Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Medial temporal lobe contributions to resting-state networks

Medial temporal lobe contributions to resting-state networks The medial temporal lobe (MTL) is a set of interconnected brain regions that have been shown to play a central role in behavior as well as in neurological disease. Recent studies using resting-state functional magnetic resonance imaging (rsfMRI) have attempted to understand the MTL in terms of its functional connectivity with the rest of the brain. However, the exact characterization of the whole-brain networks that co-activate with the MTL as well as how the various sub-regions of the MTL are associated with these networks remains poorly understood. Here, we attempted to advance these issues by exploiting the high spatial resolution 7T rsfMRI dataset from the Human Connectome Project with a data-driven analysis approach that relied on independent component analysis (ICA) restricted to the MTL. We found that four different well- known resting-state networks co-activated with a unique configuration of MTL subcomponents. Specifically, we found that different sections of the parahippocampal cortex were involved in the default mode, visual and dorsal attention networks; sections of the hippocampus in the somatomotor and default mode networks; and the lateral entorhinal cortex in the dorsal attention network. We replicated this set of results in a validation sample. These results provide new insight into how the MTL and its subcomponents contribute to known resting-state networks. The participation of the MTL in an expanded range of resting-state networks is in line with recent proposals on MTL function. Keywords Medial temporal lobe · Functional connectivity · Resting-state fMRI · Independent component analysis · Dual regression Abbreviations mEnt Medial entorhinal cortex MTL Medial temporal lobe hHi Head of the hippocampus DMN Default mode network bHi Body of the hippocampus rfMRI Resting-state functional magnetic resonance tHi Tail of the hippocampus imagingTh Thalamus proper MNI Montreal Neurological Institute 152Cd Caudate aPHG Anterior parahippocampal cortexPu Putamen pPHG Posterior parahippocampal cortexPal Pallidum lEnt Lateral entorhinal cortexHi Hippocampus Amg Amygdala Ac Accumbens area vDC Ventral DC * Niels Janssen njanssen@ull.es STS Bankssts cACC Caudal anterior cingulate Department of Cognitive, Social and Organizational cdMF Caudal middle frontal Psychology, Faculty of Psychology and Speech Therapy, Cun Cuneus University of La Laguna, San Cristóbal de La Laguna, Spain Ent Entorhinal Department of Basic Medical Sciences, Faculty FuG Fusiform of Health Sciences, University of La Laguna, San Cristóbal de La Laguna, Spain iP Inferior parietal iT Inferior temporal Institute of Biomedical Technologies, University of La Laguna, San Cristóbal de La Laguna, Spain ICG Isthmus cingulate lO Lateral occipital Instituto Universitario de Neurociencia, University of La Laguna, San Cristóbal de La Laguna, Spain Vol.:(0123456789) 1 3 996 Brain Structure and Function (2022) 227:995–1012 lOF Lateral orbitofrontal et al. 2019). Characterizing the whole-brain functional net- LgG Lingual gyrus works that co-activate with the MTL has implications for mOF Medial orbitofrontal understanding its role in health and disease. One potential mT Middle temporal reason for why this issue remains unresolved may be due PHG Parahippocampal gyrus to methodological limitations in previous studies. Here, we PCL Paracentral relied on a data-driven parcellation of the MTL using the IFGOp Pars opercularis whole-brain high spatial resolution 7T rsfMRI dataset from IFGOr Pars orbitalis the Human Connectome Project (HCP). IFGTr Pars triangularis The standard view on the connectivity between the MTL PCAL Pericalcarine and the rest of the brain is that the MTL is connected with PoG Postcentral two distinct whole-brain networks. Both anatomical and pCC Posterior cingulate functional connectivity studies have shown that MTL con- PrG Precentral nectivity is largely organized along a posterior–anterior PCun Precuneus gradient. For example, tract-tracing studies in monkeys and rACg Rostral anterior cingulate rodents have found that posterior sections of the parahip- rMF Rostral middle frontal pocampal gyrus (PHG) and posterior sections in the hip- SF Superior frontal pocampal formation show increased (mono- or poly-syn- SP Superior parietal aptic) connectivity with posterior midline regions like the ST Superior temporal retrosplenial cortex and posterior cingulate cortex, whereas SMG Supramarginal anterior sections of the parahippocampal cortex and ante- FrP Frontal pole rior sections of the hippocampal formation show increased TmP Temporal pole connectivity with anterior brain regions like the orbitofron- TTG Transverse temporal tal cortex and amygdala (Aggleton 2012; Jones and Witter Ins Insula 2007; Kobayashi and Amaral 2007, 2003; Kondo et al. 2005; Rosene & Van Hoesen 1977; Strange et al. 2014; Suzuki and Amaral 1994). Similarly, lateral sections of the entorhinal Introduction cortex (Ent) have been associated with the posterior whole- brain network, whereas medial Ent has been associated with The medial temporal lobe (MTL) has received much inter- the anterior network (Jones and Witter 2007; Strange et al. est in research and the clinic due to its key implication in 2014). More recent functional connectivity studies using memory processes (e.g., Alvarez and Squire 1994; Suzuki rsfMRI have further confirmed this bipartite organization and Amaral 2004; Squire et al. 2007), as well as due to its of MTL whole-brain connectivity. Specifically, Libby et al. involvement in several relatively common pathological (2012) found that seeds placed in posterior PHG revealed co- conditions (e.g., temporal lobe epilepsy, schizophrenia and activity with posterior midline regions like the retrosplenial Alzheimer’s disease; Douw et al. 2015; Seidman et al. 2003; cortex, precuneus, posterior cingulate and occipital cortex, Kenkhuis et al. 2019; Govindpani et al. 2020). The MTL whereas seeds placed in more anterior locations produced encompasses a number of different anatomical structures, co-activity with orbitofrontal cortex and inferior temporal primarily the parahippocampal and entorhinal cortices as cortex. Other studies placing seeds in various locations along well as the hippocampal formation. Recent resting-state the hippocampal long axis have revealed a similar separation fMRI (rsfMRI) studies have attempted to understand the of co-activity between posterior and anterior networks (Kahn MTL by considering its functional connectivity with the rest et al. 2008; Qin et al. 2016), as have studies examining the of the brain (Kahn et al. 2008; Libby et al. 2012; Qin et al. entorhinal cortex (Schröder et al. 2015). In short, evidence 2016; Ranganath and Ritchey 2012; Ritchey et al. 2015; from various sources now confirms that MTL connectivity Ruiz-Rizzo et al. 2020; Schröder et al. 2015; Wang et al. can be associated with both a posterior and an anterior net- 2016). An on-going debate regarding this issue concerns the work (Ranganath and Ritchey 2012; Ritchey et al. 2015). different whole-brain functional networks that connect to the However, three recent functional connectivity studies have MTL. Specifically, whereas the traditional view is that the presented results that have challenged this view. These studies MTL interacts mainly with two whole-brain networks (Kahn have relied on data-driven approaches to examine functional et al. 2008; Libby et al. 2012; Qin et al. 2016; Ranganath and connectivity thereby avoiding potential biases inherent in Ritchey 2012; Ritchey et al. 2015; Schröder et al. 2015; Bar- functional connectivity techniques that rely on placing seeds nett et al. 2019), other recent studies using data-driven tech- (Zuo et al. 2010). Specifically, Wang et al. (2016) examined niques have found that the MTL connects with additional the slice-by-slice connectivity between both the parahip- networks (Ruiz-Rizzo et al. 2020; Wang et al. 2016; Plachti pocampal gyrus as well as the hippocampus and the rest of 1 3 Brain Structure and Function (2022) 227:995–1012 997 the brain using a hierarchical clustering technique. Interest- the spatially restricted group ICA technique (srICA; Bless- ingly, they found that there were three connectivity clusters ing et al. 2016; Ezama et al. 2021; Formisano et al. 2004). along the parahippocampal long axis, one in posterior PHG Whole-brain ICA is frequently used to separate signal from that connected to the aforementioned posterior network, one noise in fMRI studies (e.g., Janssen and Mendieta 2020; in anterior PHG (termed perirhinal cortex) that connected Smith et al. 2013). Instead, by applying ICA to a particular to the anterior network, and one in an even more anterior brain region, the noise profile that is specific to that brain PHG location that connected to a network of regions that region will be taken into account and result in a more sensi- included the insula, post central gyrus and amygdala. Simi- tive separation of signal from noise in that region. larly, Ruiz-Rizzo et al. (2020) examined functional connec- The current study relied on high spatial resolution data tivity of the MTL as well as the amygdala using a spatially as well as a targeted analysis approach to clarify the con- restricted Independent Component Analysis (ICA) approach tributions of the MTL in existing resting-state networks. (Blessing et al. 2016; Formisano et al. 2004). They found that Specifically, whole-brain functional connectivity maps (FC clusters of activity detected inside the MTL and amygdala maps) associated with the MTL clusters found by the srICA co-activated with sets of brain regions that correlated with were computed using the Dual Regression technique (Nick- the reference networks of Allen et al. (2011). Specifically, erson et al. 2017). These whole-brain FC maps reflected in addition to the default mode network (correlation varied the large-scale co-activity with the specific MTL clusters from r = 0.12 to r = 0.48 for different MTL activity clusters), found in the previous step. Next, MTL clusters were classi- they also found that MTL was somewhat connected to other fied as signal or noise on the basis of an algorithm that relied networks like the salience (r = 0.14 ), frontal (r = 0.11 ), basal on the correlation with the 7 known resting-state networks ganglia ( r = 0.40 ) and visual networks ( r = 0.11 ). Finally, obtained by Yeo et al. (2011). We then used linear mixed a recent paper by Plachti et al. (2019) focused on the hip- effect regression analyses to calculate the relative contribu- pocampus using a consensus clustering technique and found tion of each MTL subcomponent in the different resting-state that connectivity between the hippocampus and the rest of the networks. Specic fi ally, the MTL was segmented into anterior brain was best described by 3, 5 and even 7 clusters. and posterior portions of the PHG, head, body and tail of Thus, it appears that whereas some studies have found the hippocampus and mEnt and lEnt. Finally, we relied on a that the MTL is connected to two different whole-brain test-validation approach in which results obtained in the test functional networks, others have found it is connected to dataset were validated on a second dataset. additional different networks. This empirical discrepancy may be the result of methodological limitations in the afore- mentioned studies. First, previous studies have relied on Methods relatively low spatial resolution fMRI acquisition protocols ( ∼ 3.5 mm voxels). One concern with such low resolution Participants data is that the large size voxels may be unable to accurately separate signals from different resting-state networks leading Data for all participants were downloaded from the Human to variability in the reported results. Second, previous stud- Connectome Website. The initial dataset consisted of 184 ies have not sufficiently taken into account the observation participants who had participated in the 7T data acquisition. that the MTL is affected by local magnetic field inhomoge- However, data from 12 participants were excluded due to neities that lead to reduced temporal signal-to-noise ratios the presence of specific Quality Control issues identified by (tSNR; e.g., Olman et al. 2009; Weiskopf et al. 2006). The the HCP (i.e., QC issues A, B, C, and D). The final sample, additional noise in the MTL region may hamper the detec- therefore, consisted of 172 participants, between the ages tion of MTL contributions to resting-state networks and also of 22 and 35 (104 females). Further detailed description on produce inconsistencies in the reported results. In this study, the study subjects may be found in Van Essen et al. (2012). we addressed these two limitations in three ways. First, we The data analyses were conducted in agreement with the addressed the separability of signals inside the MTL using declaration of Helsinki and with the protocol established a dataset with high spatial resolution (1.6 mm isotropic) by the Ethics Commission for Research of the Universidad acquired at a high field strength (7T). Previous studies have de La Laguna, the Comité de Ética de la Investigación y shown that compared to lower resolutions at 3T, increased Bienestar Animal. spatial resolution at 7T results in more clearly defined rest- ing-state networks (Vu et al. 2017). Second, previous studies Data acquisition and preprocessing have also shown that reducing the voxel size in fMRI data avoids partial voluming effects and leads to improved signal Data packages herein used come from the WU-Minn HCP detection in areas with low tSNR (Hyde et al. 2001; Rob- Data-1200 Subjects data set. For this experiment, we down- inson et al. 2004; Sladky et al. 2013). Finally, we relied on loaded 7T Resting-State fMRI 1.6 mm/32k FIX-Denoised 1 3 998 Brain Structure and Function (2022) 227:995–1012 (Compact) and Resting-State fMRI FIX-Denoised ms, flip angle = variable, and also resulting in 0.7 mm iso- (Extended) datasets. As per the HCP reference manual, tropic voxels. The structural images were acquired on a 3T these data were acquired by the Washington-University and Siemens Connectom Skyra scanner. The downloaded struc- Minnesota Consortium, with a Siemens Magnetom 7T MR tural packages contained the T1w and T2w images for each Scanner and a Nova32 32-channel Siemens receive head coil participant as well as the full Freesurfer output and trans- from Nova Medical. Four 16-min-long rsfMRI acquisitions formation matrices that were relevant for our downstream were acquired per subject. RsfMRI acquisitions alternated analyses (see below). For additional specific information on the direction of the phase encoding gradient, where two the pre-processing of these structural images, we refer to sessions were acquired in the posterior-to-anterior phase Glasser et al. (2013). direction and the other two in anterior-to-posterior phase direction. For the resting-state acquisitions, participants Data analysis were instructed to fix their sight on a white cross-hair over a dark background (Smith et al. 2013). MRI scanning param- The aim of this study was to explore the different whole- eters for the resting-state data were based on acquisitions brain networks that co-activate with the MTL as well as how of Gradient-Echo EPI volumes. Each volume contained 85 the different MTL subcomponents contribute to these dif- slices that were acquired with a multiband factor of 5. Slice ferent networks. To approach these objectives, our analyses thickness was 1.6 mm with no gap, the FOV was 208 × 208 were divided into four main steps. A graphical representa- mm, matrix size 130 × 130, resulting in 1.6 mm isotropic tion of the workflow is displayed in Supplementary Fig. S1. voxels. The TR was 1000 ms, echo time (TE) 22.2 ms, and The most relevant results from these analyses will be made the flip angle 45 . We used the first two of these four rsfMRI available on our github page (https:// github. com/ iamni elsja datasets as a test set (32 minutes of resting-state data), and nssen). the last two datasets as a validation set. Both test and valida- tion datasets had alternating phase directions. Segmentation of MTL into subcomponents The downloaded fMRI datasets consisted of already pre-processed functional data according to HCP minimal The first step of our analysis involved the segmentation of preprocessing pipelines ( Glasser et al. 2013). Briefly, trans- the MTL into a number of subcomponents. This segmenta- formations that reduce head motion were estimated using tion took place in a participant-specific manner, meaning FSL MCFLIRT (Jenkinson et al. 2002), fieldmap and gradi- that each segmentation took into account the unique shape ent distortion corrections were applied, and transformations of the MTL in each participant’s brain. The MTL subcom- from fMRI space to MNI space were estimated using non- ponents were the head, body and tail of the hippocampus linear transformations. Importantly, smoothing of the data (hHi, bHi, and tHi), the anterior (aPHG) and posterior was minimized in two ways: First, the transform from native (pPHG) parahippocampal gyrus, as well as the medial and to MNI space preserved the native space resolution of the lateral entorhinal cortex (lEnt and mEnt). All segmentations fMRI acquisitions, and second, all transformations (motion relied on the Desikan–Killiany cortical Atlas as well as the correction, fMRI to MNI space) were postponed, com- subcortical segmentation that is produced by Freesurfer bined and applied in a single step using sinc interpolation. and that was included with the downloaded dataset (i.e., Next, the data in MNI space were temporally filtered using the aparc+aseg atlas in Freesurfer terminology; Desikan a 2000-s high-pass filter and automatically denoised using et  al. 2006). This provides an automatic segmentation of the FIX program (Griffanti et al. 2014; Salimi-Khorshidi the brain in terms of a set of 42 brain regions that are fitted et al. 2014). This program uses semi-automatic classifica- to the unique morphology of each participant’s brain. This tion of head-motion and other artifacts which minimized the is achieved by combining prior information about the prob- potential impact of head-motion artifacts in our data (Salimi- able spatial location of a given brain area and its surround- Khorshidi et al. 2014). The final files were demeaned and ing structures with information about the morphology of had native 1.6 mm isotropic resolution in MNI space. We a specific target participant brain. This way of segmenting then extracted and regressed out the CSF and WM signal the brain into regions is, therefore, more accurate than other that was obtained from each participant’s wmparc file. atlas segmentations that are based on normalized brains. In addition, for the structural data, we downloaded the To obtain the three subdivisions for the hippocampus 3T Structural Preprocessed and 3T Structural Preprocessed we relied on the Hippocampal Subfields and Nuclei of the Extended packages. Again as per the HCP reference man- Amygdala script (v21) with the (0.7 mm) T2w image as ual, the T1w images were acquired using a 3DMPRAGE