Abstract
The superior longitudinal fasciculus (SLF) is the large white matter association tract connecting the prefrontal and posterior parietal cortices. Past studies in nonhuman primates have parcellated the SLF into three subdivisions and have outlined the specific corticocortical organization and terminations for each subdivision. However, it is difficult to characterize these structural connections in humans to the specificity of tract-tracing studies in animals. This has led to disagreement on how the SLF subdivisions are organized in the human brain, including if the dorsomedial SLF (SLF-I) is part of the cingulum subsystem. Here, we present a novel large-scale, probabilistic map of the SLF subdivisions, using high-resolution diffusion imaging data from the Human Connectome Project (HCP). We used image data from 302 adult males and 405 adult females to model the three SLF subdivisions in each hemisphere and attempted to characterize the frontal and parietal termination points for each subdivision. SLF subdivisions were successfully modeled in each subject, showing the dorsomedial-to-ventrolateral organization similar to that in nonhuman primate histological studies. We also found minimal differences between SLF-I models with and without the cingulate gyrus excluded, suggesting that the SLF-I may be a separable tract from the cingulum. Lastly, the SLF subdivisions showed differentiable associations with major cognitive domains such as memory and executive functions. While histological confirmation is needed beyond tractography, these probabilistic masks offer a first step in guiding future exploration of frontoparietal organization by providing detailed characterization of the SLF subdivisions and their potential cortical terminations.
- cortical terminations
- diffusion MRI
- frontoparietal; Human Connectome Project
- superior longitudinal fasciculus
- tractography
Significance Statement
The prefrontal and posterior parietal areas are interconnected via the SLF, which has been characterized in great detail in monkeys. However, it is difficult to map the SLF organization in the human brain, and previous diffusion MRI findings have been inconsistent. Using diffusion MRI data from 707 individuals, our probabilistic tractography revealed dorsomedial-to-ventrolateral organization of the three SLF subdivisions and their cortical terminations. Our tractography also suggests limited shared volume between the SLF-I and the cingulum, a controversy in recent literature. The SLF subdivisions also differ in their cognitive associations. As a result, we created a large-scale, high-resolution probabilistic parcellation of the SLF, representing an advancement toward standardizing the mapping of human frontoparietal structural connections for clinical and scientific research.
Introduction
The prefrontal cortex (PFC) and the posterior parietal cortex (PPC) are two higher order regions that are closely anatomically and functionally linked (Petrides and Pandya, 1984; Selemon and Goldman-Rakic, 1988; Cipolloni and Pandya, 1999). Their integrity has been shown to be critical for executive function (Goldman et al., 1971; D’Esposito and Postle, 1999) and selective attention (Corbetta et al., 1998; Bisley and Goldberg, 2003; Hagler and Sereno, 2006). The functional neuroimaging literature further suggests that the frontoparietal network serves as the foundation of cognitive control of behavior (Cole et al., 2013). However, the structural organization of the frontoparietal network is still not fully established for the human brain.
The superior longitudinal fasciculus (SLF) is a large association tract containing the reciprocal connections between the frontal and parietal cortices. While the SLF was first identified in human postmortem dissection studies as early as the 1800's by Reil and Autenrieth, detailed connection patterns between the PFC and PPC were mostly characterized using tract-tracing in nonhuman primates (NHP; Martino et al., 2013). Pioneering work by Petrides and Pandya (1984) used axonal tracing to show three SLF subdivisions, SLF-I, II, and III, organized in a dorsomedial-to-ventrolateral gradient. Overall, tract-tracing studies have defined the particular cortical and intermediate terminations for each SLF subdivision (Cipolloni and Pandya, 1999; Schmahmann et al., 2008), which are crucial for understanding the functional interdependence of frontoparietal regions (Gnadt and Andersen, 1988; Selemon and Goldman-Rakic, 1988; Funahashi et al., 1989; Chafee and Goldman-Rakic, 2000).
Due to methodological limitations, earlier neuroimaging attempts to characterize the SLF in the human brain have not achieved the level of intricacy of NHP histological studies (Janelle et al., 2022). With technological advancements, researchers have successfully subdivided the human SLF into SLF-I, II, and III, using tractography, resembling the NHP SLF organization (Makris et al., 2005; Thiebaut de Schotten et al., 2011). Despite finding cross-species similarities in SLF parcellation (Hecht et al., 2015; Rojkova et al., 2016; Conner et al., 2018), the exact human SLF organization is still in debate. Some have questioned the number of SLF subdivisions in humans (Bernal and Altman, 2010; Martino et al., 2013; Kamali et al., 2014) and, more controversially, others have questioned whether the SLF-I is a subdivision belonging to the SLF at all, or rather a part of the cingulum subsystem instead (Wang et al., 2016; Wu et al., 2016), due to their shared connectivity with the cingulate gyrus (CG) and mediofrontal anatomical location (Petrides and Pandya, 1984; Bubb et al., 2018). Further, due to the lack of high-resolution tractography, the cortical terminations of the human SLF subdivisions are yet to be specified, which has lead to inconsistent region of interest (ROI) choices for tractography and conflicting findings of SLF organization across studies (Kamali et al., 2014; Hecht et al., 2015; Wang et al., 2016). Lastly, while previous diffusion imaging studies have associated the SLF's integrity with working memory (Kinoshita et al., 2016; Jolly et al., 2017; Koshiyama et al., 2020), and executive functions (Jolly et al., 2017; Veldsman et al., 2020), the roles of SLF subdivisions in different forms of cognition remains unclear.
Here, we constructed the first high-resolution probabilistic maps of the human SLF. Our goal was to utilize the knowledge of the cortical terminations of the SLF subdivisions from past histological and tractography studies to guide ROI selection and tractography on a large diffusion dataset and to determine if the resulting SLF subdivisions in the human brain resemble the monkey tract-tracing literature. We also estimated the cortical terminations of each subdivision and attempted to address the SLF-I controversy by differentiating it from the cingulum. We further examined the relationship between cognitive performance and SLF microstructural integrity. We aimed to provide an accessible, large-scale, nuanced probabilistic map of the SLF organization that can be applied in future research in a more standard, anatomically relevant manner and to provide further characterization of the structural underpinnings of frontoparietal communication.
