Brain Topography

, Volume 25, Issue 3, pp 264–271

Connectivity-based parcellation reveals interhemispheric differences in the insula

Authors

    • Department of Biomedical Laboratory and Imaging Science, Faculty of MedicineUniversity of Debrecen Medical and Health Science Center
  • Péter P. Molnár
    • Institute of Pathology, Faculty of MedicineUniversity of Debrecen Medical and Health Science Center
  • Péter Bogner
    • Department of Radiography, Faculty of Health SciencesUniversity of Pécs
  • Monika Béres
    • Department of Biomedical Laboratory and Imaging Science, Faculty of MedicineUniversity of Debrecen Medical and Health Science Center
  • Ervin L. Berényi
    • Department of Biomedical Laboratory and Imaging Science, Faculty of MedicineUniversity of Debrecen Medical and Health Science Center
Original Paper

DOI: 10.1007/s10548-011-0205-y

Cite this article as:
Jakab, A., Molnár, P.P., Bogner, P. et al. Brain Topogr (2012) 25: 264. doi:10.1007/s10548-011-0205-y

Abstract

The aim of this work was to use probabilistic diffusion tractography to examine the organization of the human insular cortex based on the similarities of its remote projections. Forty right-handed healthy subjects (33.8 ± 12.7 years old) with no history of neurological injury were included in the study. After the spatial standardization of diffusion tensor images, insular cortical masks were delineated based on the Harvard–Oxford Cortical Atlas and were used to initiate fibertracking. Cluster analysis by the k-means algorithm was employed to partition the insular voxels into two groups that featured the most distinct distribution of connections. In order to perform volumetric comparisons, the assigned label maps were transformed back to space of the subjects’ native anatomical MR images. The outlines of the change in connectivity profile did not respect the known cytoarchitectural subdivisions and were shown to be independent from the gyral anatomy. Interhemispheric asymmetry in the volumes of connectivity-based subdivisions was observed putatively marking a leftward functional dominance of the anterior insula and its reciprocally interconnected targets which influences the size of insular area where similar connections are represented. The fractional anisotropy values were not significantly different between the hemipsheres or connectivity-based clusters; however, the mean diffusivity was higher in the anterior insula in both hemispheres.

Keywords

Insula of reilCerebral cortexDiffusion tensor imagingDiffusion magnetic resonance imaging

Introduction

The insula of Reil, located deeply within the lateral sulcus, is known to have a multifaceted sensory, motor, visceral and cognitive role and is also considered as a vestibular association area. The insular cortex is acknowledged as the anatomical representation for interoceptive awareness, i.e., the “sense of the physiological condition of the body” (Craig 2002). Its functional and anatomical diversity has been described in humans and non-human primates (Augustine 1985), with changes in cytoarchitecture that follow a rostroventral to dorsal and posterior gradient, from agranular to dysgranular and granular cortex (Augustine 1996; Mesulam and Mufson 1982). In humans, functional neuroimaging studies by means of resting-state functional MRI (fMRI) have been recently used to reinvigorate the relationship of morphology and function of the insula by demonstrating consistent changes of patterns of activation or functional connectivity (Cauda et al. 2011). Diffusion-weighted and diffusion tensor imaging offer remarkable possibilities to explicate the properties of the hindered biological diffusion (Basser and Pierpaoli 1996) while tractography depict structural connectivity within distinct brain regions (Westin et al. 2002). More complex models describe the propagation of diffusion within the gray matter (Behrens et al. 2003a). Diffusion-based techniques provide the possibility to parcellate the gray matter according to its local diffusion properties (Ziyan and Westin 2008), either by quantifying the connection strengths to predefined cortical areas (Behrens et al. 2003b), or more generally, to compute similarities between connections to remote areas (Johansen-Berg et al. 2004). This technique has already been applied to segment the insula (Nanetti et al. 2009; Cerliani et al. 2011).

