Disrupted interhemispheric communication in primary progressive aphasia: a multimodal MRI study of homotopic connectivity Authors: Yuan Tao, Kyrana Tsapkini, Brenda Rapp Published: 2026 ABSTRACT The two hemispheres of the brain are strongly connected both structurally and functionally, and disruptions to interhemispheric communication can have significant cognitive and behavioral consequences. While alterations to interhemispheric structural and/or functional connectivity have been reported in many neurological conditions, this issue has been scarcely addressed in Primary Progressive Aphasia (PPA). PPA is a neurogenerative syndrome primarily involving (left-)lateralized gray matter atrophy and typically categorized into three variants - logopenic, non-fluent, and semantic. Here we examined interhemispheric, homotopic structural and functional connectivity in 72 individuals from the three variants and age-matched controls. Structural connectivity (SC) was calculated from diffusion-weighted MRI and functional connectivity (FC) from resting-state fMRI. Both resulting connectomes were based on the same atlas of homotopic parcels. We asked: 1) Are patterns of abnormal homotopic SC and/or FC shared across PPA variants? 2) How are the FC changes related to underlying structural changes to white and gray matter? We found, first, that average homotopic SC was reduced in all variants and was related to greater behavioral disease severity. By contrast, for homotopic FC, the three variants displayed distinct patterns. Second, direct FC-SC comparisons established that, across variants, the FC and SC changes were distinct from one another and were largely dissociated from gray matter atrophy. In sum, the structural connectivity between the two hemispheres weakens in PPA with disease progression. However, the uncoupling of FC from underlying structural changes indicates that these changes reflect complex patterns of functional re-organization of interhemispheric communication in response to the disease. Introduction Primary Progressive Aphasia (PPA) is a neurogenerative syndrome that is characterized by progressive language deficits and left-lateralized gray matter atrophy. The syndrome is typically categorized into three variants based on language deficit profiles: logopenic (lvPPA), non-fluent (nfvPPA), and semantic (svPPA). In terms of neuropathology, logopenic PPA (lvPPA) typically presents with AD pathology and hence is considered to be atypical or language-variant AD, whereas non-fluent (nfvPPA) and semantic PPA (svPPA) are considered subtypes of Frontotemporal Degeneration (FTD) and are most commonly associated with tau and TDP43 pathologies respectively (Grossman, 2010; Grossman et al., 2023). In terms of the neural distribution of atrophy, lvPPA typically shows atrophy in left temporal-parietal areas, nfvPPA in the left lateral frontal lobe, and svPPA in the left anterior temporal lobe. However, abnormalities in brain connectivity, both functional (signal desynchronization) and structural (reduced white matter integrity) exhibit a distribution that extends beyond areas of gray matter atrophy (Agosta et al., 2012, Agosta et al., 2013; Agosta, Galantucci, et al., 2014; Battistella et al., 2019; Elahi et al., 2017; Englund and Brun, 1987; Guo et al., 2013; Lam et al., 2014; Madhavan et al., 2015; Mahoney et al., 2013; Mandelli et al., 2018; McMillan et al., 2012; Panman et al., 2021; Ranasinghe et al., 2017; Schwindt et al., 2013; Tao et al., 2020; Tetzloff et al., 2018a; 2018b; Zhang et al., 2009). Importantly, to our knowledge, no studies have directly examined the relationship between structural and connectivity changes in PPA. The two hemispheres of the brain are strongly connected both structurally and functionally. Structurally, the two hemispheres are connected through the largest bundle of white matter fibers in the brain, the corpus callosum. The cognitive significance of interhemispheric communication was first discovered in patients who had had full or partial callosotomies and who displayed an array of cognitive changes that were presumably derived from disruption in communication between the two hemispheres (Gazzaniga, 2000, Gazzaniga, 2005). In terms of functional connectivity, homologous regions in the two hemispheres (i.e., homotopic regions) show strong and stable BOLD signal synchronization (Shen et al., 2015). Although the exact mechanisms that drive this functional synchronization between the two hemispheres and its cognitive significance are not yet well understood, loss of homotopic FC (e.g., subsequent to stroke) has been associated with behavioral deficits (Siegel et al., 2016; Tao et al., 2022; Tao and Rapp, 2019) and homotopic FC has been found to play an important role in shaping functional lateralization (Gracia-Tabuenca et al., 2018; Jin et al., 2020; Joliot et al., 2016; Tzourio-Mazoyer et al., 2016). Reduction in interhemispheric structural connectivity (SC) or corpus callosum integrity has been reported in a number of degenerative neurological diseases such as multiple sclerosis (Hawellek et al., 2011), Parkinson's Disease (Agosta et al., 2014a; Agosta et al., 2014b; Agosta et al., 2014c; Goldman et al., 2017), Alzheimer's Disease and related dementias (Caso et al., 2015; Madhavan et al., 2015; Phillips et al., 2024; Teipel, 2003; Teipel et al., 2002; Zhang et al., 2009), and subtypes of Frontotemporal Degeneration (Agosta et al., 2012, Agosta et al., 2013; Elahi et al., 2017; Lam et al., 2014; McMillan et al., 2012; Schwindt et al., 2013; Zhang et al., 2009). Note that, in these cases, white matter damage is generally widespread and not limited to the corpus callosum. In terms of functional connectivity (FC), the picture is more complex in that both disease-related increases and decreases have been reported. For unilateral stroke, it has been proposed that decreased homotopic functional synchronization is a phenotypical feature (Klingbeil et al., 2019; Siegel et al., 2016; Tao and Rapp, 2020), and similar observations have been reported in tumor resection (Almairac et al., 2021; Coget et al., 2018). However, functional changes may be different when the structural damage is gradual, taking place over an extended period of time. In a recent study, Tao et al. (Tao et al., 2022) directly compared stroke-induced aphasia and non-fluent primary progressive aphasia (nfvPPA) - two groups with similar left-hemisphere damage and language deficit profiles but different temporal courses of neurological changes. Tao et al. found that, in contrast to the stroke group, for the nfvPPA group, functional connectivity between the two hemispheres was elevated compared to age-matched healthy controls. In other words, the two hemispheres became more synchronized in non-fluent PPA than in neurotypical controls while in stroke the opposite pattern was observed, despite their similarities in brain damage location and language deficits. These findings underscore the complexity of the relationship between interhemispheric FC and SC. In fact, although homotopic FC is supported by the underlying structural connectivity between the hemispheres (Johnston et al., 2008; O'Reilly et al., 2013; Quigley et al., 