Materials and Methods
Sample
Brain images from a sample of 718 subjects were downloaded from the Human Connectome Project Aging (HCP-A; Feinberg et al., 2010; Moeller et al., 2010; Setsompop et al., 2012; Xu, 2012; Sotiropoulos et al., 2013; Van Essen et al., 2013; Harms et al., 2018; Bookheimer et al., 2019) study on February 10, 2022. The mean age of this sample is 59.5 years (SD 14.8; range, 36–100) with 314 males and 404 females, and 17.6 (SD 2.2) mean years of education. HCP-A is one of the datasets available from the Human Connectome Project (HCP), which is a multisite, large-scale neuroimaging project which collects functional, structural, and diffusion magnetic resonance imaging (MRI) data, as well as physiological and behavioral data (Bookheimer et al., 2019). Inclusion criteria for our study was that each subject must have a diffusion scan and behavioral data. Subjects were considered to be healthy aging adults, excluding individuals who have been diagnosed and treated for neurological disorders, such as Parkinson's disease and Alzheimer's disease, or major psychiatric disorders, such as schizophrenia and severe depression, for 12 months or longer in the past 5 years. Additionally, subjects were screened for dementia and mild cognitive impairment (MCI) using both the Montreal Cognitive Assessment (MoCA; Nasreddine et al., 2005) and the Modified Telephone Interview for Cognitive Status (TICS-M; de Jager et al., 2003)
After screening (detailed below, Diffusion MRI preprocessing), 11 subjects were removed from the final analysis due to motion artifacts, resulting in a final sample size of 707 (M age = 59.5 ± 14.8; 302 males, 405 females, M EDU = 17.5 ± 2.2).
Aging cohort
We decided to use the HCP-A cohort rather than the young adult HCP cohort or the HCP Developmental cohort for several reasons. While it is possible that there are aging-related changes in the cerebral white matter (Ouyang et al., 2021), these changes seem to be manifested mostly in microstructural changes, such as those measured by DTI metrics like fractional anisotropy (FA) and mean diffusivity (MD), rather than the orientation or anatomical organization of tracts in the brain (Matijevic and Ryan, 2021; Schilling et al., 2022). Additionally, more recent exploration of white matter changes across the lifespan suggest that white matter microstructure, axon density, and myelin density peaks firmly in mid adulthood (Yeatman et al., 2014; Kiely et al., 2022), with the SLF-I volume in particular peaking later in life than most other cerebral white matter tracts (Slater et al., 2019): in the late 50's. Therefore, the HCP young adult cohort, age 22–35, would not accurately represent a fully mature SLF tract (Van Essen et al., 2013). As the HCP-A subjects are 36 and older (M = 59.5), this cohort was more appropriate for the purpose of this study. Moreover, the relative myelin and axonal health of white matter tracts does not begin to significantly drop off until very late adulthood, approximately the mid-70's and early 80's (Yeatman et al., 2014; Kiely et al., 2022). For our multiple regression analysis, we chose to control for potential age-related changes in our cerebral white matter tract data rather than potentially interpreting white matter that is likely still under development in the HCP young subject samples.
MRI data acquisition and protocols
Both structural T1 and diffusion MR scans were conducted with a Siemens 3T Prisma whole-body scanner (32-channel head coil). MR images for this HCP sample were collected from four different scan sites. T1 images, 208 sagittal slices with slice thickness of 0.8 mm, were acquired using a 4-echo multi-echo MPRAGE sequence, with repetition time (TR) of 2,500 ms, inversion time (TI) of 1,000 ms, echo time (TE) of 1.8/3.6/5.4/7.2 ms, voxel size of 0.8 × 0.8 × 0.8 mm3, flip angle of 8°, matrix of 320 × 300, and field of view (FOV) of 256 × 240 × 166 mm.
Multishell Diffusion MR sequences were collected with multiband of 4, b of 1,500 and 3,000, TR of 3,230 ms, TE of 89.2 ms, voxel size of 1.5 × 1.5 × 1.5 mm3, flip angle of 78°, matrix of 140 × 140, FOV of 210 × 210 mm, and 92 sagittal slices with slice thickness of 1.5 mm. The HCP-A diffusion protocol includes two sessions of 99 and two sessions of 98 gradient directions, with each session collecting both the anterior and posterior phase-encoding directions as well as 14 nongradient volumes (b = 0), as outlined in Harms et al. (2018). We utilized the 99 gradient direction dataset for our main analysis and SLF subdivision reconstruction and the 98 gradient direction dataset for additional validation tests (detailed below, Validation of SLF mapping).
Diffusion MRI preprocessing
Preprocessing of diffusion images was done using FMRIB Software Library (FSL; Smith et al., 2004; Woolrich et al., 2009; Jenkinson et al., 2012; https://fsl.fmrib.ox.ac.uk/fsl/fslwiki). Image data from each subject was visually inspected volume by volume for movement distortions and other artifacts. If an image had considerable visible motion distortions that affected more than one-third of the images, the subject was removed from the analysis. Susceptibility distortions were corrected using FSL's TOPUP tool, which combines the anterior-to-posterior and posterior-to-anterior phase-encoded images to correct distortions created by the phase-encoding direction during acquisition (Andersson et al., 2003; Smith et al., 2004). Both T1 and B0 images underwent brain extraction using Analysis of Function Images (AFNI) 3dSkullStrip (Cox, 1996; Cox and Hyde, 1997). Each individual's diffusion images were then corrected for eddy current distortion using FSL's EDDY tool. For additional quality control, we ran EDDY_QUAD, an automated quality control software in FSL to examine the eddy-corrected outputs individually. We applied EDDY_SQUAD again to examine the eddy-corrected outputs at the group level. Outliers detected by EDDY_SQUAD were inspected manually and were removed from the analysis if the output was deemed too distorted.
Tensors were fitted and tract-based spatial statistics were created for each individual using FSL's DTIFIT (Behrens et al., 2003, 2007). Diffusion parameters were calculated and crossing fibers were modeled for each subject using the GPU version of FSL's BEDPOSTX tool (Behrens et al., 2003, 2007; Hernández et al., 2013), which uses Markov Chain Monte Carlo sampling to build up voxel-specific distributions on diffusion parameters and prepares each subject for tractography. Linear transformations from the diffusion space to each subject's T1 space and linear transformations of the subject's T1 to MNI space (Grabner et al., 2006) were computed using FSL's FLIRT (Jenkinson and Smith, 2001; Jenkinson et al., 2002). To limit reslicing, registration of diffusion space to MNI space was performed by concatenating the diffusion space to structural space transformation matrix and the structural space to MNI space transformation matrix. This resulted in a diffusion space to MNI space transformation matrix for each subject. FLIRT was then used to apply the diffusion space to MNI space transformation matrix to each subject's diffusion image, as well as their tractography outputs, normalizing the native-space images to a standard space. For tractography analysis performance in native space, the diffusion to MNI transformation matrices were inverted and concatenated in order to resample the ROIs from MNI space to each individual's native diffusion space.
