In the present study, we identified the most probable trajectories of point-to-point segregated connections between functional attentional centers using a combination of functional magnetic resonance imaging and a novel diffusion tensor imaging–based algorithm for pathway extraction. Cortical regions activated by a visuospatial attention task were subsequently used as seeds for probabilistic fiber tracking in 26 healthy subjects. Combining probability maps of frontal and temporoparietal regions yielded a network that consisted of dorsal and ventral connections. The dorsal connections linked temporoparietal cortex with the frontal eye field and area 44 of the inferior frontal gyrus (IFG). Traveling along superior longitudinal and arcuate fascicles, these fibers are well described in relation to spatial attention. However, the ventral connections, which traveled in the white matter between insula (INS) cortex and putamen parallel to the sylvian fissure, were not previously described for visuospatial attention. Linking temporoparietal cortex with anterior INS and area 45 of IFG, these connections may provide an anatomical substrate for crossmodal cortical integration needed for stimulus perception and response in relation to current intention. The newly anatomically described integral network for visuospatial attention might improve the understanding of spatial attention deficits after white matter lesions.
It was postulated that similar to other neurocognitive functions, the anatomical basis for spatial attention is provided by a large-scale neurocognitive network (Mesulam 1990). This network consists of cortical epicenters that interconnect with each other by the anatomical channels. Therefore, effective cognitive processing requires the integrity and intactness of both cortical epicenters and their connectivity (Mesulam 1990). Nowadays, the anatomy of functional centers of attention is well established, although their anatomical connectivity model is still being developed.
Thus, Corbetta and Shulman (2002) subdivided the cortical epicenters for spatial attention into a dorsal and a ventral frontoparietal networks. The dorsal attention network is activated by directional cues and is involved in attention orientation: preparing and applying goal-directed selection for stimuli and responses. As core regions, it includes bilaterally the intraparietal sulcus (IPS), the frontal eye field (FEF), and extrastriate visual areas (V5/MT+) (Corbetta and Shulman 2002). The ventral attention network is specialized for the target detection, particularly when they are salient or unexpected. It is localized predominantly in the right hemisphere and centered around the temporoparietal junction, anterior insula (INS), and ventral frontal cortex (Corbetta and Shulman 2002).
Besides the cortical regions, the intactness of structural connectivity between cortical centers is also necessary for mental processing. The critical role of corticocortical pathways for cognitive functions was postulated by Geschwind (1965) and Mesulam (1990). Because the associative pathways link the attention-specific areas, their lesion can provoke neglect as a result of disconnection within the attentional network (Mesulam 1990). Both animal and human studies confirmed the importance of the white matter intactness for spatial processing (Gaffan and Hornak 1997; Worrall et al. 2001; Thiebaut de Schotten et al. 2005). Therefore, the structural connectivity for spatial attention and its disruption are currently a subject of intensive exploration (for a review, see Bartolomeo et al. 2007; Doricchi et al. 2008). Nowadays, the superior longitudinal fascicle (SLF) subcomponents II and III, as well as the arcuate fascicle (AF), and the inferior fronto-occipital fascicle (IFOF) are thought to provide connectivity for spatial attention (Bartolomeo et al. 2007; He et al. 2007; Schmahmann et al. 2007; Doricchi et al. 2008). Though the significance of these pathways for spatial processing was confirmed in a number of studies (Doricchi and Tomaiuolo 2003; Thiebaut de Schotten et al. 2005; He et al. 2007; Urbanski et al. 2008), there are still some issues to be resolved. It is unclear how the elements of dorsal and ventral attention systems are structurally interconnected with each other. In addition, anterior INS and ventrolateral prefrontal cortex (area 45), which are important attentional centers, are shown to be connected with the temporoparietal regions not through SLF and AF but with high probability via extreme capsule fiber system (Pandya and Kuypers 1969; Petrides and Pandya 1984; Schmahmann and Pandya 2006; Anwander et al. 2007; Frey et al. 2008; Saur et al. 2008). Being located within extreme/external capsule, these fibers are not incorporated in the current connectivity model of attention. Moreover, the function of IFOF, which is proposed to connect the occipital lobe with the frontal cortex, is at present unknown (Doricchi et al. 2008).
