Probabilistic fibre tract analysis of cytoarchitectonically defined human inferior parietal lobule areas reveals similarities to macaques

Abstract

The human inferior parietal lobule (IPL) is a multimodal brain region, subdivided in several cytoarchitectonic areas which are involved in neural networks related to spatial attention, language, and higher motor processing. Tracer studies in macaques revealed differential connectivity patterns of IPL areas as the respective structural basis. Evidence for comparable differential fibre tracts of human IPL is lacking. Here, anatomical connectivity of five cytoarchitectonic human IPL areas to 64 cortical targets was investigated using probabilistic tractography. Connection likelihood was assessed by evaluating the number of traces between seed and target against the distribution of traces from that seed to voxels in the same distance as the target. The main fibre tract pattern shifted gradually from rostral to caudal IPL: Rostral areas were predominantly connected to somatosensory and superior parietal areas while caudal areas more strongly connected with auditory, anterior temporal and higher visual cortices. All IPL areas were strongly connected with inferior frontal, insular and posterior temporal areas. These results showed striking similarities with connectivity patterns in macaques, providing further evidence for possible homologies between these two species. This shift in fibre tract pattern supports a differential functional involvement of rostral (higher motor functions) and caudal IPL (spatial attention), with probable overlapping language involvement. The differential functional involvement of IPL areas was further supported by hemispheric asymmetries of connection patterns which showed left–right differences especially with regard to connections to sensorimotor, inferior frontal and temporal areas.

Introduction

The human inferior parietal lobule (IPL) is a heterogeneous, multimodal brain region as demonstrated by functional neuroimaging and lesion mapping studies. Thus, different parts of human IPL seem to be involved in different functional brain networks, where they interact with different other cortical regions within frontal, occipital, and temporal lobe.

Rostral IPL areas bilaterally seem to be involved in higher motor functions, potentially including parts of a human mirror neuron system (Rizzolatti and Craighero, 2004; Iacoboni, 2005; Keysers and Gazzola, 2009; Caspers et al., 2010). The caudal IPL, in contrast, was shown to feature hemisphere-specific functionality. The right IPL is recruited during spatial and non-spatial attention and motor preparation tasks and conceptualised as part of the “ventral attention network” (Fink et al., 2001; Corbetta and Shulman, 2002; Corbetta et al., 2008; Jakobs et al., 2009). Contrastingly, its left counter-part is thought to form Geschwind’s area in the language network, being mainly involved in semantic and phonological processing (Geschwind, 1970; Price, 2000; Gernsbacher and Kaschak, 2003; Vigneau et al., 2006).

In macaque monkeys, electrophysiological recordings have shown evidence of a comparable functional segregation of the IPL as in humans (apart from language processing). Rostral IPL areas in this species have been shown to contain mirror neurons and participate in sensorimotor processing, whereas caudal areas are mainly involved in functions such as spatial attention, auditory-sensory integration, and visuo-motor coordination, e.g., grasping (Hyvärinen, 1982; Pandya and Seltzer, 1982; Seltzer and Pandya, 1984; Rozzi et al., 2006).

In both species, this functional heterogeneity is reflected on a cytoarchitectonic level. In humans, a recent study delineated seven cytoarchitectonically distinct areas within the IPL (Caspers et al., 2006, 2008). Five of these cover the lateral surface of the IPL in a rostro-caudal sequence (Fig. 1A). The remaining two are located in the Sylvian fissure. Comparably, the macaque IPL has been reported to consist of six main areas (Pandya and Seltzer, 1982; Gregoriou et al., 2006) . Of these, four are located on the lateral surface in a rostro-caudal sequence (Fig. 1B), the other two in the Sylvian fissure.

Fig. 1.

(A) 3D reconstruction of maximum probability maps (MPM; Eickhoff et al., 2006a) of the cytoarchitectonic areas (PFop, PFt, PF, PFm, PFcm, PGa, PGp) defined within the human IPL (Caspers et al., 2006, 2008), displayed on a lateral surface view of the MNI single subject template. (B) Schematic depiction of the six areas (PF, PFG, PG, Opt, PFop, PGop) of the macaque IPL, adopted from Pandya and Seltzer (1982) and Gregoriou et al. (2006). ce central sulucs, ias inferior arcuate sulcus, ips intraparietal sulcus, lf lateral fissure, lu lunate sulcus, poc postcentral sulcus, prc precentral sulcus, ps principal sulcus, sas superior arcuate sulcus, sts superior temporal sulcus.

Tracer studies of axonal connectivity in macaques have provided a potential link between structural heterogeneity and functional diversity of the IPL by revealing a differentiated connectivity pattern of the cortical areas in this region. Rostral areas (PF, PFG) show strong reciprocal connections to (pre-) motor, somatosensory and superior parietal areas. In contrast, caudal areas are mainly connected to higher visual areas within occipital and inferior temporal cortex (Cavada and Goldman-Rakic, 1989a,b; Felleman and Van Essen, 1991; Gregoriou et al., 2006).

In humans, the anatomical connectivity of individual IPL areas is largely unknown, although macroanatomical fibre preparations and studies on disconnection syndromes such as apraxia (Freund, 2003; Culham and Valyear, 2006), spatial neglect (Karnath, 2001; Hillis, 2006; Husain and Nachev, 2007) or aphasia (Dronkers et al., 2004) support the idea of a similar connection pattern for humans as in macaques. Diffusion tensor imaging (DTI) studies in healthy humans were indeed able to show partly different connectivity of different aspects of the IPL. Makris et al. (2005) found a partition of one of the two main fibre pathways connecting the IPL with mainly frontal regions, i.e. the superior longitudinal fascicle (SLF) which they could subdivide into four distinct parts, two of which running into the rostral and caudal aspect of the IPL, respectively. Catani et al. (2005) found a comparable partition for the other main pathway, i.e. the arcuate fascicle. They demonstrated that different parts of the arcuate fascicle reach either the rostral or the caudal aspect of the IPL, comparable to the SLF as reported by Makris et al. (2005). Focusing on a possible subdivision of the parietal cortex by means of connectivity based parcellation, Rushworth et al. (2006) showed that rostral IPL is more likely to connect to ventral premotor cortex whereas caudal IPL was more likely to connect with the parahippocampal gyrus. A recent study by Mars et al. (2011) used connectivity-based parcellation of the IPL, resulting in a comparable subdivision of this region as found by cytoarchitectonic parcellation (Caspers et al., 2006, 2008). Consecutive resting-state functional connectivity analyses showed how the IPL areas were differentially connected to premotor, prefrontal and parahippocampal areas (Mars et al., 2011). These studies provide first hints that the fibre tract pattern of human IPL is different in its many parts, at least in rostral and caudal IPL.

But it can be assumed that the functional and cytoarchitectonic heterogeneity of the IPL is also reflected by a more differentiated fibre tract pattern than a bipartition. In order to provide a precise identification of areal-specific fibre tract pattern as structural basis for the involvement in different functional networks, we assessed the fibre tracts related to five cytoarchitectonic areas of the lateral IPL using probabilistic tractography based on DTI data.

Material and methods

Data acquisition

We acquired diffusion-weighted data from 40 healthy, right-handed human subjects (20 males, mean age ± SD = 28.65 ± 5.73, range 21–42; 20 females, mean age±SD = 28.75±6.20, range 21–42) on a 3.0 T Tim-Trio Siemens whole-body scanner (Siemens, Erlangen, Germany) with a maximum gradient strength of 40 mT m⁻¹, using a 12-channel phased-array head coil for signal reception. Subjects had no history of neurological or psychiatric disease, or head injury. All subjects gave informed, written consent to participate in the study which was approved by the local Ethics Committee of the RWTH Aachen University.

Diffusion-weighted images were acquired using a twice-refocused spin-echo sequence (axial slices, slice thickness: 1.8 mm, number of slices: 75, matrix = 128×128, field of view = 230×230 mm², band-width = 1502 Hz/pixel, reconstruction using an iPAT GRAPPA-scheme, final voxel resolution of 1.8×1.8×1.8 mm³). The diffusion sensitive gradients were distributed along 60 directions in an icosahedral scheme (Jones et al., 1999). For each set of diffusion-weighted data, 60 volumes with b-value = 800 s/mm² and 7 volumes with b-value = 0 s/mm² were obtained. For each subject, the entire diffusion measurement was repeated four times in successive sessions for subsequent averaging. The total scan time for the diffusion-weighted imaging protocol was about 50 min. A scanning protocol with special focus on quality protection of the data was applied, in particular with respect to the prevention of head motion. The four data sets for each subject were obtained in two 6-minute runs each which minimised motion within a scan session as subjects could relax in-between. Head fixation was carried out using foam paddles at either side of the subjects’ head. Light in the scanner room was dimmed to avoid visual distraction of the subjects during scanning. Quality of scanning in general was checked by routinely applied phantom measurements with retrieval of signal and noise change measures within the acquired images and the background to assure stability of the scanning protocol over time.