the input (Iglesias et al. 2015). Besides segmenting the hip- protocol TI/TR/TE: 1000/2400/2.14 ms, flip angle = 80 , pocampus into a set of internal subfields, this script also resulting in 0.7 mm isotropic voxels. The T2w images were provides a segmentation of the hippocampus into its head, acquired using a 3D T2-SPACE protocol TR/TE: 3200/565 body and tail sections. In addition, anterior and posterior 1 3 Brain Structure and Function (2022) 227:995–1012 999 PHG were obtained by first extracting the parahippocam- were activated during the resting state using a data-driven pal gyrus mask from a given participant’s aparc+aseg atlas. approach. We first created MTL masks combining bilateral Next, to define aPHG and pPHG, we computed an inter - hippocampus, parahippocampal gyrus and entorhinal cor- section of a plane with the centroid point of the parahip- tices that were specific to each participant. These masks pocampal gyrus. The centroid point in the anterior–posterior were similar to those described above in that they were direction was computed as the mean voxel coordinate of derived from the participant-specific aparc+aseg atlas, but the parahippocampal mask along the y-axis. Consequently, were different because they did not distinguish between the the voxels posterior to this y-plane were defined as pPHG individual MTL subcomponents. This was because we first and voxels anterior to the plane as aPHG. To separate the aimed to detect locations inside the entire MTL that are entorhinal cortex in a medial and lateral section we com- active during the resting state without taking into account puted its centroid point as the mean coordinate along the the various MTL subcomponents. To improve accuracy x-axis and den fi ed medial and lateral sections as above. This, for the group analyses, each participant’s MTL mask was therefore, produced participant-specific MTL subcomponent increased in size by one voxel. We then multiplied the masks for head, body and tail of the hippocampus, anterior masks with the cleaned, whole-brain resting-state fMRI and posterior PHG, and medial and lateral entorhinal cortex data. This produced 4D fMRI files containing only the (see Fig. 1 for a graphical presentation of the location of the timeseries of the voxels within the MTL mask of each par- MTL and its various subcomponents for a representative ticipant. We then performed group spatially restricted ICA participant). (group srICA) over these data using FSL melodic (v3.15). Given the smaller sized dataset due to the masking, we Detection of MTL activation clusters used the default method for data reduction (i.e., an initial principal component analysis) by disabling MIGP in the The next step of the analysis had three goals. First, we melodic options (Smith et al. 2014). attempted to identify those locations of the MTL that Fig. 1 Visualization of the MTL subcomponents. Location of MTL in parahippocampal gyrus (pPHG), anterior parahippocampal gyrus the left (a) and right hemispheres (b) of the whole brain along with a (aPHG), hippocampal tail (tHi), body (bHi) and head (hHi), as well zoomed view (c) as well as a tagged zoomed view (d). Panels c and d as medial (mEnt) and lateral (lEnt) entorhinal cortex highlight the various substructures that make up the MTL, posterior 1 3 1000 Brain Structure and Function (2022) 227:995–1012 As mentioned in the Introduction, the main advantage of matrices of sizes m × n where m refers to the number of srICA compared to whole-brain ICA is that a more sensi- dimensions (1–15), and n to the number of resting-state net- tive decomposition of signals in a specific region can be works (here 7). Inspection of these 15 correlation matrices obtained. As also mentioned above, a well-known problem revealed the set of resting-state networks that frequently had with fMRI data is that the GRE EPI acquisition protocols high correlation ( r > 0.40 ) with a set of components across are sensitive to local magnetic field inhomogeneities (Devlin all dimensions. We then executed an algorithm that found a et al. 2000) which lead to variability in the tSNR between component if its maximum correlation with a given resting- the different areas of the brain (see Supplementary Fig. S2 state network was above a threshold ( r > 0.4) and if this max for an overview of the tSNR in the current dataset as well maximum correlation was sufficiently higher than the second max as evidence that MTL regions had substantially lower tSNR highest correlation ( > 1.3 ), both within the same IC and max compared to other areas). A consequence of the multivariate within the same resting-state network. On the assumption nature of ICA is that signals coming from regions with low that a larger number of dimensions leads to more fraction- tSNR are less likely to be included in a component (connec- ated component clusters, we then chose the smallest dimen- tivity) map. This is because the timecourses of regions with sion at which this algorithm produced the largest number of increased noise (i.e., low tSNR) are less likely to correlate ICs. This procedure, therefore, detected in a data-driven with the timecourses of other regions. One way to deal with fashion the optimal number of dimensions for which the this issue is to apply ICA to a specific region with a low srICA produced the largest number of voxel clusters inside tSNR. In this case, the ICA procedure will take into account the MTL that were both sensitive and specific to known the specific noise profile of that region and enable a more resting-state networks. sensitive separation between voxels clusters that correspond to noise and voxel clusters that correspond to signal. In other Group‑level analyses of whole‑brain FC words, given that the MTL is a region known to have low tSNR (Olman et al. 2009 and see Supplementary Fig. S2), To anticipate our results, the previous steps resulted in the we reasoned that srICA applied to the MTL would lead to a detection of a set of FC maps that closely corresponded more sensitive detection of signal clusters inside the MTL to a set of known resting-state networks. The next step in compared to a whole-brain ICA approach. the analyses was to assess the statistical reliability of the One aspect of ICA is that it requires a decision about observed whole-brain maps at the group level. One standard the number of dimensions under which the analysis is per- way of performing such an analysis would be to rely on a formed. Here, we used a method to determine the optimal voxel-based modeling tool such as FSL randomize. How- number of dimensions for the ICA that was developed in ever, a general problem with this approach is that it assumes our laboratory (Ezama et al. 2021). Specifically, we tested that each voxel represents the exact same brain region across across a wide range of different dimensions the relationship all participants. However, as has been discussed at length between a set of components found for a specific dimension elsewhere, there is large variability in brain morphology and a set of known resting-state networks (Yeo et al. 2011). between participants, and therefore, group-level analyses of We then chose the dimension at which this relationship was this type are sub-optimal (Anticevic et al. 2008; Fischl et al. optimal. Specifically, we first performed ICA on the same 2008). Instead, we opted for a different analysis approach dataset at dimensions ranging from 1 to 15 in a stepwise that took into account the unique morphology of each par- fashion. Next, we obtained whole-brain FC maps derived ticipant’s brain. Specifically, we created a dataset that, for from all MTL activation clusters for each dimension. These each participant, contained average co-activity values for all FC maps were obtained using the Dual Regression technique their cortical and subcortical regions from the aparc+aseg (Nickerson et al. 2017). The dual regression approach con- atlas. We obtained these data by intersecting each partici- sisted of a first regression of the IC outputs of melodic to pant-specific aparc+aseg atlas with each participant-specific the cleaned and whole-brain fMRI data for each participant. whole-brain maps obtained from Dual Regression. We then The output of this first step in dual regression are the time- fitted these data to a linear mixed effect regression model courses of each independent component for each participant. of the form: The second step in dual regression involves a regression of the time-courses associated with each participant-specific IC Z = hemisphere + FC_map × brain_region + rand (participant), (1) map against the cleaned fMRI data. This second step pro- duced the whole-brain maps that represent the FC between where hemisphere was a discrete co-variable with two levels a specific IC and the rest of the brain. (left vs right), FC_map was a factor with number of levels These whole-brain FC maps obtained for each of the 15 equal to the number of ICs corresponding to resting-state dimensions were then correlated with the 7 resting-state net- networks detected in the previous step, brain_region was a works of Yeo et  al. (2011). This led to 15 correlation 1 3 Brain Structure and Function (2022) 227:995–1012 1001 factor with number of levels equal to the sum of the number obtained from Dual Regression. These data were then fit- of cortical and subcortical regions in the aparc+aseg atlas, ted to the same statistical model as described in Equation 1, and participant was a random factor with number of lev- except that the term brain_region now referred to the seven els equal to the total number of participants (i.e., 172). The MTL subcomponents. As before, our specific interest was in dependent variable was the average Z value for each brain the interaction term of the model ( FC_map × brain_region ) region computed from the participant-specific Dual Regres- that provided a test of the null hypothesis of whether the sion maps. Importantly, this mixed-effect model included a seven MTL subcomponents were activated in the same way random effect term for participant that takes into account the across the various FC maps. However, the post hoc tests that likely between-participant variability that is inherent in these were performed when this interaction term was significant data. In addition, this modeling approach uses participant- differed from those described above. Specifically, to deter - specific masks that leads to group-level results that do not mine the relative contribution of the MTL subcomponents to violate the assumption of unique brain morphology. the different resting-state networks, we first performed pair - Within this model, our specific interest was in the interac- wise comparisons of all seven MTL subcomponents within tion term of the model ( FC_map × brain_region ) that pro- each FC map. This produced a list of 21 pairwise compari- vided a test of the null hypothesis that brain regions would sons for each FC map with a test statistic (i.e., the z ratio, not have differences in mean co-activity values across the see below) that reflected the degree to which a given MTL dier ff ent FC maps. In other words, this interaction would not subcomponent differed from another MTL subcomponent. be significant if the different activation clusters detected in These pairwise test statistics were then summed, ordered, the MTL would be connected to the exact same whole-brain and thresholded at > 0. This therefore produced for each resting-state networks. When this interaction was significant, detected resting-state network an ordered list of the relative we performed post hoc tests where we compared for each FC contributions of each MTL subcomponent. map, the average Z value for a given region versus the mean of the other regions (an “effect” contrast). This, therefore, Validation analysis produced for each FC map, a list of cortical and subcorti- cal regions from the aparc+aseg atlas that had significantly To confirm the reliability of our results, we attempted to more co-activity compared to all other regions. validate our findings in a second dataset. This validation All modeling took place in the statistical computing dataset consisted of two additional rsfMRI acquisitions from environment R (v4.0.0). Mixed effect modeling relied on the same participants included in the test dataset. The two the lme4 package (v1.1.23; Bates et  al. 2007). Results rsfMRI scans from the validation set were acquired in a dif- from these regression models are presented in the form of ferent scanning session (on a different day) as the test set, but ANOVA tables that were computed directly from the output relied on the same MRI acquisition parameters. The preproc- of the mixed effect models using the lmerTest package essing protocol used was the same as described for the test (v3.1-2; Kuznetsova et al. 2017). P values in these models dataset. To validate the results, whole brain FC maps were were computed using the Satterthwaite correction for the computed from the data in the validation set using the MTL degrees of freedom. Post hoc testing was performed using clusters obtained in the test set. This analysis, therefore, pro- the emmeans package (v1.4.6; Lenth et al. 2018) when a vides a validation of the degree to which the MTL clusters given interaction term was significant (i.e., p < 0.05 ). P val- we obtained in the spatially restricted ICA step of the analy- ues in these post hoc tests were adjusted for multiple com- sis generalize to different datasets. We quantified this step parisons using the Bonferroni method. We visualized these by computing the correlation between the whole-brain FC results using the ggseg (v1.5.4; Mowinckel and Vidal- maps in the test and validation sets, and by comparing the Piñeiro 2019), and ggpubr packages (v0.3.0; Kassambara correlations of the whole-brain FC maps with the reference 2018). networks of Yeo in the test and validation sets. Relative contributions of MTL subcomponents Results The previous srICA step provided us with several clusters inside the MTL that co-activated with different resting-state Detection of MTL activation clusters networks. The final step in the analyses was to determine the relative contributions of the seven MTL subcompo- The procedure for finding the optimal number of dimen- nents (body hippocampus, anterior PHG, etc.) to the dif- sions first returned that across all the 15 dimensions tested, ferent resting-state networks. To do this, we intersected the resting-state networks 1 (visual), 2 (somatomotor), 3 (dorsal participant-specific MTL subcomponent masks (described attention), and 7 (default mode) were most frequently found above) with each participant-specific whole-brain FC map with correlations r > 0.40 . In addition, the algorithm found 1 3 1002 Brain Structure and Function (2022) 227:995–1012 that dimension 7 was the lowest dimension at which the larg- restricted ICA occupy positions within the MTL that both est number of ICs were strongly and uniquely connected to respect and cross anatomical boundaries (e.g., IC0 seems different resting-state networks. Specifically, we found that to reflect activity in both parahippocampal gyrus as well as for dimension 7, four ICs were strongly and uniquely corre- entorhinal cortex). The statistical reliability of the observed lated with four different resting-state networks: IC0 was cor - whole-brain networks as well as how the detected activation related with the dorsal attention network ( r = 0.42 ), IC1 was clusters are distributed across the various MTL subcompo- correlated with the somatomotor network ( r = 0.53 ), IC2 nents was examined in more detail below. was correlated with the default mode network ( r = 0.59 ) , and IC3 was correlated with the visual network ( r = 0.66 ; Group‑level analyses of whole‑brain FC see Table 1 for an overview of the correlations for each IC with all networks). As can be seen in Supplementary Fig. Group-level mixed-effect regression analysis revealed main S3, strong correlations ( r > 0.40 ) were frequently found for effects of Hemisphere ( F = 21.72, p < 0.0001 ), FC 1,59508 these four networks in other dimensions, suggesting that the Map ( F = 2955.38, p < 0.0001 ), and Brain Region 3,59508 detection of these four networks was not idiosyncratic to ( F = 351.21, p < 0.0001 ). Most relevant for our pre- 43,59508 dimension 7. In addition, as can be seen in Supplementary sent purposes, there was a significant interaction between Fig. S4, we also examined dimensions 20 and 30 and this FC Map and Brain Region ( F = 589.18, p < 0.0001 ), 129,59508 did not lead to the detection of new networks. Finally, as can suggesting that co-activity values of brain regions differed be seen in Supplementary Fig. S5, the ICs that were uncor- between the various whole-brain FC maps associated with related with the reference networks were indeed unlikely to each IC. As can be seen in Table  2, post hoc analyses reflect real BOLD signal. Specifically, IC4 seems to reflect revealed that IC0 and its corresponding FC map revealed signal in CSF, IC5 in draining veins, and IC6 does not reveal regions typically associated with the dorsal-attention net- much signal in the first place (see also Griffanti et al. 2017, work like the inferior and superior parietal cortex, as well for further information on manually classifying ICs). We as lateral frontal areas. Similarly, as can be seen in Table 3, can, therefore, conclude that for our data the specic fi clusters IC1 and its associated FC map showed regions typically of voxels detected by the ICA using dimension 7 for IC0, associated with the somatomotor network like the amyg- IC1, IC2 and IC3 were optimal in connecting with the four dala and post- and pre-central gyri. In addition, as can been known resting-state networks (see also Supplementary Fig. seen in Table 4, IC2 and its corresponding FC map revealed S6 for a direct comparison in overlap between the obtained regions generally found in the default mode network, like ICs and Yeo networks). the isthmus cingulate (retrosplenial cortex), the precuneus, A visual presentation of the precise location of the four and the medial orbitofrontal cortex. Finally, as can be seen ICs along with their whole-brain group-level FC map in Table 5, IC3 and its associated FC map revealed regions derived from Dual Regression is presented in Fig. 2. As can associated with the visual network like the cuneus and peri- be seen in Fig. 2, the four ICs are located at different posi- calcarine sulcus (see also Fig. 4B and C for a visual presen- tions in the MTL and revealed contrasting FC with the rest tation of these results). of the brain, indicating the involvement in different resting- state networks. A more detailed view of the location of each Relative contributions of MTL subcomponents IC within the MTL can be seen in Fig. 3A–C. In this Figure, it can be seen that the four clusters detected by spatially Group-level mixed-effect regression analyses that exam- ined the relative contribution of the MTL subcompo- nents to each resting-state network revealed main effects of Hemisphere ( F = 45.96, p < 0.0001 ), Br ain Table 1 Table of correlations of the FC maps with resting-state net- 1,9432 works Region ( F = 1867.63, p < 0.0001 ) and FC Map 6,9432 ( F = 2378.99, p < 0.0001 ). Again, important for our Yeo et al. (2011), seven networks IC0 IC1 IC2 IC3 3,9432 present purposes, the interaction between Region and FC Visual 0.13 0.12 0.02 0.66 Map was highly significant ( F = 1442.85, p < 0.0001 ), 18,9432 Somatomotor 0.02 0.53 0.03 0.24 suggesting that average co-activity values for each MTL sub- Dorsal attention 0.42 0.05 0.01 0.16 component differed between the four ICs. Further explora- Ventral attention and salience 0.03 0.01 0.02 0.21 tion of this interaction using pairwise comparisons of the Limbic 0.04 0.10 0.00 0.02 seven MTL subcomponents within each IC and then ranking Executive control 0.20 0.02 0.11 0.00 the summed z ratios revealed the relative contributions of Default mode 0.12 0.20 0.59 0.01 each MTL subcomponent. Specifically, as can be seen in Table 6, summed z ratios in IC0 (correlated with the dorsal- Correlations that were strongly and unique correlated with specific attention network) were strongest in pPHG, then in aPHG resting-state networks are shown in bold 1 3 Brain Structure and Function (2022) 227:995–1012 1003 Fig. 2 The subset of Independent Components (IC0, first row; IC1, co-activation quantified in Z values. Note how different IC hotspots second; IC2, third; IC3, fourth) detected as signal from the spatially inside the MTL connect to different areas of the brain that highly restricted ICA (left columns, srICA), and their corresponding large- overlap with resting-state networks (see also Supplementary Fig. S6). scale functional connectivity maps derived from Dual Regression srICA spatially restricted independent component analysis, IC inde- (right columns, DR). The color gradient represents the group level of pendent component, DR dual regression and finally in lEnt. Similarly, Table  6 showed that for IC1 correlations. Specifically, the FC maps associated with (correlated with the somatomotor network), summed z ratios IC0 (DA), IC1 (SM), IC2 (DMN) and IC3 (VIS), corre- were strongest in hHi followed by bHi, and tHi. In addi- lated r = 0.95, r = 0.96, r = 0.98, r = 0.97 between the test tion, Table 6 showed that for IC2 (correlated with the default and validation set, respectively. Finally, the pattern of cor- mode network) summed z ratios in descending order were relations between the obtained FC maps in the validation ranked pPHG, aPHG, hHi and bHi. Finally, Table 6 showed set and the Yeo reference networks was highly similar (see that for IC3, summed z ratios were strongest in pPHG (see Table S1). We, therefore, conclude the MTL clusters that we also Figure 5 for a graphical presentation of these results). report are robust and generalize to different datasets. Validation results Complementary results Calculation of the whole-brain FC maps in the validation One concern with the results reported above is that we only dataset using the MTL clusters obtained from the test set compared the whole-brain FC maps with the set of 7 ref- described above showed highly similar results to those erence networks of Yeo et al. (2011). In complementary obtained in the test set (see Figure S7). Correlation of FC analyses, we examined whether our results would general- maps in the test and validation sets showed highly reliable ize to the 17 reference networks of Yeo et al. (2011), the 1 3 1004 Brain Structure and Function (2022) 227:995–1012 Fig. 3 Detailed location of the main target hotspots in the MTL Note how ICs both respect and cross anatomical boundaries suggest- detected with spatially restricted ICA. IC0 (blue), IC1 (green), IC2 ing inter-structure communication. Slices in 1 mm MNI152 space. (orange) and IC3 (red) in saggital (a), coronal (b) and axial (c) views. DA dorsal attention, SM somatomotor, DM default mode, Vis visual 10 networks in Smith et al. (2009) and the 18 networks in the highest correlation with the visual network, IC2 with Allen et al. (2011). Specifically, we correlated the whole the default mode, and IC1 with the somatomotor network for brain FC maps that we obtained for dimension 7 with these both the 17 reference networks of Yeo et al. (2011) and the additional reference networks. The results from these cor- 10 networks in Smith et al. (2009). However, IC0 (associ- relations are shown in Supplementary Tables S2–4. There ated with dorsal attention in the set of 7 networks) correlated are two noteworthy observations from these results. First, r = 0.28 the default mode network C and r = 0.20 with the the results we obtained with the 17 networks of Yeo et al. dorsal attention network A in the 17 networks of Yeo et al. (2011) and the 10 networks of Smith et al. (2009) were com- (2011), and r = 0.29 with the frontoparietal network in the parable to those obtained above. Specifically, IC3 yielded set of Smith et al. (2009). This variability likely reflects a 1 3 Brain Structure and Function (2022) 227:995–1012 1005 Table 2 Cortical and subcortical areas showing reliable co-activity they resemble the low correlations found by Ruiz-Rizzo with IC0 (correlated with Dorsal Attention network) relative to the et al. (2020) that also used this reference set. We discuss mean co-activity value of all other areas this issue in more detail below. Region z ratio p value Parahippocampal 47.72 ~0.000E+00 Discussion Inferior parietal 40.13 ~0.000E+00 Precuneus 24.03 5.577E–126 The aim of the current study was to characterize the different Fusiform 20.59 1.569E–92 resting-state networks that co-activate with the MTL as well Caudal middle frontal 17.42 2.406E–66 as detail how the different MTL subcomponents contribute Inferior temporal 16.61 2.535E–60 to these resting-state networks. We examined this issue using Isthmus cingulate 11.68 7.301E–30 the high spatial resolution 7T rsfMRI dataset from the HCP Superior parietal 9.17 2.068E–18 with a data-driven method that applied ICA in a manner that Supramarginal 8.38 2.194E–15 was restricted to the MTL. We found that during the resting Entorhinal 7.79 2.789E–13 state, our method detected four activation clusters that were Middle temporal 5.78 3.301E–07 spread across the various subcomponents of the MTL. Using Medial orbitofrontal 4.10 1.802E–03 Dual Regression and mixed effect regression techniques to p values corrected for multiple comparisons using Bonferroni correc- estimate reliability across participants, we found that these tion four activation clusters were functionally connected to four different whole-brain resting-state networks that relied on further splintering of networks and label differences between different contributions of MTL subcomponents. Specifically, the reference sets. we found that the dorsal attention network (detected with Second, the correlations between our obtained results and r = 0.42 ) relied primarily on the parahippocampal gyrus the 18 networks of Allen et al. (2011) resulted in generally and entorhinal cortex, the somatomotor network ( r = 0.53 ) low correlations (see Supplementary Table S4). Although on the hippocampus, the default mode network (r = 0.59 ) we cannot provide a clear explanation of these low correla- on both parahippocampal gyrus and hippocampus, and the tions at this point, we note that these results are not in line visual network ( r = 0.66 ) on the parahippocampal cortex with these observed with the other reference sets and that Table 3 Cortical and subcortical areas showing reliable co-activity Table 4 Cortical and subcortical areas showing reliable co-activity with IC1 (correlated with Somatomotor network) relative to the mean with IC2 (correlated with Default Mode network) relative to the mean co-activity value of all other areas co-activity value of all other areas Region z ratio p value Region z ratio p value Amygdala 55.30 ~0.000E+00 Isthmus cingulate 70.52 ~0.000E+00 Hippocampus 53.71 ~0.000E+00 Precuneus 50.65 ~0.000E+00 Postcentral 49.85 ~0.000E+00 Parahippocampal 34.93 1.301E–265 Paracentral 34.58 2.413E–260 Inferior parietal 32.82 1.256E–234 Superior temporal 27.76 5.009E–168 Rostral anterior cingulate 32.67 1.781E–232 Precentral 21.26 1.237E–98 Frontal pole 26.58 5.334E–154 Bankssts 19.23 8.143E–81 Medial orbitofrontal 26.31 7.064E–151 Medial orbitofrontal 13.27 1.406E–38 Caudal middle frontal 21.98 1.828E–105 Entorhinal 12.17 1.874E–32 Superior frontal 18.42 3.889E–74 Fusiform 11.51 5.215E–29 Middle temporal 13.09 1.656E–37 Temporal pole 10.15 1.421E–22 Hippocampus 11.37 2.489E–28 Cuneus 10.00 6.511E–22 Posterior cingulate 8.69 1.499E–16 Frontal pole 8.82 4.865E–17 Caudate 7.09 5.923E–11 Isthmus cingulate 7.39 6.378E–12 Ventral DC 6.21 2.345E–08 Middle temporal 7.19 2.866E–11 Temporal pole 4.91 3.874E–05 Transverse temporal 3.83 5.402E–03 Rostral middle frontal 4.19 1.188E–03 Parahippocampal 3.62 1.279E–02 Brain stem 3.53 1.777E–02 p values corrected for multiple comparisons using Bonferroni correc- p values corrected for multiple comparisons using Bonferroni correc- tion tion 1 3 1006 Brain Structure and Function (2022) 227:995–1012 1 3 Brain Structure and Function (2022) 227:995–1012 1007 ◂Fig. 4 Spiderplot representation of functional connectivity (estimated In particular, the posterior network is typically associated marginal coefficients) between the four IC hotspots (IC0, IC1, IC2, with posterior regions like the retrosplenial cortex, precu- IC3) and the rest of the brain (a), as well as functional connectiv- neus (Ranganath and Ritchey 2012; Ritchey et al. 2015) as ity results from group-based analyses in subcortical (b) and cortical well as with occipital areas (Libby et al. 2012; Wang et al. (c) regions for each of the different ICs (see Tables  2, 3, 4  and 5 for details) 2016). As can be seen in Tables 4 and 5 these regions were exactly among those with the highest co-activation in the set of regions linked to the visual and default mode networks (see Table 6 for details). These results were validated with found in our study. Similarly, the anterior network is typi- high replication ( r > 0.95 ) in a separate dataset. cally associated with the amygdala and orbitofrontal cortex Previous studies have reported inconsistent results on the (Ranganath and Ritchey 2012; Ritchey et al. 2015), and as number of whole-brain networks that co-activate with the can be seen in Table 3, these regions also appear among the MTL. Whereas, the classical view is that the MTL is con- most co-activated in the set of regions associated with the nected to a posterior and an anterior network (Kahn et al. somatomotor network. The current results, therefore, suggest 2008; Libby et al. 2012; Qin et al. 2016; Ranganath and that, in a resting-state context, the posterior and anterior net- Ritchey 2012; Ritchey et al. 2015; Schröder et al. 2015), works typically linked with the MTL are the visual, default more recent studies have found that the MTL relies on mode and somatomotor networks, respectively. additional networks beyond these two traditionally pro- Although the current results are in line with those studies posed (Ruiz-Rizzo et al. 2020; Wang et al. 2016; Plachti that have argued for a more expanded connectivity between et al. 2019). The current results are in line with these more the MTL and the rest of the brain (Ruiz-Rizzo et al. 2020; recent studies in that they show that the MTL is connected Wang et al. 2016; Plachti et al. 2019), they differ from these to additional resting-state networks. Specifically, the cur - previous studies in some details. First, as mentioned in the rent study shows that the MTL was connected to the dorsal Introduction, a recent study by Ruiz-Rizzo et al. (2020) con- attention, somatomotor, default mode and visual networks cluded that MTL was connected with five different resting- (see Table 1 for details). Three of these four networks have state networks. Using the reference set of 20 networks of a clear correspondence to the posterior and anterior net- Allen et al. (2011), they found that MTL connected with works previously identified. Specifically, the default mode default mode, salience, frontal, basal ganglia and visual and visual network likely reflect the previously identified networks. One general problem with this study are the posterior network, while the somatomotor network likely relatively low correlations between the observed and refer- reflects the anterior network. This interpretation rests on ence networks (i.e., visual, frontal and salience networks all the specific overlap of regions usually associated with the had r < 0.15 ). Indeed, the problem here may to be related posterior/anterior networks and with the regions found to be to the specific reference atlas because when we used the linked to the resting-state networks obtained in our study. Allen et al. (2011) atlas as a reference, correlations between Fig. 5 Spiderplot representation of co-activity values (estimated mar- from pPHG and aPHG and lateral Ent, and that IC2 (Default Mode, ginal coefficients) within the 7 MTL regions for each of the four ICs green dots) relied primarily on pPHG, aPHG, and on head and body (a), as well as the relative contribution (in terms of summed z-val- of the hippocampus. DA dorsal attention, SM somatomotor, DM ues) from each MTL region to each of the four ICs (b). Note that, default mode, Vis visual for example, IC0 (Dorsal Attention, blue dots) relies on contributions 1 3 1008 Brain Structure and Function (2022) 227:995–1012 obtained networks and the reference set were also rela- parahippocampal gyrus (aPHG, including perirhinal cortex) tively low (see Complementary Results and Supplementary and anterior sections of the hippocampus. Interestingly, our Table S4). Given that correlations were substantially higher results revealed that the somatomotor network, that included in the reference sets of Yeo et al. (2011) and Smith et al. areas typically associated with the anterior network like the (2009), future studies should be geared towards resolving amygdala and medial orbitofrontal cortex, relied primar- these rather remarkable discrepancies between sets of refer- ily on the hippocampus and not on the parahippocampal ence networks. In addition, a study by Plachti et al. (2019) gyrus (see Table 3 and Figure 3). One possible explanation found that during the resting state, functional connectivity for this is that the anterior network previously observed in between the hippocampus and the rest of the brain could low spatial resolution data represented a mix between the be described by 3, 5 or even 7 clusters. This seems at odds somatomotor and dorsal attention networks thereby explain- with our observation that the hippocampus was involved ing the involvement of the aPHG. In sum, the current results, in only two networks (default mode and somatomotor net- therefore, suggest that MTL plays a role in four different works; see also Ezama et al. 2021). However, this interpre- resting-state networks where these networks display distinct tation is complicated by the fact that Plachti et al. (2019) configurations of MTL subcomponents. focused on the internal parcellation of the hippocampus and Although the MTL is traditionally linked with episodic did not present the whole-brain functional connectivity of memory (e.g., Squire & Zola-Morgan 1991), more recent the obtained parcellations. Future studies that employ the studies have shown the involvement of this structure in a consensus clustering technique used by Plachti et al. (2019) wide range of cognitive functions such as short-term mem- while also reporting whole-brain connectivity maps should ory (Ranganath and Blumenfeld 2005), visual perception be able to resolve this issue. In short, although the current (Barense et al. 2012), attention (Aly and Turk-Browne 2016; results as well as the set of studies discussed here support the Córdova et al. 2019; Ruiz et al. 2020), and language and notion of expanded connectivity of the MTL, specific details conceptual processing (Duff and Brown-Schmidt 2012; regarding the influence of the particular reference set and Mack et al. 2016; Piai et al. 2016). For example, a recent pattern of whole-brain connectivity remain to be resolved. study by Ruiz et al. (2020) revealed that patients with MTL The current results also provide insight into how the vari- lesions showed impaired performance in an attention task ous MTL subcomponents contribute to these four resting- that relied on visual perception suggesting that the MTL state networks. Specifically, the visual network relied pri- plays a role in attention processes. The results reported here marily on posterior sections of the parahippocampal gyrus are in line with these studies in that they highlight the vari- (PHG), and the dorsal attention network primarily on a ety of resting-state networks to which the MTL contributes. posterior-to-anterior gradient along the parahippocampal Specifically, although resting-state studies can only make long-axis and lateral entorhinal cortex (pPHG–aPHG–lEnt, limited claims about function, our observation of the MTL in order of relative contribution). In addition, the default in visual and dorsal attention networks seem to be in line mode network relied on a more complex pattern of co-acti- with previous studies that have emphasized the implication vation in both parahippocampal gyrus and hippocampus of MTL in perception and attentional processes (Aly and with opposite gradients in these two structures: In the para- Turk-Browne 2016; Córdova et al. 2019; Ruiz et al. 2020). hippocampal gyrus the gradient was in the posterior–ante- The current observation that the MTL is involved in four dif- rior (pPHG–aPHG) direction; whereas in hippocampus, it ferent resting-state networks, therefore, further underscores was in the anterior–posterior (head–body of hippocampus) the notion that the MTL may be involved in a more abstract direction. Finally, the somatomotor network relied primar- type of processing (e.g., relational) that plays an important ily on the hippocampus with an anterior–posterior gradient part in many different cognitive domains. (head–body–tail of hippocampus; see Table 6, and Figure 5 Our study has several limitations. First, our conclusion for details). These results are generally consistent with those that MTL relies on four particular resting-state networks is previously observed. Specifically, the posterior network has based on the specific reference atlas of Yeo et al. (2011). traditionally been associated with pPHG and middle to pos- However, as we show in Supplementary Tables S2–4, cor- terior sections of the hippocampus (Kahn et al. 2008; Libby relations with the 17 network atlas of Yeo et al. (2011) and et al. 2012; Qin et al. 2016; Ranganath and Ritchey 2012; the 10 networks of Smith et al. (2009) yielded highly simi- Ritchey et al. 2015; Ruiz-Rizzo et al. 2020; Schröder et al. lar results. In addition, group-level FC analysis relied on 2015; Wang et al. 2016). This is in line with our observa- averages across voxels in a brain region. An obvious dis- tions that the visual and default mode networks strongly advantage in this approach is that there are limitations on co-activate with posterior sections of the parahippocampal the spatial precision with which an effect may be localized. gyrus and hippocampus. In addition, the anterior network We think such concerns may be further mitigated by using is traditionally associated with anterior sections of the procedures that segment brain