprehension and letter naming speed [25–27]. Recently, Wu et al., [28] reported correlations between immediate recall and left cingulum bundles in adolescents after mild TBI. Some studies have related FA measurements with neuropsychological deficits in adults. Nakayama et al. [29] identified a positive correlation between the Mini-Mental State Examination (MMSE) and FA in the splenium of the corpus callosum. Salmond et al. [30] found a significant correlation between diffusivity and the impairment of learning and memory in the posterior cingulate, hippocampal formation and cortical areas. Kraus et al. [31] in a sample including all grade severities, found reduced FA in the ROIs analyzed and obtained a measure of the total regions of reduced FA that negatively correlated with the three cognitive domains evaluated. Furthermore, in a mild TBI sample, Niogi et al. [32] found a significant correlation between attentional control and FA within a ROI in the corona radiata and between memory performance and FA in the ROI placed in the uncinate both in the group of mild TBI patients and the control group. Kumar et al. [18] found correlations between the corpus callosum and neuropsychological tests involving processing speed as well as visuospatial and visuperceptive tasks. Finally, Lipton et al., [33] and Miles et al. [34] found that reductions in FA in dorsolateral prefrontal cortex correlated significantly with tests of executive functions. In summary, DTI technique, in special FA measures, has been found sensitive to reflect cognitive deficits associate with TBI.

Memory is one of the functions that is most frequently impaired by TBI [35–37]. The concept of multiple memory systems and their different neuroanatomical substrates is currently accepted [38, 39]. Declarative and working memory systems are significantly impaired after traumatic brain injury (TBI). Deficits in declarative memory - the capacity for conscious recollection of facts and events - are a common consequence of head trauma that are disproportionately suffered in comparison with other cognitive functions [35, 40]. These memory difficulties improve slowly and although progress is made over the first and second year following injury, they remain apparent over time [40–42]. Neuroanatomically, declarative memory depends on the integrity of the hippocampus and its connections with the neocortex [43, 44]. In neuroimaging studies with TBI patients, declarative memory has been found to correlate negatively with hippocampal [45, 46] and fornix damage [47]. Working memory is defined as the ability to maintain and manipulate information temporarily [48]. Impairment of this memory is frequent in TBI patients given that implicated neural substrates, particularly the frontal cortex, are highly vulnerable in this type of injury. There is considerable evidence that working memory depends on network activity including the frontal and parietal regions and its connections. A meta-analysis of functional neuroimaging studies conducted by Owen et al. [49] provided strong evidence for the activation of frontal and parietal cortical regions by various versions of the n-back paradigm. The main fasciculus linking the parietal and frontal lobes is the superior longitudinal fasciculus (SLF) and hence it is likely that this has a role in working memory. Relations between the SLF and working memory deficits have been reported in multiple sclerosis [50] but not in TBI patients. To our knowledge there is no study investigating the impairment of white matter damage related to declarative and working memory deficits in a sample of severe and diffuse TBI.

The aim of this study was to investigate the role of white matter damage in declarative and working memory deficits after diffuse TBI, focusing on the main associative fasciculi [51] including those connecting the cerebral regions involved in the declarative memory and working memory networks.

Our study had two main hypotheses: firstly, that a decreased FA in the superior longitudinal fasciculi (SLF), which is presumably involved in working memory function since it links the parietal and prefrontal regions, would correlate with working memory deficits, and, secondly, that a decreased FA in the fornix, the main fasciculus interconnecting the hippocampus with the frontal lobe, would correlate with declarative memory impairment.

The present study provides evidence of the implications of TAI in declarative and working memory deficits in TBI. DTI group comparison revealed global whole brain reductions in mean FA values for patients and FA maps confirmed that almost all the major fibers were involved. Although our patients suffered global white matter integrity impairment, we found two different and restricted patterns of correlations with the FA and neuropsychological assessment. Whole brain DTI analysis showed that decreased FA throughout the brain correlated with 2-back measures but not with the Rivermead Test. Results from the ROI analyses of the main association fibers showed, as predicted, that working memory specifically correlated with the superior longitudinal fasciculi. However, it was also found to correlate with the corpus callosum, the arcuate fasciculi and with the fornix. On the other hand, declarative memory deficits only correlated with the fornix, as we had expected, and the corpus callosum. These results suggest that there are two different patterns of FA reduction related with two types of memory dysfunctions.