2003; Roland et al., 2017), synchronization of functional signals in homotopic brain regions can persist even when the callosal structural underpinning is absent (as in corpus callosum agenesis, (Tyszka et al., 2011; Quigley et al., 2003)) or severed (as in callosotomy, (O'Reilly et al., 2013; Roland et al., 2017; Uddin et al., 2008)). These observations indicate that the functional synchronization of the hemispheres can be driven by mechanisms other than structural connectivity alone. In cases in which there is gradual degeneration of the corpus callosum, such as in neurodegenerative diseases and normal aging, increased homotopic FC has sometimes been reported. For instance, in normal aging, whereas corpus callosum volume declines after middle age (Danielsen et al., 2020), homotopic FC may increase (Zuo et al., 2010), again highlighting the dissociation between homotopic structural and functional connectivity. In terms of PPA, research has been almost exclusively focused on left hemisphere changes while the right hemisphere and the relationship between the two hemispheres have been less studied. To our knowledge, (Tao et al., 2022) is the only study reporting abnormal interhemispheric (specifically homotopic) functional connectivity in PPA (non-fluent variant). In terms of interhemispheric white matter connectivity, reduction has been reported for all PPA variants (Agosta et al., 2012, Agosta et al., 2013; Caso et al., 2015; Elahi et al., 2017; Grossman et al., 2013; Lam et al., 2014; Madhavan et al., 2015; McMillan et al., 2012; Phillips et al., 2024; Schwindt et al., 2013; Zhang et al., 2009). In PPA, white matter damage has been attributed to various disease mechanisms, including axonal degeneration associated with neuronal cell body death. However, when white matter damage occurs in areas without notable gray matter atrophy, it has been attributed to the accumulation of pathological proteins in the white matter. For instance, (Agosta et al., 2012) examined gray and white matter integrity across the brain in all three PPA variants, and found that all PPA variants exhibited widespread bilateral white matter damage that included the corpus callosum. They found that white matter damage was also present beyond areas of atrophic gray matter and that the degree of white matter damage largely did not correlate with gray matter volume, supporting the hypothesis of neuropathological changes affecting the white matter tracts themselves (Agosta et al., 2014b; Neumann et al., 2007; Schofield et al., 2003). In another study that examined whole-brain structural connectivity in AD and FTD subtypes that included PPA, McMillan et al. (McMillan et al., 2012) found that while both groups showed widespread damage, the white matter damage was overall more severe for the FTD group than for the AD group (especially in the anterior corpus callosum), despite more widespread gray matter damage for the AD group. The authors argued that it might be due to the fact that FTD's underlying pathology (specifically FTD-tau and TDP43) might trigger immune responses that could have more detrimental consequences to white matter than would AD pathology (Agosta et al., 2014b, Schofield et al., 2003). However, those studies only examined white matter structural connectivity and whether and how these changes affect functional connectivity is unknown. Current study In the current study, we examined interhemispheric structural and functional connectivity and the relationship between them in Primary Progressive Aphasia (PPA) with diffusion MRI and resting-state functional MRI across the whole-brain. For constructing both the structural and the functional connectomes we used a parcellation scheme that was specifically developed for evaluating homotopic connectivity which consisted of 200 pairs of homotopic cortical parcels (Yan et al., 2023). Structural connectivity between each homotopic pair was calculated as the average fractional anisotropy value (FA) of all the white matter streamlines connecting the pair of parcels, whereas functional connectivity was calculated as the temporal synchronization of the resting-state fMRI time-series of the two parcels. We carried out three sets of analyses: 1) We first characterized both the structural and functional connectivity abnormalities of each PPA variant by comparing them to age-matched healthy controls; 2) We then evaluated the relationship between abnormalities in structural and functional connectivity for each PPA variant, examining one-to-one structural-functional correspondences between each homotopic pair; and 3) To determine if the connectivity changes could be due to gray matter damage, we examined the relationship between the degree of structural and functional connectivity abnormality and the integrity of the gray matter at the end-points of each homotopic connection, measured by high-resolution T1 structural imaging. This investigation allows us to address the following key questions: 1) Are patterns of abnormal homotopic structural and/or functional connectivity shared across all primary progressive aphasia (PPA) variants? 2) How are functional changes in interhemispheric synchronization and communication related to underlying structural changes to white and gray matter? Methods and results Participants Seventy-two participants diagnosed with PPA were recruited as part of clinical trial studies (NCT02606422, NCT04122001, NCT03887481) from 2012 to 2022. In the current investigation, their baseline, pre-treatment MRI scans were analyzed. The cohort included 29 logopenic (lvPPA), 30 non-fluent (nfvPPA), and 13 semantic variant (svPPA) individuals. Differential diagnosis was based on clinical assessment, neuropsychological and language testing using the Gorno-Tempini et al. (2011) consensus criteria. Demographic information and standard testing scores of each PPA variant are reported in Table 1 and Table S1. The variants did not differ in age, years of education, sex, time since onset, and FTLD-CDR scores. All participants were administered high-resolution T1 structural imaging, resting-state fMRI, and all but one received diffusion-weighted imaging that was used to evaluate structural connectivity (one participant was not included in analyses involving structural connectivity). MRI data were also obtained from 18 healthy controls (mean age = 65.4, SD = 4.4, range 59–75, 11 females) that were matched with the patients in age, years of education, and sex, who were evaluated in the context of the clinical trials. All participants were native English speakers and right-handed. The study was approved by the Johns Hopkins Hospital and Johns Hopkins University Institutional Review Board. All participants provided informed consent. We note that the functional homotopic connectivity findings for the nfvPPA sample were included in a previous report comparing PPA and stroke (Tao et al., 2022). Importantly, however, the previous study did not examine SC, nor the relationship between SC and FC for this sample.Table 1Demographic characteristics of the healthy controls (HC) and the three PPA variants. The group mean and standard deviation values are shown. Controls and PPA participants were matched