Tractography and probabilistic maps
Probabilistic tractography was performed using the GPU version of FSL's PROBTRACKX, which is a program that samples Bayesian distributions of diffusion directions for modeling white matter tracts ending in gray matter (Behrens et al., 2003, 2007; Hernández et al., 2013). We decided to use PROBTRACKX to run tractography as previous studies suggest that its underlying ball-and-stick model is one of the most conservative models when reconstructing streamlines between two regions (Thomas et al., 2014), minimizing the potential spurious trajectories. We considered that using a more conservative tractography method on an individual basis would minimize tracking errors in the final group-wise probabilistic maps, as our goal here is not to explore all of the potential connections between the frontal and parietal cortices, but to examine if the human SLF maps resemble past monkey histological findings. Using PROBTRACKX, we ran tractography in each subject's native space using the multiple masks option (see below for ROI mask definitions). For each subject, six sessions of tractography were conducted—one for each of the three SLF subsections in each hemisphere. Each tractography session was conducted by calculating the number of streamlines between each pair of frontal and parietal masks using the default PROBTRACKX parameters: 5,000 samples per voxel, loop checks enabled, curvature threshold of 0.2, and step length of 0.5 mm. We then visually inspected the path files and thresholded the individual tracts by dividing the path files with the combined waytotal of mask one and mask two and then thresholding the mask by 0.5% streamline density. This resulted in a thresholded probability map of each individual's 3 SLF tracts per hemisphere and provided the average probability that a given voxel belongs to the tract connecting each frontal-parietal mask pair. The resulting six SLF subdivision masks per subject were then normalized to the MNI template, binarized, summed across subjects and then divided by the total sample size, resulting in a group-wise probability map in standard space for each SLF subdivision for both hemispheres.
Frontal and parietal ROIs and exclusion/inclusion ROIs
To model the subdivisions of the SLF, the ROIs were selected carefully by considering both histological studies in NHP (Petrides and Pandya, 1984; Cipolloni and Pandya, 1999) and human DTI studies (Thiebaut de Schotten et al., 2011; Hecht et al., 2015; Conner et al., 2018). In addition, we reviewed previous studies that compared human and nonhuman primate white matter terminations of the SLF to most accurately choose the frontal and parietal ROIs (Thiebaut de Schotten et al., 2011, 2012; Hecht et al., 2015). Accordingly, we defined SLF-I as the tracts in between the superior frontal gyrus (SFG) and the superior parietal lobule (SPL). We defined SLF-II as the tracts between the middle frontal gyrus (MFG) and the angular gyrus (AG). Lastly, we defined SLF-III as the tracts between the inferior frontal gyrus (IFG) and the supramarginal gyrus (SMG).
Using past NHP studies as a guide (Petrides and Pandya, 1984; Cipolloni and Pandya, 1999; Schmahmann et al., 2008), we ensured that we provided adequate coverage of all major frontal and parietal cortical terminations. Notably, for our frontal regions, we ensured that we provided proper coverage of the full SFG, MFG, and IFG, including frontal Brodmann's areas (BA) such BA 4, 6, 8, and 44–46. Similarly, we included all areas of the parietal cortex as outlined in NHP studies (Petrides and Pandya, 1984). This includes the full SPL (BA 7) and IPS, with areas of the superior, middle, and inferior parietal lobule. The ROIs were first created using the Mindboggle atlas (Klein et al., 2005) in MNI space. The Mindboggle ROIs atlas provided adequate coverage of the three frontal gyri, including the anterior portions of the frontal lobe, which was missing in the Harvard–Oxford atlas. The Mindboggle atlas also provided clear, unambiguous boundaries for each gyrus and more precisely followed the shape of the anatomy, allowing us to better define the cortical boundaries of each ROI. However, because Mindboggle does not capture the white matter in between each region's gyral folds, we further used ITK-Snap (www.itksnap.org) to hand label the corresponding white matter for each cortical area in order to ensure proper tract creation. Additionally, we used the Harvard–Oxford cortical atlas to assist in defining the boundaries of the AG, as Mindboggle does not differentiate the AG separately from the inferior parietal cortex. See Figure 1a for the final ROIs. These ROIs are available at https://github.com/mamandola/SLF_probabilistic_map.
a, Regions of interest (ROIs) used for tractography. SLF-I ROIs (red), superior frontal gyrus and superior parietal lobule; SLF-II ROIs (blue), middle frontal gyrus and angular gyrus; SLF-III ROIs (green), inferior frontal gyrus and supramarginal gyrus. b, Uncorrected SLF subdivision volumes normalized to MNI space at 50% probability across subjects. c, SLF volume indexes corrected by ROI size.
Furthermore, we used several exclusion and waypoint ROIs to ensure we modeled only the SLF tracts in our tractography. Firstly, we included an exclusion ROI of the whole midsagittal plane to eliminate all the generated streamlines that crossed over into the opposite hemisphere. This was necessary because the SLF is an association fiber defined by connecting unilateral cortical areas within the same hemisphere. We also included an exclusion ROI along the axial plane of the temporal lobe in order to eliminate any shared tracts between the SLF-III and the arcuate fasciculus. Similar to Hecht et al. (2015), we included another exclusion ROI in the external capsule in order to limit crossing fibers. Lastly, we included a large inclusion ROI in the white matter between the frontal and parietal lobes, specifically directly underlying the central sulcus, to ensure that all streamlines connecting the frontal and parietal cortices were included. All of these inclusion and exclusion ROIs were drawn using ITK-Snap in MNI space (www.itksnap.org). For each subject's tractography run, all six frontal/parietal ROIs, inclusion ROIs, and exclusion ROIs were transformed into their native diffusion space using FSL's FLIRT. In order to more precisely map and label each cortical termination, we used the Multi-modal Human Connectome Project Atlas (HCP-MM1; Glasser et al., 2016). To categorize a cortical termination, the probabilistic maps at 30% overlap (i.e., at least 30% of participants had modeled tract in that area) for each SLF subdivision were masked by the ROIs used to run tractography to more clearly visualize gray matter terminations. We believe that 30% overlap was an appropriate threshold because it reflected locations of cortical terminations that were clearly demonstrated but was conservative enough that there were no spurious tracts which altered the overall organization of the map. If the masked probabilistic reached the gray matter, as defined by the HCP atlas, that gray matter was categorized as a cortical termination.