Today, the main tool for a noninvasive investigation of structural connectivity in humans is diffusion tensor imaging (DTI)–based fiber tracking that reconstructs macroscopic axonal organization in nervous system tissues (Basser 1997; Mori et al. 1999). This tool is valuable to visualize the white matter anatomy of the human brain per se (Catani et al. 2002; Mori et al. 2008). Recently, DTI-based fiber tracking was used to investigate the human structural connectivity within attention system (He et al. 2007; Doricchi et al. 2008). However, these pioneering studies were constrained by methodological limitations. First, the precise anatomical information about the connected cortical regions was not obtainable (Thiebaut de Schotten et al. 2008). Second, most of the studies were performed using deterministic fiber tracking that required selecting region of interest within white matter by the researcher. The selection was based on the prior knowledge of presumed connectivity for attention that biased the identification of a fascicle of the researcher's interest. That is why the current model of structural connectivity for attention system needs further investigation.
In the present study, we aimed to explore the structural connectivity for visuospatial attention both as segregated connections and the integral network. Here, we applied both functional magnetic resonance imaging (fMRI) and a novel DTI-based method to extract the pathway between key attentional regions. By means of a visuospatial attention paradigm in fMRI, we revealed the network nodes of the attention system. The determined peaks of activated clusters were subsequently used as seed points for the probabilistic fiber tracking; then the probability maps of 2 seeds were multiplied to extract the pathway between them; the resulting region-to-region maps were analyzed on a group level by averaging the individuals' maps. As the right hemisphere seems to be dominant for visuospatial attention, only the right lateralized network was investigated.
Materials and Methods
Twenty-six healthy right-handed subjects were included in the study (7 males and 19 females, age range 22–53 years old, mean 30, standard deviation 7.3). All subjects had no neurological or psychiatric history and were not taking any psychoactive medication. Written informed consent was obtained from each subject according to the Declaration of Helsinki and Ethic Committee of the University Medical Center Freiburg.
All healthy subjects were scanned with fMRI after completing a behavioral training session. To obtain the visuospatial attention activation, we used 2 sessions of a Posner-like paradigm in a mixed design, in which the task blocks (32 s) alternated with the rest condition (16 s), when the fixation cross (size, 1.4°) was displayed. During the task blocks, subjects were presented a centrally located arrow (size, 4°) for 800 ms, which pointed equally to the right and left indicating the direction of the possible target appearance. For two-thirds of the trials, the target (size, 1.3°) appeared at the indicated direction for 400 ms at 11° of visual angle from the fixation cross. The remaining one-third of the trials were null events without target. The trials were separated by a crosshair for a jittered period of 1000–2000 ms. Each task block contained 12 randomly intermixed trials: 8 events with target (4 right and 4 left) and 4 null events (2 with left and 2 with rightward arrow). The experimental session consisted of 8 task blocks. The subjects performed 2 sessions, between which their state and task performance were controlled. Overall, there were 192 trials distributed among 16 task blocks. Subjects were instructed to fixate on the crosshair and to detect the targets by pressing the button with the right thumb as fast as possible. Behavioral performance was controlled by target detection in valid events and nonresponse in “null events.”