regions to more fine grained 1 3 Brain Structure and Function (2022) 227:995–1012 1009 parcels (e.g., Schaefer et al. 2018). Finally, although we Table 6 Strongest contrast for MTL substructures for each IC compared the contributions of 7 different MTL subcom - Region IC Summed z ratio Rank Network ponents to whole-brain resting-state networks, we did not pPHG 0 366.37 1 explicitly include the perirhinal cortex. The reason for this aPHG 0 158.15 2 Dorsal attention is twofold. First, the perirhinal cortex simply does not form lEnt 0 8.44 3 part of the standard Desikan–Killany atlas that we used for hHi 1 342.85 1 our segmentations (Desikan et al. 2006). Second, the precise bHi 1 180.64 2 Somatomotor anatomical definition of perirhinal cortex remains highly tHi 1 141.79 3 disputed (Suzuki and Amaral 1994; Augustinack et  al. pPHG 2 233.97 1 2014), which complicates an accurate automatic segmenta- aPHG 2 122.92 2 Default mode tion. Thus, although the usefulness of this structure is clear hHi 2 30.40 3 (Augustinack et al. 2014; Libby et al. 2012), the integration bHi 2 10.85 4 of this structure into an automatic segmentation pipeline that pPHG 3 456.13 1 Visual also includes other MTL structures has prevented us from analyzing the contribution of this structure at this moment Contrast based on the sum of pairwise z ratio differences for all sub- in time. structures. Rank indicates the order of the substructures relative con- To conclude, the current study used a high spatial reso- tribution for each IC lution dataset to examine the different whole-brain resting- state networks that co-activate with the MTL. A secondary (Kahn et al. 2008; Libby et al. 2012; Qin et al. 2016; Ran- goal was to examine how the different MTL subcompo- nents contribute to these resting-state networks. We found ganath and Ritchey 2012; Ritchey et al. 2015; Ruiz-Rizzo et al. 2020; Schröder et al. 2015; Wang et al. 2016; Bar- that the MTL co-activates with the default mode, somato- motor, visual and dorsal attention networks. Our results nett et  al. 2019; Vincent et  al. 2006). They are in line with previous FC studies that have found that the MTL revealed that these networks are subserved by distinct con- figurations of MTL subcomponent co-activity, where the is linked to additional networks (Ruiz-Rizzo et al. 2020; Wang et al. 2016; Plachti et al. 2019) and are suggestive default mode network relied on a combination of activity in parahippocampal gyrus and hippocampus, the somato- of a functional role of MTL that goes beyond episodic memory (Aly and Turk-Browne 2016; Barense et al. 2012; motor network on the hippocampus, the visual network on the parahippocampal gyrus, and the dorsal attention Córdova et al. 2019; Duff and Brown-Schmidt 2012; Ran- ganath and Blumenfeld 2005; Ruiz et al. 2020). Finally, network on the parahippocampal gyrus and the entorhinal cortex. These results go beyond previous studies that have the current results are obtained from young healthy adults and therefore establish a baseline pattern of how various associated MTL with a posterior and an anterior network MTL subcomponents contribute to known resting-state networks. In the future, we hope to examine how these Table 5 Cortical and subcortical areas showing reliable co-activity patterns change with aging and pathology. with IC3 (correlated with Visual network) relative to the mean co- activity value of all other areas Supplementary Information The online version contains supplemen- Region z ratio p value tary material available at https://doi. or g/10. 1007/ s00429- 021- 02442-1 . Cuneus 99.93 ~0.000E+00 Acknowledgements The authors would like to thank José María Pericalcarine 89.54 ~0.000E+00 Pérez González (IMETISA) for technical support. Correspondence Lingual 77.22 ~0.000E+00 and requests for materials should be addressed to NJ (njanssen@ Superior parietal 40.98 ~0.000E+00 ull.es). Precuneus 23.29 2.205E–118 Funding Open Access funding provided thanks to the CRUE- Parahippocampal 18.32 2.276E–73 CSIC agreement with Springer Nature. This study was supported Fusiform 16.28 5.737E–58 by grants PSI2017-84933-P and PSI2017-91955-EXP to NJ, and by Lateral occipital 15.69 7.799E–54 TESIS2019010146 from the Board of Economy, Industry, Trade and Postcentral 11.46 9.415E–29 Knowledge of the Canarian Government with a European Social Fund co-financing rate to SLS. Paracentral 8.86 3.329E–17 Caudal anterior cingulate 7.90 1.164E–13 Availability of data and material Data were provided by the Human Transverse temporal 6.98 1.274E–10 Connectome Project, WU-Minn Consortium (Principal Investigators: Precentral 4.16 1.375E–03 David Van Essen and Kamil Ugurbil; 1U54MH091657) funded by the 16 NIH Institutes and Centers that support the NIH Blueprint for p values corrected for multiple comparisons using Bonferroni correc- Neuroscience Research; and by the McDonnell Center for Systems tion 1 3 1010 Brain Structure and Function (2022) 227:995–1012 Neuroscience at Washington University. The data that support the Bates D, Sarkar D, Bates MD, Matrix L (2007) The lme4 package. R findings of this study are openly available from Human Connectome package version 2:74 Project (www. human conne ctome. org). Blessing EM, Beissner F, Schumann A, Brünner F, Bär K-J (2016) A data-driven approach to mapping cortical and subcortical intrinsic functional connectivity along the longitudinal hippocampal axis. Code availability Code will be made available upon reasonable request. Hum Brain Mapp 37:462–476 Córdova NI, Turk-Browne NB, Aly M (2019) Focusing on what mat- Declarations ters: modulation of the human hippocampus by relational atten- tion. Hippocampus 29:1025–1037 Conflict of interest All authors declare that they have no conflict of Desikan RS, Ségonne F, Fischl B, Quinn BT, Dickerson BC, Blacker interest. D, Buckner RL, Dale AM, Maguire RP, Hyman BT et al (2006) An automated labeling system for subdividing the human cerebral Ethics approval The data analyses were conducted in agreement with cortex on mri scans into gyral based regions of interest. Neuroim- the declaration of Helsinki and with the protocol established by the age 31:968–980 Ethics Commission for Research of the Universidad de La Laguna, the Devlin JT, Russell RP, Davis MH, Price CJ, Wilson J, Moss HE, Mat- Comité de Ética de la Investigación y Bienestar Animal. thews PM, Tyler LK (2000) Susceptibility-induced loss of sig- nal: comparing pet and FMRI on a semantic task. Neuroimage Consent to participate Not applicable. 11:589–600 Douw L, DeSalvo MN, Tanaka N, Cole AJ, Liu H, Reinsberger C, Stuf- Consent for publication Not applicable. flebeam SM (2015) Dissociated multimodal hubs and seizures in temporal lobe epilepsy. Ann Clin Transl Neurol 2:338–352 Duff MC, Brown-Schmidt S (2012) The hippocampus and the flex- Open Access This article is licensed under a Creative Commons Attri- ible use and processing of language. Front Hum Neurosci 6:69 bution 4.0 International License, which permits use, sharing, adapta- Ezama L, Hernández-Cabrera JA, Seoane S, Pereda E, Janssen N tion, distribution and reproduction in any medium or format, as long (2021) Functional connectivity of the hippocampus and its sub- as you give appropriate credit to the original author(s) and the source, fields in resting-state networks. Eur J Neurosci 53:3378 provide a link to the Creative Commons licence, and indicate if changes Fischl B, Rajendran N, Busa E, Augustinack J, Hinds O, Yeo were made. The images or other third party material in this article are BT, Mohlberg H, Amunts K, Zilles K (2008) Cortical fold- included in the article's Creative Commons licence, unless indicated ing patterns and predicting cytoarchitecture. Cereb Cortex otherwise in a credit line to the material. If material is not included in 18:1973–1980 the article's Creative Commons licence and your intended use is not Formisano E, Esposito F, Di Salle F, Goebel R (2004) Cortex-based permitted by statutory regulation or exceeds the permitted use, you will independent component analysis of FMRI time series. Magn need to obtain permission directly from the copyright holder. To view a Reson Imaging 22:1493–1504 copy of this licence, visit http://cr eativ ecommons. or g/licen ses/ b y/4.0/ . Glasser MF, Sotiropoulos SN, Wilson JA, Coalson TS, Fischl B, Andersson JL, Xu J, Jbabdi S, Webster M, Polimeni JR et al (2013) The minimal preprocessing pipelines for the human con- nectome project. Neuroimage 80:105–124 References Govindpani K, Turner C, Waldvogel HJ, Faull RL, Kwakowsky A (2020) Impaired expression of gaba signaling components in Aggleton JP (2012) Multiple anatomical systems embedded within the the Alzheimer’s disease middle temporal gyrus. Int J Mol Sci primate medial temporal lobe: implications for hippocampal func- 21:8704 tion. Neurosci Biobehav Rev 36:1579–1596 Griffanti L, Salimi-Khorshidi G, Beckmann CF, Auerbach EJ, Douaud Allen EA, Erhardt EB, Damaraju E, Gruner W, Segall JM, Silva RF, G, Sexton CE, Zsoldos E, Ebmeier KP, Filippini N, Mackay CE Havlicek M, Rachakonda S, Fries J, Kalyanam R et al (2011) A et al (2014) ICA-based artefact removal and accelerated FMRI baseline for the multivariate comparison of resting-state networks. acquisition for improved resting state network imaging. Neuroim- Front Syst Neurosci 5:2 age 95:232–247 Alvarez P, Squire LR (1994) Memory consolidation and the medial Griffanti L, Douaud G, Bijsterbosch J, Evangelisti S, Alfaro-Almagro temporal lobe: a simple network model. Proc Natl Acad Sci F, Glasser MF, Duff EP, Fitzgibbon S, Westphal R, Carone D 91:7041–7045 et al (2017) Hand classification of FMRI ICA noise components. Aly M, Turk-Browne NB (2016) Attention stabilizes representations in Neuroimage 154:188–205 the human hippocampus. Cereb Cortex 26:783–796 Hyde JS, Biswal BB, Jesmanowicz A (2001) High-resolution FMRI Anticevic A, Dierker DL, Gillespie SK, Repovs G, Csernansky JG, Van using multislice partial k-space gr-epi with cubic voxels. Magn Essen DC, Barch DM (2008) Comparing surface-based and vol- Reson Med 46:114–125 ume-based analyses of functional neuroimaging data in patients Iglesias JE, Augustinack JC, Nguyen K, Player CM, Player A, Wright with schizophrenia. Neuroimage 41:835–848 M, Roy N, Frosch MP, McKee AC, Wald LL et al (2015) A com- Augustinack JC, Magnain C, Reuter M, van der Kouwe AJ, Boas D, putational atlas of the hippocampal formation using ex  vivo, Fischl B (2014) MRI parcellation of ex vivo medial temporal lobe. ultra-high resolution MRI: application to adaptive segmentation Neuroimage 93:252–259 of in vivo mri. Neuroimage 115:117–137 Barense MD, Groen II, Lee AC, Yeung L-K, Brady SM, Gregori M, Janssen N, Mendieta CCR (2020) The dynamics of speech motor con- Kapur N, Bussey TJ, Saksida LM, Henson RN (2012) Intact trol revealed with time-resolved FMRI. Cereb Cortex 30:241–255 memory for irrelevant information impairs perception in amnesia. Jenkinson M, Bannister P, Brady M, Smith S (2002) Improved optimi- Neuron 75:157–167 zation for the robust and accurate linear registration and motion Barnett AJ, Man V, McAndrews MP (2019) Parcellation of the hip- correction of brain images. Neuroimage 17:825–841 pocampus using resting functional connectivity in temporal lobe Jones BF, Witter MP (2007) Cingulate cortex projections to the para- epilepsy. Front Neurol 10:920 hippocampal region and hippocampal formation in the rat. Hip- pocampus 17:957–976 1 3 Brain Structure and Function (2022) 227:995–1012 1011 Kahn I, Andrews-Hanna JR, Vincent JL, Snyder AZ, Buckner RL Ruiz NA, Meager MR, Agarwal S, Aly M (2020) The medial temporal (2008) Distinct cortical anatomy linked to subregions of the lobe is critical for spatial relational perception. J Cogn Neurosci medial temporal lobe revealed by intrinsic functional connectiv- 32:1780–1795 ity. J Neurophysiol 100:129–139 Ruiz-Rizzo AL, Beissner F, Finke K, Müller HJ, Zimmer C, Pasquini Kassambara A (2018) ggpubr:“ggplot2” based publication ready plots. L, Sorg C (2020) Human subsystems of medial temporal lobes R package version 0.1, 7 extend locally to amygdala nuclei and globally to an allostatic- Kenkhuis B, Jonkman LE, Bulk M, Buijs M, Boon BD, Bouwman interoceptive system. Neuroimage 207:116404 FH, Geurts JJ, van de Berg WD, van der Weerd L (2019) 7t MRI Salimi-Khorshidi G, Douaud G, Beckmann CF, Glasser MF, Griffanti allows detection of disturbed cortical lamination of the medial L, Smith SM (2014) Automatic denoising of functional MRI temporal lobe in patients with Alzheimer’s disease. NeuroImage data: combining independent component analysis and hierarchi- Clin 21:101665 cal fusion of classifiers. Neuroimage 90:449–468 Kobayashi Y, Amaral DG (2003) Macaque monkey retrosplenial cor- Schaefer A, Kong R, Gordon EM, Laumann TO, Zuo X-N, Holmes tex: II. cortical afferents. J Comp Neurol 466:48–79 AJ, Eickhoff SB, Yeo BT (2018) Local-global parcellation of the Kobayashi Y, Amaral DG (2007) Macaque monkey retrosplenial cor- human cerebral cortex from intrinsic functional connectivity MRI. tex: III. cortical efferents. J Comp Neurol 502:810–833 Cereb Cortex 28:3095–3114 Kondo H, Saleem KS, Price JL (2005) Differential connections of Schröder TN, Haak KV, Jimenez NIZ, Beckmann CF, Doeller CF the perirhinal and parahippocampal cortex with the orbital and (2015) Functional topography of the human entorhinal cortex. medial prefrontal networks in macaque monkeys. J Comp Neurol Elife 4:e06738 493:479–509 Seidman LJ, Pantelis C, Keshavan MS, Faraone SV, Goldstein JM, Kuznetsova A, Brockhoff PB, Christensen RH et al (2017) lmertest Horton NJ, Makris N, Peter F, Caviness VS, Tsuang MT (2003) package: tests in linear mixed effects models. J Stat Softw 82:1–26 A review and new report of medial temporal lobe dysfunction as Lenth R, Singmann H, Love J, Buerkner P, Herve M (2018) Emmeans: a vulnerability indicator for schizophrenia: a magnetic resonance estimated marginal means, aka least-squares means. R package imaging morphometric family study of the parahippocampal version 1:3 gyrus. Schizophr Bull 29:803–830 Libby LA, Ekstrom AD, Ragland JD, Ranganath C (2012) Differential Sladky R, Baldinger P, Kranz GS, Tröstl J, Höflich A, Lanzenberger connectivity of perirhinal and parahippocampal cortices within R, Moser E, Windischberger C (2013) High-resolution functional human hippocampal subregions revealed by high-resolution func- MRI of the human amygdala at 7 t. Eur J Radiol 82:728–733 tional imaging. J Neurosci 32:6550–6560 Smith SM, Fox PT, Miller KL, Glahn DC, Fox PM, Mackay CE, Filip- Mack ML, Love BC, Preston AR (2016) Dynamic updating of hip- pini N, Watkins KE, Toro R, Laird AR et al (2009) Correspond- pocampal object representations reflects new conceptual knowl- ence of the brain’s functional architecture during activation and edge. Proc Natl Acad Sci 113:13203–13208 rest. Proc Natl Acad Sci 106:13040–13045 Mowinckel AM, Vidal-Piñeiro D (2019) Visualisation of brain sta- Smith SM, Beckmann CF, Andersson J, Auerbach EJ, Bijsterbosch tistics with r-packages ggseg and ggseg3d. arXiv preprint J, Douaud G, Duff E, Feinberg DA, Griffanti L, Harms MP et al arXiv:1912.08200, (2013) Resting-state FMRI in the human connectome project. Nickerson LD, Smith SM, Öngür D, Beckmann CF (2017) Using dual Neuroimage 80:144–168 regression to investigate network shape and amplitude in func- Smith SM, Hyvärinen A, Varoquaux G, Miller KL, Beckmann CF tional connectivity analyses. Front Neurosci 11:115 (2014) Group-PCA for very large FMRI datasets. Neuroimage Olman CA, Davachi L, Inati S (2009) Distortion and signal loss in 101:738–749 medial temporal lobe. PLoS One 4:e8160 Squire LR, Zola-Morgan S (1991) The medial temporal lobe memory Piai V, Anderson KL, Lin JJ, Dewar C, Parvizi J, Dronkers NF, Knight system. Science 253:1380–1386 RT (2016) Direct brain recordings reveal hippocampal rhythm Squire LR, Wixted JT, Clark RE (2007) Recognition memory and underpinnings of language processing. Proc Natl Acad Sci the medial temporal lobe: a new perspective. Nat Rev Neurosci 113:11366–11371 8:872–883 Plachti A, Eickhoff SB, Hoffstaedter F, Patil KR, Laird AR, Fox PT, Strange BA, Witter MP, Lein ES, Moser EI (2014) Functional organi- Amunts K, Genon S (2019) Multimodal parcellations and exten- zation of the hippocampal longitudinal axis. Nat Rev Neurosci sive behavioral profiling tackling the hippocampus gradient. Cereb 15:655–669 Cortex 29:4595–4612 Suzuki WL, Amaral DG (1994) Perirhinal and parahippocampal cor- Qin S, Duan X, Supekar K, Chen H, Chen T, Menon V (2016) Large- tices of the macaque monkey: cortical afferents. J Comp Neurol scale intrinsic functional network organization along the long 350:497–533 axis of the human medial temporal lobe. Brain Struct Funct Suzuki WA, Amaral DG (2004) Functional neuroanatomy of the medial 221:3237–3258 temporal lobe memory system. Cortex Ranganath C, Blumenfeld RS (2005) Doubts about double dissocia- Van Essen DC, Ugurbil K, Auerbach E, Barch D, Behrens T, Bucholz tions between short-and long-term memory. Trends Cogn Sci R, Chang A, Chen L, Corbetta M, Curtiss SW et al (2012) The 9:374–380 human connectome project: a data acquisition perspective. Neu- Ranganath C, Ritchey M (2012) Two cortical systems for memory- roimage 62:2222–2231 guided behaviour. Nat Rev Neurosci 13:713–726 Vincent JL, Snyder AZ, Fox MD, Shannon BJ, Andrews JR, Raichle Ritchey M, Libby LA, Ranganath C (2015) Cortico-hippocampal sys- ME, Buckner RL (2006) Coherent spontaneous activity identi- tems involved in memory and cognition: the PMAT framework. fies a hippocampal-parietal memory network. J Neurophysiol Progress Brain Res 219:45–64 96:3517–3531 Robinson S, Windischberger C, Rauscher A, Moser E (2004) Opti- Vu AT, Jamison K, Glasser MF, Smith SM, Coalson T, Moeller S, mized 3 t epi of the amygdalae. Neuroimage 22:203–210 Auerbach EJ, Uğurbil K, Yacoub E (2017) Tradeoffs in pushing Rosene DL, Van Hoesen GW (1977) Hippocampal efferents reach the spatial resolution of FMRI for the 7t human connectome pro- widespread areas of cerebral cortex and amygdala in the rhesus ject. Neuroimage 154:23–32 monkey. Science 198:315–317 Wang S-F, Ritchey M, Libby LA, Ranganath C (2016) Functional con- nectivity based parcellation of the human medial temporal lobe. Neurobiol Learn Mem 134:123–134 1 3 1012 Brain Structure and Function (2022) 227:995–1012 Weiskopf N, Hutton C, Josephs O, Deichmann R (2006) Optimal EPI evaluation using ICA and dual regression approach. Neuroimage parameters for reduction of susceptibility-induced bold sensitiv- 49:2163–2177 ity losses: a whole-brain analysis at 3 t and 1.5 t. Neuroimage 33:493–504 Publisher's Note Springer Nature remains neutral with regard to Yeo BT, Krienen FM, Sepulcre J, Sabuncu MR, Lashkari D, Hollin- jurisdictional claims in published maps and institutional affiliations. shead M, Roffman JL, Smoller JW, Zöllei L, Polimeni JR et al (2011) The organization of the human cerebral cortex estimated by intrinsic functional connectivity. J Neurophysiol 106:1125–1165 Zuo X-N, Kelly C, Adelstein JS, Klein DF, Castellanos FX, Milham MP (2010) Reliable intrinsic connectivity networks: test-retest 1 3 http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Brain Structure and Function Springer Journals

Medial temporal lobe contributions to resting-state networks

Download PDF
18 pages

Loading next page...