Cingulum and SLF-I differentiation
To test if the cingulum and SLF-I are separable tracts, we repeated the tractography sessions for each subject using the same protocol described above, except that we included the whole cingulate gyrus (CG) from the Harvard–Oxford cortical atlas at 25% probability as an additional component of the exclusion ROI. This way, if any streamlines were generated passing through the CG, it would be deleted from the model. As the cingulum is the main white matter tract innervating the CG and makes contact with the entirety of the CG (for review, see Bubb et al., 2018), this would minimize the possibility of streamlines that belong to the cingulum being attributed to the SLF-I. With two tractography runs, one with the CG as an exclusion ROI and one without, we were able to compare the two modeled SLF-I tracts directly and assess the extent to which the cingulum and the SLF-I display shared subcortical white matter volume in the probabilistic SLF map. Further, to test whether the cingulum and SLF-I had substantial shared volume, we first calculated the volumetric overlap between our modeled SLF-I tracts and the cingulum. The cingulum ROI was created using the John Hopkins University (JHU) DTI-based White Matter Atlas (Oishi et al., 2009). The SLF-I masks consisted of each individual's binarized tractography run normalized to MNI space. Note that this is a different mask from the group-wise probabilistic masks of the SLF-I—these are the individual masks that have been normalized to MNI space, binarized, and thresholded by 0.5% of the waytotal per individual. With this liberal threshold, we assessed the extent to which the relatively unfiltered SLF-I masks for each individual displayed volumetric overlap with the cingulum mask, as these were much more sensitive to shared volume than the 50% overlap group-wise probabilistic maps. To calculate the percentage of shared volume for each subject, we divided the number of shared voxels between the SLF-I mask and cingulum ROI by the total volume of the subject's SLF-I mask.
Statistical analysis
Firstly, we tested whether the three SLF subdivisions across the two hemispheres differed in volumetric characteristics by performing a 3 × 2 repeated-measures analysis of variance (ANOVA). The volume of each SLF mask was measured in cubic millimeters (mm3) after normalizing each individual's mask to the MNI template to control for individual head size. To also correct for differences in ROI size on SLF subdivision models, we divided the volume of each individual tract by the total volume of its corresponding frontal and parietal ROIs and repeated the 3 × 2 repeated-measures ANOVA. For example, each subject's modeled right SLF-I subdivision was normalized to MNI space and then the volume extracted. This volume was then divided by the total volume of the right SFG and right SPL ROIs used for SLF-I tractography. To evaluate if inclusion of the CG affected the volumetric characteristics of the SLF-I, we also conducted a 2 × 2 ANOVA testing for the main effect of CG inclusion/exclusion on SLF-I volume in both hemispheres.
Further, we explored whether the specific SLF subdivisions were related to particular cognitive functions across subjects. Cognitive indices for each major cognitive domain were estimated using the behavioral data available from the HCP database. Following protocols in previous HCP studies (Weintraub et al., 2013), we separated the behavioral data into the following six domains: Executive Function/Inhibitory Control and Attention, Executive Function/Task Shifting, Working Memory, Episodic Memory, Language, and Processing Speed (Table 1). The individual cognitive scores of the subjects were z-scored across subjects for each task, and those within each cognitive domain were summed, resulting in a composite score for each cognitive domain per individual. These cognitive domain indices were validated using a confirmatory factor analysis (CFA), and positive loadings were confirmed for each domain (Table 1). The CFA was produced using the CFA function in the factor-analyzer module (https://pypi.org/project/factor-analyzer/), which utilizes tools from both sci-kit learn and the R psych library.
Cognitive tasks in the HCP-A sample
For each cognitive domain, we constructed a multiple regression model to examine the relationship between individual differences in microstructural integrity of SLF subdivisions and cognitive performance across subjects. Microstructural integrity of the SLF subdivisions was measured by using the subdivision group-level masks at 50% overlap and extracting the average FA value per subdivision from each subject. To control for shared variance and to determine if a particular subsection of the SLF contributed more to cognitive performance, all three subdivision FA values were included in all regression models as separate predictor variables. Age, sex, years of education, and scan site were included as confounding variables. Separate models were made for each hemisphere, resulting in 12 total models.
Validation of SLF mapping
Several past studies challenged the reliability of tractography (Maier-Hein et al., 2017; Schilling et al., 2021). Though tractography has the potential to be highly anatomically accurate with carefully selected anatomical ROIs (Schilling et al., 2020), it is difficult to make a definitive claim of any neuroanatomical representation without the use of dissection or NHP histological cross-validation. Therefore, we took several steps to test the reproducibility of our data and validate our SLF masks. Firstly, to test if our results were robust over datasets, we performed a second run of the full tractography protocol detailed above using the 98 gradient direction dataset from the same HCP-A subjects and generated a second group-wise probabilistic map for each SLF subdivision (N = 703, with 15 individuals excluded here because of head motion and artifacts in this validation dataset). To check for alignment and spatial similarity between the two versions of the SLF subvision masks, a Dice coefficient (Dice, 1945) was calculated for each pair of the group-wise probabilistic maps. A coefficient of 0.7 and above would be considered good alignment (Dice, 1945; Dauguet et al., 2007; Timmers et al., 2016; Wilson et al., 2017; Kreilkamp et al., 2019).
Additionally, we ran an exploratory analysis using constrained spherical deconvolution (CSD) tractography using MRTrix3 (Tournier et al., 2012, 2019). Probabilistic tractography with the iFOD2 algorithm was performed on 50 randomly selected subjects (M age = 59.3, SD = 14.3, 27F/18M) from the same HCP-A cohort, using default parameters. We performed tractography for each SLF subdivision with the frontal ROI as the seed and the corresponding parietal ROI as the inclusion image, and vice versa. The final subdivision tract was obtained by summing the streamlines from both runs, and streamline density maps were created and thresholded in an identical manner to the PROBTRACKX masks. We then generated the group-level CSD probabilistic maps and again calculated the Dice similarity coefficient between the CSD/PROBTRACKX masks. Lastly, we recalculated cingulum and SLF-I shared volume per subject using the CSD-derived streamline density maps, as past literature suggests that ROI-guided CSD is more sensitive to potential trajectories compared with the PROBTRACKX ball-and-stick tractography (Thomas et al., 2014), meaning that the CSD-derived maps may be more sensitive to shared volume.
Results
To better understand frontal-parietal organization, we created a probabilistic tractography map of the three SLF subdivisions in each hemisphere using high-resolution diffusion MRI data from a large-scale cohort. We also addressed a controversy in the recent literature by designing tractography parameters to determine if the SLF-I and cingulum are distinguishable white matter tracts. For all 707 subjects, tractography successfully modeled all three SLF subdivisions for both hemispheres. Figure 2 shows the resulting whole-brain probabilistic maps for each SLF subdivision at 50% subject overlap for the cohort. The estimated cortical terminations for each SLF subdivision are outlined in Table 2.
