Choice reaction time performance correlates with diffusion anisotropy in white matter pathways supporting visuospatial attention

Methods

Participants. Twelve young healthy participants (four male, eight female; mean age, 23.2 ± 1.6 years; minimum age, 19.7 years; maximum age, 26.4 years) were studied. All participants were right handed. All participants provided informed written consent by the guidelines of the Massachusetts General Hospital internal review board. The participants gave their informed consent in writing before each of the sessions.

CRT Paradigm. The participants were tested on a self-paced CRT task. The task was the baseline task for a larger study examining implicit motor sequence learning. Four empty squares were presented horizontally on a computer monitor using the e-prime software package (Psychology Software Tools, Pittsburgh). The squares were white on a black background. Participants rested the index and middle fingers of both hands on two keys of a four-key response board.

In each trial, one of the four squares (the target square) was filled in solid. The participant responded by pressing the corresponding key on the response board. The targets were presented in a random order with the only exception that no single location was presented twice in a row. The task was repeated in six blocks of 72 trials for a total of 432 trials. Before the session, the participants were instructed to respond as quickly as possible without sacrificing error, and to not correct errors. Between blocks, the participant rested for a 45-s rest period, during which the participant was instructed to make no response. The CRT for block 1 provided an index of native CRT and the ΔCRT between blocks 1 and 6 provided a measure of within-session improvement. The reaction time scores for the participants are given in Fig. 1. No significant correlations were observed between CRT and age (r_s = 0.34, P = not significant) or between ΔCRT and age (r_s = 0.43, P = not significant).

Image Acquisition. Diffusion tensor MRI scans were obtained on all 12 participants. The scans were performed on a Siemens Sonata 1.5 Tesla MRI scanner (Erlangen, Germany) located at the Athinoula A. Martinos Center for Biomedical Imaging at Massachusetts General Hospital. The DTI acquisition used single-shot EPI (34) and a twice-refocused spin echo sequence (35). The imaging parameters were repetition time (TR) = 9,000 ms, echo time (TE) = 68 ms, b = 700 s/mm², g_max = 26 mT/m, seven directions (one T2 plus six diffusion-weighted), eight averages, total acquisition time 8 min 33 seconds. The diffusion gradient directions were obtained from the six nonopposed edge-centers of a cube in q space.

Sixty axial oblique slices were acquired. The slices were oriented in the intercommissural (AC-PC) plane. The field of view was 256 × 256 mm and the matrix size was 128 × 128 to give 2 × 2 mm in-plane resolution. The slice thickness was 2 mm (0-mm skip). The mean postaveraged signal-to-noise ratio of the T2 and diffusion-weighted images was 35 and 18, respectively. The eddy current distortions between diffusion weightings were typically less than ≈1-2 voxels.

Preprocessing. Correction for motion and residual eddy current distortion was applied by registering all of the images to the first acquired T2 image. The registration was performed by using flirt with a 12-df global affine transformation and a mutual information cost function (36). flirt is available through the FSL software library (www.fmrib.ox.ac.uk/fsl). Trilinear interpolation was used for the resampling. The diffusion tensor, associated eigensystem, and FA metric were calculated for each voxel as described (16, 37).

Atlas-Based Analysis. The correlation between CRT and FA was measured by using complementary atlas-based and manual region of interest (ROI) approaches. The atlas-based analysis was performed by normalizing the individual FA volumes to Montreal Neurological Institute (MNI) space by using flirt (36). The MNI normalization was performed by registering each individual T2 volume to a skull-stripped version of the MNI 152-subject T2 template (38). The skull-stripping was performed by using the brain bet program (39) from FSL. The transformation from this registration step was then applied to the individual FA volume.

The MNI-normalized FA volumes were smoothed by using a 3D Gaussian kernel with 6-mm full-width at half-maximum (FWHM) and 6-mm spatial extent. The FA volumes were smoothed to increase the statistical sensitivity to correlations equal to or larger than the kernel width. The smoothing width of 6-mm FWHM (= 3 voxel widths) was selected based on numerical evidence that three-voxel FWHM provides sufficient smoothing for 3D Gaussian random fields (40). Voxels with trace diffusion >6 μm²/ms were not included in the smoothing operation in an effort to minimize the partial volume contribution from cerebrospinal fluid. MNI-normalized FA volumes were also produced by using nearest neighbor interpolation and no smoothing to obtain normalized FA volumes immune from the effects of resampling and smoothing.