Our report aims to describe two regions within the left and right insula that are defined by clustering insular image voxels based on their distant cortical connections. Furthermore, our observations provide evidence for interhemispheric variability of the clusters in terms of spatial location, overall volume and micro-structural properties of diffusion.

Materials and methods

Subjects

Forty right-handed healthy subjects (19 males; 33.8 ± 12.7 years old) with no history of neurological injury participated in the study. All participants gave informed written consent to procedures approved by the Institutional Review Board.

Data acquisition

All images were obtained using a 1.5T MRI system (Siemens, Erlangen, Germany) equipped with an eight channel phased array head coil. Diffusion weighted (DWI) data were acquired using a single-shot pulsed gradient spin echo EPI sequence (TR = 10000 ms, TE = 118 ms); 12 different directions of diffusion-encoding magnetic gradients were used, with a b-value of 1000 s/mm2. Each volume consisted of 55 transverse slices, slice thickness 3 mm, voxel size: 0.97 × 0.97 mm. A three-dimensional high resolution anatomical scan of the whole brain (17.2 × 23 cm FOV, 384 × 512 matrix, 0.45 × 0.45 × 0.83 mm voxels) was acquired with a T1-weighted MPRAGE sequence.

Image processing

To characterize the intra-voxel properties of diffusion, probability distributions of fiber directions were analyzed for each voxel using the FMRIB Diffusion Toolbox in the FSL software package (University of Oxford). This processing step allowed to perform probabilistic diffusion tractography that estimates the pathways passing through any given voxel. Due to the number of diffusion-encoding gradients, we judged to calculate a model with one fiber population per voxel.

Spatial standardization was performed for the DTI and 3DT1 images to allow comparison of results across subjects. Non-linear registration of the subjects’ fractional anisotropy (FA) images and a FA template (FMRIB58, distributed with FSL) image was achieved by executing the FNIRT algorithm of the FMRIB’s FSL package. Subsequently, anatomical images and tractography results were stored in a standard MNI152 neuroimaging space.

Left and right insular masks of the entire insular cortex were defined by utilizing the Harvard-Oxford cortical atlas. A T1-weighted image template in the MNI152 space was used to review and refine the borders of the insular cortex. The final region of interest (ROI) comprised only the band of gray matter voxels surrounded by the extreme capsule and the periinsular sulci.

Parcellation of the insula

Tractography was initiated from the reviewed insular cortical masks in the MNI152 space. The connection strength between each “seed” voxel and every remote brain voxel was estimated as the probability of tracts reach their target through a trajectory guided by the model of local diffusion characteristics. A non-linear registration was used to map the coordinates of seed voxels to the space of the diffusion images and then to project the tractograms back to the standard space. Similarly to a number of studies on connectivity-based segmentation, the connection probability estimates were stored in an M * N array (M: seed voxels, N: low-resolution brain voxels), a correlation metric between each row of the array was determined (i.e. the similarity of remote connection patterns) and cross-correlations were expressed in an M * M matrix (Klein et al. 2007; Jbabdi et al. 2009). Cluster analysis by the k-means algorithm was utilized to partition the M seed voxels into two groups that featured the most distinct distribution of connections. It has been shown that the segmentation of the human cortex based on patterns of connectivity is strongly influenced by the utilized clustering method; Nanetti et al. suggested to use a repeated k-means algorithm (optimal times of repetition: 256). Our method is based on a random initialization; it is followed by searching for the next cluster which is the furthest away from the previous, this is repeated until K number of centers are found. This way of initialization allows to robustly determine identical clusters and multiple repetition is not justified. Finally, cluster membership labels were mapped back to the reference space for each subject using the inverted spatial transformation of the standardization procedure.