Definition of seed and target areas

The five cytoarchitectonic areas on the lateral surface of the IPL (in rostro-caudal sequence: PFt, PF, PFm, PGa, PGp; Caspers et al., 2006, 2008) were used as seed areas. Definition of the seed volumes was based on probabilistic cytoarchitectonic maps (Caspers et al., 2008) as implemented in the Düsseldorf-Jülich histological atlas (JHA; Zilles and Amunts, 2010) using the SPM Anatomy Toolbox (http://www.fz-juelich.de/inm/spm_anatomy_toolbox; Eickhoff et al., 2005). The rest of the cerebral cortex was subdivided in different targets, again based on the JHA where available. For those regions that have not yet been cytoarchitectonically defined, we used the macroscopic probabilistic Harvard–Oxford cortical structural atlas (HOCSA) as included in FSL (Oxford Centre for Functional Magnetic Resonance Imaging of the Brain, http://www.fmrib.ox.ac.uk/fsl/data/atlas-descriptions.html) for target definition. In total, 64 target areas were defined in each hemisphere (Table 1), for which locations are schematically depicted in Fig. 2.

Table 1.

Target areas for connectivity analysis of IPL seed areas.

	Target area	Atlas	Labelling
Frontal lobe	Broca’s speech region, BA 44	JHA	BA 44
	Broca’s speech region, BA 45	JHA	BA 45
	Primary motor area 4a	JHA	4a
	Primary motor area 4p	JHA	4p
	Premotor cortex, BA 6	JHA	BA 6
	Frontal medial cortex	HOCSA	FMed
	Frontal operculum	HOCSA	FOperc
	Frontal orbital cortex	HOCSA	FOrbit
	Frontal pole	HOCSA	FPole
	Middle frontal gyrus	HOCSA	FMid
	Superior frontal gyrus	HOCSA	FSup
	Anterior cingulate cortex	HOCSA	CingAnt
	Posterior cingulate cortex	HOCSA	CingPost
	Paracingulate cortex	HOCSA	ParaCing
Primary somatosensory cortex (SI)	Area 1	JHA	1
	Area 2	JHA	2
	Area 3a	JHA	3a
	Area 3b	JHA	3b
Secondary somatosensory cortex (SII)	Area OP1	JHA	OP1
	Area OP2	JHA	OP2
	Area OP3	JHA	OP3
	Area OP4	JHA	OP4
Insula	Dysgranular insula Id1	JHA	Id1
	Granular insula Ig1	JHA	Ig1
	Granular Insula Ig2	JHA	Ig2
Parietal lobe	Superior parietal area 5Ci	JHA	5Ci
	Superior parietal area 5L	JHA	5L
	Superior parietal area 5M	JHA	5M
	Superior parietal area 7A	JHA	7A
	Superior parietal area 7M	JHA	7M
	Superior parietal area 7PC	JHA	7PC
	Superior parietal area 7P	JHA	7P
	Intraparietal area hIP1	JHA	hIP1
	Intraparietal area hIP2	JHA	hIP2
	Intraparietal area hIP3	JHA	hIP3
Occipital lobe	Visual area hOc1, BA 17	JHA	hOc1
	Visual area hOc2, BA 18	JHA	hOc2
	Visual area hOc3A	JHA	hOc3A
	Visual area hOc3d	JHA	hOc3d
	Visual area hOc3v	JHA	hOc3v
	Visual area hOc4	JHA	hOc4
	Visual area hOc5	JHA	hOc5
	Lateral inferior occipital cortex	HOCSA	OLatInf
	Lateral superior occipital cortex	HOCSA	OLatSup
	Lingual gyrus	HOCSA	Ling
	Occipital fusiform gyrus	HOCSA	OFusi
Temporal lobe	Auditory area TE1.0	JHA	TE1.0
	Auditory area TE1.1	JHA	TE1.1
	Auditory area TE1.2	JHA	TE1.2
	Auditory area TE3	JHA	TE3
	Inferior temporal gyrus, anterior part	HOCSA	TInfAnt
	Inferior temporal gyrus, posterior part	HOCSA	TInfPost
	Inferior temporal gyrus, temporooccipital part	HOCSA	TInfTempocc
	Middle temporal gyrus, anterior part	HOCSA	TMidAnt
	Middle temporal gyrus, posterior part	HOCSA	TMidPost
	Middle temporal gyrus, temporooccipital part	HOCSA	TMidTempocc
	Superior temporal gyrus, anterior part	HOCSA	TSupAnt
	Superior temporal gyrus, posterior part	HOCSA	TSupPost
	Temporal fusiform cortex, anterior part	HOCSA	TFusiAnt
	Temporal fusiform cortex, posterior part	HOCSA	TFusiPost
	Temporal occipital fusiform cortex	HOCSA	TOFusi
	Planum temporale	HOCSA	PT
	Temporal Pole	HOCSA	TPole
	Posterior parahippocampal gyrus	HOCSA	ParaHippoPost

Areal naming as provided in the atlases and respective publications. JHA: Jülich histological atlas, HOCSA: Harvard–Oxford cortical structural atlas. Labelling of JHA areas according to publications: Amunts et al. (1999, 2000), Choi et al. (2006),Eickhoff et al. (2006b,c),Geyer (2004), Geyer et al. (1996, 1999, 2000),Grefkes et al. (2001),Kurth et al. (2010),Malikovic et al. (2007), Morosan et al. (2001), Rottschy et al. (2007), Scheperjans et al. (2008a,b).

Fig. 2.

Visualisation of seed (green) and target areas (red) on the (A) lateral 3D reconstructed and (B) medial surface view of the MNI single subject template. For labelling conventions of areas, cf. Table 1.

Seeds and targets were defined separately for each hemisphere by using the centre of each region in a maximum probability map (MPM; Eickhoff et al., 2006a) included in the JHA and the HOCSA atlases. Centres were defined as the top 10% of the underlying probability that are also represented by the respective MPM. To perform diffusion-image analysis the ensuing area definitions had to be transformed from MNI standard space into the individual diffusion space (Eickhoff et al., 2010), being the reason why only the centre representation of the MPM has been chosen to avoid overlapping seed and target regions in individual diffusion spaces. Transformation fields were generated by normalising the mean b = 0 image from each subject to the MNI tissue probability maps (TPM) using a segmentation-based approach (Ashburner and Friston, 2005). By using the combination of a transformation from the MNI single subject template into the TPM space with the inverse transformation of the individual b = 0 image, we transformed seeds and targets from the template brain into the individual diffusion spaces. All diffusion-weighted images, including the b = 0 images, were spatially realigned. All 28 b = 0 images of each subject (7 b = 0 images of4 scanning sessions) were then averaged to provide an anatomical reference for consecutive registration of all diffusion-weighted images to the individual mean b = 0 image which allowed for a consecutive application of the transformation procedure to all diffusion-weighted images. The representations of seeds and targets in individual diffusion space were then mapped on the grey–white matter interface with restriction to fractional anisotropy (FA) values of at least 0.25. This procedure was necessary to provide sufficient directional information for the tracking algorithm to proceed. Within the grey matter, directional information is poorly available due to low anisotropy. In particular, transforming the regions of interest (seeds and targets) onto the grey–white matter interface resulted in mean FA-values of around 0.4 within the transformed region of interest, thus providing sufficient directional information for consecutive tracking. The grey–white matter interface was gauged based on segmentation of the average of the 28 b = 0 images for each subject by means of unified segmentation (Ashburner and Friston, 2005) which provided a robust and reliable basis for defining the interface between the two tissue classes grey and white matter. By adding an FA threshold of 0.25 to the grey matter segment, it was possible to only identify those grey matter voxels that were located close to the grey matter and construct of the respective interface.

Image analysis

Data were preprocessed using SPM 5 (Wellcome Department of Imaging Neuroscience, London, UK, http://www.fil.ion.ucl.ac.uk/) and FSL 4.1 (http://www.fmrib.ox.ac.uk/fsl). After correction for eddy currents and head motion, the four data sets (sessions) for each subject were averaged to improve signal-to-noise ratio. Data sets were additionally checked for motion artefacts or other acquisition-related quality compromises (ghosting, signal drop-outs etc). But due to a scanning protocol with focus on optimal acquisition of high quality data, only one data set had to be removed from further analysis, leaving 39 data sets which showed minimal motion and no other artefacts for consecutive analysis. Brain segmentation into three compartments (grey matter, white matter, cerebrospinal fluid (CSF)) was then performed on the averaged b = 0 images for each participant using the unified segmentation approach (Ashburner and Friston, 2005).