fMRI Scan Acquisition and Data Analysis
T2*-weighted echo planar images (EPI) with blood oxygen level–dependent contrast (matrix size 64 × 64, voxel size 3 × 3 × 3 mm3, 36 slices, no gap) were obtained using a 3-T TRIO MRI System (Siemens, Erlangen, Germany). The repetition time (TR) was 2190 ms, echo time (TE) 30 ms, and flip angle 75°. Anatomical images were acquired with a sagittal magnetization prepared rapid acquisition gradient echo sequence (TR = 2200 ms, TE = 2.15 ms, flip angle = 12°) with a matrix of 256 × 256 × 176. The fMRI data were corrected for motion and distortion across runs using a reference volume acquired in a previous scan (Zaitsev et al. 2004, 2006). The first 4 volumes were discarded to allow for T1 equilibration effects. The subsequent image processing of the fMRI data was performed with SPM5 (http://www.fil.ion.ucl.ac.uk/spm/software/spm5/). The images were corrected for different acquisition times between slices with reference to the middle slice. Using normalization parameters from the segmentation of the individual anatomical images (Ashburner and Friston 2005), the images were normalized to the standard Montreal Neurological Institute (MNI) space and resampled to 3 × 3 × 3 mm3 voxels. Then, smoothing was applied with a Gaussian kernel of 9-mm full-width half maximum. A general linear model analysis was performed with a block design of the functional runs, using the canonical hemodynamic response function. The task blocks were modeled as one main regressor to analyze the main effect of interest at the single-subject level. To capture the extra variance introduced by the motor response, it was modeled as another regressor orthogonalized with respect to the main regressor of interest. Six movement parameters were used as additional regressors. In addition, data were high-pass filtered at 1/128 Hz. At a group level, a 1-sample t-test was performed to find the peak voxels of the activated clusters (family-wise error [FWE] corrected, P < 0.05) that were subsequently used as seed points for probabilistic fiber tracking. The anatomical specification of the activated clusters was obtained using the Anatomy toolbox implemented in SPM5: http://www.fz-juelich.de/ime/spm_anatomy_toolbox (Eickhoff et al. 2005). The determined regions of interest in MNI space were inverse normalized for every subject, so that the functional nodes of the group analysis were translated to the individual space. The DTI data were acquired directly after the fMRI session.
Diffusion-Weighted Magnetic Resonance Imaging Acquisition and Fiber Tracking
Diffusion-weighted images were acquired using a diffusion-sensitive single-shot spin echo EPI sequence with the following parameters: TR = 10 000 ms, TE = 94 ms, voxel size 2 × 2 × 2 mm3, number of slices = 69, no gap. For each of the 61 diffusion encoding directions, an effective b value of 1000 s/mm2 was used. In addition, 9 volumes without diffusion weighting (b value = 0 s/mm2) equally distributed throughout the scan were acquired. Similar to the fMRI data, motion and distortion correction were applied during reconstruction by using a previously acquired reference scan (Zaitsev et al. 2004, 2006). The local diffusion parameters were calculated using the DTI and Fiber Tool (http://www.uniklinik-freiburg.de/mr/live/arbeitsgruppen/diffusion/fibertools_en.html). The seed points for probabilistic fiber tracking consisted of spheres of 4-mm radius centered in the peaks of fMRI group activation that were inverse normalized into individual space. Fiber tracking was performed with a Monte Carlo simulation of random walks algorithm similar to the probabilistic index of connectivity (PICo) approach (Kreher et al. 2008). The tracking area included 1) the white matter to avoid tracking across anatomical borders and 2) a rim of gray matter to ensure the contact of the cortical seeds with the white matter. Starting from a seed point, the random walk was performed repetitively, so that from each seed voxel 105 random walkers were propagated within individual's tracking area. Using this algorithm, we obtained the probability maps for all seed points, which contained information about PICos and the main direction of the random walk vectors. To extract the long association pathway between temporoparietal and frontal seed points on a single-subject level, we multiplied pairwise the corresponding probability maps (Kreher et al. 2008). This multiplication step takes into account the frequency, with which the voxel was visited from both seeds, and the direction of the random walk vectors from both seeds (Supplementary Fig. S1). The conjunction area of 2 probabilistic maps cannot be simply considered as fiber track between 2 seeds because it consists of 1) fibers connecting 2 seeds (“connecting fibers,” i.e., fibers of interest) and 2) fibers connecting these seeds with a third area (“merged fibers,” i.e., fibers of no-interest). In the algorithm by Kreher et al. (2008), these areas are differentiated by taking into account additional information about the direction of the random walk. This is performed by controlling the direction of the elementary steps of random walker using the principal eigenvector of the diffusion tensor as the reference: connecting fibers of interest