Participants
In this retrospective analysis we studied data from healthy subjects from the Swedish BioFINDER-2 study (Skåne University Hospitals, Sweden [NCT03174938]), which was approved by the Regional Ethical Committee in Lund, Sweden, (EPN file number 2016/1053). Participants were enrolled between 2014 and 2021 following attainment of written consent in accordance with the Declaration of Helsinki. For further study details, see http://biofinder.se and (Palmqvist, Janelidze et al. 2020). Briefly, study participants were recruited using the following inclusion criteria: (i.) absence of cognitive symptoms, (ii.) Mini-Mental State Examination (MMSE) score of 26-30 at baseline, (iii.) not fulfilling criteria for mild cognitive impairment or dementia according to DSM-5 (American Psychiatric Association 2013), (iv.) absence of active psychological or psychiatric disease and (v.) fluency in Swedish. Additional exclusion criteria used in the present study were: (i.) age ≥ 60 years old, (ii.) an abnormal CSF amyloid-ß42/40 ratio, described in the Supplementary Material, (iii.) a high volume of white matter hyperintensities, (> 3 standard deviations from the cohort mean) identified as described in the Supplementary Material, and (iv.) poor MRI image quality obscuring identification of the PCS.
Magnetic Resonance Image Acquisition
MRI scans were performed on a MAGNETOM Prisma 3T scanner (Siemens Healthineers, Erlangen, Germany), equipped with a 64-channel head coil. A T1w MPRAGE (magnetization-prepared rapid gradient-echo) sequence was acquired with the following acquisition parameters: repetition time: 1900 ms; echo time: 2.54 ms; echo spacing: 7.3 ms; voxel size: 1x1x1 mm3 and field of view: 256x256x176 mm3. GRAPPA (generalized autocalibrating partially parallel acquisitions33) was applied with acceleration factor of 2 and 24 reference lines. A single-shot echo-planar imaging sequence was used to acquire 104 diffusion-weighted imaging volumes (repetition time: 3500 msec; echo time: 73 msec; resolution: 2x2x2 mm3; field of view 220x220x124 mm3; b values range: 0, 100, 1000, and 2500 sec/mm2 distributed over 2, 6, 32, and 64 directions; twofold parallel acceleration and partial Fourier factor=7/8). A second diffusion MRI scan was also obtained with a reverse phase-encoding and 7 gradient directions (1 x b = 0 and 6 x b = 1000 s/mm2) for correction of susceptibility-induced distortions. A T2-weighted FLAIR scan (repetition time: 5000; echo time 393 ms, same resolution and FoV as for the T1-weighted image) was also acquired. Spontaneous blood oxygen level-dependent (BOLD) oscillations were acquired with a gradient-echo planar sequence (eyes closed, in-plane resolution = 3 × 3 mm2, slice thickness = 3.6 mm, repetition time = 1020 ms, echo time = 30 ms, flip-angle = 63°, 462 dynamic scans, 7.85 min)
Paracingulate Sulcus Measurement and Classification Criteria
Individuals were grouped in accordance with hemispheric presence of a PCS. PCS presence was identified via manual classification of structural T1 MRI data according to an adapted version or Garrison’s established protocol for PCS classification (Garrison, Fernyhough et al. 2015), which has been used and described previously (Harper, Lindberg et al. 2022, Harper, de Boer et al. 2023) and is documented in full in the Supplementary Material. Briefly, potential PCS, meeting classification criteria were manually traced and measured. As is standard amongst classification protocols hemispheres with a PCS ≥ 20mm in length were categorised as possessing a “present” PCS whereas hemispheres failing to meet these criteria were deemed to possess an “absent” PCS (Ono, Kubik et al. 1990, Yücel, Stuart et al. 2002, Le Provost, Bartres-Faz et al. 2003, Garrison, Fernyhough et al. 2015, Del Maschio, Sulpizio et al. 2019). Sulcation ratings were performed independently by two raters, LH and AS, who were blinded to individuals’ demographic data. Disagreement between raters was resolved by consensus.
MRI data processing
MPRAGE images were imported into MANGO (Multi-image Analysis GUI, v 4.0, http://ric.uthscsa.edu/mango/mango.html, The University of Texas Health Science Center) software and prepared, aligning the x axis in the sagittal plane with the bicommissural line (AC–PC). Further y and z axis rotational corrections were performed in order to ensure optimal orientation for analysis.
Tract Segmentation
The diffusion weighted data were processed using a combination of open-source algorithms. In brief the acquired images were denoised and the Gibbs ringing artifacts were removed using MRtrix3 (Tournier, Smith et al. 2019) routines. Correction for susceptibility induced distortions, using images acquire with opposite phase polarities, motion and Eddy currents was performed employing FSL Top-up (Andersson, Skare et al. 2003) and Eddy (Andersson and Sotiropoulos 2016) (FMRIB Software Library, version 6.0.4; Oxford, United Kingdom). Parametric maps of mean diffusivity (MD), fractional anisotropy (FA), axial diffusivity (AD) and radial diffusivity (RD) were computed using DIPY (Henriques, Correia et al. 2021) routines (https://dipy.org/) Following pre-processing of diffusion MRI scans, white matter tracts were segmented using, TractSeg (Wasserthal, Neher et al. 2018). Both the 72 tracts definition included in TractSeg and the 42 tracts definition derived from Xtract (Warrington, Bryant et al. 2020) were used in order to improve internal validity. Furthermore, the Xtract method divides the cingulate bundle into three distinct tracts offering a more focused analysis of white matter contiguous with the PCS. Diffusivity metrics and tract volumes were analysed in accordance with ipsilateral hemispheric PCS presence in the superior longitudinal fasciculus I (SLF-I) [both segmentation methods], the cingulum (CG) [TractSeg] and the dorsal (CBD), pre-genual (CBG), and temporal (CBT) cingulum [Xtract]. Further method description and quality control measures are documented in the Supplementary Material. Tract segmentations example are displayed in Figure 2.