Diffusion probabilistic tractography was then performed using the Diffusion Toolbox FDT (version 4.0.4) implemented in FSL (Behrens et al., 2003a,b; Smith et al., 2004). Possible multiple fibre orientations in each voxel were estimated according to Behrens et al. (2007). The CSF-segment obtained from segmentation of the individual b = 0 images was used as exclusion mask for probabilistic tractography to avoid tracts passing through ventricles or sulci. The grey matter-white matter interface mask was used as stop mask for the tracking algorithm to stop after reaching a relevant target voxel. Otherwise, tracts might be prone to further proceed until they reach the CSF which would cause exclusion of this tract from further analysis.

Probabilistic tractography was carried out from each seed area using 100,000 samples for each seed voxel to improve signal-to-noise ratio and create a stable probability distribution. For each voxel in the brain, the number of samples (probabilistic tracts) passing through it was registered, together with the respective distance from the seed voxel along the tract. For visualisation purposes only, all individual tractography results were transformed into anatomical MNI space and then averaged across subjects.

Statistical analysis

Statistical analysis was performed using MATLAB (The Math-Works). To quantitatively compare the connection likelihood of the individual IPL areas with different targets in spite of the varying distances between them, we first recorded the number of traces (samples from probabilistic tractography) for all voxels within the entire mask of each target (Fig. 3A).

Fig. 3.

(A) Schematic overview of the statistical approach used for identifying terminal zone of a particular seed-target connection (for details, see text). S: Seed, T: Target, $\stackrel{-}{D}$: distance between seed and target voxels, exemplarily given for one highlighted voxel (yellow square in termination zone of target T), $\stackrel{-}{D}+1$, $\stackrel{-}{D}-1$: uncertainty of tract length around $\stackrel{-}{D}$, ${\stackrel{-}{F}}_{\mathit{ST}}$: defined mean frequency of traces between S and T. (B) Histogram showing an exemplary frequency distribution of traces running from a seed to one target with visualisation of the cutoff point for the identification of the termination zone. N_voxel (y-axis): number of voxels in the target region which featured a particular trace-count as indicated by N_Traces on the x-axis (following probabilistic tractography with 100,000 samples run from each seed voxel). (C) Reference distribution of traces derived from drawing a random sample of voxels at the same distance as the target voxels in the termination zone. N_voxel: number of voxels which showed a certain frequency of traces in the randomisation procedure to generate the reference distribution.

For the analysis of relevant traces, it has to be considered that cortical targets are irregularly shaped in three dimensional space due to the gyral and sulcal pattern of the respective region. Thus, the fibre tracts that can be assessed by DTI might terminate in only a small part of the target, i.e. that part of the target which is located closest to the seed. Moreover, the employed tractography analysis favours target voxels which are located close to the white matter as directional information is poor in more superficial grey matter due to low anisotropy values, even if DTI-based tractography is generally able to also reach voxels far within the grey matter.

Thus, most fibre tracts should predominantly reach the target in this kind of “termination zone”, due to technical and biological constraints. To the contrary, the majority of voxels showed lower frequencies of traces since these voxels were only rarely hit by incoming traces when running the probabilistic tractography (left part of Fig. 3B). Analysis of tract counts was consequently restricted to this termination zone, excluding voxels within the target that only reflect stray connectivity (i.e. voxels which were rarely hit due to probabilistic phenomena), but may severely bias further analysis.

By restricting analysis to that termination zone, those voxels within each target mask which were reached most consistently in the probabilistic tractography were identified. It has to be emphasised that this was performed separately for each seed-target combination, i.e., a particular target was allowed to have distinct termination zones for the different seed areas.

When examining the distribution of trace frequencies for the different seed-target combinations, i.e. how many voxels were reached by which number of traces (trace density), two patterns became evident (Fig. 3B). In most instances, one or a few clusters of target voxels showed considerably higher trace density than all others, indicative of a distinct termination zone. Algorithmically, voxels forming this termination zone were identified using a stepwise threshold decrease. Initially, the 10 voxels with the highest trace density were considered. This minimum cluster-size was introduced to reduce the effect of outliers and potential technical artefacts. The threshold for inclusion was then successively lowered until the voxel to be included next showed a trace density less than half as high as the median trace density of the voxels already included. In cases where no such break could be identified, the procedure was stopped after inclusion of 5% of all target voxels.

The number of traces passing through the different voxels in the termination zone was averaged to yield the trace density ${\stackrel{-}{F}}_{\mathit{ST}}$ representing the connectivity between the particular seed (S) and that target (T). When trying to compare these values across different seed-target combinations or when trying to interpret whether an observed F_ST is indicative of an anatomical connection, one faces two problems. First, as 100,000 traces originated from each (individual) seed voxel, the number of traces will increase with the size of the seed volume. This holds true for any given number of samples. A reliable comparison between different seed-target connections would therefore be aggravated. Thus, a correction for this potentially Biasing factor ‘seed size’ is needed. Second, the number of traces is also highly dependent on the distance (along the tract) between the seed and the target. In particular, the further away a target is located from a particular seed, the lower the trace count that it will reach the target, as the number of potential “wrong turns” increases. Hence, low density values for a distant seed-target pair may actually reflect stronger connectivity than high values for a proximate pair. In fact, stray samples close to the seed result in higher trace counts than “true” connectivity at a more distant location.

In order to correct for these two potential biases (i.e. seed size and distance), we referenced the observed mean trace density ${\stackrel{-}{F}}_{\mathit{ST}}$ against a reference distribution reflecting the empirically expected trace density in the same distance to the respective seed. This would ameliorate the influence of both Biasing factors on further analysis, since the reference distribution will be drawn from voxels at the same distance from the same seed as the target. The distance between a seed and a target was defined as the physiological distance, i.e. following the natural way of the fibre tracts with all curves around sulci. Thus, the distance is no straight-line distance in 3D, but fitted to the anatomy. Thus, only voxels which were reached by at least one trace were included in any further analysis, yielding a more conservative approach as compared to including also those voxels with a trace count of zero. Actually, different voxels in the termination zone may be located in slightly different distances from the seed which was accommodated by recording the distances between each voxel in the termination zone and the seed. Subsequently, for each voxel of the termination zone, we identified all voxels within a whole brain grey–white matter interface mask that were located at the same distance from the seed as that target voxel. The trace densities of these voxels then provided a reference distribution for the number of probabilistic traces that may be expected in the same distance from the seed as that particular voxel of the termination zone. To account for potential rounding errors in distance $\stackrel{-}{D}$ of each target voxel, the voxels for the reference distribution were drawn from a distance interval +/−1 voxel around the computed distance $\stackrel{-}{D}$ (Fig. 3A).

To establish the reference distribution for the mean trace density ${\stackrel{-}{F}}_{\mathit{ST}}$, we used a randomisation procedure. For each voxel in the termination zone, one trace density from the reference voxels as defined above was selected independently at random. This procedure thus yielded as many trace density values as were obtained from the termination zone that had the same distance from the seed. These random trace densities were then averaged to yield an estimate for the mean trace density of a random set of the same number of voxels in the same distance as the termination zone. As this mean trace density was computed from a voxel that were matched with those constituting the termination zone in terms of distance along the tracts, it reflects an at-chance trace density in the same range. This procedure of drawing the same number of samples and consecutive averaging of the trace counts observed for these (random but matched) voxels was repeated 10⁶ times yielding (due to the central limit theorem) a normal distribution for the mean trace density in the termination zone. In other words, we obtained a reference distribution of at-chance (mean) trace counts across an equally large number of voxels as contained in the seed, which were located in the same distance from the target. Depending on the distance from the seed, a differing number of reference voxels existed. At the relevant distances of up to about 120–130 steps from the seed, there were at least 3000 to 5000 reference voxels from which the samples for the reference distribution could be chosen (Suppl. Fig. S1).

