Parcellation of left parietal tool representations by functional connectivity

Abstract

Manipulating a tool according to its function requires the integration of visual, conceptual, and motor information, a process subserved in part by left parietal cortex. How these different types of information are integrated and how their integration is reflected in neural responses in the parietal lobule remains an open question. Here, participants viewed images of tools and animals during functional magnetic resonance imaging (fMRI). k-Means clustering over time series data was used to parcellate left parietal cortex into subregions based on functional connectivity to a whole brain network of regions involved in tool processing. One cluster, in the inferior parietal cortex, expressed privileged functional connectivity to the left ventral premotor cortex. A second cluster, in the vicinity of the anterior intraparietal sulcus, expressed privileged functional connectivity with the left medial fusiform gyrus. A third cluster in the superior parietal lobe expressed privileged functional connectivity with dorsal occipital cortex. Control analyses using Monte Carlo style permutation tests demonstrated that the clustering solutions were outside the range of what would be observed based on chance 'lumpiness' in random data, or mere anatomical proximity. Finally, hierarchical clustering analyses were used to formally relate the resulting parcellation scheme of left parietal tool representations to previous work that has parcellated the left parietal lobule on purely anatomical grounds. These findings demonstrate significant heterogeneity in the functional organization of manipulable object representations in left parietal cortex, and outline a framework that generates novel predictions about the causes of some forms of upper limb apraxia.

Keywords: Apraxia; Functional connectivity; Manipulable objects; Parietal cortex; fMRI.

Figure 1

Regions showing Tool Preferences in a whole-brain analysis (random effects). The regions identified included the left parietal lobule, the left dorsal occipital cortex, the left and right medial fusiform gyrus, the left posterior middle/inferior temporal gyrus, and the left ventral premotor cortex—all replicating previous findings (see text for discussion).

Figure 2

K-means clustering solution. A. The left medial fusiform gyrus was used as a seed from the ventral visual pathway. The cluster intersection map projected on the inflated surface of the left hemisphere is shown with the resulting clusters: Inferior Parietal ROI (green), the Intraparietal Sulcus ROI (red), and the Superior Parietal ROI (blue). The results of the ROI-based functional connectivity analysis indicate that the left Inferior Parietal ROI expressed privileged functional connectivity with ventral premotor cortex, the left Anterior Intraparietal Sulcus ROI expressed privileged functional connectivity with the left medial fusiform gyrus seed, and the left Superior Parietal ROI expressed privileged functional connectivity with left dorsal occipital cortex. B. The left posterior middle temporal gyrus was used as a seed from the ventral visual pathway. The cluster intersection map is projected on the inflated surface of the left hemisphere. In the ROI-based functional connectivity analysis, the left Inferior Parietal ROI expressed privileged functional connectivity with the left ventral premotor cortex, the left Intraparietal Sulcus ROI expressed privileged functional connectivity with the left posterior middle temporal gyrus, and the left Superior Parietal ROI expressed privileged functional connectivity with the left dorsal occipital cortex. (p < .05; ** p < .01; *** p < .001). Error bars plot the standard error of the mean, across participants.

Figure 3

Euclidean distance differences for simulated and ‘real’ clusters within left parietal cortex. A. The histograms plot the distribution of between-cluster and within-cluster Euclidean distances for analyses over randomly shuffled data (see also Supplemental Figure 2 for further analyses involving the simulated data). B. In order to have a principled way of comparing the Euclidean distance difference between the within- and between-cluster voxel pairings for the ‘real’ data, we pooled all within-cluster Euclidean distances (collapsing across clusters) into one distribution (blue), and we pooled all between-cluster Euclidean distances (collapsing across clusters) into a second distribution (red). We fit normal distributions to the histograms in order to show the magnitude of the size difference between the within- and between-cluster Euclidean distance distributions in the ‘real’ clusters. C. The mean magnitude difference among all within- and between-cluster voxel pairs are plotted for ‘real’ and simulated data. Because ‘real’ clusters were of different sizes, we took the average within-cluster distance, and subtracted that value from two separate between-cluster values in order to obtain a mean Euclidean distance difference (i.e., to calculate the mean Euclidean distance, we computed all pairwise Euclidean distances among voxels in the three parietal clusters, then subtracted the mean distance among all voxels in different clusters from the mean distance among all voxels in the same cluster. The same approach was taken for the simulated data. Error bars represent standard deviation over the three mean Euclidean distance difference scores.

Figure 4

Comparison of Caspers’ and colleagues’ anatomical parcellation scheme, Ruschel and colleagues’ anatomical parcellation scheme, and the parietal clusters we have reported. A. Overlap between our parietal clusters and the clusters reported by Caspers and colleagues (2013). The Inferior Parietal ROI partially overlaps with areas PFt and PF; the Intraparietal Sulcus ROI minimally overlaps with area PF; the Superior Parietal ROI does not overlap with any of the anatomically-based inferior parietal clusters, but likely overlaps with regions IPS3 and/or IPS4 from Konen and Kastner (2008). B. Dendrograms representing the similarity in functional connectivity among the anatomically-derived clusters of Caspers and colleagues and the left parietal clusters we reported (left panel), and among the DTI-based clusters of Ruschel and colleagues and the left parietal clusters we reported (right panel).