have an opposite main traversing direction of the random walk vectors and merged fibers, in contrast, have the same direction of the random walk vectors (Supplementary Fig. S1). Therefore, the multiplication algorithm is able to suppress the merged fibers and extract the connecting fibers. Finally, the multiplication method enables the extraction of the most probable direct pathway between 2 cortical seed regions without using a priori knowledge about the presumed course. The values on the combined map represent a voxelwise estimation of the probability indices that a voxel is a part of the connecting fiber bundle of interest (probability index forming part of the bundle of interest [PIBI]) (Kreher et al. 2008). Using multiplication algorithm, all temporoparietal maps were combined permutational with all frontal maps. This resulted in 16 combined region-to-region maps for each subject. These maps were further processed with SPM5 spatial tool including: 1) rescaling of the minimum and maximum from the original maps to the range between 0 and 1; 2) normalization to the standard MNI space; and 3) smoothing with a Gaussian kernel of 3-mm full-width at half maximum. To obtain the anatomy of the group, the individuals’ region-to-region maps were averaged between all subjects. This resulted in 16 mean maps. To show the composite network, we averaged 4 mean maps of each frontal seed. Due to the nature of the probabilistic tracking, the multiplication algorithm for all connections, and the subsequent spatial processing, the maps contain a very large number of voxels with near-zero values and only a minor fraction of voxels with values of interest. The generated empirical analysis defined the voxels of interest as those with PIBIs >10−6. To separate voxels within fiber tracks and those that are likely to contain only noise, we demonstrated only those with value >0.0148, which corresponded to the voxels of interest with upper 5% of PIBIs.
All subjects completed the fMRI experiment successfully and detected all targets with a mean reaction time of 323.30 ± 90.62 ms (325.80 ± 93.34 and 320.79 ± 89.59 ms for left and right targets, respectively). Only 4 subjects gave false responses (the number of their false-positive responses were 1, 6, 1, and 1).
The group analysis of the fMRI data identified regions involved in the spatial processing (Fig. 1, peak coordinates are reported in Table 1). The recruitment of the bilateral middle temporal areas (area V5/MT+), IPS, and the crossing of the caudal superior frontal gyrus and precentral sulcus (FEF) showed the involvement of the dorsal attention system. Right lateralized activated areas of the temporoparietal cortical junction consisting of the supramarginal gyrus (SMG) and the caudal superior temporal gyrus (STG), the anterior INS, and the caudal inferior frontal gyrus (IFG) pars opercularis (IFGoper) and pars triangularis (IFGtri) were identified as the ventral attention system. We used the peaks of the activated clusters as seed regions for the probabilistic fiber tracking.
|Medial temporal gyrus/inferior occipital gyrus||V5/MT+||54||−63||3||10.28|
|Superior temporal gyrus||STG||63||−36||21||9.86|
|Medial frontal gyrus/precentral gyrus||FEF||51||3||48||9.99|
|Pars opercularis (area 44)||IFGoper||54||12||12||6.38|
|Pars triangularis (area 45)||IFGtri||51||21||−3||6.49|
|Medial temporal gyrus/inferior occipital gyrus||V5/MT+||54||−63||3||10.28|
|Superior temporal gyrus||STG||63||−36||21||9.86|
|Medial frontal gyrus/precentral gyrus||FEF||51||3||48||9.99|
|Pars opercularis (area 44)||IFGoper||54||12||12||6.38|
|Pars triangularis (area 45)||IFGtri||51||21||−3||6.49|
Fiber Tracking Results
To assess the structural connectivity for visuospatial attention, we first analyzed the most probable trajectories of all 16 segregated long association fibers between temporoparietal and frontal regions (Fig. 2). Relative to the upper edge of the INS, identified connections traveled either dorsally or ventrally. The dorsal trajectories had fibers linking all temporoparietal regions with FEF and IFGoper. In contrast, ventral trajectories had fibers that connected temporoparietal regions with INS and IFGtri. Thus, the frontal seeds determined ventral or dorsal course for all temporoparietal–frontal connections independent from their posterior ends (Fig. 3). But posterior regions determined the disposition pattern of segregated connections relative to each other: each frontal seed had the same order of fibers’ temporoparietal ends (Fig. 3). Thus, in posterior segments of all fiber bundles, IPS/SMG connections were the most medial and STG connections were the most lateral, while V5/MT+ fibers were in an intermediate position (Fig. 3). Additionally, the STG fibers formed an arch around the caudal part of the sylvian fissure (Fig. 3). Fibers from V5/MT+ took always the upper rostral course to the inferior parietal lobule and then continued either in the dorsal–rostral direction (within FEF and IFGoper fiber bundles) or in the ventral direction (within INS and IFGtri fiber bundles). In other words, whereas frontal cortical regions determined the ventral or dorsal fiber's trajectories, the temporoparietal cortical regions specified the disposition of connections to each other.