 
/lp/springer-journals/medial-temporal-lobe-contributions-to-resting-state-networks-0gQZcj30CT

References (87)

Publisher
Springer Journals
Copyright
Copyright © The Author(s) 2022
ISSN
1863-2653
eISSN
1863-2661
DOI
10.1007/s00429-021-02442-1
Publisher site
See Article on Publisher Site

Abstract

The medial temporal lobe (MTL) is a set of interconnected brain regions that have been shown to play a central role in behavior as well as in neurological disease. Recent studies using resting-state functional magnetic resonance imaging (rsfMRI) have attempted to understand the MTL in terms of its functional connectivity with the rest of the brain. However, the exact characterization of the whole-brain networks that co-activate with the MTL as well as how the various sub-regions of the MTL are associated with these networks remains poorly understood. Here, we attempted to advance these issues by exploiting the high spatial resolution 7T rsfMRI dataset from the Human Connectome Project with a data-driven analysis approach that relied on independent component analysis (ICA) restricted to the MTL. We found that four different well- known resting-state networks co-activated with a unique configuration of MTL subcomponents. Specifically, we found that different sections of the parahippocampal cortex were involved in the default mode, visual and dorsal attention networks; sections of the hippocampus in the somatomotor and default mode networks; and the lateral entorhinal cortex in the dorsal attention network. We replicated this set of results in a validation sample. These results provide new insight into how the MTL and its subcomponents contribute to known resting-state networks. The participation of the MTL in an expanded range of resting-state networks is in line with recent proposals on MTL function. Keywords Medial temporal lobe · Functional connectivity · Resting-state fMRI · Independent component analysis · Dual regression Abbreviations mEnt Medial entorhinal cortex MTL Medial temporal lobe hHi Head of the hippocampus DMN Default mode network bHi Body of the hippocampus rfMRI Resting-state functional magnetic resonance tHi Tail of the hippocampus imagingTh Thalamus proper MNI Montreal Neurological Institute 152Cd Caudate aPHG Anterior parahippocampal cortexPu Putamen pPHG Posterior parahippocampal cortexPal Pallidum lEnt Lateral entorhinal cortexHi Hippocampus Amg Amygdala Ac Accumbens area vDC Ventral DC * Niels Janssen njanssen@ull.es STS Bankssts cACC Caudal anterior cingulate Department of Cognitive, Social and Organizational cdMF Caudal middle frontal Psychology, Faculty of Psychology and Speech Therapy, Cun Cuneus University of La Laguna, San Cristóbal de La Laguna, Spain Ent Entorhinal Department of Basic Medical Sciences, Faculty FuG Fusiform of Health Sciences, University of La Laguna, San Cristóbal de La Laguna, Spain iP Inferior parietal iT Inferior temporal Institute of Biomedical Technologies, University of La Laguna, San Cristóbal de La Laguna, Spain ICG Isthmus cingulate lO Lateral occipital Instituto Universitario de Neurociencia, University of La Laguna, San Cristóbal de La Laguna, Spain Vol.:(0123456789) 1 3 996 Brain Structure and Function (2022) 227:995–1012 lOF Lateral orbitofrontal et al. 2019). Characterizing the whole-brain functional net- LgG Lingual gyrus works that co-activate with the MTL has implications for mOF Medial orbitofrontal understanding its role in health and disease. One potential mT Middle temporal reason for why this issue remains unresolved may be due PHG Parahippocampal gyrus to methodological limitations in previous studies. Here, we PCL Paracentral relied on a data-driven parcellation of the MTL using the IFGOp Pars opercularis whole-brain high spatial resolution 7T rsfMRI dataset from IFGOr Pars orbitalis the Human Connectome Project (HCP). IFGTr Pars triangularis The standard view on the connectivity between the MTL PCAL Pericalcarine and the rest of the brain is that the MTL is connected with PoG Postcentral two distinct whole-brain networks. Both anatomical and pCC Posterior cingulate functional connectivity studies have shown that MTL con- PrG Precentral nectivity is largely organized along a posterior–anterior PCun Precuneus gradient. For example, tract-tracing studies in monkeys and rACg Rostral anterior cingulate rodents have found that posterior sections of the parahip- rMF Rostral middle frontal pocampal gyrus (PHG) and posterior sections in the hip- SF Superior frontal pocampal formation show increased (mono- or poly-syn- SP Superior parietal aptic) connectivity with posterior midline regions like the ST Superior temporal retrosplenial cortex and posterior cingulate cortex, whereas SMG Supramarginal anterior sections of the parahippocampal cortex and ante- FrP Frontal pole rior sections of the hippocampal formation show increased TmP Temporal pole connectivity with anterior brain regions like the orbitofron- TTG Transverse temporal tal cortex and amygdala (Aggleton 2012; Jones and Witter Ins Insula 2007; Kobayashi and Amaral 2007, 2003; Kondo et al. 2005; Rosene & Van Hoesen 1977; Strange et al. 2014; Suzuki and Amaral 1994). Similarly, lateral sections of the entorhinal Introduction cortex (Ent) have been associated with the posterior whole- brain network, whereas medial Ent has been associated with The medial temporal lobe (MTL) has received much inter- the anterior network (Jones and Witter 2007; Strange et al. est in research and the clinic due to its key implication in 2014). More recent functional connectivity studies using memory processes (e.g., Alvarez and Squire 1994; Suzuki rsfMRI have further confirmed this bipartite organization and Amaral 2004; Squire et al. 2007), as well as due to its of MTL whole-brain connectivity. Specifically, Libby et al. involvement in several relatively common pathological (2012) found that seeds placed in posterior PHG revealed co- conditions (e.g., temporal lobe epilepsy, schizophrenia and activity with posterior midline regions like the retrosplenial Alzheimer’s disease; Douw et al. 2015; Seidman et al. 2003; cortex, precuneus, posterior cingulate and occipital cortex, Kenkhuis et al. 2019; Govindpani et al. 2020). The MTL whereas seeds placed in more anterior locations produced encompasses a number of different anatomical structures, co-activity with orbitofrontal cortex and inferior temporal primarily the parahippocampal and entorhinal cortices as cortex. Other studies placing seeds in various locations along well as the hippocampal formation. Recent resting-state the hippocampal long axis have revealed a similar separation fMRI (rsfMRI) studies have attempted to understand the of co-activity between posterior and anterior networks (Kahn MTL by considering its functional connectivity with the rest et al. 2008; Qin et al. 2016), as have studies examining the of the brain (Kahn et al. 2008; Libby et al. 2012; Qin et al. entorhinal cortex (Schröder et al. 2015). In short, evidence 2016; Ranganath and Ritchey 2012; Ritchey et al. 2015; from various sources now confirms that MTL connectivity Ruiz-Rizzo et al. 2020; Schröder et al. 2015; Wang et al. can be associated with both a posterior and an anterior net- 2016). An on-going debate regarding this issue concerns the work (Ranganath and Ritchey 2012; Ritchey et al. 2015). different whole-brain functional networks that connect to the However, three recent functional connectivity studies have MTL. Specifically, whereas the traditional view is that the presented results that have challenged this view. These studies MTL interacts mainly with two whole-brain networks (Kahn have relied on data-driven approaches to examine functional et al. 2008; Libby et al. 2012; Qin et al. 2016; Ranganath and connectivity thereby avoiding potential biases inherent in Ritchey 2012; Ritchey et al. 2015; Schröder et al. 2015; Bar- functional connectivity techniques that rely on placing seeds nett et al. 2019), other recent studies using data-driven tech- (Zuo et al. 2010). Specifically, Wang et al. (2016) examined niques have found that the MTL connects with additional the slice-by-slice connectivity between both the parahip- networks (Ruiz-Rizzo et al. 2020; Wang et al. 2016; Plachti pocampal gyrus as well as the hippocampus and the rest of 1 3 Brain Structure and Function (2022) 227:995–1012 997 the brain using a hierarchical clustering technique. Interest- the spatially restricted group ICA technique (srICA; Bless- ingly, they found that there were three connectivity clusters ing et al. 2016; Ezama et al. 2021; Formisano et al. 2004). along the parahippocampal long axis, one in posterior PHG Whole-brain ICA is frequently used to separate signal from that connected to the aforementioned posterior network, one noise in fMRI studies (e.g., Janssen and Mendieta 2020; in anterior PHG (termed perirhinal cortex) that connected Smith et al. 2013). Instead, by applying ICA to a particular to the anterior network, and one in an even more anterior brain region, the noise profile that is specific to that brain PHG location that connected to a network of regions that region will be taken into account and result in a more sensi- included the insula, post central gyrus and amygdala. Simi- tive separation of signal from noise in that region. larly, Ruiz-Rizzo et al. (2020) examined functional connec- The current study relied on high spatial resolution data tivity of the MTL as well as the amygdala using a spatially as well as a targeted analysis approach to clarify the con- restricted Independent Component Analysis (ICA) approach tributions of the MTL in existing resting-state networks. (Blessing et al. 2016; Formisano et al. 2004). They found that Specifically, whole-brain functional connectivity maps (FC clusters of activity detected inside the MTL and amygdala maps) associated with the MTL clusters found by the srICA co-activated with sets of brain regions that correlated with were computed using the Dual Regression technique (Nick- the reference networks of Allen et al. (2011). Specifically, erson et al. 2017). These whole-brain FC maps reflected in addition to the default mode network (correlation varied the large-scale co-activity with the specific MTL clusters from r = 0.12 to r = 0.48 for different MTL activity clusters), found in the previous step. Next, MTL clusters were classi- they also found that MTL was somewhat connected to other fied as signal or noise on the basis of an algorithm that relied networks like the salience (r = 0.14 ), frontal (r = 0.11 ), basal on the correlation with the 7 known resting-state networks ganglia ( r = 0.40 ) and visual networks ( r = 0.11 ). Finally, obtained by Yeo et al. (2011). We then used linear mixed a recent paper by Plachti et al. (2019) focused on the hip- effect regression analyses to calculate the relative contribu- pocampus using a consensus clustering technique and found tion of each MTL subcomponent in the different resting-state that connectivity between the hippocampus and the rest of the networks. Specic fi ally, the MTL was segmented into anterior brain was best described by 3, 5 and even 7 clusters. and posterior portions of the PHG, head, body and tail of Thus, it appears that whereas some studies have found the hippocampus and mEnt and lEnt. Finally, we relied on a that the MTL is connected to two different whole-brain test-validation approach in which results obtained in the test functional networks, others have found it is connected to dataset were validated on a second dataset. additional different networks. This empirical discrepancy may be the result of methodological limitations in the afore- mentioned studies. First, previous studies have relied on Methods relatively low spatial resolution fMRI acquisition protocols ( ∼ 3.5 mm voxels). One concern with such low resolution Participants data is that the large size voxels may be unable to accurately separate signals from different resting-state networks leading Data for all participants were downloaded from the Human to variability in the reported results. Second, previous stud- Connectome Website. The initial dataset consisted of 184 ies have not sufficiently taken into account the observation participants who had participated in the 7T data acquisition. that the MTL is affected by local magnetic field inhomoge- However, data from 12 participants were excluded due to neities that lead to reduced temporal signal-to-noise ratios the presence of specific Quality Control issues identified by (tSNR; e.g., Olman et al. 2009; Weiskopf et al. 2006). The the HCP (i.e., QC issues A, B, C, and D). The final sample, additional noise in the MTL region may hamper the detec- therefore, consisted of 172 participants, between the ages tion of MTL contributions to resting-state networks and also of 22 and 35 (104 females). Further detailed description on produce inconsistencies in the reported results. In this study, the study subjects may be found in Van Essen et al. (2012). we addressed these two limitations in three ways. First, we The data analyses were conducted in agreement with the addressed the separability of signals inside the MTL using declaration of Helsinki and with the protocol established a dataset with high spatial resolution (1.6 mm isotropic) by the Ethics Commission for Research of the Universidad acquired at a high field strength (7T). Previous studies have de La Laguna, the Comité de Ética de la Investigación y shown that compared to lower resolutions at 3T, increased Bienestar Animal. spatial resolution at 7T results in more clearly defined rest- ing-state networks (Vu et al. 2017). Second, previous studies Data acquisition and preprocessing have also shown that reducing the voxel size in fMRI data avoids partial voluming effects and leads to improved signal Data packages herein used come from the WU-Minn HCP detection in areas with low tSNR (Hyde et al. 2001; Rob- Data-1200 Subjects data set. For this experiment, we down- inson et al. 2004; Sladky et al. 2013). Finally, we relied on loaded 7T Resting-State fMRI 1.6 mm/32k FIX-Denoised 1 3 998 Brain Structure and Function (2022) 227:995–1012 (Compact) and Resting-State fMRI FIX-Denoised ms, flip angle = variable, and also resulting in 0.7 mm iso- (Extended) datasets. As per the HCP reference manual, tropic voxels. The structural images were acquired on a 3T these data were acquired by the Washington-University and Siemens Connectom Skyra scanner. The downloaded struc- Minnesota Consortium, with a Siemens Magnetom 7T MR tural packages contained the T1w and T2w images for each Scanner and a Nova32 32-channel Siemens receive head coil participant as well as the full Freesurfer output and trans- from Nova Medical. Four 16-min-long rsfMRI acquisitions formation matrices that were relevant for our downstream were acquired per subject. RsfMRI acquisitions alternated analyses (see below). For additional specific information on the direction of the phase encoding gradient, where two the pre-processing of these structural images, we refer to sessions were acquired in the posterior-to-anterior phase Glasser et al. (2013). direction and the other two in anterior-to-posterior phase direction. For the resting-state acquisitions, participants Data analysis were instructed to fix their sight on a white cross-hair over a dark background (Smith et al. 2013). MRI scanning param- The aim of this study was to explore the different whole- eters for the resting-state data were based on acquisitions brain networks that co-activate with the MTL as well as how of Gradient-Echo EPI volumes. Each volume contained 85 the different MTL subcomponents contribute to these dif- slices that were acquired with a multiband factor of 5. Slice ferent networks. To approach these objectives, our analyses thickness was 1.6 mm with no gap, the FOV was 208 × 208 were divided into four main steps. A graphical representa- mm, matrix size 130 × 130, resulting in 1.6 mm isotropic tion of the workflow is displayed in Supplementary Fig. S1. voxels. The TR was 1000 ms, echo time (TE) 22.2 ms, and The most relevant results from these analyses will be made the flip angle 45 . We used the first two of these four rsfMRI available on our github page (https:// github. com/ iamni elsja datasets as a test set (32 minutes of resting-state data), and nssen). the last two datasets as a validation set. Both test and valida- tion datasets had alternating phase directions. Segmentation of MTL into subcomponents The downloaded fMRI datasets consisted of already pre-processed functional data according to HCP minimal The first step of our analysis involved the segmentation of preprocessing pipelines ( Glasser et al. 2013). Briefly, trans- the MTL into a number of subcomponents. This segmenta- formations that reduce head motion were estimated using tion took place in a participant-specific manner, meaning FSL MCFLIRT (Jenkinson et al. 2002), fieldmap and gradi- that each segmentation took into account the unique shape ent distortion corrections were applied, and transformations of the MTL in each participant’s brain. The MTL subcom- from fMRI space to MNI space were estimated using non- ponents were the head, body and tail of the hippocampus linear transformations. Importantly, smoothing of the data (hHi, bHi, and tHi), the anterior (aPHG) and posterior was minimized in two ways: First, the transform from native (pPHG) parahippocampal gyrus, as well as the medial and to MNI space preserved the native space resolution of the lateral entorhinal cortex (lEnt and mEnt). All segmentations fMRI acquisitions, and second, all transformations (motion relied on the Desikan–Killiany cortical Atlas as well as the correction, fMRI to MNI space) were postponed, com- subcortical segmentation that is produced by Freesurfer bined and applied in a single step using sinc interpolation. and that was included with the downloaded dataset (i.e., Next, the data in MNI space were temporally filtered using the aparc+aseg atlas in Freesurfer terminology; Desikan a 2000-s high-pass filter and automatically denoised using et  al. 2006). This provides an automatic segmentation of the FIX program (Griffanti et al. 2014; Salimi-Khorshidi the brain in terms of a set of 42 brain regions that are fitted et al. 2014). This program uses semi-automatic classifica- to the unique morphology of each participant’s brain. This tion of head-motion and other artifacts which minimized the is achieved by combining prior information about the prob- potential impact of head-motion artifacts in our data (Salimi- able spatial location of a given brain area and its surround- Khorshidi et al. 2014). The final files were demeaned and ing structures with information about the morphology of had native 1.6 mm isotropic resolution in MNI space. We a specific target participant brain. This way of segmenting then extracted and regressed out the CSF and WM signal the brain into regions is, therefore, more accurate than other that was obtained from each participant’s wmparc file. atlas segmentations that are based on normalized brains. In addition, for the structural data, we downloaded the To obtain the three subdivisions for the hippocampus 3T Structural Preprocessed and 3T Structural Preprocessed we relied on the Hippocampal Subfields and Nuclei of the Extended packages. Again as per the HCP reference man- Amygdala script (v21) with the (0.7 mm) T2w image as ual, the T1w images were acquired using a 3DMPRAGE the input (Iglesias et al. 2015). Besides segmenting the hip- protocol TI/TR/TE: 1000/2400/2.14 ms, flip angle = 80 , pocampus into a set of internal subfields, this script also resulting in 0.7 mm isotropic voxels. The T2w images were provides a segmentation of the hippocampus into its head, acquired using a 3D T2-SPACE protocol TR/TE: 3200/565 body and tail sections. In addition, anterior and posterior 1 3 Brain Structure and Function (2022) 227:995–1012 999 PHG were obtained by first extracting the parahippocam- were activated during the resting state using a data-driven pal gyrus mask from a given participant’s aparc+aseg atlas. approach. We first created MTL masks combining bilateral Next, to define aPHG and pPHG, we computed an inter - hippocampus, parahippocampal gyrus and entorhinal cor- section of a plane with the centroid point of the parahip- tices that were specific to each participant. These masks pocampal gyrus. The centroid point in the anterior–posterior were similar to those described above in that they were direction was computed as the mean voxel coordinate of derived from the participant-specific aparc+aseg atlas, but the parahippocampal mask along the y-axis. Consequently, were different because they did not distinguish between the the voxels posterior to this y-plane were defined as pPHG individual MTL subcomponents. This was because we first and voxels anterior to the plane as aPHG. To separate the aimed to detect locations inside the entire MTL that are entorhinal cortex in a medial and lateral section we com- active during the resting state without taking into account puted its centroid point as the mean coordinate along the the various MTL subcomponents. To improve accuracy x-axis and den fi ed medial and lateral sections as above. This, for the group analyses, each participant’s MTL mask was therefore, produced participant-specific MTL subcomponent increased in size by one voxel. We then multiplied the masks for head, body and tail of the hippocampus, anterior masks with the cleaned, whole-brain resting-state fMRI and posterior PHG, and medial and lateral entorhinal cortex data. This produced 4D fMRI files containing only the (see Fig. 1 for a graphical presentation of the location of the timeseries of the voxels within the MTL mask of each par- MTL and its various subcomponents for a representative ticipant. We then performed group spatially restricted ICA participant). (group srICA) over these data using FSL melodic (v3.15). Given the smaller sized dataset due to the masking, we Detection of MTL activation clusters used the default method for data reduction (i.e., an initial principal component analysis) by disabling MIGP in the The next step of the analysis had three goals. First, we melodic options (Smith et al. 2014). attempted to identify those locations of the MTL that Fig. 1 Visualization of the MTL subcomponents. Location of MTL in parahippocampal gyrus (pPHG), anterior parahippocampal gyrus the left (a) and right hemispheres (b) of the whole brain along with a (aPHG), hippocampal tail (tHi), body (bHi) and head (hHi), as well zoomed view (c) as well as a tagged zoomed view (d). Panels c and d as medial (mEnt) and lateral (lEnt) entorhinal cortex highlight the various substructures that make up the MTL, posterior 1 3 1000 Brain Structure and Function (2022) 227:995–1012 As mentioned in the Introduction, the main advantage of matrices of sizes m × n where m refers to the number of srICA compared to whole-brain ICA is that a more sensi- dimensions (1–15), and n to the number of resting-state net- tive decomposition of signals in a specific region can be works (here 7). Inspection of these 15 correlation matrices obtained. As also mentioned above, a well-known problem revealed the set of resting-state networks that frequently had with fMRI data is that the GRE EPI acquisition protocols high correlation ( r > 0.40 ) with a set of components across are sensitive to local magnetic field inhomogeneities (Devlin all dimensions. We then executed an algorithm that found a et al. 2000) which lead to variability in the tSNR between component if its maximum correlation with a given resting- the different areas of the brain (see Supplementary Fig. S2 state network was above a threshold ( r > 0.4) and if this max for an overview of the tSNR in the current dataset as well maximum correlation was sufficiently higher than the second max as evidence that MTL regions had substantially lower tSNR highest correlation ( > 1.3 ), both within the same IC and max compared to other areas). A consequence of the multivariate within the same resting-state network. On the assumption nature of ICA is that signals coming from regions with low that a larger number of dimensions leads to more fraction- tSNR are less likely to be included in a component (connec- ated component clusters, we then chose the smallest dimen- tivity) map. This is because the timecourses of regions with sion at which this algorithm produced the largest number of increased noise (i.e., low tSNR) are less likely to correlate ICs. This procedure, therefore, detected in a data-driven with the timecourses of other regions. One way to deal with fashion the optimal number of dimensions for which the this issue is to apply ICA to a specific region with a low srICA produced the largest number of voxel clusters inside tSNR. In this case, the ICA procedure will take into account the MTL that were both sensitive and specific to known the specific noise profile of that region and enable a more resting-state networks. sensitive separation between voxels clusters that correspond to noise and voxel clusters that correspond to signal. In other Group‑level analyses of whole‑brain FC words, given that the MTL is a region known to have low tSNR (Olman et al. 2009 and see Supplementary Fig. S2), To anticipate our results, the previous steps resulted in the we reasoned that srICA applied to the MTL would lead to a detection of a set of FC maps that closely corresponded more sensitive detection of signal clusters inside the MTL to a set of known resting-state networks. The next step in compared to a whole-brain ICA approach. the analyses was to assess the statistical reliability of the One aspect of ICA is that it requires a decision about observed whole-brain maps at the group level. One standard the number of dimensions under which the analysis is per- way of performing such an analysis would be to rely on a formed. Here, we used a method to determine the optimal voxel-based modeling tool such as FSL randomize. How- number of dimensions for the ICA that was developed in ever, a general problem with this approach is that it assumes our laboratory (Ezama et al. 2021). Specifically, we tested that each voxel represents the exact same brain region across across a wide range of different dimensions the relationship all participants. However, as has been discussed at length between a set of components found for a specific dimension elsewhere, there is large variability in brain morphology and a set of known resting-state networks (Yeo et al. 2011). between participants, and therefore, group-level analyses of We then chose the dimension at which this relationship was this type are sub-optimal (Anticevic et al. 2008; Fischl et al. optimal. Specifically, we first performed ICA on the same 2008). Instead, we opted for a different analysis approach dataset at dimensions ranging from 1 to 15 in a stepwise that took into account the unique morphology of each par- fashion. Next, we obtained whole-brain FC maps derived ticipant’s brain. Specifically, we created a dataset that, for from all MTL activation clusters for each dimension. These each participant, contained average co-activity values for all FC maps were obtained using the Dual Regression technique their cortical and subcortical regions from the aparc+aseg (Nickerson et al. 2017). The dual regression approach con- atlas. We obtained these data by intersecting each partici- sisted of a first regression of the IC outputs of melodic to pant-specific aparc+aseg atlas with each participant-specific the cleaned and whole-brain fMRI data for each participant. whole-brain maps obtained from Dual Regression. We then The output of this first step in dual regression are the time- fitted these data to a linear mixed effect regression model courses of each independent component for each participant. of the form: The second step in dual regression involves a regression of the time-courses associated with each participant-specific IC Z = hemisphere + FC_map × brain_region + rand (participant), (1) map against the cleaned fMRI data. This second step pro- duced the whole-brain maps that represent the FC between where hemisphere was a discrete co-variable with two levels a specific IC and the rest of the brain. (left vs right), FC_map was a factor with number of levels These whole-brain FC maps obtained for each of the 15 equal to the number of ICs corresponding to resting-state dimensions were then correlated with the 7 resting-state net- networks detected in the previous step, brain_region was a works of Yeo et  al. (2011). This led to 15 correlation 1 3 Brain Structure and Function (2022) 227:995–1012 1001 factor with number of levels equal to the sum of the number obtained from Dual Regression. These data were then fit- of cortical and subcortical regions in the aparc+aseg atlas, ted to the same statistical model as described in Equation 1, and participant was a random factor with number of lev- except that the term brain_region now referred to the seven els equal to the total number of participants (i.e., 172). The MTL subcomponents. As before, our specific interest was in dependent variable was the average Z value for each brain the interaction term of the model ( FC_map × brain_region ) region computed from the participant-specific Dual Regres- that provided a test of the null hypothesis of whether the sion maps. Importantly, this mixed-effect model included a seven MTL subcomponents were activated in the same way random effect term for participant that takes into account the across the various FC maps. However, the post hoc tests that likely between-participant variability that is inherent in these were performed when this interaction term was significant data. In addition, this modeling approach uses participant- differed from those described above. Specifically, to deter - specific masks that leads to group-level results that do not mine the relative contribution of the MTL subcomponents to violate the assumption of unique brain morphology. the different resting-state networks, we first performed pair - Within this model, our specific interest was in the interac- wise comparisons of all seven MTL subcomponents within tion term of the model ( FC_map × brain_region ) that pro- each FC map. This produced a list of 21 pairwise compari- vided a test of the null hypothesis that brain regions would sons for each FC map with a test statistic (i.e., the z ratio, not have differences in mean co-activity values across the see below) that reflected the degree to which a given MTL dier ff ent FC maps. In other words, this interaction would not subcomponent differed from another MTL subcomponent. be significant if the different activation clusters detected in These pairwise test statistics were then summed, ordered, the MTL would be connected to the exact same whole-brain and thresholded at > 0. This therefore produced for each resting-state networks. When this interaction was significant, detected resting-state network an ordered list of the relative we performed post hoc tests where we compared for each FC contributions of each MTL subcomponent. map, the average Z value for a given region versus the mean of the other regions (an “effect” contrast). This, therefore, Validation analysis produced for each FC map, a list of cortical and subcorti- cal regions from the aparc+aseg atlas that had significantly To confirm the reliability of our results, we attempted to more co-activity compared to all other regions. validate our findings in a second dataset. This validation All modeling took place in the statistical computing dataset consisted of two additional rsfMRI acquisitions from environment R (v4.0.0). Mixed effect modeling relied on the same participants included in the test dataset. The two the lme4 package (v1.1.23; Bates et  al. 2007). Results rsfMRI scans from the validation set were acquired in a dif- from these regression models are presented in the form of ferent scanning session (on a different day) as the test set, but ANOVA tables that were computed directly from the output relied on the same MRI acquisition parameters. The preproc- of the mixed effect models using the lmerTest package essing protocol used was the same as described for the test (v3.1-2; Kuznetsova et al. 2017). P values in these models dataset. To validate the results, whole brain FC maps were were computed using the Satterthwaite correction for the computed from the data in the validation set using the MTL degrees of freedom. Post hoc testing was performed using clusters obtained in the test set. This analysis, therefore, pro- the emmeans package (v1.4.6; Lenth et al. 2018) when a vides a validation of the degree to which the MTL clusters given interaction term was significant (i.e., p < 0.05 ). P val- we obtained in the spatially restricted ICA step of the analy- ues in these post hoc tests were adjusted for multiple com- sis generalize to different datasets. We quantified this step parisons using the Bonferroni method. We visualized these by computing the correlation between the whole-brain FC results using the ggseg (v1.5.4; Mowinckel and Vidal- maps in the test and validation sets, and by comparing the Piñeiro 2019), and ggpubr packages (v0.3.0; Kassambara correlations of the whole-brain FC maps with the reference 2018). networks of Yeo in the test and validation sets. Relative contributions of MTL subcomponents Results The previous srICA step provided us with several clusters inside the MTL that co-activated with different resting-state Detection of MTL activation clusters networks. The final step in the analyses was to determine the relative contributions of the seven MTL subcompo- The procedure for finding the optimal number of dimen- nents (body hippocampus, anterior PHG, etc.) to the dif- sions first returned that across all the 15 dimensions tested, ferent resting-state networks. To do this, we intersected the resting-state networks 1 (visual), 2 (somatomotor), 3 (dorsal participant-specific MTL subcomponent masks (described attention), and 7 (default mode) were most frequently found above) with each participant-specific whole-brain FC map with correlations r > 0.40 . In addition, the algorithm found 1 3 1002 Brain Structure and Function (2022) 227:995–1012 that dimension 7 was the lowest dimension at which the larg- restricted ICA occupy positions within the MTL that both est number of ICs were strongly and uniquely connected to respect and cross anatomical boundaries (e.g., IC0 seems different resting-state networks. Specifically, we found that to reflect activity in both parahippocampal gyrus as well as for dimension 7, four ICs were strongly and uniquely corre- entorhinal cortex). The statistical reliability of the observed lated with four different resting-state networks: IC0 was cor - whole-brain networks as well as how the detected activation related with the dorsal attention network ( r = 0.42 ), IC1 was clusters are distributed across the various MTL subcompo- correlated with the somatomotor network ( r = 0.53 ), IC2 nents was examined in more detail below. was correlated with the default mode network ( r = 0.59 ) , and IC3 was correlated with the visual network ( r = 0.66 ; Group‑level analyses of whole‑brain FC see Table 1 for an overview of the correlations for each IC with all networks). As can be seen in Supplementary Fig. Group-level mixed-effect regression analysis revealed main S3, strong correlations ( r > 0.40 ) were frequently found for effects of Hemisphere ( F = 21.72, p < 0.0001 ), FC 1,59508 these four networks in other dimensions, suggesting that the Map ( F = 2955.38, p < 0.0001 ), and Brain Region 3,59508 detection of these four networks was not idiosyncratic to ( F = 351.21, p < 0.0001 ). Most relevant for our pre- 43,59508 dimension 7. In addition, as can be seen in Supplementary sent purposes, there was a significant interaction between Fig. S4, we also examined dimensions 20 and 30 and this FC Map and Brain Region ( F = 589.18, p < 0.0001 ), 129,59508 did not lead to the detection of new networks. Finally, as can suggesting that co-activity values of brain regions differed be seen in Supplementary Fig. S5, the ICs that were uncor- between the various whole-brain FC maps associated with related with the reference networks were indeed unlikely to each IC. As can be seen in Table  2, post hoc analyses reflect real BOLD signal. Specifically, IC4 seems to reflect revealed that IC0 and its corresponding FC map revealed signal in CSF, IC5 in draining veins, and IC6 does not reveal regions typically associated with the dorsal-attention net- much signal in the first place (see also Griffanti et al. 2017, work like the inferior and superior parietal cortex, as well for further information on manually classifying ICs). We as lateral frontal areas. Similarly, as can be seen in Table 3, can, therefore, conclude that for our data the specic fi clusters IC1 and its associated FC map showed regions typically of voxels detected by the ICA using dimension 7 for IC0, associated with the somatomotor network like the amyg- IC1, IC2 and IC3 were optimal in connecting with the four dala and post- and pre-central gyri. In addition, as can been known resting-state networks (see also Supplementary Fig. seen in Table 4, IC2 and its corresponding FC map revealed S6 for a direct comparison in overlap between the obtained regions generally found in the default mode network, like ICs and Yeo networks). the isthmus cingulate (retrosplenial cortex), the precuneus, A visual presentation of the precise location of the four and the medial orbitofrontal cortex. Finally, as can be seen ICs along with their whole-brain group-level FC map in Table 5, IC3 and its associated FC map revealed regions derived from Dual Regression is presented in Fig. 2. As can associated with the visual network like the cuneus and peri- be seen in Fig. 2, the four ICs are located at different posi- calcarine sulcus (see also Fig. 4B and C for a visual presen- tions in the MTL and revealed contrasting FC with the rest tation of these results). of the brain, indicating the involvement in different resting- state networks. A more detailed view of the location of each Relative contributions of MTL subcomponents IC within the MTL can be seen in Fig. 3A–C. In this Figure, it can be seen that the four clusters detected by spatially Group-level mixed-effect regression analyses that exam- ined the relative contribution of the MTL subcompo- nents to each resting-state network revealed main effects of Hemisphere ( F = 45.96, p < 0.0001 ), Br ain Table 1 Table of correlations of the FC maps with resting-state net- 1,9432 works Region ( F = 1867.63, p < 0.0001 ) and FC Map 