in age, gender, and years of education (age: p = 0.31, gender: p = 0.85, education: p = 0.16).HC (N = 18)lvPPA (N = 29)nfvPPA (N = 30)svPPA (N = 13)Age65.4 (4.4)65.9 (9.0)68.8 (6.2)67.3 (7.2)Gender (N of female)1117138Education (years)17.4 (3.2)16 (1.78)16 (3.59)16 (2.64)Time since onset (years)NA3.3 (2.3)3.7 (2.4)4.7 (2.7)FTLD-CDR sum of boxes (scale 0–18)NA5.7 (3.9)4.9 (3.8)7.7 (5.0)FTLD-CDR Language (scale 0–3)NA1.6 (0.8)1.7 (0.7)2.0 (0.6)FTLD-CDR: Frontotemporal Lobar Degeneration Clinical Dementia Rating (Knopman et al., 2008). Table 1Demographic characteristics of the healthy controls (HC) and the three PPA variants. The group mean and standard deviation values are shown. Controls and PPA participants were matched in age, gender, and years of education (age: p = 0.31, gender: p = 0.85, education: p = 0.16).HC (N = 18)lvPPA (N = 29)nfvPPA (N = 30)svPPA (N = 13)Age65.4 (4.4)65.9 (9.0)68.8 (6.2)67.3 (7.2)Gender (N of female)1117138Education (years)17.4 (3.2)16 (1.78)16 (3.59)16 (2.64)Time since onset (years)NA3.3 (2.3)3.7 (2.4)4.7 (2.7)FTLD-CDR sum of boxes (scale 0–18)NA5.7 (3.9)4.9 (3.8)7.7 (5.0)FTLD-CDR Language (scale 0–3)NA1.6 (0.8)1.7 (0.7)2.0 (0.6)FTLD-CDR: Frontotemporal Lobar Degeneration Clinical Dementia Rating (Knopman et al., 2008). Demographic characteristics of the healthy controls (HC) and the three PPA variants. The group mean and standard deviation values are shown. Controls and PPA participants were matched in age, gender, and years of education (age: p = 0.31, gender: p = 0.85, education: p = 0.16). FTLD-CDR: Frontotemporal Lobar Degeneration Clinical Dementia Rating (Knopman et al., 2008). MRI data acquisition MRI data were collected at the F.M. Kirby Research Center for Functional Brain Imaging (Baltimore, MD, USA) with a Phillips 3 T scanner. The T1-weighted structural MRI acquisition parameters were as follows: TR = 8.1 ms, TE = 3.7 ms, FOV = 224 × 160 × 180 mm (ap, fh, rl), flip angle = 8°, voxel dimension = 1 × 1 × 1 mm, data matrix = 224 × 224 × 160. The diffusion-weighted imaging (DWI) scans were acquired with the following parameters: TR/TE = 7.324/75 ms, SENSE factor = 2.5, FOV = 212 × 212, matrix size = 96 × 96 reconstructed to 256 × 256. The DWI scans had 32 directions with b = 700 s/mm2, and one volume with no diffusion weighting (b = 0 s/mm2). For the resting-state fMRI (rs-fMRI) scans, because of some protocol changes during the long recruitment period, there was some variability in the scan durations: for the majority of the participants (n = 44) scan duration was 8.75 min (n = 45, 63%), for 12 participants (18.%) scan duration was 6.5 min and for 14 participants (19%) it was 11.875 min. The acquisition parameters of the resting-state fMRI were as follows: TR = 2500 ms, TE = 30 ms, FOV = 240 × 141 × 240 mm (ap, fh, rl), flip angle = 75°, voxel dimension = 3 × 3 × 3 mm, data matrix = 80 × 80 × 47. All healthy controls were scanned with the same scanning parameters in the same scanner as the PPA participants. For the resting-state fMRI (rs-fMRI) scans, as for the PPA participants, the majority of the control participants' scans lasted 8.75 min (n = 11, 61%), while 1 (5%) lasted 6.5 min, 3 (17%) lasted 11.875 min, and 3 (17%) lasted 7.5 min. Note that because we applied scrubbing (see Section 3.3.2 for data preprocessing), the final numbers of volumes of each participant differed from each other. Hence, we chose to use all the available volumes for each participant despite the different durations. Neuroimaging data processing Atlas Given the goal of investigating interhemispheric connectivity in PPA, to construct both the functional and structural connectivity matrices we used an atlas that optimized homotopic parcels (Yan et al., 2023). The atlas contained 200 pairs of homotopic parcels, and, thus, 200 homotopic connections. However, given that for the tractography methods we applied, structural homotopic connectivity can be quite sparse for certain areas, we ended up limiting our analyses to the 63 homotopic connections for which at least a quarter of the control participants had non-zero streamlines (see Section 3.3.3 for details). Resting-state data preprocessing and functional connectivity (FC) calculation The resting-state fMRI data were first preprocessed with fmriprep 1.4.1. with its default parameters (https://fmriprep.org/en/1.4.1/workflows.html#bold-preprocessing). The preprocessed time-series were then used to construct the functional connectivity matrices with the Python toolbox nilearn (https://nilearn.github.io/stable/index.html). Specifically, the following confounds calculated by fmriprep were regressed out from the time-series: average timeseries of the white matter and CSF and their derivatives; 12 motion parameters; the top three temporal principal components; and discrete cosine-basis regressors served as high-pass filter. A low-pass filter of 0.8 Hz was also applied. Scrubbing was carried out based on FD and DVARS such that TRs of FD > 0.5 or DVARS>1.5, plus any TRs that were not part of a series of five consecutive TRs were scrubbed, as were first 3 volumes of the scan. Note that the final number of TRs after scrubbing did not differ between groups (ps > 0.22, see Supplementary Material 1). Functional connectivity was calculated as the pairwise Pearson correlation matrix using the averaged time-series of the atlas parcels (see 3.3.1), with Fisher's z transformation. Diffusion data preprocessing and structural connectivity (SC) calculation Diffusion image preprocessing was carried out with ExploreDTI (https://www.exploredti.com), a Matlab-based toolbox. Preprocessing included correcting for Gibbs ringing artefacts, signal drift correction, eddy currents, head motion, and EPI distortions. Structural connectivity construction was performed with MRtrix3 (Tournier et al., 2019). A whole-brain streamline tractography was constructed with the deterministic algorithm SD_STREAM combined with Anatomically-Constrained-Tractography (ACT) with backtracking (Tournier et al., 2012). The resulting streamlines were optimized for streamline density using SIFT2 (Smith et al., 2015b). The T2-weighted image was linearly co-registered to the T1-weighted structural image, and the atlas was reversely co-registered to the structural image to construct the structural connectivity (SC) matrix. Specifically, the SC matrix was weighted by the FA values such that the connectivity strength of a structural connection corresponded to the mean FA of all the streamlines for each parcel, weighted by the mean FA of the vertices that compose of each streamline (Smith et al., 2015a). We limited our analyses of structural connections to those for which at least a quarter of the control group (i.e., 5 participants) had non-zero values. This resulted in 63, out of a possible 200, homotopic connections which were mainly located in the frontal and the parietal lobes. The 63 ROIs and their geometric centers are listed in Table S2. To ensure the quality of the structural connectome, we randomly selected 5 connections from the 63 in five randomly selected subjects from each group to visually inspect the connecting streamlines to make sure that there were no obvious errors. Those