Analysis and visualization of connectivity-based subdivisions

Based on the contours of each connectivity cluster, three-dimensional objects were formed and for each model the center-of-gravity point (COG) was determined in the MNI152 stereotactical space. Spatial consistency was measured by the scatter of COGs and the variability of cluster volumes. For volumetric comparisons, each insular volume and its clustered partitions were transformed back to the subject’s native anatomical space by using the inverted transformation of the standardization step (i.e. registration of the MPRAGE T1 images to the MNI152 T1 template). A population-averaged representation of the partitioning was computed by assigning the label value to each reference-space insular voxel that was most likely to be found in the individual cluster maps, i.e. the mode of the 4D object was determined. This 3D dataset of the most common cluster assignment was then used to create a 3D mesh for each cluster. To demonstrate the intersubject variability of the discovered clusters, we calculated images representing the 95th, 90th, 50th, 10th and 5th percentiles of the label assignments. Our secondary goal was to evaluate whether the connectivity-based partitions present different diffusion microenvironment; to achieve this, the fractional anisotropy (FA) and mean diffusivity (MD) was provided for the anterior and posterior clusters in both hemispheres; the calculation was done using the standard equations provided elsewhere (Basser and Pierpaoli 1996). An erosion with a 3 × 3 × 3 voxel box kernel was performed on the cluster masks; we assume that this operation reduced the influence of partial volume effect by the adjacent corticospinal fluid voxels.

Visualization of fiber tract anatomy from the connectivity-based insular subdivisions

To reveal the anatomical correspondence of the fiber tracts emerging from the newly defined insular subdivisions, we performed probabilistic tractography for each subject, initiated from the voxels representing the discovered insular clusters. For each voxel in the brain in each subject, a label was assigned indicating whether that voxel is most likely to be connected to the first cluster (label:1), second cluster (label:2) or no connections to the insula (label:0). Corresponding maps were summed over the subjects and the resulting back-projected tract distributions were visualized (i.e. separately computing maps projected from the 1st and 2nd clusters), in the same way as represented in a figure by Menke et al. (2010).

Results

Based on the variability of remote connections, two insular subregions were successfully identified in both hemispheres. The stereotactical coordinates of the COGs were consistent, with low deviation from the mean coordinates of each cluster (Table 1). The population-averaged cluster map was controlled for correspondence to major anatomical landmarks (Fig. 1) and cytoarchitectonic map of the human insula (Fig. 2). Anterior (AI) and posterior divisions (PI) were depicted, the former extending to approximately one-third of the antero-posterior (AP) length of the insula, delimited by a curved plane perpendicular to the AP axis. In both hemispheres, the AI comprised the limen of the insula and the anterior short gyri enclosed by the precentral insular sulcus; this partition also included the antero-ventral part of the long insular gyri. The connectivity-based outlines did not present a clear correspondence to the cytoarchitectonic subdivisions; however, a match between the AI and the agranular subdivision plus antero-dorsal dysgranular area is observed. The 3D visualization of intersubject variability in cluster volumes is available in the Supplementary Fig. 1.
Table 1

Stereotactical coordinates (mm) of the connectivity-based insular clusters in the MNI152 space

 

Left insula

Right insula

Anterior cluster

Posterior cluster

Anterior cluster

Posterior cluster

X

−35.33 ± 0.36

−38.11 ± 0.40

37.34 ± 0.49

38.67 ± 0.38

Y

11.72 ± 2.27

−8.71 ± 2.45

10.58 ± 3.88

−5.79 ± 3.01

Z

−3.83 ± 1.47

4.02 ± 1.20

−4.20 ± 3.73

3.77 ± 2.07

https://static-content.springer.com/image/art%3A10.1007%2Fs10548-011-0205-y/MediaObjects/10548_2011_205_Fig1_HTML.gif
Fig. 1

Connectivity-based clusters of the human insular cortex. Top row: sagittal T1-weighted MR image overlaid with the connectivity-based cortical subdivisions averaged through 40 subjects. Black outline: posterior insula (PI), white outline: anterior insula (AI). Bottom row: three-dimensional mesh representing the averaged anterior (white) and posterior insular connectivity clusters, overlay: major insular sulci. Rectangles: center-of-gravity points of the AI subdivision of individual subjects in the standard stereotactical space, crosses: center-of-gravity of the PI subdivision of individual subjects

https://static-content.springer.com/image/art%3A10.1007%2Fs10548-011-0205-y/MediaObjects/10548_2011_205_Fig2_HTML.gif
Fig. 2