6,9432 ( F = 2378.99, p < 0.0001 ). Again, important for our Yeo et al. (2011), seven networks IC0 IC1 IC2 IC3 3,9432 present purposes, the interaction between Region and FC Visual 0.13 0.12 0.02 0.66 Map was highly significant ( F = 1442.85, p < 0.0001 ), 18,9432 Somatomotor 0.02 0.53 0.03 0.24 suggesting that average co-activity values for each MTL sub- Dorsal attention 0.42 0.05 0.01 0.16 component differed between the four ICs. Further explora- Ventral attention and salience 0.03 0.01 0.02 0.21 tion of this interaction using pairwise comparisons of the Limbic 0.04 0.10 0.00 0.02 seven MTL subcomponents within each IC and then ranking Executive control 0.20 0.02 0.11 0.00 the summed z ratios revealed the relative contributions of Default mode 0.12 0.20 0.59 0.01 each MTL subcomponent. Specifically, as can be seen in Table 6, summed z ratios in IC0 (correlated with the dorsal- Correlations that were strongly and unique correlated with specific attention network) were strongest in pPHG, then in aPHG resting-state networks are shown in bold 1 3 Brain Structure and Function (2022) 227:995–1012 1003 Fig. 2 The subset of Independent Components (IC0, first row; IC1, co-activation quantified in Z values. Note how different IC hotspots second; IC2, third; IC3, fourth) detected as signal from the spatially inside the MTL connect to different areas of the brain that highly restricted ICA (left columns, srICA), and their corresponding large- overlap with resting-state networks (see also Supplementary Fig. S6). scale functional connectivity maps derived from Dual Regression srICA spatially restricted independent component analysis, IC inde- (right columns, DR). The color gradient represents the group level of pendent component, DR dual regression and finally in lEnt. Similarly, Table  6 showed that for IC1 correlations. Specifically, the FC maps associated with (correlated with the somatomotor network), summed z ratios IC0 (DA), IC1 (SM), IC2 (DMN) and IC3 (VIS), corre- were strongest in hHi followed by bHi, and tHi. In addi- lated r = 0.95, r = 0.96, r = 0.98, r = 0.97 between the test tion, Table 6 showed that for IC2 (correlated with the default and validation set, respectively. Finally, the pattern of cor- mode network) summed z ratios in descending order were relations between the obtained FC maps in the validation ranked pPHG, aPHG, hHi and bHi. Finally, Table 6 showed set and the Yeo reference networks was highly similar (see that for IC3, summed z ratios were strongest in pPHG (see Table S1). We, therefore, conclude the MTL clusters that we also Figure 5 for a graphical presentation of these results). report are robust and generalize to different datasets. Validation results Complementary results Calculation of the whole-brain FC maps in the validation One concern with the results reported above is that we only dataset using the MTL clusters obtained from the test set compared the whole-brain FC maps with the set of 7 ref- described above showed highly similar results to those erence networks of Yeo et al. (2011). In complementary obtained in the test set (see Figure S7). Correlation of FC analyses, we examined whether our results would general- maps in the test and validation sets showed highly reliable ize to the 17 reference networks of Yeo et al. (2011), the 1 3 1004 Brain Structure and Function (2022) 227:995–1012 Fig. 3 Detailed location of the main target hotspots in the MTL Note how ICs both respect and cross anatomical boundaries suggest- detected with spatially restricted ICA. IC0 (blue), IC1 (green), IC2 ing inter-structure communication. Slices in 1 mm MNI152 space. (orange) and IC3 (red) in saggital (a), coronal (b) and axial (c) views. DA dorsal attention, SM somatomotor, DM default mode, Vis visual 10 networks in Smith et al. (2009) and the 18 networks in the highest correlation with the visual network, IC2 with Allen et al. (2011). Specifically, we correlated the whole the default mode, and IC1 with the somatomotor network for brain FC maps that we obtained for dimension 7 with these both the 17 reference networks of Yeo et al. (2011) and the additional reference networks. The results from these cor- 10 networks in Smith et al. (2009). However, IC0 (associ- relations are shown in Supplementary Tables S2–4. There ated with dorsal attention in the set of 7 networks) correlated are two noteworthy observations from these results. First, r = 0.28 the default mode network C and r = 0.20 with the the results we obtained with the 17 networks of Yeo et al. dorsal attention network A in the 17 networks of Yeo et al. (2011) and the 10 networks of Smith et al. (2009) were com- (2011), and r = 0.29 with the frontoparietal network in the parable to those obtained above. Specifically, IC3 yielded set of Smith et al. (2009). This variability likely reflects a 1 3 Brain Structure and Function (2022) 227:995–1012 1005 Table 2 Cortical and subcortical areas showing reliable co-activity they resemble the low correlations found by Ruiz-Rizzo with IC0 (correlated with Dorsal Attention network) relative to the et al. (2020) that also used this reference set. We discuss mean co-activity value of all other areas this issue in more detail below. Region z ratio p value Parahippocampal 47.72 ~0.000E+00 Discussion Inferior parietal 40.13 ~0.000E+00 Precuneus 24.03 5.577E–126 The aim of the current study was to characterize the different Fusiform 20.59 1.569E–92 resting-state networks that co-activate with the MTL as well Caudal middle frontal 17.42 2.406E–66 as detail how the different MTL subcomponents contribute Inferior temporal 16.61 2.535E–60 to these resting-state networks. We examined this issue using Isthmus cingulate 11.68 7.301E–30 the high spatial resolution 7T rsfMRI dataset from the HCP Superior parietal 9.17 2.068E–18 with a data-driven method that applied ICA in a manner that Supramarginal 8.38 2.194E–15 was restricted to the MTL. We found that during the resting Entorhinal 7.79 2.789E–13 state, our method detected four activation clusters that were Middle temporal 5.78 3.301E–07 spread across the various subcomponents of the MTL. Using Medial orbitofrontal 4.10 1.802E–03 Dual Regression and mixed effect regression techniques to p values corrected for multiple comparisons using Bonferroni correc- estimate reliability across participants, we found that these tion four activation clusters were functionally connected to four different whole-brain resting-state networks that relied on further splintering of networks and label differences between different contributions of MTL subcomponents. Specifically, the reference sets. we found that the dorsal attention network (detected with Second, the correlations between our obtained results and r = 0.42 ) relied primarily on the parahippocampal gyrus the 18 networks of Allen et al. (2011) resulted in generally and entorhinal cortex, the somatomotor network ( r = 0.53 ) low correlations (see Supplementary Table S4). Although on the hippocampus, the default mode network (r = 0.59 ) we cannot provide a clear explanation of these low correla- on both parahippocampal gyrus and hippocampus, and the tions at this point, we note that these results are not in line visual network ( r = 0.66 ) on the parahippocampal cortex with these observed with the other reference sets and that Table 3 Cortical and subcortical areas showing reliable co-activity Table 4 Cortical and subcortical areas showing reliable co-activity with IC1 (correlated with Somatomotor network) relative to the mean with IC2 (correlated with Default Mode network) relative to the mean co-activity value of all other areas co-activity value of all other areas Region z ratio p value Region z ratio p value Amygdala 55.30 ~0.000E+00 Isthmus cingulate 70.52 ~0.000E+00 Hippocampus 53.71 ~0.000E+00 Precuneus 50.65 ~0.000E+00 Postcentral 49.85 ~0.000E+00 Parahippocampal 34.93 1.301E–265 Paracentral 34.58 2.413E–260 Inferior parietal 32.82 1.256E–234 Superior temporal 27.76 5.009E–168 Rostral anterior cingulate 32.67 1.781E–232 Precentral 21.26 1.237E–98 Frontal pole 26.58 5.334E–154 Bankssts 19.23 8.143E–81 Medial orbitofrontal 26.31 7.064E–151 Medial orbitofrontal 13.27 1.406E–38 Caudal middle frontal 21.98 1.828E–105 Entorhinal 12.17 1.874E–32 Superior frontal 18.42 3.889E–74 Fusiform 11.51 5.215E–29 Middle temporal 13.09 1.656E–37 Temporal pole 10.15 1.421E–22 Hippocampus 11.37 2.489E–28 Cuneus 10.00 6.511E–22 Posterior cingulate 8.69 1.499E–16 Frontal pole 8.82 4.865E–17 Caudate 7.09 5.923E–11 Isthmus cingulate 7.39 6.378E–12 Ventral DC 6.21 2.345E–08 Middle temporal 7.19 2.866E–11 Temporal pole 4.91 3.874E–05 Transverse temporal 3.83 5.402E–03 Rostral middle frontal 4.19 1.188E–03 Parahippocampal 3.62 1.279E–02 Brain stem 3.53 1.777E–02 p values corrected for multiple comparisons using Bonferroni correc- p values corrected for multiple comparisons using Bonferroni correc- tion tion 1 3 1006 Brain Structure and Function (2022) 227:995–1012 1 3 Brain Structure and Function (2022) 227:995–1012 1007 ◂Fig. 4 Spiderplot representation of functional connectivity (estimated In particular, the posterior network is typically associated marginal coefficients) between the four IC hotspots (IC0, IC1, IC2, with posterior regions like the retrosplenial cortex, precu- IC3) and the rest of the brain (a), as well as functional connectiv- neus (Ranganath and Ritchey 2012; Ritchey et al. 2015) as ity results from group-based analyses in subcortical (b) and cortical well as with occipital areas (Libby et al. 2012; Wang et al. (c) regions for each of the different ICs (see Tables  2, 3, 4  and 5 for details) 2016). As can be seen in Tables 4 and 5 these regions were exactly among those with the highest co-activation in the set of regions linked to the visual and default mode networks (see Table 6 for details). These results were validated with found in our study. Similarly, the anterior network is typi- high replication ( r > 0.95 ) in a separate dataset. cally associated with the amygdala and orbitofrontal cortex Previous studies have reported inconsistent results on the (Ranganath and Ritchey 2012; Ritchey et al. 2015), and as number of whole-brain networks that co-activate with the can be seen in Table 3, these regions also appear among the MTL. Whereas, the classical view is that the MTL is con- most co-activated in the set of regions associated with the nected to a posterior and an anterior network (Kahn et al. somatomotor network. The current results, therefore, suggest 2008; Libby et al. 2012; Qin et al. 2016; Ranganath and that, in a resting-state context, the posterior and anterior net- Ritchey 2012; Ritchey et al. 2015; Schröder et al. 2015), works typically linked with the MTL are the visual, default more recent studies have found that the MTL relies on mode and somatomotor networks, respectively. additional networks beyond these two traditionally pro- Although the current results are in line with those studies posed (Ruiz-Rizzo et al. 2020; Wang et al. 2016; Plachti that have argued for a more expanded connectivity between et al. 2019). The current results are in line with these more the MTL and the rest of the brain (Ruiz-Rizzo et al. 2020; recent studies in that they show that the MTL is connected Wang et al. 2016; Plachti et al. 2019), they differ from these to additional resting-state networks. Specifically, the cur - previous studies in some details. First, as mentioned in the rent study shows that the MTL was connected to the dorsal Introduction, a recent study by Ruiz-Rizzo et al. (2020) con- attention, somatomotor, default mode and visual networks cluded that MTL was connected with five different resting- (see Table 1 for details). Three of these four networks have state networks. Using the reference set of 20 networks of a clear correspondence to the posterior and anterior net- Allen et al. (2011), they found that MTL connected with works previously identified. Specifically, the default mode default mode, salience, frontal, basal ganglia and visual and visual network likely reflect the previously identified networks. One general problem with this study are the posterior network, while the somatomotor network likely relatively low correlations between the observed and refer- reflects the anterior network. This interpretation rests on ence networks (i.e., visual, frontal and salience networks all the specific overlap of regions usually associated with the had r < 0.15 ). Indeed, the problem here may to be related posterior/anterior networks and with the regions found to be to the specific reference atlas because when we used the linked to the resting-state networks obtained in our study. Allen et al. (2011) atlas as a reference, correlations between Fig. 5 Spiderplot representation of co-activity values (estimated mar- from pPHG and aPHG and lateral Ent, and that IC2 (Default Mode, ginal coefficients) within the 7 MTL regions for each of the four ICs green dots) relied primarily on pPHG, aPHG, and on head and body (a), as well as the relative contribution (in terms of summed z-val- of the hippocampus. DA dorsal attention, SM somatomotor, DM ues) from each MTL region to each of the four ICs (b). Note that, default mode, Vis visual for example, IC0 (Dorsal Attention, blue dots) relies on contributions 1 3 1008 Brain Structure and Function (2022) 227:995–1012 obtained networks and the reference set were also rela- parahippocampal gyrus (aPHG, including perirhinal cortex) tively low (see Complementary Results and Supplementary and anterior sections of the hippocampus. Interestingly, our Table S4). Given that correlations were substantially higher results revealed that the somatomotor network, that included in the reference sets of Yeo et al. (2011) and Smith et al. areas typically associated with the anterior network like the (2009), future studies should be geared towards resolving amygdala and medial orbitofrontal cortex, relied primar- these rather remarkable discrepancies between sets of refer- ily on the hippocampus and not on the parahippocampal ence networks. In addition, a study by Plachti et al. (2019) gyrus (see Table 3 and Figure 3). One possible explanation found that during the resting state, functional connectivity for this is that the anterior network previously observed in between the hippocampus and the rest of the brain could low spatial resolution data represented a mix between the be described by 3, 5 or even 7 clusters. This seems at odds somatomotor and dorsal attention networks thereby explain- with our observation that the hippocampus was involved ing the involvement of the aPHG. In sum, the current results, in only two networks (default mode and somatomotor net- therefore, suggest that MTL plays a role in four different works; see also Ezama et al. 2021). However, this interpre- resting-state networks where these networks display distinct tation is complicated by the fact that Plachti et al. (2019) configurations of MTL subcomponents. focused on the internal parcellation of the hippocampus and Although the MTL is traditionally linked with episodic did not present the whole-brain functional connectivity of memory (e.g., Squire & Zola-Morgan 1991), more recent the obtained parcellations. Future studies that employ the studies have shown the involvement of this structure in a consensus clustering technique used by Plachti et al. (2019) wide range of cognitive functions such as short-term mem- while also reporting whole-brain connectivity maps should ory (Ranganath and Blumenfeld 2005), visual perception be able to resolve this issue. In short, although the current (Barense et al. 2012), attention (Aly and Turk-Browne 2016; results as well as the set of studies discussed here support the Córdova et al. 2019; Ruiz et al. 2020), and language and notion of expanded connectivity of the MTL, specific details conceptual processing (Duff and Brown-Schmidt 2012; regarding the influence of the particular reference set and Mack et al. 2016; Piai et al. 2016). For example, a recent pattern of whole-brain connectivity remain to be resolved. study by Ruiz et al. (2020) revealed that patients with MTL The current results also provide insight into how the vari- lesions showed impaired performance in an attention task ous MTL subcomponents contribute to these four resting- that relied on visual perception suggesting that the MTL state networks. Specifically, the visual network relied pri- plays a role in attention processes. The results reported here marily on posterior sections of the parahippocampal gyrus are in line with these studies in that they highlight the vari- (PHG), and the dorsal attention network primarily on a ety of resting-state networks to which the MTL contributes. posterior-to-anterior gradient along the parahippocampal Specifically, although resting-state studies can only make long-axis and lateral entorhinal cortex (pPHG–aPHG–lEnt, limited claims about function, our observation of the MTL in order of relative contribution). In addition, the default in visual and dorsal attention networks seem to be in line mode network relied on a more complex pattern of co-acti- with previous studies that have emphasized the implication vation in both parahippocampal gyrus and hippocampus of MTL in perception and attentional processes (Aly and with opposite gradients in these two structures: In the para- Turk-Browne 2016; Córdova et al. 2019; Ruiz et al. 2020). hippocampal gyrus the gradient was in the posterior–ante- The current observation that the MTL is involved in four dif- rior (pPHG–aPHG) direction; whereas in hippocampus, it ferent resting-state networks, therefore, further underscores was in the anterior–posterior (head–body of hippocampus) the notion that the MTL may be involved in a more abstract direction. Finally, the somatomotor network relied primar- type of processing (e.g., relational) that plays an important ily on the hippocampus with an anterior–posterior gradient part in many different cognitive domains. (head–body–tail of hippocampus; see Table 6, and Figure 5 Our study has several limitations. First, our conclusion for details). These results are generally consistent with those that MTL relies on four particular resting-state networks is previously observed. Specifically, the posterior network has based on the specific reference atlas of Yeo et al. (2011). traditionally been associated with pPHG and middle to pos- However, as we show in Supplementary Tables S2–4, cor- terior sections of the hippocampus (Kahn et al. 2008; Libby relations with the 17 network atlas of Yeo et al. (2011) and et al. 2012; Qin et al. 2016; Ranganath and Ritchey 2012; the 10 networks of Smith et al. (2009) yielded highly simi- Ritchey et al. 2015; Ruiz-Rizzo et al. 2020; Schröder et al. lar results. In addition, group-level FC analysis relied on 2015; Wang et al. 2016). This is in line with our observa- averages across voxels in a brain region. An obvious dis- tions that the visual and default mode networks strongly advantage in this approach is that there are limitations on co-activate with posterior sections of the parahippocampal the spatial precision with which an effect may be localized. gyrus and hippocampus. In addition, the anterior network We think such concerns may be further mitigated by using is traditionally associated with anterior sections of the procedures that segment brain