To understand the connectivity on the level of the integral network, we grouped connections in respect to their frontal ends in to FEF, IFGoper, INS, and IFGtri fiber bundles (Fig. 4). The FEF and IFGoper fiber bundles had dorsal trajectories that were adjacent but topographically separated. From the temporoparietal cortex, they traveled within the white matter of the inferior parietal lobule, then crossed the corona radiata and separated to take FEF or IFGoper direction. The FEF fiber bundle was medial and above the IFGoper track. The later one traveled laterally and below from the FEF bundle, mainly parallel and adjacent to the upper edge of the INS (Fig. 5).
IFGtri and INS fiber bundles followed predominantly the ventral course and mainly overlapped in their trajectories. Both bundles traveled within the white matter between the INS cortex and putamen (Fig. 5) with the maximum of probability projected to the perisylvian space and superior temporal gyrus at z = 0 (Fig. 2). Due to the spatial resolution of the method, it was impossible to attribute their trajectory specifically to either extreme or external capsules.
In the present study, we showed the most probable pattern of structural connectivity for visuospatial attention both as segregated fiber bundles and integral network in the standard MNI space.
The present fMRI paradigm aimed at activating the visuospatial processing areas, including both the orientation and target detection components. The obtained pattern of activation in the present experiment is generally in line with the results of the previous studies (Corbetta et al. 2000; Perry and Zeki 2000). We chose to use null trials as opposed to invalid trials to ensure that there were attention-related trials without motor responses. This enabled us to disambiguate between responses evoked by visual attention and motor responses per se.
The combination of fMRI and DTI-based fiber tracking allowed us to obtain anatomically precise regions of interest on the level of the functional system. Additionally, defining seeds with fMRI makes it possible to consider the group anatomical and functional characteristics. Because it is known that one functional center can consist of cytoarchitectural heterogenic parts, which have different pattern of connectivity, the precise localization of functional centers is crucial. The use of regions of interest, obtained in other studies, may lead to a systematic error also due to the different spatial preprocessing. Because the previous studies demonstrated high cytoarchitecture intersubject variability especially for the higher order frontal areas (Amunts et al. 1999; Fischl et al. 2007), the functional seeds definition in our experiment eliminates possible methodological shortcomings of fiber tracking due to this cytoarchitectural incongruence.
In general, the combination of fMRI and DTI is extremely difficult particularly because of their different physical principles. We were able to obtain sufficiently precise spatial registration between fMRI-defined peaks of activation and DTI reconstruction due to the following procedures. The seed spheres (8 mm diameter) were settled within both the white and gray matter, which also composed the tracking area. Together with precise inverse normalization of fMRI activation peaks into individual space, this resulted in an accurate combination of both fMRI and DTI fiber tracking.