The correlation between CRT and FA was computed for each voxel in atlas space. The correlation coefficient was computed by using nonparametric Spearman rank regression (41). Nonparametric Spearman rank regression was used instead of a parametric Pearson regression because of the potential nonnormality of the FA distribution across individuals and the lack of a prior hypothesis that CRT and FA are linearly correlated. The Spearman correlation coefficient (r_s) map was visualized as an overlay on either the MNI single individual T1 template, the group averages FA volume, or a diffusion tensor cuboid map from a representative participant. The representative participant was chosen as the one with the most voxels with the median FA. Voxel-level correlations greater than r_s > 0.7 were deemed “high.”

Multiple Comparisons Correction. Multiple comparisons correction was applied at the cluster level by using the Monte Carlo permutation method (42). The corrected significance was computed based on the size of voxel clusters at the P < 0.005 level. The Monte Carlo calculation used 10⁴ trials. Contiguous clusters ≥20 voxels in size (corresponding to a volume of 0.160 cc) were reported. Corrected significance levels of P < 0.05 were deemed “significant.”

The permutation method was also used to compute the corrected significance thresholds at the voxel level. The corrected voxel-level significance thresholds were reported but were not applied to threshold the data. The voxel-level thresholds were not applied so that the anatomic continuity of the correlation maps could be visualized.

ROI Analysis. Confirmatory ROIs were defined in each individual participant's MNI-normalized T2 volume. The FA volumes were MNI-normalized to account for differences in head orientation and to provide optimal matching of regions across individuals. The FA values were sampled from FA volumes that had been MNI-normalized by using nearest neighbor resampling and had not been smoothed, thus eliminating the smoothing effects of resampling and kernel smoothing.

ROIs were placed in a total of 10 areas. The areas were selected based on significant clusters identified in the atlas-based analysis, or to serve as control regions. The 10 anatomic areas included the optic radiations of each hemisphere, the genu and splenium of the corpus callosum, the anterior and posterior thalamus in each hemisphere, and the posterior limb of the internal capsule in each hemisphere. All ROIs were created from small swaths within the WM to attempt to reduce partial volume contamination. The posterior limb of the internal capsule was included as an ROI to test for an effect in motor WM.

The ROIs were defined as follows. The optic radiation was defined in the axial view on a slice approximating the center of the splenium of the corpus callosum and the ventricles displaying the most lateral extent, and drawn lateral to the lateral ventricle, spanning the most posterior aspect of the ventricle, from the region where the posterior callosal fibers intersect with the optic radiations to the most lateral aspect of the posterior ventricular region. The genu and the splenium of the corpus callosum were drawn in the axial view, defined as a line of voxels spanning a portion of the medial/lateral extent of the approximate center of each region of the callosum. Thalamic ROIs were created as a 3D crosshairs (seven voxels) to attempt to sample from a homogeneous region within the structure, and were drawn on an axial slice approximating the center of the thalamus placed in the anterior- and posterior-most regions while staying completely within the structure. The posterior limb of the internal capsule was drawn as a line of voxels down the center of the structure in the axial view on a slice approximating the center of the genu of the corpus callosum, where the inflection between the anterior limb and posterior limb was greatest.

Results

The participants' native CRT (block 1) and within-session improvement in CRT (block 1 minus block 6) are shown in Fig. 1. Table 1 provides a list of all anatomic regions where significant (corrected P < 0.05) correlations between CRT and FA were observed. Significant clusters of positive correlation between native CRT and FA were found in the right posterior thalamus, right optic radiation, right medial precuneus WM, and left superior temporal sulcus WM. Negative correlations were found in left superior temporal sulcus WM and the left parietal operculum. The positive correlations for within-session improvement in CRT localized to the left posterior thalamus and right precuneus WM, and the negative correlations localized to right superior cuneus WM and right superior temporal sulcus WM.

The ROI analysis showed that native CRT was significantly correlated with FA in the right optic radiation but not the left (Table 2 and Fig. 2). Within-session improvement in CRT was significantly correlated with FA in the right optic radiation and both hemispheres of the posterior thalamus (Table 2). No significant correlations were observed for the corpus callosum ROIs or the poster limb of the internal capsule.

The individual FA maps showed a clear increase in FA in the right optic radiation with increasing CRT (Fig. 3). The correlation between CRT and FA localized to the right visual pathway (Fig. 4), extending from the lateral geniculate nucleus through Meyer's loop to the optic radiation, terminating at the junction between the optic radiation and the posterior forceps of the corpus callosum. The correlation maps for within-session improvement in CRT are shown in Fig. 5A. The diffusion tensor maps in Fig. 5B show that the ΔCRT-FA correlation lies in the posterior thalamus at the border between the pulvinar and lateral geniculate nucleus.

Discussion

DTI in young healthy adults revealed significant correlations between CRT and FA in right thalamus, and right visual, right medial precuneus, and left superior temporal WM. The ROI analysis for within-session improvement in CRT indicated correlations in both hemispheres of the posterior thalamus in the region of the lateral geniculate and pulvinar nuclei.