regions to more fine grained 1 3 Brain Structure and Function (2022) 227:995–1012 1009 parcels (e.g., Schaefer et al. 2018). Finally, although we Table 6 Strongest contrast for MTL substructures for each IC compared the contributions of 7 different MTL subcom - Region IC Summed z ratio Rank Network ponents to whole-brain resting-state networks, we did not pPHG 0 366.37 1 explicitly include the perirhinal cortex. The reason for this aPHG 0 158.15 2 Dorsal attention is twofold. First, the perirhinal cortex simply does not form lEnt 0 8.44 3 part of the standard Desikan–Killany atlas that we used for hHi 1 342.85 1 our segmentations (Desikan et al. 2006). Second, the precise bHi 1 180.64 2 Somatomotor anatomical definition of perirhinal cortex remains highly tHi 1 141.79 3 disputed (Suzuki and Amaral 1994; Augustinack et  al. pPHG 2 233.97 1 2014), which complicates an accurate automatic segmenta- aPHG 2 122.92 2 Default mode tion. Thus, although the usefulness of this structure is clear hHi 2 30.40 3 (Augustinack et al. 2014; Libby et al. 2012), the integration bHi 2 10.85 4 of this structure into an automatic segmentation pipeline that pPHG 3 456.13 1 Visual also includes other MTL structures has prevented us from analyzing the contribution of this structure at this moment Contrast based on the sum of pairwise z ratio differences for all sub- in time. structures. Rank indicates the order of the substructures relative con- To conclude, the current study used a high spatial reso- tribution for each IC lution dataset to examine the different whole-brain resting- state networks that co-activate with the MTL. A secondary (Kahn et al. 2008; Libby et al. 2012; Qin et al. 2016; Ran- goal was to examine how the different MTL subcompo- nents contribute to these resting-state networks. We found ganath and Ritchey 2012; Ritchey et al. 2015; Ruiz-Rizzo et al. 2020; Schröder et al. 2015; Wang et al. 2016; Bar- that the MTL co-activates with the default mode, somato- motor, visual and dorsal attention networks. Our results nett et  al. 2019; Vincent et  al. 2006). They are in line with previous FC studies that have found that the MTL revealed that these networks are subserved by distinct con- figurations of MTL subcomponent co-activity, where the is linked to additional networks (Ruiz-Rizzo et al. 2020; Wang et al. 2016; Plachti et al. 2019) and are suggestive default mode network relied on a combination of activity in parahippocampal gyrus and hippocampus, the somato- of a functional role of MTL that goes beyond episodic memory (Aly and Turk-Browne 2016; Barense et al. 2012; motor network on the hippocampus, the visual network on the parahippocampal gyrus, and the dorsal attention Córdova et al. 2019; Duff and Brown-Schmidt 2012; Ran- ganath and Blumenfeld 2005; Ruiz et al. 2020). Finally, network on the parahippocampal gyrus and the entorhinal cortex. These results go beyond previous studies that have the current results are obtained from young healthy adults and therefore establish a baseline pattern of how various associated MTL with a posterior and an anterior network MTL subcomponents contribute to known resting-state networks. In the future, we hope to examine how these Table 5 Cortical and subcortical areas showing reliable co-activity patterns change with aging and pathology. with IC3 (correlated with Visual network) relative to the mean co- activity value of all other areas Supplementary Information The online version contains supplemen- Region z ratio p value tary material available at https://doi. or g/10. 1007/ s00429- 021- 02442-1 . Cuneus 99.93 ~0.000E+00 Acknowledgements The authors would like to thank José María Pericalcarine 89.54 ~0.000E+00 Pérez González (IMETISA) for technical support. Correspondence Lingual 77.22 ~0.000E+00 and requests for materials should be addressed to NJ (njanssen@ Superior parietal 40.98 ~0.000E+00 ull.es). Precuneus 23.29 2.205E–118 Funding Open Access funding provided thanks to the CRUE- Parahippocampal 18.32 2.276E–73 CSIC agreement with Springer Nature. This study was supported Fusiform 16.28 5.737E–58 by grants PSI2017-84933-P and PSI2017-91955-EXP to NJ, and by Lateral occipital 15.69 7.799E–54 TESIS2019010146 from the Board of Economy, Industry, Trade and Postcentral 11.46 9.415E–29 Knowledge of the Canarian Government with a European Social Fund co-financing rate to SLS. Paracentral 8.86 3.329E–17 Caudal anterior cingulate 7.90 1.164E–13 Availability of data and material Data were provided by the Human Transverse temporal 6.98 1.274E–10 Connectome Project, WU-Minn Consortium (Principal Investigators: Precentral 4.16 1.375E–03 David Van Essen and Kamil Ugurbil; 1U54MH091657) funded by the 16 NIH Institutes and Centers that support the NIH Blueprint for p values corrected for multiple comparisons using Bonferroni correc- Neuroscience Research; and by the McDonnell Center for Systems tion 1 3 1010 Brain Structure and Function (2022) 227:995–1012 Neuroscience at Washington University. The data that support the Bates D, Sarkar D, Bates MD, Matrix L (2007) The lme4 package. R findings of this study are openly available from Human Connectome package version 2:74 Project (www. human conne ctome. org). Blessing EM, Beissner F, Schumann A, Brünner F, Bär K-J (2016) A data-driven approach to mapping cortical and subcortical intrinsic functional connectivity along the longitudinal hippocampal axis. Code availability Code will be made available upon reasonable request. Hum Brain Mapp 37:462–476 Córdova NI, Turk-Browne NB, Aly M (2019) Focusing on what mat- Declarations ters: modulation of the human hippocampus by relational atten- tion. Hippocampus 29:1025–1037 Conflict of interest All authors declare that they have no conflict of Desikan RS, Ségonne F, Fischl B, Quinn BT, Dickerson BC, Blacker interest. D, Buckner RL, Dale AM, Maguire RP, Hyman BT et al (2006) An automated labeling system for subdividing the human cerebral Ethics approval The data analyses were conducted in agreement with cortex on mri scans into gyral based regions of interest. Neuroim- the declaration of Helsinki and with the protocol established by the age 31:968–980 Ethics Commission for Research of the Universidad de La Laguna, the Devlin JT, Russell RP, Davis MH, Price CJ, Wilson J, Moss HE, Mat- Comité de Ética de la Investigación y Bienestar Animal. thews PM, Tyler LK (2000) Susceptibility-induced loss of sig- nal: comparing pet and FMRI on a semantic task. Neuroimage Consent to participate Not applicable. 11:589–600 Douw L, DeSalvo MN, Tanaka N, Cole AJ, Liu H, Reinsberger C, Stuf- Consent for publication Not applicable. flebeam SM (2015) Dissociated multimodal hubs and seizures in temporal lobe epilepsy. Ann Clin Transl Neurol 2:338–352 Duff MC, Brown-Schmidt S (2012) The hippocampus and the flex- Open Access This article is licensed under a Creative Commons Attri- ible use and processing of language. Front Hum Neurosci 6:69 bution 4.0 International License, which permits use, sharing, adapta- Ezama L, Hernández-Cabrera JA, Seoane S, Pereda E, Janssen N tion, distribution and reproduction in any medium or format, as long (2021) Functional connectivity of the hippocampus and its sub- as you give appropriate credit to the original author(s) and the source, fields in resting-state networks. Eur J Neurosci 53:3378 provide a link to the Creative Commons licence, and indicate if changes Fischl B, Rajendran N, Busa E, Augustinack J, Hinds O, Yeo were made. The images or other third party material in this article are BT, Mohlberg H, Amunts K, Zilles K (2008) Cortical fold- included in the article's Creative Commons licence, unless indicated ing patterns and predicting cytoarchitecture. Cereb Cortex otherwise in a credit line to the material. If material is not included in 18:1973–1980 the article's Creative Commons licence and your intended use is not Formisano E, Esposito F, Di Salle F, Goebel R (2004) Cortex-based permitted by statutory regulation or exceeds the permitted use, you will independent component analysis of FMRI time series. Magn need to obtain permission directly from the copyright holder. To view a Reson Imaging 22:1493–1504 copy of this licence, visit http://cr eativ ecommons. or g/licen ses/ b y/4.0/ . Glasser MF, Sotiropoulos SN, Wilson JA, Coalson TS, Fischl B, Andersson JL, Xu J, Jbabdi S, Webster M, Polimeni JR et al (2013) The minimal preprocessing pipelines for the human con- nectome project. Neuroimage 80:105–124 References Govindpani K, Turner C, Waldvogel HJ, Faull RL, Kwakowsky A (2020) Impaired expression of gaba signaling components in Aggleton JP (2012) Multiple anatomical systems embedded within the the Alzheimer’s disease middle temporal gyrus. Int J Mol Sci primate medial temporal lobe: implications for hippocampal func- 21:8704 tion. Neurosci Biobehav Rev 36:1579–1596 Griffanti L, Salimi-Khorshidi G, Beckmann CF, Auerbach EJ, Douaud Allen EA, Erhardt EB, Damaraju E, Gruner W, Segall JM, Silva RF, G, Sexton CE, Zsoldos E, Ebmeier KP, Filippini N, Mackay CE Havlicek M, Rachakonda S, Fries J, Kalyanam R et al (2011) A et al (2014) ICA-based artefact removal and accelerated FMRI baseline for the multivariate comparison of resting-state networks. acquisition for improved resting state network imaging. Neuroim- Front Syst Neurosci 5:2 age 95:232–247 Alvarez P, Squire LR (1994) Memory consolidation and the medial Griffanti L, Douaud G, Bijsterbosch J, Evangelisti S, Alfaro-Almagro temporal lobe: a simple network model. Proc Natl Acad Sci F, Glasser MF, Duff EP, Fitzgibbon S, Westphal R, Carone D 91:7041–7045 et al (2017) Hand classification of FMRI ICA noise components. Aly M, Turk-Browne NB (2016) Attention stabilizes representations in Neuroimage 154:188–205 the human hippocampus. Cereb Cortex 26:783–796 Hyde JS, Biswal BB, Jesmanowicz A (2001) High-resolution FMRI Anticevic A, Dierker DL, Gillespie SK, Repovs G, Csernansky JG, Van using multislice partial k-space gr-epi with cubic voxels. Magn Essen DC, Barch DM (2008) Comparing surface-based and vol- Reson Med 46:114–125 ume-based analyses of functional neuroimaging data in patients Iglesias JE, Augustinack JC, Nguyen K, Player CM, Player A, Wright with schizophrenia. Neuroimage 41:835–848 M, Roy N, Frosch MP, McKee AC, Wald LL et al (2015) A com- Augustinack JC, Magnain C, Reuter M, van der Kouwe AJ, Boas D, putational atlas of the hippocampal formation using ex  vivo, Fischl B (2014) MRI parcellation of ex vivo medial temporal lobe. ultra-high resolution MRI: application to adaptive segmentation Neuroimage 93:252–259 of in vivo mri. Neuroimage 115:117–137 Barense MD, Groen II, Lee AC, Yeung L-K, Brady SM, Gregori M, Janssen N, Mendieta CCR (2020) The dynamics of speech motor con- Kapur N, Bussey TJ, Saksida LM, Henson RN (2012) Intact trol revealed with time-resolved FMRI. Cereb Cortex 30:241–255 memory for irrelevant information impairs perception in amnesia. Jenkinson M, Bannister P, Brady M, Smith S (2002) Improved optimi- Neuron 75:157–167 zation for the robust and accurate linear registration and motion Barnett AJ, Man V, McAndrews MP (2019) Parcellation of the hip- correction of brain images. Neuroimage 17:825–841 pocampus using resting functional connectivity in temporal lobe Jones BF, Witter MP (2007) Cingulate cortex projections to the para- epilepsy. Front Neurol 10:920 hippocampal region and hippocampal formation in the rat. Hip- pocampus 17:957–976 1 3 Brain Structure and Function (2022) 227:995–1012 1011 Kahn I, Andrews-Hanna JR, Vincent JL, Snyder AZ, Buckner RL Ruiz NA, Meager MR, Agarwal S, Aly M (2020) The medial temporal (2008) Distinct cortical anatomy linked to subregions of the lobe is critical for spatial relational perception. J Cogn Neurosci medial temporal lobe revealed by intrinsic functional connectiv- 32:1780–1795 ity. J Neurophysiol 100:129–139 Ruiz-Rizzo AL, Beissner F, Finke K, Müller HJ, Zimmer C, Pasquini Kassambara A (2018) ggpubr:“ggplot2” based publication ready plots. L, Sorg C (2020) Human subsystems of medial temporal lobes R package version 0.1, 7 extend locally to amygdala nuclei and globally to an allostatic- Kenkhuis B, Jonkman LE, Bulk M, Buijs M, Boon BD, Bouwman interoceptive system. Neuroimage 207:116404 FH, Geurts JJ, van de Berg WD, van der Weerd L (2019) 7t MRI Salimi-Khorshidi G, Douaud G, Beckmann CF, Glasser MF, Griffanti allows detection of disturbed cortical lamination of the medial L, Smith SM (2014) Automatic denoising of functional MRI temporal lobe in patients with Alzheimer’s disease. NeuroImage data: combining independent component analysis and hierarchi- Clin 21:101665 cal fusion of classifiers. Neuroimage 90:449–468 Kobayashi Y, Amaral DG (2003) Macaque monkey retrosplenial cor- Schaefer A, Kong R, Gordon EM, Laumann TO, Zuo X-N, Holmes tex: II. cortical afferents. J Comp Neurol 466:48–79 AJ, Eickhoff SB, Yeo BT (2018) Local-global parcellation of the Kobayashi Y, Amaral DG (2007) Macaque monkey retrosplenial cor- human cerebral cortex from intrinsic functional connectivity MRI. tex: III. cortical efferents. J Comp Neurol 502:810–833 Cereb Cortex 28:3095–3114 Kondo H, Saleem KS, Price JL (2005) Differential connections of Schröder TN, Haak KV, Jimenez NIZ, Beckmann CF, Doeller CF the perirhinal and parahippocampal cortex with the orbital and (2015) Functional topography of the human entorhinal cortex. medial prefrontal networks in macaque monkeys. J Comp Neurol Elife 4:e06738 493:479–509 Seidman LJ, Pantelis C, Keshavan MS, Faraone SV, Goldstein JM, Kuznetsova A, Brockhoff PB, Christensen RH et al (2017) lmertest Horton NJ, Makris N, Peter F, Caviness VS, Tsuang MT (2003) package: tests in linear mixed effects models. J Stat Softw 82:1–26 A review and new report of medial temporal lobe dysfunction as Lenth R, Singmann H, Love J, Buerkner P, Herve M (2018) Emmeans: a vulnerability indicator for schizophrenia: a magnetic resonance estimated marginal means, aka least-squares means. R package imaging morphometric family study of the parahippocampal version 1:3 gyrus. Schizophr Bull 29:803–830 Libby LA, Ekstrom AD, Ragland JD, Ranganath C (2012) Differential Sladky R, Baldinger P, Kranz GS, Tröstl J, Höflich A, Lanzenberger connectivity of perirhinal and parahippocampal cortices within R, Moser E, Windischberger C (2013) High-resolution functional human hippocampal subregions revealed by high-resolution func- MRI of the human amygdala at 7 t. Eur J Radiol 82:728–733 tional imaging. J Neurosci 32:6550–6560 Smith SM, Fox PT, Miller KL, Glahn DC, Fox PM, Mackay CE, Filip- Mack ML, Love BC, Preston AR (2016) Dynamic updating of hip- pini N, Watkins KE, Toro R, Laird AR et al (2009) Correspond- pocampal object representations reflects new conceptual knowl- ence of the brain’s functional architecture during activation and edge. Proc Natl Acad Sci 113:13203–13208 rest. Proc Natl Acad Sci 106:13040–13045 Mowinckel AM, Vidal-Piñeiro D (2019) Visualisation of brain sta- Smith SM, Beckmann CF, Andersson J, Auerbach EJ, Bijsterbosch tistics with r-packages ggseg and ggseg3d. arXiv preprint J, Douaud G, Duff E, Feinberg DA, Griffanti L, Harms MP et al arXiv:1912.08200, (2013) Resting-state FMRI in the human connectome project. Nickerson LD, Smith SM, Öngür D, Beckmann CF (2017) Using dual Neuroimage 80:144–168 regression to investigate network shape and amplitude in func- Smith SM, Hyvärinen A, Varoquaux G, Miller KL, Beckmann CF tional connectivity analyses. Front Neurosci 11:115 (2014) Group-PCA for very large FMRI datasets. Neuroimage Olman CA, Davachi L, Inati S (2009) Distortion and signal loss in 101:738–749 medial temporal lobe. PLoS One 4:e8160 Squire LR, Zola-Morgan S (1991) The medial temporal lobe memory Piai V, Anderson KL, Lin JJ, Dewar C, Parvizi J, Dronkers NF, Knight system. Science 253:1380–1386 RT (2016) Direct brain recordings reveal hippocampal rhythm Squire LR, Wixted JT, Clark RE (2007) Recognition memory and underpinnings of language processing. Proc Natl Acad Sci the medial temporal lobe: a new perspective. Nat Rev Neurosci 113:11366–11371 8:872–883 Plachti A, Eickhoff SB, Hoffstaedter F, Patil KR, Laird AR, Fox PT, Strange BA, Witter MP, Lein ES, Moser EI (2014) Functional organi- Amunts K, Genon S (2019) Multimodal parcellations and exten- zation of the hippocampal longitudinal axis. Nat Rev Neurosci sive behavioral profiling tackling the hippocampus gradient. Cereb 15:655–669 Cortex 29:4595–4612 Suzuki WL, Amaral DG (1994) Perirhinal and parahippocampal cor- Qin S, Duan X, Supekar K, Chen H, Chen T, Menon V (2016) Large- tices of the macaque monkey: cortical afferents. J Comp Neurol scale intrinsic functional network organization along the long 350:497–533 axis of the human medial temporal lobe. Brain Struct Funct Suzuki WA, Amaral DG (2004) Functional neuroanatomy of the medial 221:3237–3258 temporal lobe memory system. Cortex Ranganath C, Blumenfeld RS (2005) Doubts about double dissocia- Van Essen DC, Ugurbil K, Auerbach E, Barch D, Behrens T, Bucholz tions between short-and long-term memory. Trends Cogn Sci R, Chang A, Chen L, Corbetta M, Curtiss SW et al (2012) The 9:374–380 human connectome project: a data acquisition perspective. Neu- Ranganath C, Ritchey M (2012) Two cortical systems for memory- roimage 62:2222–2231 guided behaviour. Nat Rev Neurosci 13:713–726 Vincent JL, Snyder AZ, Fox MD, Shannon BJ, Andrews JR, Raichle Ritchey M, Libby LA, Ranganath C (2015) Cortico-hippocampal sys- ME, Buckner RL (2006) Coherent spontaneous activity identi- tems involved in memory and cognition: the PMAT framework. fies a hippocampal-parietal memory network. J Neurophysiol Progress Brain Res 219:45–64 96:3517–3531 Robinson S, Windischberger C, Rauscher A, Moser E (2004) Opti- Vu AT, Jamison K, Glasser MF, Smith SM, Coalson T, Moeller S, mized 3 t epi of the amygdalae. Neuroimage 22:203–210 Auerbach EJ, Uğurbil K, Yacoub E (2017) Tradeoffs in pushing Rosene DL, Van Hoesen GW (1977) Hippocampal efferents reach the spatial resolution of FMRI for the 7t human connectome pro- widespread areas of cerebral cortex and amygdala in the rhesus ject. Neuroimage 154:23–32 monkey. Science 198:315–317 Wang S-F, Ritchey M, Libby LA, Ranganath C (2016) Functional con- nectivity based parcellation of the human medial temporal lobe. Neurobiol Learn Mem 134:123–134 1 3 1012 Brain Structure and Function (2022) 227:995–1012 Weiskopf N, Hutton C, Josephs O, Deichmann R (2006) Optimal EPI evaluation using ICA and dual regression approach. Neuroimage parameters for reduction of susceptibility-induced bold sensitiv- 49:2163–2177 ity losses: a whole-brain analysis at 3 t and 1.5 t. Neuroimage 33:493–504 Publisher's Note Springer Nature remains neutral with regard to Yeo BT, Krienen FM, Sepulcre J, Sabuncu MR, Lashkari D, Hollin- jurisdictional claims in published maps and institutional affiliations. shead M, Roffman JL, Smoller JW, Zöllei L, Polimeni JR et al (2011) The organization of the human cerebral cortex estimated by intrinsic functional connectivity. J Neurophysiol 106:1125–1165 Zuo X-N, Kelly C, Adelstein JS, Klein DF, Castellanos FX, Milham MP (2010) Reliable intrinsic connectivity networks: test-retest 1 3

Journal

Brain Structure and Function – Springer Journals

Published: Apr 1, 2022

Keywords: Medial temporal lobe; Functional connectivity; Resting-state fMRI; Independent component analysis; Dual regression

There are no references for this article.