Correspondence between insular anatomy and various subdivision approaches and imaging methods. Top left: clustering based on similarities of structural connectivity (DTI), 2-way clustering. Top right: clustering based on similarities of structural connectivity (DTI), 3-way clustering. Bottom left: three systems of functional connectivity identified using fMRI (schematic drawing based on the paper by Deen et al. 2011). Bottom right: major cytoarchitectural domains of the human insula. aps anterior periinsular sulcus, sis short insular sulcus, pcis precentral insular sulcus, cis central insular sulcus, pis postcentral insular sulcus, sps superior periinsular sulcus, ips inferior periinsular sulcus, AI, PI anterior, posterior insula, MI dorsomedial insula (in 3-way clustering), vAI ventro-anterior insula, dAI dorso-anterior insula, Ia agranular, Id dysgranular, Ig granular insula, G hypergranular subdivision, VENs von Economo neurons. Image credits: Bottom right: Unpublished work by Gallay et al. (2011, with permission)

The total volume of insular gray matter was not different between the hemispheres. In the left hemisphere, the anterior division of the insular gray matter was found to be significantly larger than the posterior cluster (difference: 34.5%) while on the right side the two partitions were equal in volume. This asymmetry, as expressed by the AI volume to PI volume ratio, showed significant interhemispheric differences.

For each connectivity cluster, the scalar properties of intra-voxel diffusion were determined. The degree of diffusion anisotropy, as expressed by the FA value did not present significant interhemispheric variability and neither was different measured on the AI/PI clusters in the left hemisphere. The FA value of the right AI was significantly higher than the PI. In both hemispheres, the mean diffusivity was consistently and significantly larger in the anterior connectivity partition. Cluster volumes and regional diffusion properties are summarized in Table 2.
Table 2

Volumes, volume ratios of the connectivity-based insular clusters in the subjects’ native space; regional values of diffusion characteristics (MEAN ± SD)

 

Left insula

Right insula

Anterior cluster

Posterior cluster

Total

AI/PI ratio

Anterior cluster

Posterior cluster

Total

AI/PI ratio

Volume (mm3)

3912 ± 946

3466 ± 1027

7378 ± 951

1.35 ± 0.9

3560 ± 719

3848 ± 842

7409 ± 828

0.99 ± 0.38

Sig. of antero-posterior volume difference

P = 0.047

  

P = 0.104

  

Sig. of left/right volume asymmetry

P = 0.065

P = 0.073

P = 0.879

P = 0.027

    

Fractional anisotropy

0.181 ± 0.01

0.178 ± 0.02

0.18 ± 0.01

1.02 ± 0.09

0.182 ± 0.02

0.174 ± 0.01

0.18 ± 0.01

1.05 ± 0.09

Sig. of antero-posterior FA difference

P = 0.382

  

P = 0.01

  

Mean diffusivity (*10−3 mm/s2)

1.13 ± 0.1

0.97 ± 0.06

1.05 ± 0.08

1.17 ± 0.14

1.17 ± 0.15

0.932 ± 0.05

1.052 ± 0.08

1.26 ± 0.16

Sig. of antero-posterior MD difference

P < 0.001

  

P < 0.001

  
In both hemispheres, tract distributions from the AI revealed connections with the temporal and occipital lobe, opercular and orbitofrontal cortex, triangular and opercular parts of the inferior frontal gyrus. The density of pathways approaching the orbitofrontal and inferior frontal cortex appeared larger in left hemisphere. The PI subdivision showed extensive connections to the parietal and frontal lobes, predominantly to parts of the somatosensory, motor and premotor cortices adjacent to the midline. An overlap of AI/PI connections in the occipital lobe was noted. Cingular and thalamic connections of each connectivity-based cluster were only observed in a small number of cases; the AI connections projecting to the MD nucleus and the PI reaching the ventrolateral thalamus. Images showing the major domains of connections emerging from each insular subdivision are demonstrated in Fig. 3.
https://static-content.springer.com/image/art%3A10.1007%2Fs10548-011-0205-y/MediaObjects/10548_2011_205_Fig3_HTML.gif
Fig. 3