The likelihood of observing ${\stackrel{-}{F}}_{\mathit{ST}}$ by chance, given the distance of this target from the seed, was then expressed as the proportion of the realisations within the reference distribution that showed less or equal trace densities, being a quantitative description of this proportion (Fig. 3C). We thus evaluated the mean number of traces between seed and target (termination zone) against a reference distribution reflecting the by-chance distribution of mean trace counts in such set of voxels at the same distance from the seed. Note, that even though this assessment was based on a reference distribution, no inference was sought at this stage. Rather, these likelihoods were converted into standard Z-scores to allow statistical inference on their consistency across subjects in a second-level group analysis. A Z-score of 0 indicates that it is equally likely to find a higher or a lower mean trace density at that distance by chance. A positive Z-score points towards a termination zone that receives more traces than expected by chance in the same distance from the seed. In contrast, negative Z-scores would be found for targets that show a lower mean trace density than expected by chance in the same distance. It has to be noted that the probabilities obtained from comparison with the reference distribution do not reflect connection probabilities per se (how likely is the target connected) but rather the probabilities for excess of the expected trace number (how (un-) likely would one see the observed trace count at chance in the same distance). Importantly, the former probability would certainly decrease with increasing distance from the seed due to increasing uncertainty and “fanning out” of the tracts. The current approach does not attempt to correct for this phenomenon. Instead, it implements a reference distribution of how many trace counts could be expected by chance at a given distance and describes the observed effects relative to this reference.

Consistency of these findings across subjects was then tested for by means of one-sample T-tests performed separately for each seed-target combination (p<0.001; Bonferroni-corrected for multiple comparisons). Together with the individual Z-scores, this consistency test provides a measure for the likelihood of a given connection across subjects. Thus, the results of the present analysis will be referred to by calling them a connection which one should be more or less confident about.

Further statistical analysis was performed on the confidence measures (individual Z-scores) from each seed to every target area. Different target areas were grouped into anatomically and functionally defined groups to test for differences in anatomical connectivity from IPL seed areas to cortical regions involved in different functional systems. Repeated-measures ANOVAs of connection likelihood values were used to test for influences of the within-subject factors of “IPL seed” (PFt, PF, PFm, PGa, PGp), and “hemisphere” (left, right), as well as the interaction between them on connection likelihood. In particular, the main effect of the factor “IPL seed” assessed whether there was a significant difference between the five IPL seeds with respect to their connection likelihood to that group of targets. The main effect of factor “hemisphere” tested whether the connection likelihood between the five seeds and that assessed target differed between the two hemispheres. The interaction between both factors tested for differences in connection likelihood between left and right hemisphere that were conditioned upon the IPL seed considered.

Reference to macaque data

In order to facilitate a comparison of the results of the present analysis with those from tracer studies in macaques, the relevant literature on connectivity of the inferior parietal lobule in macaque cortex has been reviewed and summarised in a figure (Fig. 10 of the Discussion) using the same layout as for the current results (Fig. 5). To this end, different labelling systems had to be combined and displayed on the schematic drawing of a typical macaque hemisphere in Fig. 10 adopted from Schmahmann and Pandya (2006), with minor changes according to Felleman and Van Essen (1991) and Luppino et al. (1993).

Fig. 10.

Schematic drawings of (A) the lateral and (B) the mesial part of a macaque hemisphere, indicating those connections between the four lateral IPL areas and cortical targets that were found to be connected to either one the IPL areas. IPL areas are marked in four different colours: PF red, PFG green, PG yellow, Opt blue. Coloured dots beneath each target region indicate the presence of a connection to the respective seed region. Areas within the depth of a sulcus were placed along its rim to facilitate readability. Such depiction pertains to the following areas: areas Ig, and RI within the depth of the Sylvian fissure on the insular cortex; areas AIP, VIP, MIP, and LIP within the intraparietal sulcus; parts of areas 46v, and 46d within the depth of the principal sulcus; areas STPa, STPp, FST, MSTd, MSTl, TPO, TEm, and TEpv within the depth of the superior temporal sulcus. Connections depicted here were obtained from and named according to the relevant literature (cf. text) and by reference to the CoCoMac database (http://www.cocomac.org; Stephan et al., 2001).

Fig. 5.

Schematic drawings of (A) the lateral and (B) the mesial part of a human hemisphere, indicating those connections between the five IPL seed areas and cortical targets that were consistently (across subjects) found to be more likely than expected by chance for lateral (A) and mesial (B) target areas. IPL seed areas are marked in five different colours: PFt violet, PF red, PFm green, PGa yellow, PGp blue. Coloured dots beneath each target region indicate the presence of a connection to the respective seed region. For labelling conventions of areas, cf. Table 1.

Results depicted in the figure and within the text of the Discussion were derived from the following publications on IPL connectivity of macaque cortex: Pandya and Seltzer (1982), Petrides and Pandya (1984), Cavada and Goldman-Rakic (1989a,b), Andersen et al. (1990), Neal et al. (1990a,b), Felleman and Van Essen (1991), Luppino et al. (1993), Lewis and Van Essen (2000), Matelli and Luppino (2001), Zhong and Rockland (2003), Gregoriou et al. (2006), Rozzi et al. (2006), Schmahmann and Pandya (2006), Petrides and Pandya (2009) as well as the CoCoMac database (http://www.cocomac.org; Stephan et al., 2001).

The parcellation scheme of the macaque IPL as referred to in the Discussion of the present manuscript was adopted from today’s most widely used parcellation of Pandya and Seltzer (1982). This includes areas PF, PFG, PG, and Opt in rostro-caudal sequence, for which differential connection patterns have been shown by several groups. Since not all publications used this parcellation scheme, the results of the respective publications had to be adapted accordingly. Whenever possible, this was based on the ORT algorithm in CoCoMac (Stephan et al., 2001), in other cases by consensus across the literature and among the investigators. Connections reported for area 7b (rostral IPL) were attributed in most cases to areas PF and PFG, whereas connections reported for area 7a (caudal IPL) were attributed to PG and Opt. If rostral or caudal subareas of 7b or 7a were explicitly distinguished and could be identified as either one of the four areas of Pandya and Seltzer, the reported connection was specifically attributed to the respective area.

Results

Fibre tract pattern of IPL areas

The courses of the (hemisphere-specific) fibre tract patterns for the individual IPL seed areas are visualised in Fig. 4. It should be noted that the probabilistic nature of the tractography as well as the inter-individual variability of fibre tract patterns contributes to substantial uncertainty of these group-averaged tracts.

Fig. 4.

Probabilistic tractography results for five IPL seed areas, ordered from top-left to bottom-right in a rostral sequence and overlaid with the respective cytoarchitectonic seed region (green), for (A) left and (B) right hemisphere. Probabilistic fibre tracts, projected back from individual diffusion spaces into anatomical MNI space, are displayed with regard of the uncertainty of tract traces (among others due to interindividual variability), as revealed by the probabilistic tracking algorithm. The density of traces is colour-coded from opaque yellow for low to transparent red for high uncertainties. Traces are projected onto the sagittal and transversal planes of the MNI single subject template for anatomical orientation. c caudal, d dorsal, r rostral, v ventral.

Visual inspection of the tractography patterns indicates that homologous regions on either hemisphere show a largely similar course. The most conspicuous differences in fibre tract pattern can be observed between rostral (PFt, PF, PFm) and caudal (PGa, PGp) areas of the IPL. Tracts originating from rostral IPL areas tend to run mainly in direction of the inferior frontal, postcentral, and superior parietal regions. They predominantly follow the course of the superior longitudinal (SLF) and arcuate fascicles (AF). In contrast, tracts from caudal IPL areas preferentially follow fibre pathways in direction of lateral occipital and temporal areas, at this joining the inferior longitudinal fascicle via anterior parts of the SLF. Additionally, fibres from the latter region also reach the inferior frontal cortex by a more ventral pathway through the extreme capsule. Interestingly, this differentiation approximately matches the putative location of the border between Brodmann’s areas 40 (rostral, corresponding topographically to the PF areas) and 39 (caudal, PG areas). It should be emphasised, however, that this change in the associated fibre tract patterns is not abrupt but rather shows a gradient when moving from rostral to caudal areas of the IPL with an accentuation at the transition between PFm and PGa.

This visualisation of the probabilistic fibre traces provides qualitative information about possible major pathways connecting the individual areas of the IPL to the rest of the brain, but requires a statistical analysis for quantification of the results.

Seed-target connections of IPL areas

To quantify the respective connectivity patterns, mean trace density in the termination zone obtained from probabilistic tractography for each seed-target combination was assessed relative to the expected trace density at the same distance and statistically tested against its inter-individual variability, providing a measure for the connection likelihood for a given seed-target combination. The findings of these analyses are summarised in Fig. 5 which illustrates those connections of the individual IPL areas that were consistently expressed with higher connection likelihood than expected given their length (p<0.001 for the random-effects inference following Bonferroni-correction for multiple comparisons).