random samples are shown in Supplementary Fig. 1. Gray matter volume Gray matter atrophy was identified with voxel-based morphometry (VBM) analysis using FSL-VBM (Smith et al., 2004). The T1-weighted MPRAGE images first underwent skull-stripping and gray matter segmentation, then were co-registered to the MNI space. Consistent with FSL recommendations, a study-specific template was created by averaging images from an equal number of participants from the control group and each PPA variant. Then all gray matter images were non-linearly co-registered to this study-specific template and modulated for local expansion and contraction. Finally, the co-registered images were smoothed with an isotropic Gaussian kernel with a 3 mm sigma before entered into group statistics. Each PPA variant was compared to the control group using the permutation-based method implemented in FSL-VBM, multiple comparisons correction was conducted using the threshold-free cluster enhance (TFCE) method (Smith and Nichols, 2009). To examine the relationship between gray matter integrity and the structural and functional connectivity strengths (see below Section 3.4.3), we extracted the gray matter volume values for each of the same parcels using the co-registered, modulated, and smoothed gray matter images produced by the VBM analysis described above. Statistical analyses Analysis 1: characterizing abnormalities in structural and functional homotopic connectivity Analysis 1 A: Quantifying homotopic SC/FC abnormalities in each PPA variant. We compared each PPA variant to the control group in terms of both their structural (SC) and functional (FC) homotopic connectivity via linear regression modeling. Specifically, we constructed two models, one for SC and one for FC. For each model, the dependent variable was the global average SC or FC value for each participant, averaging across the 63 connections for which at least a quarter of the control group had non-zero values, see Sec 3.3.3). The independent variable of interest was the 4-level categorical variable Group (controls and the three variants). For the FC analysis, in-scanner motion (averaged relative root-mean-squared) was included as a covariate. Analysis 1B: Examining the relationship between homotopic SC/FC and clinical severity. We examined the relationship between homotopic SC and FC values and behavioral disease severity as indexed by the Frontotemporal Lobar Degeneration Clinical Dementia Rating (FTLD-CDR) score, sum of boxes (hereafter referred to as CDR, (Knopman et al., 2008)). Specifically, we evaluated two linear regression models in which the dependent variable was each participant's whole-brain average homotopic SC or FC and predictors were the CDR score, age, and in-scanner motion for FC. Analysis 1C: Examining the spatial distribution of homotopic SC/FC abnormalities across the PPA variants. To evaluate the spatial distributions of the SC and FC abnormalities across the PPA variants, we first carried out the same regression modeling as in Analysis 1 A that compared each PPA variant to the controls, but for each of the 63 connections for SC and FC. This yielded a t-value for each of the SC and FC connections for each of the three variants that quantified the SC/FC abnormality of each PPA variant. Those t-values were used in subsequent analyses (Analyses 2 and 3). Here, in order to examine whether the three PPA variants exhibited similar or different SC/FC abnormality patterns, we grouped the resulting t-values into 7 resting-state-networks (RSNs, see Table S2) as proposed in the same atlas that provided the homotopic parcels (Yan et al., 2023): dorsal attention network (DAN), sensorimotor network (SMN), visual network (VIS), ventral attention network (VAN), default mode network (DMN), limbic network, and frontoparietal network (FPN). For each network, t-values (quantifying degree of SC and FC abnormality) for the three variants were compared to one another using one-way-ANOVA, FDR correction for 7 models was applied (alpha<0.05). Analysis 2: evaluating the correspondence between structural and functional homotopic connectivity abnormalities We examined two aspects of the relationship between structural and functional homotopic connectivity abnormalities: their magnitudes and spatial distributions. Note that these analyses examine abnormality (i.e., the difference between PPA and healthy controls) using the t-values from the models described in Section 3.4.1, Analysis 1C. These t-values quantified the extent to which a PPA group differed from controls (as opposed to raw FC and SC connectivity values). In other words, we aimed to examine the correspondence between structural and functional abnormality, rather than the basic correspondence between structural and functional connectivity strengths. Analysis 2 A: Comparing the magnitudes of SC and FC abnormalities. For each variant, we compared the distributions of the t-values that indexed the magnitude of structural and functional connectivity abnormalities of the 63 connections (from Analysis 1C), using the Kolmogorov-Smirnov test, a non-parametric test that evaluates if two empirical distributions are drawn from the same distribution. The results were FDR corrected. Analysis 2B: Comparing the spatial distributions of SC and FC abnormalities. To evaluate if there were differences in the spatial distribution of FC and SC abnormalities for each variant, we examined the similarity of the spatial distributions across the brain for the structural and functional connections by calculating the Pearson correlation between the t-values (from Analysis 1C) of the corresponding structural and functional connections. P-values for these three correlations were determined using 1000 permutation tests that involved shuffling the t-values while maintaining the same degree of spatial autocorrelation using BrainSMASH (Burt et al., 2020). Results were FDR corrected. Analysis 3: examining the relationship between structural and functional homotopic connectivity abnormalities and gray matter atrophy To examine the relationship between SC and FC abnormalities and gray matter atrophy, we first calculated the degree of gray matter atrophy for each parcel by comparing the average parcel-wise gray matter volume of each PPA variant to the control group values. This produced a t-value for each parcel of each PPA variant that quantified the degree of gray matter atrophy (GM). Note that because a homotopic connection has two endpoints (one in each hemisphere), there were two GM values associated with each homotopic connection. We conducted linear regression analyses to quantify the amount of variance in the SC/FC abnormalities that could be explained by the GM changes: We examined two models for each variant, one for SC and one for FC, using the same SC/FC connection-wise t-values described above (from Analysis 1C) as the dependent variables and the GM t-values as an independent variable. The amount of variance explained was indexed by the r-squared value of each model. To specifically evaluate if gray matter atrophy in the left and the right hemisphere contributed differently to SC/FC, we carried out an additional set of identical analyses that