Cross-sectional images demonstrating the fiber tract anatomy of the connectivity-based insular subdivisions, overlaid on a standard neuroimaging MRI template (MNI152). Connection probability estimates (n = 40, averaged) from the anterior and posterior insular subdivision depicted in three different sagittal images

Discussion

Early endeavors to map the human cortex, such as works by K. Brodmann and Von Economo (1925), discovered a limited agreement between macroscopic structures of the brain (gyral or sulcal anatomy) and the organization defined by fine microstructural features like the cytoarhitecture. The same ambiguity is present when attempting to define cortical areas based on similarities in activation patterns when performing executive or cognitive tasks, evidence comes from a large number of neuroimaging studies employing functional MRI. Revitalized by tractography and related techniques, the hodological (i.e. connectionist) approach identifies cortical regions that receive or send out similar connections (Johansen-Berg et al. 2004; Klein et al. 2007; Jbabdi et al. 2009). Such parcellations potentially generate interest by exploring the human connectional neuroanatomy, nonetheless facilitating the understanding of the cortical representations of major neurocognitive networks.

We used diffusion tractography data to reveal changes in insular connectivity profile by executing a k-means clustering method that labels adjacent areas based on the similarities in the distribution of remote connections. As an initial hypothesis, the algorithms were forced to search for two segments in the insular gray matter. The impetus for this assumption was that most studies on functional connectivity utilizing resting-state fMRI (r-fMRI) predominantly described a twofold division of the insula into an anteroventral and posterodorsal cluster (Cauda et al. 2011), however, a threefold functional parcellation was also suggested (Deen et al. 2011). Parcellating the insula based on diffusion tensor tractography demonstrated a gradual change in tractography patterns with a rostrocaudal trajectory (Cerliani et al. 2011).

Insular domains with similar structural connectivities

We conclude that the DTI-based segmentation greatly overlaps with the same depictions of studies using fMRI (Cauda et al. 2011; Deen et al. 2011); and it is noteworthy that connections of the ventral part of the long insular gyri and the anterior short insular gyri are similar, this coherence was more pronounced on the structural connectivity segmentations where a larger proportion of the long gyri were included in the area denoted as anterior insula (AI). In addition to the two-way segmentation described in this work, we performed segmentation of the insula cortex into three regions, similarly to the work by Deen et al. (2011). This experiment also demonstrated a rostrocaudal difference in connectivity patterns, marking an anterior, dorsomedial and dorsal-posterior cluster (Supplementary Fig. 2.). There is evidence from primates (Augustine 1985; Augustine 1996; Mesulam and Mufson 1982; Gallay et al. 2011) and humans that the anterior insula presents a significantly different cytoarchitecture as well as afferent and efferent connectivity than the posterior division. The AI, as defined by its connections, embodied the agranular and part of dysgranular insula which is known to be interconnected to the frontal, orbitofrontal cortex (Carmichael and Price 1994) and the amygdaloid body in macaque monkeys (Ray and Price 1993; Stefanacci and Amaral 2000).

Connectivity data of the human insula is relatively sparse and limited to observations from r-fMRI measurements or depictions of anatomical connectivity by means of diffusion tensor tractography. A study by Cauda et al. (2011) concluded that the ventralmost anterior insula is functionally interconnected (i.e., shows temporal correlation of activation patterns) to the rostral anterior cingulate cortex, middle and inferior frontal cortex and the temporoparietal cortex while the dorsal posterior insula is connected to various cortical targets like the dorsal-posterior cingulate, premotor, supplementary motor, temporal and occipital cortex. While tract tracing studies from primates describe complex, region-specific thalamic projections to both the anterior and posterior insula (Guldin and Markowitsch 1984), a human r-fMRI study conflicts with such observations by reporting less pronounced or non-existing posterior insular connection with the thalamus (Cauda et al. 2011). When controlling the fiber tract anatomy from each connectivity cluster in both hemispheres, we discovered tracts anatomy similar to those of r-fMRI studies.