We found that superior longitudinal fasciculi damage is related with working memory but not with declarative memory deficits. These correlations were expectable since the longitudinal fasciculi connect the associative frontal and parietal regions involved in working memory functions [70–72, 49]. The correlation between working memory deficit and the superior longitudinal was also described in multiple sclerosis, pathology that also involves white matter damage [50].

In our sample, FA reductions of corpus callosum correlated with both working and declarative memory impairments. In declarative memory the correlations were seen in the posterior region whereas in working memory the correlations involved anterior and posterior regions, thus again these results point to differential patterns of correlations for both types of memory impairment.

According our results, declarative memory impairment did not depend on diffuse white matter damage since no correlations between FA maps or mean values and declarative memory values were seen. However, the ROI analysis revealed that the fornix FA impairment correlated with the Rivermead test. This result is in agreement with the role of the damage of the hippocampus and its connections in declarative memory deficits in TBI [45, 46].

Our declarative memory results partially agree with those obtained by Salmond et al. [30]. Using a voxel-based analysis with SPM tools, these authors found a significant positive correlation between declarative memory and diffusivity in the left hippocampal formation, the left posterior cingulate, and the left frontal, temporal and occipital regions. The more widespread pattern of correlations observed in their study can be explained by the use of FDR correction, which is more liberal than the FWE correction used in ours [73]. Correlations between FA values in the fornix and declarative memory impairment have been also observed in patients with multiple sclerosis [74]. Other studies investigating FA correlations with declarative memory functions in mild TBI samples have reported significant correlations with the uncinate fasciculi [30, 32] and the cingulum [28]. Although we found decreased FA in these fasciculi, correlations did not reach statistical significance. These discrepancies may be explained by the varying grade of severity of the samples, the difference in the memory tests used, and DTI methodological differences.

In the present study, working memory deficits also correlated with the fornix in both the whole brain analysis and the ROI analyses. There is some evidence from fMRI studies that the hippocampus is involved in working memory functions in healthy subjects [75–78]. Moreover, several animal studies also suggest a role for the hippocampus in working memory [79–81]. Anatomically, prefrontal regions involved in working memory tasks receive projections from the hippocampus [82, 83] and are connected directly to the ventral hippocampus and indirectly to the dorsal hippocampus via the thalamus [84–86]. This structural connectivity supports the idea that the hippocampus has a role in working memory functioning as suggested by our findings.

Finally, significant correlations were observed between the PTA variable and white matter integrity. Whole brain map analysis showed that PTA is an excellent index predictor of the degree of impairment of the major white matter tracts and association fibers. These results suggest that the recovery of memory functions is dependent on the integrity of the complex neocortical regions. Unlike previous studies [13, 14, 17], no correlations were found between GCS and FA maps or mean FA values. This result was to be expected as the fact that all our patients had severe TBI meant that GCS variability would not be sufficient to reach statistical significance.

Our study has certain limitations and our results should be regarded as preliminary. The small sample size and its specific diffuse characteristics may preclude the generalization of the results. The presence of mixed focal and diffuse pathology frequently observed in severe TBI may confound the mapping of neural and behavioral changes in these patients. As our study sample excluded significant cortical pathology, the cognitive impairment observed is more likely to be due to the diffuse pathology alone. Nevertheless, we cannot exclude the possibility that reductions in gray matter in several subcortical structures are also influencing memory deficits in TBI.

This DTI study suggests that declarative and working memory deficits in diffuse TBI patients are related to differential patterns of FA reduction. Working memory impairment reflects the diffuse white matter damage affecting large scale networks such as the superior longitudinal fasciculi, whereas declarative memory deficits seem to be the result of more local disruption of the cerebral circuitry.