Resting State Functional MRI pre-processing
Resting state functional MRI data pre-processing was performed using a pipeline composed of FSL(Jenkinson, Beckmann et al. 2012), AFNI(Cox 1996) and ANTS(Avants, Tustison et al. 2014). Anatomical processing involved skull stripping, segmentation of CSF, white and grey matter, and normalization to MNI152 space(Grabner, Janke et al. 2006). Following bulk motion and slice timing correction, nuisance regression compensated white matter/CSF signal, physiological noise(Behzadi, Restom et al. 2007), motion parameters(Johnstone, Ores Walsh et al. 2006), and scanner drift. Finally, the functional data were band-pass filtered (0.01–0.1 Hz) and transformed to MNI space. Frames causing outliers in total frame-to-frame signal variation (75 percentile + 1.5 interquartile range) were censored (Power, Barnes et al. 2012). Subjects with a mean/maximum framewise displacement exceeding 0.3/3.0mm were excluded. The processed functional MRI data were resampled to 6 x 6 x 6 mm3 and masked with grey matter derived from a cortical resting-state network atlas (Thomas Yeo, Krienen et al. 2011), Harvard-Oxford subcortical atlas (Desikan, Ségonne et al. 2006). The variance stabilized Fisher z-transformed Pearson correlation between the resulting grey matter BOLD voxel time series yielded our functional connectivity measure.
Statistical Analysis
Tract Segmentation analysis
Diffusivity metrics and tract volume analyses were performed in R software (R Version 4.2.1 CoreTeam 2016, https://www.r-project.org/) using general linear models, including age, sex, and handedness as covariates in all models. In addition, individual’s total intracranial volume was included as a covariate in all models analysing tract volume. As these analyses were explorative correction for multiple comparisons was not performed.
Seed-based Functional Connectivity analysis
The Salience/Ventral Attention, Default mode and Visual networks were defined geographically according to network parcel locations defined by the Schaefer 200 parcel 7 network atlas (Schaefer, Kong et al. 2017), further descriptions are provided in the Supplementary Material.
Functional connectivity (FC) analysis was performed using Pearson correlation coefficients between the mean time series of the 200 seeds corresponding to the 200 parcels of the Schaefer 200 parcel 7 network atlas. FC’s were converted into z-scores to improve normality using Fisher r-to-z transformation. Individuals z-scores were then averaged across ROIs relating to the predefined networks of interest. Finally, GLMs were fitted according to group averaged z-scores determined by ipsilateral PCS presence, controlling for the effects of age, sex, and handedness. Significance was identified at P = 0.05.
Voxel-based Functional Connectivity analysis
A medial frontal lobe region of interest (ROI) was created for each hemisphere using the Schaefer 200 parcel 7 network atlas (Schaefer, Kong et al. 2018). Selected parcels were those overlapping the predicted location of the PCS in MNI-152 space (Grabner, Janke et al. 2006). ROIs are detailed in Supplementary Figure 1.
Voxel wise whole brain connectivity in 6 x 6 x 6 mm3 space was evaluated using a two-step procedure. First the mean connectivity of all voxels was calculated using Persons r correlation.
The functional connectome was then restricted with a network mask corresponding to high connectivity with the medial frontal lobe ROIs by thresholding the all-subject-mean connectivity of all subjects at a correlation corresponding to P = 0.0001 (given the number of frames in the rs-fMRI time series). Cortical ROIs corresponding to this network mask, which included the bilateral anteromedial frontal cortices as well as portions of the insular, lateral temporal, parietal, and posterior cingulate cortices were then drawn on the resulting voxel-wise link density maps, see Figure 3. These regions are part of the DMN and SN resting state networks, which both overlap with the source region. As scattered connectivity was obtained with subcortical regions of the basal ganglia and hippocampus/amygdala, these structures were added to the ROI set using the anatomical definitions according to the Harvard-Oxford subcortical atlas (Desikan, Ségonne et al. 2006) and not by manual delineation. Note that the tracing of these regions only affected the visualization and labelling in the resulting connectograms and that the network mask used in the calculation was applied to the links and not the voxels.
In the second step, the whole brain functional connectome was limited to the identified regions connecting strongly to the ROIs and entered into a network based statistic component calculation (Zalesky, Fornito et al. 2010) comprising a connected set of links on which connectivity differed in accordance with ipsilateral hemispheric PCS presence, based on a binarized connectivity graph at a threshold of P < 0.001 (given group sizes), controlled for the effects of age, sex and handedness. Results of significant network components with altered connectivity and summarizing connectograms are displayed in Figure 4.
Data Availability
Anonymized data will be shared by request from a qualified academic investigator for the sole purpose of replicating procedures and results presented in the article if data transfer is in agreement with relevant legislation on the general data protection regulation and decisions and by the relevant Ethical Review Boards, which should be regulated in a material transfer agreement.