All areas showed consistent connections to inferior frontal (44, 45, frontal operculum, orbitofrontal cortex, frontal pole) and posterior insular regions. The visually apparent shift of predominant connectivity patterns from rostral to caudal IPL areas is also well supported by the statistical analysis. Rostral areas PFt and PF mainly showed consistent connections with targets located in the inferior frontal, somatosensory and superior parietal cortices while reaching only posterior parts of the temporal cortex. Intermediate areas of the IPL (PFm, PGa) likewise show reliable connections to the aforementioned targets, but additionally feature consistent fibre tracts to middle and superior frontal as well as anterior temporal regions. The most caudal area (PGp), finally, again connects with the inferior frontal cortex and, like PFm and PGa, to the entire temporal lobe. In contrast to the other areas, however, there is also significantly expressed connectivity to auditory and temporo-occipital areas.

The visualisation of the fibre tract patterns (Fig. 4) already suggested inter-hemispheric asymmetries. Thus, the fibre tracts that showed consistently greater connection likelihood than expected by chance were further analysed by means of repeated-measures ANOVAs. Fig. 6 summarises the connections that were found to be more pronounced in either hemisphere following correction for multiple comparisons (p<0.05).

Fig. 6.

Schematic drawings of a typical lateral view of a (A) left and (B) right human hemisphere, indicating those connections for which repeated-measures ANOVA revealed a significant left–right asymmetry effect. Only those connections are displayed which were consistently (across subjects) found to be more likely than expected by chance as visualised in Fig. 5. IPL seed areas are marked in five different colours: PFt violet, PF red, PFm green, PGa yellow, PGp blue. Coloured dots beneath each target region indicate the presence of a connection to the respective seed region. For labelling conventions of areas, cf. Table 1.

Rostral IPL areas PFt and PF showed more consistent connections with sensorimotor and superior parietal areas in the left hemisphere, whereas in the right hemisphere, more likely connections existed with secondary somatosensory, posterior insular and inferior opercular and orbito-frontal areas. Caudal IPL areas PGa and PGp had more consistent connections with intraparietal, superior parietal, posterior insular, lateral occipital and auditory areas in the left hemisphere, whereas more likely connections with posterior temporal, temporo-occipital, and inferior frontal areas existed within the right hemisphere. A complete list of the results of the ANOVA can be found in Supplementary Table S1.

While this binarisation into presence or absence of consistent evidence for a connecting fibre tract provides a useful overview over the connectivity of the human IPL, it may fail to reflect finer levels of the rostro-caudal gradients suggested by visual inspection. In order to reveal such gradients and provide more fine-graded information, the relative confidence for each connection has to be considered. Evidently, this analysis again has to accommodate distance-related effects. Connection likelihood of a particular seed-target combination was thus defined by averaging the Z-score (expressing above-chance connectivity) across subjects, reflecting the average chance that the observed trace density would be obtained in the same distance in our group of subjects. Summarising these mean Z-scores for each seed-target connection in a polar plot provided a connectivity fingerprint for each assessed IPL area (cf. Passingham et al., 2002) in which higher values denote stronger evidence for above-chance connectivity.

Comparing the left- and right-hemispheric fingerprints for each seed region again indicated a comparable pattern of connectivity for both hemispheres. Fingerprints of the left hemisphere (Fig. 7) show more consistent connections to temporal areas for all IPL seed areas, probably reflecting the dominance of the arcuate fascicle in the left hemisphere. The shift of fibre tract pattern from rostral to caudal IPL areas is again confirmed by the more fine-grained information provided in the connectivity fingerprints. While most rostral area PFt showed predominant connections to superior parietal cortex and somatosensory regions (major weight of the fingerprint in lower left quadrant of the polar plot), most caudal area PGp featured strong connections to auditory and temporal areas (major weight of the fingerprint in upper right quadrant of polar plot; Figs. 7, 8). Between these two extremes, the connectivity fingerprints for areas PF, PFm, and PGa reflect a relatively continuous shift in connectivity patterns. In particular, the fingerprint of area PF, i.e., the area just caudal to PFt in rostral IPL, closely resembled that of the latter area. The fingerprint of area PFm in the centre of the IPL showed a balanced connectivity to the parietal-sensorimotor and the temporal group. The fingerprint of caudally adjacent area PGa already showed a preponderance of the upper right quadrant.

Fig. 7.

Connectivity fingerprints for the left hemispheric seed regions. Results for consistency testing of fibre tract patterns (across subjects) for each seed-target combination are displayed as Z-scores, reflecting the chance that the observed trace density was higher (positive Z-scores) than expected according to the reference distribution. To clearly highlight the relevant connections (positive Z-scores) of the IPL seed areas, all non-positive (zero or negative) Z-scores (i.e. the chance that observed trace density was lower than or equally high as expected at the same distance) were set to zero. Labelling of areas as introduced in Table 1. Black circle within each polar plot marks a Z-score of 4.80, corresponding to a p-value of 0.001 (corrected for multiple comparisons), to indicate significance of each seed-target connection as a measure of consistency across subjects.

Fig. 8.

Connectivity Fingerprints for each seed region in the right hemisphere. Results for consistency testing of fibre tract patterns (across subjects) for each seed-target combination are displayed as Z-scores, reflecting the chance that the observed trace density was higher (positive Z-scores) than expected based on the reference distribution. To clearly highlight the relevant connections (positive Z-scores) of the IPL seed areas, all non-positive (zero or negative) Z-scores (i.e. the chance that observed trace density was lower than or equally high as expected at the same distance) were set to zero. Labelling of areas as introduced in Table 1. Black circle within each polar plot marks a Z-score of 4.80, corresponding to a p-value of 0.001 (corrected for multiple comparisons), to indicate significance of each seed-target connection as a measure of consistency across subjects.

Connections of IPL seed areas to anatomically- and functionally-defined target groups

Entering the mean Z-scores across subjects into repeated-measures ANOVAs allowed for statistical confirmation of shift in predominant fibre tract patterns from rostral to caudal IPL seed areas (Fig. 9, Tables 2 and 3).

Fig. 9.

Fibre tract strengths between IPL seed areas (x-axis) and anatomically and functionally relevant groups of targets for left (black bars) and right hemisphere (white bars). Individual fibre tract densities to all target areas, separated for left and right hemispheres, can be found in Supplementary Figs. S1–S4. Connection strengths are displayed as mean Z-scores across subjects (averaged across targets within each group), derived from the comparison for each seed-target connection with the reference distribution. Error bars denote standard error. For each plot, the statistics from the repeated-measures ANOVA are displayed beneath the heading of the plot. Grouping of target areas is indicated in Table 2. Fibre tract strengths between IPL seed areas (x-axis) and anatomically and functionally relevant groups of targets for left (black bars) and right hemisphere (white bars). Individual fibre tract densities to all target areas, separated for left and right hemispheres, can be found in Supplementary Figs. S1–S4. Connection strengths are displayed as mean Z-scores across subjects (averaged across targets within each group), derived from the comparison for each seed-target connection with the reference distribution. Error bars denote standard error. For each plot, the statistics from the repeated-measures ANOVA are displayed beneath the heading of the plot. Grouping of target areas is indicated in Table 2.

Table 2.

Grouping of target areas in relation to anatomical and functional criteria.

Group name	Included areas
Frontal	FMed, FMid, FOperc, FOrbit, FPole, FSup
Broca	44, 45
Motor	4a, 4p, 6
Primary somatosensory (SI)	3a, 3b, 1, 2
Secondary somatosensory (SII)	OP1, OP2, OP3, OP4
Cingulate	ACC, PCC, ParaCing
Superior parietal	5L, 5M, 5Ci, 7PC, 7A, 7M, 7P
Intraparietal sulcus (IPS)	hIP1, hIP2, hIP3
Insula	Id1, Ig1, Ig2
Auditory	TE1.0, TE1.1, TE1.2, TE3
Temporal anterior	TSup Ant, TMidAnt, TInfAnt, TFusiAnt, TPole
Temporal posterior	TSupPost, TMidPost, TMidTempocc, TInfPost, TInfTempocc, TFusiPost, TOFusi
Early visual	V1, V2, V3A, V3d, V3v, V4v
Higher visual	V5, OLatSup, OLatInf

Labelling of areas as introduced in Table 1.

Table 3.

Statistics of connections between seeds and groups of targets.