included the independent variable hemisphere and the interaction between GM and hemisphere. We then evaluated whether the overall association (across the three variants) between GM and SC was statistically different from the assoication between GM and FC using permutation test. Note that we combined all three variants here because they exhibited similar FC vs. SC patterns (see Section 4.4). To do so, we did the following: (1) We used the regression model structure described above, but included all three variants in each SC and FC model and calculated the “observed R-squared difference” between the two models; (2) Swapped the SC and FC values per participant for a random subset of participants, then evaluated the same regression models and calculated the differences between “pseudo R-squared values”; (3) Repeated step 2 for 10 K times to build the null-distribution of “pseudo R-squared difference values”; (4) The “observed R-squared difference” value from step 1 was compared against the 10 K null-distribution to obtain the p-value indicating if GM explained different amounts of variance in SC and FC. Results Participants The three PPA variants exhibited distributions of gray matter atrophy that are typically associated with each of the variants (Fig. 1): Logopenic PPA (lvPPA) showed significant gray matter reduction in the left temporal and parietal lobes, atrophy of the non-fluent PPA (nfvPPA) group was mainly found in the left frontal lobe, and semantic PPA (svPPA) showed left-lateralized temporal lobe atrophy. The PPA participants were also evaluated with a comprehensive battery of cognitive and language tests. The results showed the typical behavioral patterns that have been associated with each variant (Table S1).Fig. 1Gray matter atrophy of each PPA variant calculated with voxel-based morphometry (VBM). Top to bottom: logopenic (lvPPA), non-fluent (nfvPPA), semantic (svPPA).Fig. 1 Gray matter atrophy of each PPA variant calculated with voxel-based morphometry (VBM). Top to bottom: logopenic (lvPPA), non-fluent (nfvPPA), semantic (svPPA). Analysis 1: abnormalities in structural and functional homotopic connectivity in PPA Analysis 1 A. Global abnormality. In terms of brain-wide average structural homotopic connectivity (SC), all three variants showed significantly lower FA values than controls (Fig. 2a. lv: p = 0.005, nfv: p = 0.0007, sv: p = 0.059). There were no differences between the PPA variants (lv-nfv: p = 0.472, lv-sv: p = 0.601, nfv-sv: p = 0.278). In terms of brain-wide average functional homotopic connectivity (FC), the non-fluent variant showed significantly higher average values than the controls (p = 0.045) while the average values for the other two variants did not differ from those of controls (lv: p = 0.752, sv: p = 0.811, Fig. 2b). The nfvPPA values were significantly higher than those of the lvPPA group, while neither of the other two variant comparisons were statistically significant (lv-nfv: p = 0.008, lv-sv: p = 0.587, nfv-sv: p = 0.118).Fig. 2Average homotopic structural (a) and functional connectivity (b). All PPA variants showed lower structural connectivity (SC) than controls, and the non-fluent variant PPA showed higher functional connectivity (FC) than the controls. The error bars depict 95% confidence interval. ∼p < 0.1, *p < 0.05, **:p < 0.01,Fig. 2 Average homotopic structural (a) and functional connectivity (b). All PPA variants showed lower structural connectivity (SC) than controls, and the non-fluent variant PPA showed higher functional connectivity (FC) than the controls. The error bars depict 95% confidence interval. ∼p < 0.1, *p < 0.05, **:p < 0.01, Analysis 1B: Relationship between homotopic SC, FC and clinical severity. In terms of the relationship of FC and SC abnormalities with disease severity, regression analysis showed that lower global homotopic SC was related to greater disease severity, as measured behaviorally, i.e., higher CDR scores (p = 0.0204, Fig. 3), as well as to older age (p = 0.0007). We also tested if there was an interaction between PPA variant and CDR and found no significant interaction (p = 0.87). Since both CDR and age were included in the regression model, the results indicate that the two factors independently contributed to lower homotopic SC. In contrast, global homotopic FC was not significantly related to either CDR (p = 0.95, Fig. 3) nor age (p = 0.48). There was also no significant variant by CDR interaction (p = 0.50).Fig. 3Relationship between average homotopic structural (SC,a) and functional connectivity (FC,b) with disease severity measured by the frontotemporal lobar degeneration clinical dementia rating score (CDR, sum of boxes). Lower homotopic SC was significantly associated with greater disease severity (i.e., higher CDR scores). No effects were found for homotopic FC.Fig. 3 Relationship between average homotopic structural (SC,a) and functional connectivity (FC,b) with disease severity measured by the frontotemporal lobar degeneration clinical dementia rating score (CDR, sum of boxes). Lower homotopic SC was significantly associated with greater disease severity (i.e., higher CDR scores). No effects were found for homotopic FC. Analysis 1C: Spatial distribution of homotopic SC and FC abnormalities across the PPA variants. We also examined the spatial distributions of homotopic SC and FC abnormality by comparing each PPA variant to the controls at each homotopic connection, with the distributions of the t-values of those comparisons shown in Fig. 4a. As already seen with the whole-brain average results, all three variants exhibited widespread reduction of homotopic SC. However, the effects of homotopic FC differed across the variants such that nfvPPA showed widespread increases relative to controls, whereas the other two variants displayed both increases and decreases (Fig. 4a). To statistically compare these effects across variants, we grouped the t-values into 7 resting-state networks (Yan et al., 2023, Table S2) and compared the distributions of the t-values across the variants for each network (Fig. 4b). For SC, none of the networks exhibited significant variant difference whereas for FC, 5 out of the 7 networks did (FDR corrected ps < 0.05). These results confirm that the three PPA variants showed similar distributions of homotopic SC reduction but largely differed in terms of the distribution of homotopic FC changes. Similar effects were observed when we grouped the t-values by lobes (Supplementary Materials 2).Fig. 4Spatial distribution of the homotopic structural (SC) and functional connectivity (FC) abnormalities. a) Connection-wise t-values for each PPA variant compared to the healthy control group (HC) for each of the 63 homotopic connections. Negative t-values (blue) indicate PPA values were lower than HC and positive t-values (red) indicate that PPA values were higher than HC. Note that the aim of this figure was to display the whole brain patterns and hence no thresholding was applied and all connections that were analyzed are visualized. b) t-values in (a) grouped by resting-state network defined in Yan et al. (2023), note that the dashed lines indicate the zero t-value (i.e., level of the control