Whole-brain probabilistic maps of the SLF subdivisions at 50% subject overlap. a, Axial view. b, Coronal view. c, Sagittal view. Red, SLF-I; blue, SLF-II; green, SLF-III. Color bars correspond to percent overlap between subject SLF masks, each thresholded by 0.5%.
Specific terminations of the SLF subdivisions
Three SLF subdivisions
SLF-I
This dorsomedial subdivision of SLF was successfully identified in each hemisphere for every subject in the sample. Figure 3, b and c, shows the SLF-I at a probabilistic threshold of 50% overlap among the subjects, and Figure 3a displays the most pronounced terminations in the frontal and parietal lobes. In the frontal lobe, the left SLF-I terminated in the posterior portion of the SFG, with its termination closest to the medial premotor cortex (BA 6) and subregion BA 6a and with some terminations approaching the supplementary motor area (SMA). The left SLF-I extended posteriorly, making up a large portion of the white matter tract inferior to the primary motor cortex (dorsal portions of BA 4) and the anterior somatosensory cortex (BA 3), and it terminated close to the anteromedial and anterolateral posterior parietal cortex (BA 7). The posterior end of the left SLF-I also displayed smaller terminations near medial lateral intraparietal area (LIP), medial ventral intraparietal area (VIP), and lateral precuneus visual area (PCV). The right SLF-I followed largely the same tractography pattern, though the parietal terminations were largely in the PCV. There was a minor amount of overlap between the inferior portions of the right SLF-I and the dorsolateral splenium of the corpus callosum. Notably, there were little or negligible terminations approaching any portions of the cingulate gyrus or the cingulum tract for either the left or the right SLF-I, which is discussed in the section below.
Cortical terminations and whole-brain illustration of SLF-I. a, The most pronounced cortical terminations in the frontal (left) and the parietal (right) lobes. Cortical boundaries were determined using the multimodal HCP atlas (Glasser et al., 2016). The SLF-I showed terminations in the medial frontal lobe, with its most clear terminations in multiple locations within the premotor cortex, BA 6. Within the parietal lobe, the clearest terminations are found in the superior parietal lobule, BA 7. b, SLF-I map at 50% subject overlap shown in coronal view, illustrating terminations in the PCV. c, SLF-I map at 50% subject overlap shown in sagittal view. The top row shows the lateral to medial slices in the right hemisphere, and the bottom row shows the slices in the left hemisphere.
SLF-II
Laterally, the SLF-II was also successfully segmented in each hemisphere for every subject. Figure 4, b and c, shows the SLF-II at a probabilistic threshold of 50% overlap among the subjects. The SLF-II displayed frontal terminations closest to BA 8a (Fig. 4a). The terminations were particularly evident in the dorsolateral PFC, (dorsolateral MFG—BA 8Av and 8c), with terminations reaching the frontal eye fields (FEF). The anterior portions of SLF-II bordered the perimeter of the inferior frontal junction (IFJ). The SLF-II also displayed dorsolateral frontal terminations in the posterior MFG, posterior to BA 8 and ventral to the FEF, potentially in MFG area 55b, as suggested by some recent multimodal studies (Glasser et al., 2016; Hazem et al., 2021). Posteriorly, the SLF-II traveled ventromedially underneath the primary somatosensory cortex (anterior BA 2) and reached posterior parts of the inferior parietal cortex (area PG, BA 39). The SLF-II also had smaller terminations reaching the anterior and medial portions of the inferior parietal (PFm, BA 40), the lateral portions of the inferior parietal lobe (IP), and medial portions of the temporal-parietal-occipital junction (TPOJ). The right SLF-II largely follows the same organization, though there seems to be a larger proportion of terminations close to the anteroventral portions of the dorsolateral PFC (BA 8A) and the posterior MFG (BA 8 and MFG area 55b) compared with the left SLF-II.
Cortical terminations and whole-brain illustration of the SLF-II. a, The most pronounced cortical terminations in the frontal (left) and the parietal (right) lobes. Cortical boundaries were determined using the multimodal HCP atlas (Glasser et al., 2016). The SLF-II showed more lateral terminations when compared with the SLF-I. In particular, there were clear terminations in the dorsolateral prefrontal cortex, BA 8. There were also terminations in the posterior MFG in area 55b and the FEF. Within the parietal cortex, there were terminations in the posterior portions of the inferior parietal lobe, particularly in area PF and PG. b, The whole-brain SLF-II map at 50% subject overlap in the coronal view, illustrating terminations in BA 8. c, SLF-II map at 50% subject overlap shown in sagittal view, including some of the terminations in the FEF. The top row shows the lateral to medial slices starting at the right hemisphere, and the bottom row shows the slices in the left hemisphere.
SLF-III
The ventrolateral subdivision of the SLF was successfully segmented in each hemisphere for every subject. Figures 5, b and c, shows the full SLF-III at 50% overlap among subjects. The left SLF-III in the frontal lobe had the most pronounced terminations reaching the pars opercularis of the IFG, particularly in BA 44 (Fig. 5a) and 45, including the dorsal portions of Broca's area. The most anterior terminations of the SLF-III reached the pars triangularis (BA 45) but did not reach the pars orbitalis (BA 47). There were also terminations reaching the ventral premotor cortex (BA 6, Fig. 5a) and posterior IFJ. Extending posteriorly, the SLF-III passed inferior to the primary somatosensory cortex (anterior BA 2) and anterior part of the inferior parietal lobule (IPL, hlP2, and hlP1) and largely terminated close to the dorsal and dorsomedial portions of the retro-insular cortex (RI) and the rostral half of the inferior parietal cortex (PF). It displayed large terminations reaching the parietal operculum, with notable terminations in the SMG [perisylvian language area (PSL) and superior temporal visual (STV) subregions], OP, and parietal areas, PF and PFcm. The right SLF-III follows the same organization, though it displayed slightly fewer terminations in the pars triangularis of the IFG.
Cortical terminations and whole-brain illustration of the SLF-III. a, The most pronounced cortical terminations in the frontal (left) and the parietal (right) lobes. Cortical boundaries were determined using the multimodal HCP atlas (Glasser et al., 2016). The most clear terminations of the SLF-III in the frontal lobe were in the IFG pars opercularis, BA 44 and FOP. There were also strong terminations in the ventral portions of BA 6. Within the parietal lobe, there were clear terminations in the supramarginal gyrus, particularly in parietal area PF, and certain terminations in the retro-insular cortex. b, The whole-brain SLF-III map at 50% subject overlap in the coronal view. c, The whole-brain SLF-III map at 50% subject overlap in the sagittal view, illustrating terminations in the SMG, super temporal visual area (STV), and pars triangularis (BA 45). The top row shows the lateral to medial slices starting at the right hemisphere, and the bottom row shows the slices in the left hemisphere.