Group name	Main effect ‘Seed’	Main effect ‘Hemi’	Interaction ‘Seed×Hemi’
Frontal	< 0.001	< 0.001	0.004
Broca	0.003	0.002	0.005
Motor	<0.001	0.700	0.013
Primary somatosensory (SI)	<0.001	0.004	0.014
Secondary somatosensory (SII)	<0.001	0.018	0.002
Cingulate	<0.001	0.497	0.026
Superior parietal	<0.001	0.444	0.002
Intraparietal sulcus (IPS)	<0.001	<0.001	<0.001
Insula	<0.001	0.192	<0.001
Auditory	<0.001	0.058	0.002
Temporal anterior	<0.001	0.289	0.736
Temporal posterior	<0.001	<0.001	0.932
Early visual	<0.001	<0.001	0.951
Higher visual	<0.001	0.010	0.178

Groups of areas as defined in Table 2. Values in the columns Main effect ‘Seed’, Main effect ‘Hemi’, and Interaction ‘Seed×Hemi’ denote the p-values of the repeated-measures ANOVA. Significant results for p<0.001 are marked in bold type. Hemi: Hemisphere.

Targets were summarised into anatomically or functionally defined groups as follows. First, target areas within different anatomical brain regions were grouped together (frontal, parietal, occipital, temporal). Where meaningfully possible, groups were further subdivided with respect to functional systems (e.g. motor, primary and secondary somatosensory, early and higher visual, auditory) based on knowledge from neuroimaging, lesion data, and non-human primates. Such analysis for different groups of targets provided information about the involvement of the IPL seed areas in different functional systems. The connection likelihoods for each seed are displayed for each target individually in Supplementary Figs. S2–S5.

For all target groups, there was a significant main effect of factor “IPL seed” (Fig. 9), indicating that for each given target region connection likelihood was differentially expressed between the individual seed areas. This result thus statistically confirmed the above summary of the connectivity fingerprints (Figs. 7 and 8). While prefrontal areas showed a slight increase in connection likelihood from rostral to caudal IPL areas, Broca’s region had highest fibre tract densities in the intermediate areas PF, PFm, and PGa, with lower connection likelihood to area PFt. Secondary (SII) and particularly primary (SI) somatosensory cortex, to the contrary, showed a pronounced decrease in fibre tract density from PFt to PGp (Fig. 9, first two rows). A similar, though less pronounced pattern was seen for the superior parietal cortex while the anterior intraparietal sulcus (Choi et al., 2006; Scheperjans et al., 2008a,b) featured an abrupt drop in connection likelihood from the anterior PF- to the posterior PG-areas.

The posterior insula (Kurth et al., 2010), in contrast, showed an increase in connection likelihood from rostral to caudal IPL seed areas. This pattern was even more pronounced for the auditory and temporal regions, which all showed a strong increase in fibre tract density from rostral to caudal IPL areas. In addition, it should be noted that the overall connection likelihood with the IPL is substantially higher for posterior temporal target areas as compared to the anterior temporal and auditory ones (Fig. 9, fourth row). Within the occipital lobe, two groups of areas could be differentiated: Early visual areas (V1–V4v) showed generally low connection likelihood with a slight decrease from rostral to caudal IPL areas whereas higher visual areas (on lateral occipital cortex up to V5) showed a pronounced increase in fibre tract density from rostral to caudal IPL areas (Fig. 9, last row).

Discussion

The fibre tract patterns of five cytoarchitectonic areas on the lateral surface of the human inferior parietal lobule were assessed by means of probabilistic tractography. Connection likelihood was analysed by evaluating the number of traces reaching a particular target against those reaching other voxels in the same distance from the seed. Random-effects inference then delineated connections that were consistently (across subjects) expressed with higher than expected densities. Across IPL areas, most consistent connections existed with inferior frontal, posterior temporal and insular targets. Connectivity patterns, however, shifted gradually from rostral to caudal IPL areas. Whereas rostral areas featured most likely connections to somatosensory and superior parietal areas, more caudal areas were predominantly connected to the temporal lobe. Connectivity patterns of IPL areas favoured a left–right asymmetry with regard to different target areas of somatosensory, superior parietal (left hemisphere), and inferior frontal, orbito-frontal and temporal areas (right hemisphere).

Methodological considerations

Any kind of tractography study faces the problem of false positives or negatives due to limits in effective resolution, resulting in non-resolved crossing/kissing fibres. This can cause tractography traces to get absorbed into bigger fibre bundles like the superior longitudinal or arcuate fascicle (Ciccarelli et al., 2003; Klein et al., 2009). This problem is ameliorated by using probabilistic (instead of deterministic) tractography methods with crossing fibre models (Behrens et al., 2007) . Moreover, tractography results are highly dependent on the distance between seed and target (Behrens et al., 2003a,b). This may already impair relative comparisons between individual results, the procedure currently most frequently used in tractography studies (e.g. Behrens et al., 2003a; Rushworth et al., 2006; Uddin et al., 2010). More importantly, it makes inference about the relevance of a particular trace density for a given seed-target combination hardly possible. Thus, it is difficult to infer whether a particular observed trace-density is indeed indicative of a fibre connection between these two regions.

We aimed at addressing these two problems by the present algorithm by introducing a reference distribution of traces for each seed-target combination. The reference distribution was set depending on the distance between seed and target, thus accounting for the possible confounding factor “distance” within further analysis. Furthermore, this distribution provided a reference to evaluate each seed-target connection separately by testing if traces between seed and target were more consistently expressed (across subjects) than expected at the same distance from the seed. Thus, for each seed-target connection, we could evaluate if connection likelihood was consistently above chance.

Nevertheless, it still has to be emphasised that DTI does not provide information about anatomical connectivity sensu stricto, i.e. synaptic connectivity as revealed by invasive tracer studies in macaques, which will be discussed in the next section. Rather, this method assesses the presence and confidence about macroanatomical fibre bundles between seeds and targets.

By using probabilistic atlases for delineation of seed and target areas together with probabilistic tractograhpy, it was possible to stringently use probabilistic approaches throughout the current study. Like probabilistic tractography for fibre paths, probabilistic atlases accommodate inter-subject variability and hence allow inclusion of uncertainty estimates into the investigation. Combining two different probabilistic atlases of the human cerebral cortex (JHA and HCSA) allowed for whole brain coverage. Some regions, however, were covered in a more coarse fashion than others due to the current limitation of available atlases. For example, the middle and superior frontal gyrus were only covered by a single area each. Based on different parcellation schemes, it could be assumed that these two regions can at least be separated into areas 8, 9, and 46. As known from studies in macaques, these areas can be further subdivided, supporting the notion that a comparably fine-grained parcellation can be assumed for the human prefrontal cortex as well. Areas 46v, 46d, 8A, 8B, and 9 of the macaque prefrontal cortex show a differential connection pattern with the IPL areas (see discussion below). Thus, it can be assumed that such parcellation might yield a differential connectivity pattern also in humans. Our results may hence represent a first overview of the global connectivity patterns of inferior parietal areas rather than a fine-grained description of pre-frontal connections. By the time reliable information on a functionally relevant parcellation of the human prefrontal cortex will be available, future studies might add further insight into the patterns of anatomical connections between human prefrontal and inferior parietal areas.

It has to be further emphasised that our approach with definition of seed and target areas is different from the connectivity-based parcellation which aims at parcellation of a cortical region based on its connectivity pattern to every other voxel in the brain (Anwander et al., 2007; Behrens et al., 2003a; Eickhoff et al., 2011; Johansen-Berg et al., 2004) In these approaches, connectivity was assessed without a priori anatomical suppositions about potentially connected areas, thus allowing for an assumption-free investigation of brain connectivity. Our approach in turn is based on a priori knowledge on the location of cortical areas which limits the analysis to previously defined regions. Hua et al. (2009) described an approach which combined the ideas of both seed-target and connectivity-based parcellation approaches: First, they obtained overall probability distributions from all voxels inside the brain. In a second step, they drew regions of interest across the cortex. Each resulting trace of the first analysis was post-hoc assigned to one of the fibre tracts known to be connected with the respective region of interest which was passed by the tractography trace. Thus, Hua et al. (2009) also used a priori anatomical knowledge for definition of regions of interest. But in contrast to the approach of the present study, their regions of interest were not used as seed regions for tractography, but rather as targets through which existing tracts should go. Thus, they also identified those target regions which were reached by certain traces, but based on overall tract distributions within the brain. Conversely, we focused on traces which originated in different predefined seed regions.

Connectivity patterns of macaque IPL areas and homology considerations

The overall pattern revealed by our results bears close resemblance to the axonal connection patterns found in tracer studies on macaques (Fig. 10), as both species showed a comparable rostrocaudal shift in preferential connectivity.