group). Top: Homotopic SC; Bottom: Homotopic FC. Distributions for the three variants were compared with one-way ANOVA. *:p < 0.05 after FDR correction, DAN: dorsal attention network; SMN: Sensorimotor network; VIS: visual network; VAN: ventral attention network; DMN: default mode network; LIMB: limbic network; FPN: frontoparietal network.Fig. 4 Spatial distribution of the homotopic structural (SC) and functional connectivity (FC) abnormalities. a) Connection-wise t-values for each PPA variant compared to the healthy control group (HC) for each of the 63 homotopic connections. Negative t-values (blue) indicate PPA values were lower than HC and positive t-values (red) indicate that PPA values were higher than HC. Note that the aim of this figure was to display the whole brain patterns and hence no thresholding was applied and all connections that were analyzed are visualized. b) t-values in (a) grouped by resting-state network defined in Yan et al. (2023), note that the dashed lines indicate the zero t-value (i.e., level of the control group). Top: Homotopic SC; Bottom: Homotopic FC. Distributions for the three variants were compared with one-way ANOVA. *:p < 0.05 after FDR correction, DAN: dorsal attention network; SMN: Sensorimotor network; VIS: visual network; VAN: ventral attention network; DMN: default mode network; LIMB: limbic network; FPN: frontoparietal network. Analysis 2: evaluating the correspondence between homotopic SC and FC abnormalities Analysis 2 A: Comparing the magnitudes of SC and FC abnormalities. Within each variant, visual inspection indicated that the effects of homotopic SC and FC differed from each other in terms of their polarity (i.e., being higher or lower than controls). In fact, statistically, for all three PPA variants, the t-value distributions indexing SC and FC abnormality were significantly different from one another (Fig. 5, top, FDR corrected, alpha<0.05), such that the SC t-values were more negative (lower than controls) than FC t-values.Fig. 5Correspondence between the distributions of homotopic structural (SC) and functional connectivity (FC) abnormalities for each PPA variant. Abnormality is quantified as t-values from the comparison between the PPA variant and healthy controls for each of the 63 homotopic connections, shown in the brain plot in Fig. 4a. Left to right: logopenic (lvPPA), non-fluent (nfvPPA), semantic (svPPA). Top: Histograms of the SC and FC t-values. All variants exhibited significantly different distributions of t-values (indexing degree of abnormality) for SC and FC, such that the t-values for SC were more negative than the t-values for FC. Bottom: Pearson correlations between SC and FC t-values. Each data-point in the scatterplots indicates one homotopic connection, with its FC t-value on the x-axis and its SC t-value on the y-axis. All variants exhibited low correlations, indicating little correspondence between the homotopic connections' degree of structural and functional abnormality.Fig. 5 Correspondence between the distributions of homotopic structural (SC) and functional connectivity (FC) abnormalities for each PPA variant. Abnormality is quantified as t-values from the comparison between the PPA variant and healthy controls for each of the 63 homotopic connections, shown in the brain plot in Fig. 4a. Left to right: logopenic (lvPPA), non-fluent (nfvPPA), semantic (svPPA). Top: Histograms of the SC and FC t-values. All variants exhibited significantly different distributions of t-values (indexing degree of abnormality) for SC and FC, such that the t-values for SC were more negative than the t-values for FC. Bottom: Pearson correlations between SC and FC t-values. Each data-point in the scatterplots indicates one homotopic connection, with its FC t-value on the x-axis and its SC t-value on the y-axis. All variants exhibited low correlations, indicating little correspondence between the homotopic connections' degree of structural and functional abnormality. Analysis 2B: Comparing the spatial distributions of SC and FC abnormalities. In terms of the spatial correspondence between the homotopic SC and FC changes, the correlations of t-values of corresponding connections were low and non-significant for all variants (Fig. 5, bottom. lv: r = 0.15; nfv: r = 0.07; sv: r = 0.27; n.s. after FDR correction), indicating that, for homotopic connections, lower SC integrity did not necessarily lead to abnormal FC. In other words, there was no reliable relationship between the integrity of the structural connectivity of a connection between two homologous areas and the degree of the functional synchronization between the same two areas. Analysis 3: examining the relationship between homotopic functional and structural connectivity and gray matter atrophy We evaluated whether the magnitude of structural (SC) and functional connectivity (FC) abnormalities could be explained by the magnitude of gray matter (GM) atrophy at their end-point regions in either hemisphere. Table 2 reports the amount of variance in SC and FC explained by GM atrophy for each of the three variants (i.e., model r-squared values). As can be seen, GM only explained a small fraction of the variability, ranging from 1.32% to 10.29%. After correction for multiple comparisons, only the SC models of the non-fluent and the semantic variant were statistically significant (FDR corrected, alpha<0.05). Furthermore, the relationships between SC/FC and GM did not differ between hemispheres (ps > 0.05, FDR corrected).Table 2Percentage of variance in homotopic structural (SC) and functional connectivity (FC) changes explained by gray matter (GM) atrophy for each of the variants. The r-squared value of each model is reported. Analysis results (see text for details) showed that while the individual effect sizes were very small, the overall difference between amount of variance explained by GM atrophy was statistically higher for SC than for FC (p = 0.02). *:p < 0.05, FDR corrected.SCFClvPPA3.28%1.32%nfvPPA7.55%*3.39%svPPA10.29%*4.33% Table 2Percentage of variance in homotopic structural (SC) and functional connectivity (FC) changes explained by gray matter (GM) atrophy for each of the variants. The r-squared value of each model is reported. Analysis results (see text for details) showed that while the individual effect sizes were very small, the overall difference between amount of variance explained by GM atrophy was statistically higher for SC than for FC (p = 0.02). *:p < 0.05, FDR corrected.SCFClvPPA3.28%1.32%nfvPPA7.55%*3.39%svPPA10.29%*4.33% Percentage of variance in homotopic structural (SC) and functional connectivity (FC) changes explained by gray matter (GM) atrophy for each of the variants. The r-squared value of each model is reported. Analysis results (see text for details) showed that while the individual effect sizes were very small, the overall difference between amount of variance explained by GM atrophy was statistically higher for SC than for FC (p = 0.02). *:p < 0.05, FDR corrected. As can be seen in Table 2, for all PPA variants, the values of SC and FC variance explained by GM atrophy were higher for SC than FC, suggesting a closer relationship between GM atrophy and SC