We used a two-factor repeated-measures ANOVA to test for volumetric differences among the SLF subdivisions across the two hemispheres (Fig. 1b). The volumes were extracted using each individual's normalized SLF subdivision masks. There was a significant main effect of subdivision (F(2,1412) = 471.4, p < 0.001), and post hoc t tests revealed that SLF-I displayed the largest volume, followed by SLF-II, with SLF-III showing the smallest volume in both hemispheres (t's(1412) = 10.9–21.3, p's < 0.001). The main effect of hemisphere (F(1,706) = 266.3, p < 0.001) and hemisphere by subdivision interaction (F(2,1412) = 44.4, p < 0.001) were also significant, with larger right hemisphere subdivisions (t(706) = −9.0, −8.9, and −3.7; p's < 0.001, 0.001, and 0.01 for SLF-I, II, and III, respectively). However, when corrected by corresponding ROI volumes, the subdivision effect reversed. While the effects of subdivision (F(2,1412) = 3,260.6, p < 0.001), hemisphere (F(2,706) = 119.9, p < 0.001), and the hemisphere by subdivision interaction (F(2,1412) = 43.2, p < 0.001; Fig. 1b) were significant, post hoc tests revealed a larger corrected SLF-III volume, followed by SLF-II, and then SLF-I (t's(1412) = −17.3–51.4, p's < 0.001). This is unsurprising, as the SFG ROIs used to guide SLF-I tractography was the largest, followed by the MFG ROI, and then the IFG ROI. The hemisphere effect stayed consistent for SLF-II and III (p's < 0.001), but not SLF-I (t(1412) = 0.006, p > 0.05).
Distinguishing SLF-I and the cingulum
To clarify the degree of shared volume between the SLF-I and the cingulum, we conducted a second tractography analysis with the CG included as an exclusion ROI. Any modeled tracts crossing into the CG were terminated or excluded from the resulting SLF-I. The extent of shared volume between the cingulum ROI and SLF-I was quantified by calculating the percentage of shared voxels between each individual's SLF-I mask and the cingulum mask as a ratio of their total SLF volume (Fig. 6a).
SLF-I probabilistic maps at 50% subject overlap with and without the cingulate gyrus (CG) as an exclusion ROI. a, Red, SLF-I without the CG as an exclusion ROI; blue, SLF-I with the CG as an exclusion ROI; purple, overlap between the two probabilistic maps. The maps were largely indistinguishable along the tract. The two areas of largest differentiation are displayed in the magnified sagittal sections. b, The distribution of overlap between the SLF-I and the cingulum, using the individual SLF-I tractography maps in each subject's native space. The majority of subjects show 0% overlap between the SLF-I and the cingulum, and the average percent overlap for all subjects was 0.18% for the right SLF-I, and 0.04% for the left SLF-I, suggesting that the two tracts are distinguishable. c, Violin plots comparing the volume of the SLF-I with and without the cingulum included in tractography.
The average percentage of shared volume between the SLF-I and cingulum for the entire sample was minimal: 0.18% for the right SLF-I and 0.04% for the left SLF-I (Fig. 6a,b). Among all, 77.5 and 58.8% of subjects showed no shared volume at all between the cingulum and the SLF-I in the left and right hemisphere, respectively (Fig. 6b). Despite the tiny volumetric overlap, the differences in SLF-I volume were significant with the CG inclusion versus exclusion (F(1,706) = 184.6, p < 0.001), with the right SLF-I seeming to gain more volume than the left SLF-I when the CG was not excluded from tractography (Fig. 6c). This effect is unsurprising, as the CG is a large region to be employed as an exclusion ROI and likely overcorrects crossing fibers between the two tracts. It is worth acknowledging that the average decrease in volume when excluding the CG was 1.9% for the left SLF-I and 2.8% for the right SLF-I. Taken together, since less than half of the subjects showed only minimal or negligible shared volume with the cingulum, these data suggest that the SLF-I and the cingulum are separable systems.
SLF subdivision validation
We tested if the SLF tractography results were reproducible using different dataset (99- vs 98-direction diffusion images) and methodology (PROBTRACKX vs CSD). First, we reran tractography on the 98 gradient direction dataset from the same sample and found the resulting SLF maps was highly similar to those from the 99-direction dataset, with an average Dice coefficient of 0.97 (range, 0.96–0.98). The trajectory, shape, and cortical termination patterns of the two sets of SLF subdivisions were virtually identical, with minimal or negligible visual difference between them. Second, we conducted additional tractography using CSD on a smaller sample and found that the spatial similarity between the PROBTRACKX and CSD-derived probabilistic maps of the three SLF subdivisions were moderately high to high with an average Dice coefficient of 0.73 (range, 0.70–0.77), with some small variation in the posterior terminations. The CSD-derived masks were on average larger than the PROBTRACKX-derived masks, which is to be expected, as tractography derived by CSD is typically more sensitive to potential trajectories (Thomas et al., 2014). Further, the CSD-derived individual SLF-I tracts also demonstrated negligible shared volume with the cingulum, with an average volumetric overlap of 0.07 and 0.27% in the left and right hemisphere, respectively. Again, over half of the subjects showed no shared volume between the cingulum and the SLF at all (31/50 for the left SLF, 25/50 for the right SLF). In sum, both the 98-direction dataset and CSD-derived SLF masks were comparable with the main tractography analysis.
SLF subdivisions and cognitive functions
To explore whether an SLF subdivision is more closely related with particular cognitive domains, we used multiple regression models to examine the relationship between individual differences in SLF microstructural integrity and cognitive performance across subjects (Fig. 7). Microstructural integrity was measured as the average FA of each SLF subdivision masked using the cohort's 50% overlap tractography map. Each regression model included all three SLF subdivisions from one hemisphere as predictor variables to control for variance shared between the subdivisions, as not surprisingly they were closely correlated with each other (r = 0.43–0.84). Age, sex, years of education, and scan site were included as covariates in each model.
Relationship between the microstructural integrity of SLF subdivisions and cognitive task performance in five domains. Each graph represents the bivariate relationship between the FA values of an SLF subdivision at 50% overlap and cognitive performance indexes across subjects. Red, SLF-I; blue, SLF-II; gray, SLF-III. Bolded r values correspond to significant FA coefficients in the corresponding multiple regression models.