In macaques, rostral IPL areas mainly connected to (pre-) motor, inferior frontal, somatosensory, superior parietal and posterior temporal areas (Fig. 10). A similar pattern was found here for rostal human IPL areas PFt and PF, hinting at a possible homology between human areas PFt/PF and macaque areas PF/PFG, respectively. This is further supported by a recent tracing study (Petrides and Pandya, 2009) showing that macaque area PF is mainly connected to ventral premotor cortex (area 6v), whereas intermediate areas PFG and PG preferentially connected to areas 44 and 45. A similar pattern was found here for human IPL as rostral-most area PFt showed weaker connectivity to areas 44 and 45 than intermediate areas PF, PFm, and PGa (Fig. S2).

Caudal IPL areas in macaques showed connections to more dorsal parietal areas, the prefrontal cortex (in particular true for area PG) as well as numerous areas within the whole temporal lobe (in particular for Opt; Fig. 10). This pattern strongly resembled that of the two caudal-most areas of human IPL (areas PGa and PGp), suggesting a possible homology between human PGa/PGp and macaque PG/Opt, respectively.

Intermediate human IPL area PFm which does not have an obvious homologue in macaques showed characteristics of both rostral and caudal connection patterns. Interestingly, cytoarchitectonic features of this region also share characteristics of rostral and caudal IPL areas (Caspers et al., 2006). It may thus be speculated that this intermediate area in human IPL is a new evolutionary feature that may have derived from a more subtle subdivision of adjacent PFG or PG in the macaque IPL (Krubitzer, 2009). This view is further supported by recent evidence that especially central parts of the IPL may have evolved more pronouncedly in humans than other parts of the brain (Husain and Nachev, 2007; Rushworth et al., 2009; Simon et al., 2004).

Despite these congruencies, there are also notable differences in fibre tract patterns between human and macaque IPL, e.g. with respect to occipital connections. Whereas for macaques no connections between IPL and primary and secondary visual areas V1 and V2 were observed, the results of our present study suggest such fibre tracts in humans. Higher visual areas as V3 and V4 connect to caudal IPL areas in the macaque whereas in humans connections are also found with the more rostral IPL (e.g., for targets OLatSup, OLatInf). To provide a possible explanation for these differences, one has to consider several aspects.

It has been argued that the macaque visual system follows a rather strict hierarchy: While earlier visual areas such as V1 and V2 first project to higher extrastriate areas such as V3, V4, DP, and LIP, these areas in turn provide the relevant visually related connections with caudal IPL areas (Andersen et al., 1990). It might be assumed that a comparable hierarchical processing stream exists in humans, making the existence of any direct tract between early visual and IPL areas unlikely. This suspicion is strengthened by the notion that caudal rather than rostral IPL areas are primarily involved in visual processing (cf. Mountcastle et al., 1975; Hyvärinen, 1982 in macaques; e.g., Fink et al., 2001 in humans). On the other hand, however, a recent meta-analysis indicated that rostral IPL areas and higher visual areas are conjointly recruited by action observation (Caspers et al., 2010), matching the existence of mirror neurons in the rostral IPL of macaques (Rizzolatti, 2005). How such visual input reaches the rostral IPL, however, is still unclear. The present data suggest that connections from earlier visual areas may contribute to this functional interaction even though such connections have not yet been demonstrated in macaques. Such shortcuts may hold a particular relevance for the rapid, “on-line” processing of observed actions, in particular in the context of social interactions. The finding that not only mirror neurons, but also grasping neurons were found to be located within macaque area PFG (Rozzi et al., 2008) further supports this notion since grasping-related actions highly rely on visual input.

From a methodological perspective, however, one also has to consider the relative advantages and drawbacks of tracer studies (in macaques) and DTI. Whereas tracing methods reveal axonal connectivity and yield a high specificity, their sensitivity may be limited by the selection of injection sites and methods, incomplete tracer uptake or spread, unexpectedly slow transport (or to early sacrifice), the histological method used to detect the labelled brain sites and the (usual) lack of whole brain coverage (Köbbert et al., 2000; Stephan et al., 2001). In contrast, DTI is particularly prone to false positives due to limited resolution and limited specificity. This pertains in particular to the assimilation of traces into major fibre bundles that are a frequent source of false positive results. Independent of evolutionary differences, interpretation of the current finding thus faces the challenge that it is present in the more sensitive but distinctly less specific method but absent in the specific but somewhat less sensitive approach. In summary, we would thus cautiously suggest that connections between visual areas and anterior IPL would be physiologically reasonable as a fast framework for action observation and may warrant further investigation by specifically targeted tracing experiments. Until such confirmation, however, the possibility of false positives in tractography has to be considered an at least equally likely scenario.

A similar argument pertains to the connections to the frontal pole found in the present study. For the macaque, it must remain tentative if such connections are indeed absent since review of the relevant literature (Petrides and Pandya, 1984; Cavada and Goldman-Rakic, 1989b; Andersen et al., 1990; Rozzi et al., 2006) together with information from the CoCoMac database did not provide final evidence about their presence or absence. Nevertheless, a recent study demonstrated nonexisting connections between IPL and anterior prefrontal regions in macaques (Petrides and Pandya, 2007) . But it was also argued that in particular prefrontal cortex showed very strong expansion in the evolution from macaques to humans (Semendeferi et al., 2001). A comparable suggestion was raised for the IPL in humans as compared to macaques (Husain and Nachev, 2007; Simon et al., 2004), a concept which was particularly stressed for the central parts of IPL (see above; Rushworth et al., 2009). In a recent study, Mars et al. (2011) could show that the IPL, its central part in particular, was functionally connected with anterior prefrontal cortex whereas no such interaction was found in macaques. In the present study, we also found such connections between IPL areas and the frontal pole region which further supports these recent findings of resting-state functional connectivity and evolutionary considerations and provides evidence for a direct anatomical connection between these two cortical regions.

With regard to other prefrontal regions, the current study is certainly limited by the lack of reliable target regions. This limits the comparison to a juxtaposition of the connections of the ‘superior frontal gyrus’ target with a summary of results for macaque areas 8 and 9. Whereas all but the rostral-most IPL area were found to connect with the superior frontal gyrus, mainly caudal areas of the macaque IPL seemed to project to superior frontal areas 8B and 9. Likewise, for the middle frontal gyrus, the present human data mainly fit the macaque data of target areas 46d and 8A, but not area 46v. In both cases, we would resort to caution regarding evolutionary interpretations, since no reliable more fine-grained and functionally relevant parcellation of human prefrontal cortex is currently available. The issue of prefrontal connections of IPL areas thus warrants further investigation once such information becomes available in humans.

On a more general note, our results again highlight the difficulty of establishing inter-species differences in the concurrent presence of diverging methods and potential evolutionary changes. One approach to address this problem is to use the same method (DTI) in both species, as done in studies on the arcuate fasciculus (Rilling et al., 2008), prefrontal-cerebellar circuits (Ramnani et al., 2006), and prefrontal connections (Croxson et al., 2005). The studies show good agreement between human and macaque DTI data, but point to considerable inter-species differences in connectivity patterns even when using a comparable method. On the other hand, Dyrby et al. (2007) and Schmahmann et al. (2007) demonstrated that while many connections obtained from invasive tracing methods could also be replicated by means of DTI tractography in the same animal, others could not. Are thus all differences between humans and primates related to limitations of one (or both) technique? We would argue against this view, since evolutionary effects that result in the evident differences between macaque and man also warrant consideration. While humans share many aspects of cortical organisation with non-human primates, uniquely human features may not be neglected (Finlay et al., 2001; Kaas, 2006; Kaas and Preuss, 2003; Striedter, 2005). This may hold true particularly for the regions showing the most pronounced differences as discussed above, the prefrontal cortex (which is vastly expanded and presumably re-organised in humans, cf. Semendeferi et al., 2001) and the occipital lobe, which also has a considerable differential morphology in humans and macaques due to different importance of the visual system in these two species. While primary visual cortex (V1) in macaques occupies a major part of the lateral occipital cortex, the same region in humans is mainly located on the mesial surface within the calcarine sulcus. The same holds true for the extrastriate visual areas V2–V5 which in the macaque are located within the lunate sulcus and on the prelunate gyrus, bringing them close to the caudal end of the IPL (Kaas, 2006; Ungerleider and Haxby, 1994), whereas in humans these areas are located more caudo-ventrally (ventral visual stream) and caudo-dorsally (dorsal visual stream) (Amunts et al., 2000; Malikovic et al., 2007; Rottschy et al., 2007). Finally, differential connectivity pattern of comparable regions may also result from differential cortical structure and folding patterns as white matter tracts had to adapt to the gross morphological differences as well (Kaas, 2000).