The right visual and right parietal cortices are specialized for visuospatial attention (43-46). The pulvinar nucleus, located in the posterior thalamus, is a key node in the visuospatial attention network (47). Superior temporal gyrus has been implicated in visual awareness (48) and voluntary control of visuospatial attention (49). No significant correlations were observed in the posterior limb of the internal capsule or the corpus callosum for either experiment. These results indicate that native performance and within-session improvement in CRT are associated with variations in the WM, supporting visuospatial attention as opposed to pathways supporting motor movement or interhemispheric transmission.

Madden and colleagues (31) reported a correlation between RT on a visual oddball task and FA in the splenium of the corpus callosum. The present study found that RT-FA correlations are also present in the thalamus and projection and association pathways, are lateralized, and also hold for within-session improvement in RT. No significant correlations were found in the splenium of the corpus callosum, which could be due to differences in the behavioral paradigm or experimental protocol for this study and ref. 31.

The physiological mechanisms for the observed CRT-FA correlation require substantial further investigation. FA is influenced by a number of factors including, but not limited to, myelination, axon diameter, axon density, and ultrastructure (14). The correlation may be due to the influence of myelin thickness on both water self-diffusion (14) and NCV (9). The myelin hypothesis would predict a negative correlation between CRT and FA because increased myelin thickness would cause increased FA and faster NCV, which would in turn result in faster RT. However, the CRT-FA correlations were predominately positive, particularly in the right optic radiation and the posterior thalamus (Tables 1 and 2).

The positivity of the CRT-FA correlation may be due to the effect being mediated by factors other than myelin. Significant RT-FA correlations were observed in the thalamus despite its low myelin content compared to WM (50). In both WM and thalamus, physiological factors other than myelin that could influence the correlation include intravoxel fiber crossing or axon diameter. In anatomic regions containing intravoxel fiber crossing (51), increased FA of an individual fiber population can result in a decrease in the overall FA (52, 53). For example, in the junction between the optic radiation and the posterior forceps (Figs. 3 and 4) a selective increase in the FA of the optic radiation could result in a decrease in the overall FA. The fiber crossing effect might also be a factor in Meyer's loop where the rapidly bending fibers produce intravoxel orientational dispersion (54). The relationship between composite fiber architecture and FA can be clarified with diffusion MRI methods which can resolve intravoxel structure such as q-space imaging (55) and high angular resolution diffusion imaging (51, 54).

The CRT-FA correlations may also be caused by interindividual differences in axon diameter distribution. Large caliber axons, e.g., magnocellular axons, have faster NCVs, which could result in faster CRT. The relationship between axon diameter and FA is not fully known, although increased axon diameter could result in lower FA due to an increase in the mobility of water in the intraaxonal compartment (56).

The positive correlation may also be due to interference effects, although this mechanism is regarded as unlikely. Faster NCV in the optic radiation could result in slower CRT because of interference between the early visual processing components and the subsequent target identification and motor response programming components of the CRT task. This hypothesis could be tested by measuring the correlation between FA and simple RT or the latency for the P100 visual evoked response.

The correlations reported here may potentially reflect an associative as opposed to a causal relationship. It will be necessary to investigate the influence of other physiological factors such as the length of the optic radiation and the surface area and thickness of primary visual cortex. Detecting changes in FA with learning or skill acquisition would also address the issue of causality.

The correlation between CRT and FA was assessed by using both an atlas-based analysis and an ROI approach. The atlas-based analysis and the ROI analysis provided complementary information. For example, for the ΔCRT experiment, the right posterior thalamus failed to achieve significance at the cluster level (Table 1), but the correlation was significant in the ROI analysis (Table 2).

The present study examined the correlation between FA and behavioral RT performance. It will also be possible to extend this method to examine the correlation between FA and more direct measures of information processing speed obtained with electroencephalography (EEG) or magnetoencephalography (MEG). For example, correlating FA with event-related potential latency or phase offset for synchronous brain activity measured on EEG or MEG could identify WM pathways that mediate interindividual differences in information processing speed for specific cognitive tasks.

It remains to be determined to what extent the CRT-FA is caused by genetic or epigenetic factors. Pfefferbaum et al. (57) found that FA in the corpus callosum in old age is genetically regulated, with the FA of the splenium of the corpus callosum being more strongly regulated than that of the genu. RTs have been found to be moderately to highly heritable (58). In light of the correlation between CRT and FA, the heritability of both CRT and FA points to the question of whether the two measures share common genetic factors. It should be possible to address this question by examining the correlation between CRT and FA in a twin design. Identifying the genetic and epigenetic contributions to CRT-FA correlation will help shed light on the fundamental role of WM physiology in human cognitive performance.