abnormalities than for FC. In fact, the statistical permutation-based analysis confirmed that amount of variance that GM atrophy explained for SC damage was significantly higher than for FC (p = 0.02). Overall, the results of this analysis indicate that gray matter atrophy was not the primary driver of the observed disease-related SC and FC connectivity changes, and especially not for FC. Discussion In this study, we examined interhemispheric, homotopic structural and functional connectivity in three primary progressive aphasia (PPA) variants. We found: (1) Homotopic structural connectivity (SC) was reduced in all three variants (Fig. 2a), and, lower SC was related to older age and greater disease severity as measured behaviorally (FTLD-CDR, Fig. 3a); (2) With regard to homotopic functional connectivity (FC), only the non-fluent variant showed significantly elevated brain-wide average values relative to healthy age-matched individuals whereas, for the other variants, average FC values did not differ from controls (Fig. 2b); (3) For each variant, the FC changes did not correspond to the SC changes (Fig. 4, Fig. 5); and (4) The FC and SC changes could not explained by gray matter atrophy at the end-points of the homotopic connections (Table 2). In sum, the results indicate that, in PPA, the structural connectivity between the two hemispheres becomes weaker as the disease progresses. However, the functional communication between the hemispheres does not seem to be closely coupled with the structural properties of underlying gray or white matter and may, therefore, reflect re-organization of interhemispheric functional communication and synchronicity that occurs in response to the disease. Dissociation between structural and functional connectivity In the current study, for the three main primary progressive aphasia (PPA) variants we found a significant dissociation between the patterns of their homotopic structural (SC) and functional connectivity (FC) abnormalities. With regard to SC, all three variants showed reduced homotopic SC compared to the age-matched control group (Fig. 2a). Moreover, lower homotopic SC was linked to greater behavioral disease severity across all variants (Fig. 3). However, despite largely similar homotopic SC changes (except for the dorsal attention network, Fig. 4b), the three variants exhibited different homotopic FC changes (Fig. 2 and Fig. 4). Similarly, within each variant, the homotopic FC abnormalities were statistically different from their homotopic SC abnormalities in terms of both magnitude and distribution (Fig. 5). As discussed in the Introduction, the relationship between interhemispheric SC and FC is complex such that loss of SC (surgically or developmentally) does not necessary result in loss of FC (Tyszka et al., 2011; O'Reilly et al., 2013; Quigley et al., 2003; Roland et al., 2017; Uddin et al., 2008) These observations echo the general finding in healthy normal brains that the overall correspondence between SC and FC is only moderate (e.g., r ∼ =0.5), meaning that there is still considerable amount of FC that is not directly driven by SC (Hagmann et al., 2008; Honey et al., 2007; Shen et al., 2012, Shen et al., 2015; Zimmermann et al., 2016). Interhemispheric SC-FC dissociation has also been observed in healthy aging where interhemispheric SC declines after late middle age (Danielsen et al., 2020) but interhemispheric FC instead increases at older ages (Zuo et al., 2010). In a study that examined both interhemispheric SC and FC within a group of healthy older individuals, the authors found that age-related homotopic SC decrease was actually related to higher homotopic FC (Avelar-Pereira et al., 2020). The SC-FC dissociation observed here in the interhemispheric connectivity for all three PPA variants (see Fig. 5) provides further evidence of SC-FC dissociation, indicating that functional connectivity can persist or re-organize even in the face of irreversible structural damage. However, the nature of the mechanisms that drive the different FC re-organization patterns remains an important topic for future research. Dissociation between connectivity changes and GM damage PPA is typically considered a disease that primarily affects left hemisphere cortical gray matter. However, white matter changes in PPA have also been reported. In the left hemisphere, although the overall distribution of white matter damage generally corresponds to the distribution of gray matter atrophy, many studies have also identified considerable white matter damage beyond the areas of gray matter atrophy in all three PPA variants (Agosta et al., 2012, Agosta et al., 2013; Elahi et al., 2017; Englund and Brun, 1987; Lam et al., 2014; Madhavan et al., 2015; Mahoney et al., 2013; McMillan et al., 2012; Panman et al., 2021; Schwindt et al., 2013; Tetzloff et al., 2018; Zhang et al., 2009). In the current study, we too found that the decrease of white matter, structural connectivity (SC) between homotopic regions only showed a weak correspondence with gray matter atrophy at the end-point regions, with no more than 10% of the variance in white matter (FA) abnormalities explained by gray matter atrophy (Table 2). Specifically, the strongest effect was seen in the semantic variant, which could be due to their relatively more severe and generally greater degree of bilateral atrophy. Although some of the white-matter damage may be due to Wallerian degeneration secondary to cell body death, current evidence suggests that neuropathological deposits (e.g., of tau and TDP43) may cause primary axonal degeneration that leads to the white matter damage seen outside the atrophic regions (Agosta et al., 2012, Agosta et al., 2014b, McMillan et al., 2012; Neumann et al., 2007; Panman et al., 2021; Phillips et al., 2024; Schofield et al., 2003; Vogel et al., 2023). For instance, Panman et al. (Panman et al., 2021) modelled the chronological sequence of pathological events in FTD participants, including nfvPPA, and found that serum neurofilament light chain (NFL), which indicates axonal degeneration, and DTI-based white matter tract damage were biomarkers that preceded gray matter atrophy. Similarly, the distribution of homotopic functional connectivity (FC) abnormalities that we observed also did not correspond to the distribution of gray matter atrophy, explaining less than 5% of the variance across the PPA variants (Table 2). The general dissociation is consistent with previous findings that functional changes in PPA can be seen beyond atrophic regions and do not necessarily correlate with the magnitude of gray matter atrophy (Agosta et al., 2014c; Battistella et al., 2019; Guo et al., 2013; Mandelli et al., 2018; Ranasinghe et al., 2017; Tao et al., 2020). The underlying mechanisms leading to this dissociation between FC changes and gray matter integrity are not yet clear. One possible cause of the functional changes in non-atrophic regions may be biological changes that we are not measuring or detecting, such as metabolic changes or pathological aggregation at the cellular/sub-cellular level. Another possibility is that atrophy and functional changes elsewhere in the brain may cause “downstream” functional