The relevance of SLF and AF to the attention network was derived from previous animal studies because these tracks connect temporoparietal and posterior frontal cortices (Schmahmann and Pandya 2006; Schmahmann et al. 2007). This explains why the few DTI human studies of structural connectivity for attention concentrated on visualizing already widely accepted and well-described SLF and AF (He et al. 2007; Thiebaut de Schotten et al. 2008), and the white matter between INS and putamen was not used as starting regions for tracking. Pathways were usually defined with deterministic fiber tracking algorithm as a single route for determined regions of interest and failed to reveal the fasciculi branches (Jones 2008). These limitations constrained the previous studies in investigation of the integral anatomical network for attention.
Instead of reconstructing just a single trajectory from a given seed point with the deterministic method, we used probabilistic fiber tracking algorithm, which presents the information about the distribution of possible fiber orientations for the current seed point. Although with this approach false positives and false negatives are just as problematic as in deterministic DTI tracking, probabilistic tractography informs the user as to how likely it is that a given pathway be found “through the data set” (Jones 2008). Additionally, using the multiplication algorithm that suppresses the false-positive results (Kreher et al. 2008), we were able to extract the most probable direct pathway between 2 cortical seed regions without any assumptions concerning the fiber course (Kreher et al. 2008).
Another limitation of many DTI-based fiber tracking studies is caused by sharp degradation of fractional anisotropy close to the cortex. This makes the cortical projection sites of the white matter fiber track not obtainable. We dealt with the problem by 1) using a mask for the tracking area, which is a combination of the white and gray matter; 2) using a probabilistic fiber tracking; 3) using a novel algorithm for pathway extraction. Often the tracking area is defined by applying a threshold on anisotropy leading to the tracking only within the white matter. We also used a white matter mask to avoid the tracking across anatomical borders. However, to ensure the contact of the cortical seed regions with the white matter, we included a thin rim of the gray matter in the mask. This extended white matter mask enables tracking from the cortical region of interest. In contrast to the deterministic algorithm, the applied probabilistic fiber tracking algorithm is not so dependent on having a well-defined principal eigenvector. The random walkers propagate through the area of degraded fractional anisotropy without selecting a preferable direction, starting to track a fiber after crossing over this critical region. This facilitates the launching of trajectories from, and reconstruction of trajectories into, areas of low anisotropy such as a gray matter (Jones 2008). In addition, the applied tracking algorithm evaluates best the indices of connection probability in the middle of the fiber bundle (Kreher et al. 2008).
The limited spatial resolution of the DTI technique does not allow to differentiate directly between fibers passing through and those starting/stopping in the cortical seed regions. However, the propagated random walkers tend to choose the direction of the principle eigenvector. This direction is determined by the fibers that start/stop in the seed region, which are the fibers we are interested in. In contrast, the tracking algorithm does not reconstruct the fibers, which pass through the seed region and course parallel and adjacent to the cortex. The reason is a suppression of sharp turns in the random walker trajectories and the ineffectiveness of the tractography in the regions adjacent to the cortex. In the present study, false-positive results are suppressed additionally with the multiplication step and application of a threshold to the resulting maps.
Taking into account all methodological issues, it is necessary to note that the obtained probability maps (Fig. 2) show the most “probable” trajectories between frontal and temporoparietal attention relevant areas. The identified connectivity model was established without any a priori settings about presumed trajectories, and it does not provide any assumptions of connections' strength. Considering that DTI-based fiber tracking is sensitive to artifacts due to crossing, kissing, and fending of the fibers, it is necessary to evaluate our data with respect to the primate and/or postmortem studies.
The Anatomical Attribution of Connections
In the present study, we showed an integral network for visuospatial attention formed by dorsal and newly described ventral fiber bundles. The dorsal bundles, linking temporoparietal regions with FEF and IFGoper, correspond to SLF and AF. Both SLF and AF travel within the white matter of the inferior parietal lobule and deep in the upper shoulder of the sylvian fissure (Schmahmann and Pandya 2006; Mori et al. 2008). The SLF can be divided into 3 subcomponents from dorsal to ventral and medial to lateral resulting in SLF I, II, and III (Makris et al. 2005; Schmahmann and Pandya 2006; Schmahmann et al. 2007). Due to their disposition, we assigned the FEF fascicle to SLF II and IFGoper fascicle to SLF III. This pattern of FEF and IFGoper connectivity with temporoparietal regions was previously demonstrated in other studies in nonhuman primates and humans (Pandya and Kuypers 1969; Makris et al. 2005; Petrides and Pandya 2006; Rushworth et al. 2006; Frey et al. 2008). The AF corresponds to the STG connections within the dorsal bundles (Makris et al. 2005; Schmahmann et al. 2007).