Executive function—inhibition
Regression models were significant for both hemispheres (left; F(9,684) = 23.77, p < 0.001, R2 = 0.238; right: F(9,684) = 22.27, p < 0.001, R2 = 0.227). SLF-I was predictive of inhibitory executive functions, where both the right (β = 5.68, p = 0.003) and left (β = 4.57, p = 0.013) FA coefficients reached significance. Additionally, the left SLF-III was closely associated with inhibitory executive function, showing the largest coefficient (β = 6.30, p = 0.001), though the right SLF-III (β = 3.28, p = 0.111) did not reach significance.
Executive function—shifting/attention
Both the left (F(9,684) = 10.14, p < 0.001, R2 = 0.116) and right (F(9,684) = 9.53, p < 0.001, R2 = 0.110) hemisphere regression models were significant. The FA of the left SLF-I was positively associated with attention/shifting task performance across subjects (β = 4.89, p = 0.019), and the correlation with left SLF-III FA was approaching significance (β = 3.90, p = 0.061). None of the subdivision FA coefficients for the right SLF reached significance (β's = −0.67–3.65, p's > 0.05).
Working memory
Though both the left (F(9,684) = 14.73, p < 0.001, R2 = 0.160) and right (F(9,684) = 14.53, p < 0.001, R2 = 0.158) hemisphere regression models were significant, none of the subdivision's FA coefficients reached significance (β's = −1.15–1.75, p's > 0.05). However, it is worth to note that all the SLF subdivisions except the right SLF-III displayed a bivariate correlation between FA and working memory at some level with relatively weak effects (r's = 0.106–0.200, p's < 0.01). Additionally, when averaging the FA across subdivisions, the SLF FA coefficients did not reach significance in either the left (β's = 6.28, p > 0.05) or right (β's = 0.76, p > 0.05) hemispheres.
Episodic memory
The regression models for both hemispheres were significant (left: F(9,684) = 27.74, p < 0.001, R2 = 0.268; right: F(9,684) = 26.03, p < 0.001, R2 = 0.256). The FA of SLF-I was closely associated with episodic memory performance across subjects, with both the left (β = 21.76, p = 0.010) and the right (β = 19.46, p = 0.028) displaying strong effect sizes. None of the other subdivisions displayed significant correlations with episodic memory, suggesting that, within the SLF, SLF-I was more uniquely associated with episodic memory.
Processing speed
Both the left (F(9,684) = 29.42, p < 0.001, R2 = 0.276) and right (F(9,684) = 30.02, p < 0.001, R2 = 0.280) hemisphere regression models were significant. The right SLF-III was the only FA coefficient that reached significance (β = 3.23, p = 0.015).
Language
Both the left (F(9,684) =13.91, p < 0.001, R2 = 0.155) and right (F(9,684) = 13.48, p < 0.001, R2 = 0.151) hemisphere regression models were significant, although none of the subdivision FA coefficients reached significance (β’s = −2.76–3.52, p’s > 0.05).
Results were similar when using a 75% group mask, though for processing speed, the right SLF-I FA coefficient was approaching significance (β = 1.62, p = 0.084), and the right SLF-III FA coefficient marginally did not reach significance (β = 1.82, p = 0.058). The left SLF-II FA coefficient also showed significant association with the language composite scores.
In sum, the SLF-I seemed to drive most of the individual differences in cognitive behaviors, particularly in the domains of episodic memory and executive functions. The SLF-III seemed to play a greater role with inhibitory control and processing speed. There were also potential associations between the left SLF-II and language ability.
Discussion
Here we present the first large-scale, high-resolution probabilistic anatomical mapping of the SLF in the human brain and characterize the specific frontal and parietal terminations for each subdivision that resemble monkey tract-tracing findings (Petrides and Pandya, 1984). We were able to further address recent controversy by showing that the differences in the SLF-I derived from tractography including and excluding the CG were minimal, suggesting that the SLF-I and the cingulum may represent two separable systems. The findings of differentiable levels of cognitive contributions among the subdivisions further support the notion that the SLF should be treated as a nuanced and subdivided tract, rather than a single, large homogenous structure.
Mapping SLF organization
The connections between the prefrontal and parietal cortices are highly nuanced, which has presented a challenge for studies seeking to characterize the specific anatomical and functional links between the two cortices and their implications for behavioral outcomes. With recent technological advancements, our tractography analysis revealed the SLF organization and potential cortical terminations in greater detail, resembling NHP histological studies (Petrides and Pandya, 1984; Cipolloni and Pandya, 1999; Schmahmann et al., 2008) and a recent smaller-scale human study that used deterministic tractography to specify SLF terminations (Conner et al., 2018).
The overall human SLF organization resembled the NHP model, with three subdivisions following a dorsomedial-to-ventrolateral gradient (Fig. 8). Our human SLF-I showed similar characteristics as in NHP, connecting the BA 7 and dorsal and medial portions of BA 6, with the latter near the SMA (Petrides and Pandya, 1984). However, while we did find some terminations approaching the dorsal anterior cingulate, they were less pronounced compared with the NHP (Petrides and Pandya, 1984). We also did not find terminations related to the primary motor cortex as in the NHP. Rather, our SLF-I tract seemed to pass through BA 4 as part of the underlying white matter.
Schematic representation of the cortical termination points for the three SLF subdivisions. Termination points were labeled using the multimodal HCP atlas by Glasser et al. (2016). These SLF cortical terminations follow a clear dorsomedial to ventrolateral organization, replicating tract-tracing studies in NHP. Red, SLF-I; blue, SLF-II; green, SLF-III.
Our SLF-II also resembled that of the NHP, particularly for the AG and MFG connections. As in the NHP (Petrides and Pandya, 1984), our tractography showed that PG in caudal IPL was connected to BA 8. Perhaps due to tractography's limitation of reaching the gray matter, we did not find SLF-II terminations reaching BA 46, as shown in the NHP (Petrides and Pandya, 1984; Cipolloni and Pandya, 1999; Schmahmann et al., 2008). However, it reached the white matter underlying BA 46, suggesting that the human SLF may also innervate the more anterior portions of the PFC. It is also possible that there was not enough overlap among subjects in a high-probability mask, as dorsolateral PFC is one of the most variable brain areas across individuals (Gordon et al., 2017; Bruno et al., 2022). Nevertheless, terminations approaching other parts of dorsolateral PFC (BA 8, 55b) and FEF were evident in our SLF-II.