changes in the non-atrophic regions (Cope et al., 2018; Jones et al., 2016; Tao et al., 2020). It is worth noting that the FC changes did show greater dissociation from gray matter atrophy than did the SC changes (see Table 2), consistent with the hypothesis that FC changes largely reflect re-organization of the brain's functional communications and hence are even less likely to follow the pattern of gray matter atrophy than SC changes. In sum, what seems clear is that the functional changes that are triggered by the underlying structural changes are dynamic and complex and are not easily traced back to specific atrophy loci. Similarities and differences in interhemispheric effects across the PPA variants The three PPA variants exhibited similar changes in homotopic structural connectivity (SC) but differed in terms of their functional connectivity (FC) changes. With regard to homotopic SC, all three variants showed a similar pattern of reduction that was related to disease severity (FTLD-CDR) (Fig. 2, Fig. 3). In terms of the spatial distribution of these SC changes, as shown in Fig. 4b, there were no significant differences among the variants. In contrast, for homotopic FC, the three variants showed very different patterns, raising the possibility that different mechanisms could be at play in the different variants. The non-fluent PPA group showed widespread FC elevation and a significant global effect (Fig. 2, Fig. 4), whereas logopenic and semantic variants exhibited more variability in terms of the distribution of elevated and reduced FC across the brain (Fig. 4). Quantitative analyses showed that the homotopic FC changes of the variants differed from each other for 5 out of the 7 networks (Fig. 4b), with the non-fluent variant differing the most from the other two. These findings raise the possibility that while the SC changes may be driven by mechanisms that are shared across the variants, FC changes may be driven by mechanism/s specific to each variant. One possible variant-specific factor is different underlying neuropathologies, although without neuropathological data, we cannot draw any conclusions in this regard. Another possible factor is different distributions of selective neural vulnerability of different pathologies, i.e., different brain areas are affected earlier and most severely in the different variants (Seeley et al., 2009). Depending on the functional role of the affected area/s, its dysfunction can have different “downstream” functional consequences. For instance, Cope et al. (Cope et al., 2018) found that the relationship between tau burden and functional connectivity went in opposite directions in Alzheimer's Disease and progressive supranuclear palsy (PSP), an FTD variant that has a similar underlying neuropathology as non-fluent PPA. They found that greater tau burden weakens FC in AD but strengthens FC in PSP. The authors argued that the difference could be due to both the different types of tau involved and the different spatial distributions of tau deposition in the two diseases. The contributions of specific types of neuropathology and their spatial distributions could be related to our current finding of higher homotopic FC in non-fluent PPA (an FTD-variant) with frontal pathology than in logopenic PPA (an AD-variant) with more posterior pathology. In sum, our findings are consistent with the possibility that the interhemispheric, homotopic SC reduction may be due to shared neuropathological factors, whereas the homotopic FC changes in each variant may be due to differences in neuropathology profiles and/or selective vulnerability of different brain regions to the disease. Limitations One limitation of this study arises from an intrinsic neuroimaging challenge of identifying interhemispheric, homotopic structural connectivity for lateral brain regions, because the white matter fibers connecting lateral homotopic areas have to travel long distances and encounter many crossing fibers. This issue not only arises for in vivo diffusion imaging but also for post-mortem fiber tracking studies (Jarbo et al., 2012). Moreover, the diffusion imaging acquisition parameters of the current study are relatively low (b = 700, 32 directions), due to the fact that the data collection period began some time ago, when these parameters were standard in clinical practice. As a result, we were likely to have more false negatives and, consequently, most connections examined here were relatively medial. Improvements in image acquisition parameters will be necessary for overcoming this limitation in the future. Another limitation is the lack of neuropathological data. Hence, we do not know the exact neuropathologies present in our cohort and thus cannot attribute the variant differences we observed to different pathologies with any confidence. Additionally, the sample size of the semantic variant is relatively small, and hence effects may be less reliable for this group. Finally, we were unable to fully evaluate if multilingualism had any influence on our findings, due to incomplete records of the information. Conclusions In this study we report, for the first time, on interhemispheric structural and functional connectivity within the same individuals in a large number of individuals across the three variants of primary progressive aphasia (PPA). Our findings add to the literature on both neurological disease and neurologically healthy individuals documenting the dissociation between structural and functional connectivity, extending previous findings to interhemispheric connectivity. We also report a dissociation between both structural and functional interhemispheric connectivity and the integrity of the gray matter areas they connect. These findings, as well as the different patterns of FC and SC across three PPA variants, raise questions about the underlying mechanisms of functional and structural changes in the context of neuropathology. Future work will be required to elucidate the nature of the mechanisms that give rise to these patterns, to understand their behavioral consequences, and to evaluate their implications for intervention, including neuromodulatory approaches. CRediT authorship contribution statement Yuan Tao: Writing – original draft, Visualization, Investigation, Formal analysis, Conceptualization. Kyrana Tsapkini: Writing – review & editing, Supervision, Project administration, Investigation, Funding acquisition. Brenda Rapp: Writing – review & editing, Supervision, Investigation, Conceptualization. Ethics declaration Written informed consent to take part in the study and to publish the article has been obtained from all participants or their legal representatives. The privacy rights of participants have been observed. This study was performed in compliance with relevant laws, regulatory frameworks and guidelines where the research took place. This study was approved by the Johns Hopkins University. (Approval No. IRB00229164; IRB00326681; IRB00201027) Funding This work was supported by grants from the 10.13039/100000002National Institutes of Health through award R01 DC014475, R01 AG068881, and R01AG075404 to KT. Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.