The arrangement of the ventral long association fibers, connecting temporoparietal cortex with IFGtri and INS, is more complex. The course of the connections within the white matter between putamen and the INS cortex corresponds to the extreme and external capsules. Because of the limited spatial resolution of DTI, it is not possible to differentiate between these anatomical structures (Mori et al. 2008). But nonhuman primate studies showed that the extreme capsule contained association fibers, whereas the external capsule was composed of corticosubcortical pathways (Schmahmann and Pandya 2006). The obtained pattern of structural connectivity for IFGtri and INS with temporoparietal cortex was described previously in animals (Pandya and Kuypers 1969; Jones and Powell 1970; Petrides and Pandya 1984, 2002; Cavada and Goldman-Rakic 1989; Seltzer and Pandya 1989; Schmahmann and Pandya 2006; Roberts et al. 2007). The ventral pattern of connectivity for area 45 (IFGtri) was confirmed in humans using high angular resolution diffusion imaging (Anwander et al. 2007; Frey et al. 2008; Saur et al. 2008) and on postmortem brain (Burgel et al. 2006). In the present study, the highest indices of the ventral connections probability corresponded to the lower third of the extreme and external capsules. According to a prevalent nomenclature, the lowest part of the white matter between INS and putamen is occupied by IFOF that links the frontobasal cortex with the parieto-occipital cortex (Catani et al. 2002; Burgel et al. 2006). However, other authors see IFOF as nonexistent, arguing that the first horizontal component of IFOF is formed by the inferior and medial longitudinal fascicles, whereas its rostral ascending limb corresponds to extreme capsule (Schmahmann and Pandya 2006). The ventral connections in the present study are direct and uninterrupted, and they connect the frontal lobe not with occipital structures but with the temporal and parietal cortices. We propose therefore that their course coincides with the rostral part of IFOF or extreme/external capsule fiber system.
The Functional Role of the Revealed Connections
An anatomical connection obtains its functional significance through the interaction of the linked cortical regions. The SLF and AF are major white matter pathways linking the parietal and frontal lobes (Schmahmann and Pandya 2006). They provide the means by which the prefrontal cortex can regulate the focusing of attention within different parts of space. The FEF is considered to be essential for the selection between different aspects of the visual, auditory, and somatomotor environment based on conditional operations (Petrides 2005). SLF II connects FEF and temporoparietal cortex, including IPS; it is therefore critical for the orientation of attention to a determined location (Petrides and Pandya 2006). Whereas IFGoper receives and processes information about peripersonal space and participates in coding of object location (Mesulam 1999; Rizzolatti et al. 2002), SLF III provides this region with higher order somatosensory input (Schmahmann and Pandya 2006). AF was supposed to be a means by which the prefrontal cortex can receive the influence of audiospatial information (Schmahmann et al. 2007). However, in our study, the caudal STG activation was obtained by the visuospatial attention task, indicating the relevance of AF to the declared multimodal network for the detection of changes in the sensory environment (Downar et al. 2000).
IFGtri and INS fiber bundles with ventral trajectories have not been described so far for visuospatial attention, though both cortical regions are essential components of the ventral attention system (Downar et al. 2000; Corbetta and Shulman 2002). IFGtri corresponds to area 45, which is a part of the ventrolateral prefrontal cortex. Information processing in relation to current intentions is attributed to the ventrolateral prefrontal cortex. This region also provides selection among multiple competing stimuli with filtering out of unwanted signals (Everling et al. 2002; Petrides 2005; Bar et al. 2006; Hampshire et al. 2008). Right anterior INS is integral to self-awareness and may underlie a conscious representation of self (Craig 2002; Critchley et al. 2004; Karnath et al. 2005). This region is also a part of a “core” taskset system (Dosenbach et al. 2006). The multifunctional role of both IFGtri and INS could be realized only in the interaction with other temporoparietal areas. The new described ventral connections provide the wiring for such interaction.