Overall, SLF-III deviated the most from the NHP tract-tracing studies, particularly in the frontal terminations. Compared with the NHP, our human SLF-III reached more ventral PFC regions. The NHP SLF-III reached dorsally to the ventral bank of MFG (BA 46; Petrides and Pandya, 1984), whereas the most dorsal termination of our human SLF-III appeared close to the ventral IFJ (BA 45). The main frontal terminations resembling the NPH SLF-III were ventral BA 6 and the pars opercularis and pars triangularis. Petrides and Pandya (1984) also reported terminations from parietal area PF to the frontal operculum, which was the clearest parietal termination in our SLF-III, among other terminations reaching the parietal operculum and SMG. Interestingly, our SLF-III mask extended more into the IFJ (BA 45) than the NHP and was characterized by fewer terminations in the CG.
Distinguishing SLF-I and cingulum
Several factors caused challenges in using tractography to distinguish SLF-I and cingulum: both tracts partially innervate the medial frontal cortex and project to the PPC; the cingulum is composed of both long and short U-shaped fibers (Bubb et al., 2018) that branch off continuously along the cingulum, making it difficult to separate from other nearby fiber tracts; and both human (Thiebaut de Schotten et al., 2011) and NHP studies (Petrides and Pandya, 1984) showed that SLF-I has terminations in the anterior cingulate. Thus, some conclude that SLF-I is a part of the cingulum (Wang et al., 2016, Wu et al., 2016) or that it is not part of the SLF system at all (De Benedictis et al., 2014). Our findings suggest more limited shared volume than previously suggested, with 0% shared volume between the two tracts in over half of the cohort, and the rest showing <2% shared volume. Additionally, excluding the CG in our tractography analysis made negligible changes to the overall SLF organization.
The confusion between SLF-I and cingulum may stem from limitations of tractography in combination with differing ROI selection between studies, since the distinction between them has not been debated in the NHP tract-tracing literature. Indeed, the projection of the cingulum was shown to be more ventromedial through the subcortical white matter compared with the SLF-I and terminates in CG and retrosplenial areas in NHPs (Mufson and Pandya, 1984; Petrides and Pandya, 1984; Vogt and Pandya, 1987; Kobayashi and Amaral, 2007), with only sparse projections to SLF-I related frontal areas (Morris et al., 1999; Kobayashi and Amaral, 2007).
Taken together, our findings suggest that SLF-I terminates in more dorsal areas of the frontal cortex, following the general dorsomedial-to-ventrolateral organization of the SLF, and is a separable tract with limited shared volume with the cingulum. It is possible, however, that the SLF could be considered a supracingulate fasciculus in humans, as noted by Wang et al. (2016).
SLF and cognitive functions
Our findings suggest that the integrity of SLF-I, and to a lesser extent, SLF-III contribute the most to individual differences in executive function and episodic memory. This corroborates previous neuroimaging studies that emphasized the SLF's overall role in executive functions (Jolly et al., 2017; Veldsman et al., 2020; Ribeiro et al., 2024). While most previous studies linked SLF-I with working memory (Thiebaut de Schotten et al., 2005; Barbey et al., 2013), few studies have associated the SLF as a whole with episodic memory performance (Smith et al., 2011; Lockhart et al., 2012).
Surprisingly, we did not find any SLF subdivisions associated with working memory, although previous lesion studies suggest a critical role of SLF-I (Thiebaut de Schotten et al., 2005; Barbey et al., 2013), SLF-II (Thiebaut de Schotten et al., 2005; Cochereau et al., 2020), and SLF-III (Papagno et al., 2017; Cochereau et al., 2020) in visuospatial and verbal working memory. It is possible that a more specific working memory measure than the general HCP-A measure is necessary to distinguish the role of SLF subdivisions in this domain. Further research with more sophisticated cognitive task designs may further distinguish the roles of SLF subdivisions in memory and executive functions.
Limitations
There are several limitations to this study. Mainly, it is important to note that we cannot definitively claim neuroanatomical findings or direct replication of histological findings without tissue validation. While previous studies demonstrate that guided tractography with carefully selected ROIs can be anatomically accurate (Schilling et al., 2020), it is well established that a high number of false trajectories is inevitable when using probabilistic tractography (Maier-Hein et al., 2017). Ideally, to address the fundamental question of human neuroanatomy, future studies should cross-validate tractography findings with cross-species tractography, microdissection (Martino et al., 2013 and Wang et al., 2016) and NHP tract tracing. Another limitation is that our ROI selection was guided by anatomical assumptions, which relied on the histological literature to inform our frontal and parietal ROI placement. Placing these ROIs inherently restricts the resulting streamlines to between these regions and presupposes the existence of the SLF-I in the human brain, potentially biasing our results. Further, it is important to note that lack of volumetric overlap, and tractography at large, cannot fully address the SLF-I and cingulum controversy. Indeed, many tracts show volumetric overlap while still being separable and distinct, such as the SLF-III and arcuate fasciculus (Janelle et al., 2022). While the distinction of these two tracts may be an issue of neuroanatomical categorization or nomenclature, direct tissue validation is needed to add further insight to this issue.
Recent human neuroimaging studies have also utilized monkey neuroanatomical and neurophysiological findings to elucidate the cognitive and organizational properties of the human brain (Rolls et al., 2022), including the higher auditory and language system (Rauschecker et al., 1995; Kaas and Hackett, 2000; Rauschecker and Scott, 2009).
Studies like these and this study point to the possibilities of using animal histological findings to guide human brain mapping, potentially furthering understanding of human cognition and behavior.
In conclusion, we generated probabilistic maps of the SLF subdivisions in the human brain. We also specified the potential cortical terminations of SLF subdivisions in detail, which closely resemble tract-tracing findings, and distinguished the SLF-I as likely a separable tract from the cingulum. Lastly, our exploratory analysis suggests that the SLF subdivisions may have different contributions to executive and memory functions. These tools take us further into establishing an anatomical base for studying frontoparietal functional and structural organization. The SLF subdivision probabilistic maps and ROIs are available at https://github.com/mamandola/SLF_probabilistic_map.
Footnotes
We thank Dr. Ryan Wales for proofreading the manuscript and Elizabeth Slattery for performing quality assessment of the validation dataset. Research reported in this publication was supported by the National Institute On Aging of the National Institutes of Health under Award Number U01AG052564 and by funds provided by the McDonnell Center for Systems Neuroscience at Washington University in St. Louis. The HCP-Aging 2.0 Release data used in this report came from DOI: 10.15154/1520707.
The authors declare no competing financial interests.
- Correspondence should be addressed to Matthew Amandola at matthew.amandola.1{at}vanderbilt.edu or Hoi-Chung Leung at hoi-chung.leung{at}stonybrook.edu.