We identified two major limitations in our study protocol. The number of diffusion-weighting directions only allowed to estimate one fiber population per voxel, this inherently affects the result of probabilistic tracking of connections. While it is generally acknowledged that brain voxels tend to have multiple fiber directions (e.g. as crossing-fibers), a study on the added-value of multi-orientation models concluded that secondary fibers become less important when performing connectivity-based segmentations; e.g., in the thalamus (Behrens et al. 2007). We confirm that the change of connectivity patterns in the insula follows a anterorostral-caudal trajectory which is gradual, this would require more sophisticated algorithms to reproducibly define connectivity-based subdivisions.

Laterality of connection clusters

Our findings imply that connections of the anterior insula have larger leftward representation relative to the total insular gray matter volume; this leftward dominance of prefrontal and frontal connections coincides with the observations on forebrain asymmetry (Craig 2005) or lateralization of prefrontal activations, the latter already described as a biomarker for ingestive behaviour (Ochner et al. 2009). Our result of a larger connectome of the left AI partially conflicts with the observation by Cauda et al. (2011) where it was shown that the anterior cluster is rightward dominant; however, this type of interhemispheric dominance is marked by the strength (degree of temporal correlation) of functional and not structural connectivity. In contrast to a previous study using high-resolution MRI scans to assess the structural asymmetries (Watkins et al. 2001), we reported no interhemispheric differences of the overall volume of insular gray matter. It is disputable whether our findings of larger left anterior cluster can be attributed to the structural asymmetry of the AI shown by Watkins et al., as our interpretation of the anterior division was independent of gyral anatomy. Further support for our results on interhemispheric differences comes from a study by Cao et al. (2003) which demonstrated a marked L-R asymmetry of anisotropy (i.e. the “orderliness” of diffusion) of the subinsular white matter, implying a putative interhemipsheric asymmetry in the trajectory or density of pathways emerging from or projecting to the insula. The higher mean diffusivity, which is a directionless descriptor of the magnitude of diffusion, hallmarks different water microenvironment in the anterior insula albeit not displaying pronounced interhemispheric variability.

Conclusion

Structural connectivity-based parcellation of the insular cortex reveals a markedly larger anterior connectivity cluster in the left hemisphere. This implies a leftward functional dominance of the anterior insula and its reciprocally interconnected targets which inherently influences the size of cortical area where similar connections are represented.

Acknowledgements

The authors gratefully acknowledge the technical support of Saad Jbabdi (Centre for Functional Magnetic Imaging of the Brain, University of Oxford).

Supplementary material

10548_2011_205_MOESM1_ESM.tiff (577 kb)
Supplementary Fig. 1. Intersubject variability of DTI connectivity-based clustering of the insular cortex. 3D surfaces were generated by accessing the 95th, 90th, 50th, 10th and 5th percentile volumes of each cluster assignment across the population (n = 40). Major anatomical landmarks have been illustrated (for description, see Fig. 3.). (TIFF 576 kb)
10548_2011_205_MOESM2_ESM.tiff (785 kb)
Supplementary Fig. 2. Three-way segmentation of the insula based on its structural connectivity patterns. Visualization with a cross-sectional image of the MNI152 T1-weighted image template and 3D surfaces generated using the label asssignments. Major anatomical landmarks have been illustrated (for description, see Fig. 3.). Red object: anterior insula (AI’), teal object: dorsomedial insula (MI), blue object: posterior insula (PI). (TIFF 784 kb)

Copyright information

© Springer Science+Business Media, LLC 2011