Our data showed not only connections within the dorsal or the ventral cortical attention systems, which were proposed by Corbetta and Shulman (Corbetta and Shulman 2002), but the trajectories of the tracks that subserve the integration between them. Fox et al. (Fox et al. 2006) proposed that ventral and dorsal attention systems could functionally interact through right middle frontal gyrus. However, experimental studies of visual perception and attention pointed to the existence of the direct connections between dorsal and ventral cortical attentional centers, for example, between IPS and IFGtri (Lumer et al. 1998; Buschman and Miller 2007). These tracks are necessary to direct the corticocortical interactions for bottom-up or top-down control. To our mind, the structural link between IPS and area 45 is of high importance for visuospatial processing. Although this connection has been identified in nonhuman primate (Jones and Powell 1970; Schmahmann and Pandya 2006), it has not yet been demonstrated in humans. At the same time, we obtained a low likeliness for the direct connection between SMG and IFGtri (Fig. 2), in contrast to the high probability for the direct link between SMG and IFGoper. This is additional evidence for different connectivity signatures of the human parietal regions (Rushworth et al. 2006; Tomassini et al. 2007).
Another important point is that our results are consonant with the parcellation of Broca's area (Anwander et al. 2007; Frey et al. 2008). Although Broca's area has been broadly investigated, its right analogue was abandoned. We demonstrated that Broca's analogue in the right hemisphere also could be subdivided into areas with distinct connectivity signatures: right IFGoper, which corresponds to area 44, and IFGtri, which corresponds to area 45, have different connectivity pattern. Connectional patterns are likely to define behavior of cortical areas (Schmahmann and Pandya 2008) that is why right IFGoper and IFGtri seem to have distinct functional contribution in visuospatial processing. Our results hence pointed to the necessity of further precise investigation into the right IFG function that is usually mentioned as the functionally homogeneous area in respect to visuospatial attention.
Knowledge of structural connectivity definitely contributes to our understanding of the disconnection syndromes. We would like to emphasize that connections between one frontal seed and different temporoparietal regions mostly overlap in their course (Fig. 3). This means that there is a very low likelihood of disruption of only one connection via subcortical damage. In case of the white matter lesion in the clinical practice, we might meet more often with the damage of several temporoparietal–frontal connections. Thus the white matter lesion between INS and putamen might disrupt all connections linking temporoparietal regions with INS and IFGtri.
The present study was not aimed at investigating neglect because it was performed on healthy subjects and with a paradigm, which is not sensible for spatial neglect. However, due to the involvement of IFGtri and INS in spatial processing, we can speculate about the functional significance of the ventral connections also for neglect. The clinical importance of the ventral IFGtri and INS pathways was confirmed in patients with an isolated infarct in the right INS and adjacent white matter, who neglected multimodal stimuli (Berthier et al. 1987) or had multimodal extinction (Manes et al. 1999). In 140 right hemisphere stroke patients, a lesion of perisylvian structures was found as critical for neglect (Karnath et al. 2004). This region overlaps with the connections we found to link temporoparietal cortex with IFGoper, IFGtri, and INS. Perisylvian lesions have been demonstrated to be critical for different types of neglect in other studies as well (Samuelsson et al. 1997; Corbetta et al. 2005; Committeri et al. 2007). However, the specific analysis and clinical data from neglect patients would be required to address the link of the described attention network for spatial neglect.
German Academic Exchange Service (Deutscher Akademischer Austausch Dienst); Deutsche Forschungsgemeinschaft (We1352/14-1); Bundesministerium für Bildung und Forschung (01GW0661).
Conflict of Interest: None declared.