Considering these potential pitfalls when comparing connectivity patterns of humans and macaques, it seems even more important to take a systems perspective and use a large number of target areas to establish connectivity profiles of several seed areas. Such an approach provides a complementary approach to the connectivity-based parcellation used by Mars et al. (2011). Using the present approach of predefined seed and target areas, we could show that connectivity of human IPL areas shifted gradually from rostral to caudal. We thus amended the findings of Mars et al. (2011) who demonstrated that a similar parcellation of the IPL as based on cytoarchitecture (Caspers et al., 2006, 2008) was possible by means of DTI. Within the present study, we could elucidate how these assumed differences manifested with regard to differential connections to numerous target areas. Furthermore, we could show, in spite of the discussed differences that a similar connection pattern as in macaques (Gregoriou et al., 2006; Rozzi et al., 2006) also existed in humans. This supported the notion of a differential involvement of IPL areas in functional cortical networks which was already assumed based on electrophysiological recordings from different parts of macaque IPL (Hyvärinen, 1982; Mountcastle et al., 1975; Seltzer and Pandya, 1984).

Relation to functional segregation

The potential human mirror neuron system has received particular interest in the study of IPL functions (Rizzolatti and Craighero, 2004; lacoboni, 2005; Keysers and Gazzola, 2009). In a recent meta-analysis, it was shown that particularly the rostral-most area PFt was strongly engaged in action observation and imitation (Caspers et al., 2010). In these tasks, PFt co-activated with a region in the caudal-most aspect of ventral premotor cortex (vPMC) (comparable to the macaque mirror neuron system; Rizzolatti, 2005). This view is supported by human DTI studies highlighting vPMC to rostral IPL connections (Rushworth et al., 2006; Tomassini et al., 2007) and a study by Kelly et al. (2010) on resting-state co-activation pattern of Broca’s region, and more specifically by a recent macaque study of Petrides and Pandya (2009) showing differential connections from rostral IPL to different vPMC areas. PFt furthermore co-activated with SI, superior parietal and posterior temporal regions. The present study demonstrated that indeed rostral IPL areas showed consistent connections with Broca’s region. But more importantly, area PFt also showed consistent anatomical connections with SI, superior and intraparietal, and posterior temporal areas. We would thus argue that these fibre tracts represent the structural scaffold for the previously described functional “mirror neuron” network (Caspers et al., 2010). This notion is further supported by the lateralization structure of the connections of rostral IPL: connections with sensorimotor regions were more consistently found in the left as compared to the right hemisphere in the present study. For the potential human mirror neuron network, it was argued that such higher motor functions would likely be assumed to be supported by a bilateral network of brain areas (Iacoboni and Dapretto, 2006) which was supported by a recent meta-analysis (Caspers et al., 2010). But there have also been findings of a functional left-lateralization within this network (Rizzolatti and Arbib, 1998; Aziz-Zadeh et al., 2004). Based on the results of the present study, we would argue that the connections between rostral IPL and areas relevant within this potential mirror neuron network are more consistently expressed within the left hemisphere. It has to be noted that all our subjects were right-handed, thus introducing a potential bias towards the contralateral left hemisphere for any sensorimotor processing. It has furthermore to be stressed that mirror-related activity within IPL may also exhibit a right-lateralization, depending on the task demands (Biermann-Ruben et al., 2008). Such overriding of a potentially present general leftward-lateralization was in particular observed in the context of emotion-processing (Carr et al., 2003). Thus, it might be speculated that the predominance of left hemispheric anatomical connections as found in the present study only partially account for functional phenomena within the potential human mirror-neuron network. Findings of bilateral or right-lateralized activations within IPL might demonstrate that functional demands and the specific experimental context might be at least as important in determining functional recruitment as the structural basis of anatomical connections.

Within the language network, areas of the (left) caudal IPL, corresponding to Geschwind’s area (Geschwind, 1970), have been related to semantic and phonological processing, especially during reading (Price, 2000; Vigneau et al., 2006). Presumably, the main fibre pathway connecting Broca’s, Geschwind’s, and Wernicke’s area is the arcuate fasciculus (Catani et al., 2005; Parker et al., 2005; Frey et al., 2008; Saur et al., 2008). These studies also demonstrated an additional ventral fibre system via the external or extreme capsule. The present study showed that particularly PGa and PGp are connected to Broca’s region via this ventral route.

Overall, all IPL areas showed quite consistent connections with Broca’s region. This finding suggested that language-related functions might involve different parts of the IPL, specialised for different language-related aspects. With regard to the potential human mirror-neuron system within rostral IPL, these consistent connections to Broca’s region might be regarded as evidence for the assumed relation between the mirror neurons and the development of language (Rizzolatti and Arbib, 1998). Within the IPL, more caudal areas might relate to reading aspects (Price, 2000), which would be supported by the current finding of additional visual input to these areas. Their connection to Broca’s region and temporal though not auditory cortex furthermore point to an involvement in vocalisation (Catani et al., 2005; Saur et al., 2008). Moreover, PGp showed consistent connection patterns with auditory areas. It may thus represent the key IPL node for auditory language processing, a view supported by functional studies (Price, 2000; Saur et al., 2008). With respect to a hemispheric asymmetry of connectivity of these two caudal IPL areas, the present study revealed higher likelihood of connections to auditory, higher visual, and frontal opercular areas in the left hemisphere. It might be speculated how such connection pattern especially within the left hemisphere might relate to language functions. Geschwind (1965,1970) not only highlighted the aspect of reading to be located within rostral parts of the angular gyrus, but furthermore stressed that this left IPL hub would integrate visual and auditory word forms to generate speech (Price, 2000). In functional studies, it became evident that some aspects of the semantic processing network were modality specific, thus having an auditory (auditory cortex) and a visual component (higher visual areas, inferior posterior temporal cortex). These earlier areas were found to be involved in either semantic or phonological processing (Vigneau et al., 2006). The predominant connections to both auditory and higher visual areas in the left hemisphere, as found in the present study, might support the notion that these areas integrate these different modalities for semantic or phonological processing, independent from input modality.

But the leftward asymmetry of connections of the caudal IPL areas also comprises connections to superior parietal and intraparietal areas. These latter areas have consistently been implicated in number magnitude processing. During more complex processing steps such as calculation, the left caudal IPL is strongly involved as well (Dehaene et al., 1998; Fias et al., 2003; Nieder, 2005; Ansari, 2008; Dehaene, 2009) . With the predominance of left hemispheric connections to the intraparietal and superior parietal areas found in the present study, caudal IPL area PGa might reflect the relevant part of the angular gyrus which is involved in number processing. The respective underlying fibre tracts provide the anatomical basis for a close communication pathway between these areas, with a potential link to the language system with vocalisation of the numbers (see above).

Areas of right IPL are part of the ventral attention network as proposed by Corbetta and Shulman (2002, 2008). This network involves IPL, posterior temporal and inferior/middle frontal areas. Within the IPL, damage to areas close to the temporo-parietal junction (TPJ) is associated with the emergence of spatial neglect (Hillis, 2006). This would presumably match the location of areas PFm and PGa. In the present study, it could be shown that these areas featured consistent connections to posterior temporal and inferior frontal areas (like the other IPL areas), but particularly also to middle frontal cortex. Based on this fibre tract pattern which matches the co-activation pattern of respective functional studies (Corbetta and Shulman, 2002; Corbetta et al., 2008), it can be assumed that especially areas PFm and PGa play an important role in the ventral attention network, possibly in combination with caudal most area PGp. Further strong connections existed with lateral superior and intraparietal areas, providing a possible link to the dorsal part of the attention network (Corbetta and Shulman, 2002). Besides some remarks on a possible bilateral distribution of this network, especially the ventral attention network was most consistently found in the right hemisphere. Taking into account the right lateralization of consistent fibre tracts from these caudal IPL areas to posterior temporal, temporo-occipital and inferior frontal areas, the present study might provide the potential anatomical connectional basis of how the areas of the ventral attention network might interact. Based on the fibre connections found to be dominant in the right hemisphere, it seems reasonable to assume that information between posterior temporal and inferior frontal regions are transported via caudal IPL. The right hemisphere analogue of the arcuate fasciculus of the left hemisphere might provide the necessary pathways here.

The good agreement between anatomical fibre tract patterns of IPL areas and their possible functional roles provides strong evidence for their differential involvement in specific neural networks. The investigation of the functional and potentially context-specific relevance of these connections as revealed by effective connectivity analyses represents an important topic for future studies on connectivity of human IPL.

Supplementary materials related to this article can be found online at doi:10.1016/j.neuroimage.2011.06.027.