Longitudinal functional network connectivity changes across the clinical stages of C9orf72 hexanucleotide repeat expansion carriers

Abstract

INTRODUCTION

Intrinsic functional connectivity network abnormalities in C9orf72 hexanucleotide repeat expansion carriers emerge during the asymptomatic phase, yet longitudinal studies remain limited. We examined cross‐sectional abnormalities and longitudinal connectivity changes across clinical stages.

METHODS

We analyzed task‐free functional magnetic resonance imaging (fMRI) and structural MRI data in 36 asymptomatic (aSxC9), 17 prodromal (proC9), and 29 symptomatic (SxC9) carriers, and 107 healthy controls (HCs). Functional networks previously found altered in C9orf72, including salience, sensorimotor, default mode, and medial pulvinar thalamic networks, were examined. Associations between longitudinal connectivity and gray matter decline with baseline neurofilament light chain (NfL) concentrations and symptom severity were assessed.

RESULTS

aSxC9 and SxC9 showed longitudinal connectivity changes within specific networks. In aSxC9, connectivity changes correlated with baseline NfL. In proC9 and SxC9, changes in connectivity and gray matter were associated with baseline NfL and symptom severity.

DISCUSSION

C9orf72 expansion carriers demonstrate stage‐specific network connectivity changes.

Highlights

• Stage‐specific longitudinal connectivity changes were detected in C9orf72.

• aSxC9 and proC9 lacked detectable gray matter (GM) decline, alongside limited decline in SxC9.

• For aSxC9, connectivity network changes correlated with baseline neurofilament light chain (NfL).

• Connectivity/GM changes in proC9 and SxC9 correlate with NfL and symptom severity.

1. INTRODUCTION

Understanding the earliest systems‐level changes in genetic frontotemporal dementia (FTD) is a critical step toward forecasting symptom onset and tracking disease progression. Intrinsic connectivity networks (ICNs), as measured by task‐free‐functional magnetic resonance imaging (tf‐fMRI), show distinct abnormalities in genetic frontotemporal lobar degeneration (FTLD), even during the asymptomatic stage when atrophy is subtle or absent. ¹ , ² , ³ , ⁴ , ⁵ The most common FTLD genetic variant, a hexanucleotide repeat expansion in C9orf72, most often presents with behavioral variant frontotemporal dementia (bvFTD), amyotrophic lateral sclerosis (ALS), or bvFTD with motor neuron disease (bvFTD‐MND). ⁶ , ⁷ Most C9orf72 tf‐fMRI connectivity studies to date have been limited by cross‐sectional designs, with small sample sizes, and/or a focus on a single clinical stage. ² , ⁸ , ⁹ , ¹⁰ , ¹¹ , ¹² , ¹³ , ¹⁴ , ¹⁵

Cross‐sectional studies have identified ICN alterations in C9orf72 expansion carriers (C9) with bvFTD (C9orf72‐bvFTD), ALS (C9orf72‐ALS), or FTD‐MND. ⁸ , ⁹ , ¹⁰ , ¹¹ , ¹² , ¹³ , ¹⁴ , ¹⁵ In a previous study of 14 C9orf72‐bvFTD with or without MND, ⁸ connectivity disruptions (i.e., reduced connectivity compared to controls) were identified in the salience network (SN), which is involved in processing emotionally salient stimuli ¹⁶ ; default mode network (DMN), associated with memory and meta‐cognitive functions ¹⁷ , ¹⁸ ; and sensorimotor network (SMN), essential for motor functioning. ¹⁹ These ICNs have also been implicated in sporadic bvFTD and MND. ²⁰ , ²¹ In addition, C9orf72‐bvFTD demonstrated ICN disruption to the medial pulvinar nucleus of the thalamus, ⁸ an association nucleus involved in social cognition and in guiding visual attention toward salient stimuli. ²² Asymptomatic C9 also show functional alterations in these same networks, ² , ⁹ , ¹¹ appearing decades before expected symptom onset and in the absence of substantial atrophy. In contrast to these early connectivity anomalies, gray matter (GM) reductions in asymptomatic carriers are subtle, with atrophy typically becoming evident during the symptomatic stage. ¹⁴ , ²³ , ²⁴ , ²⁵ , ²⁶ , ²⁷ , ²⁸ , ²⁹ , ³⁰

Longitudinal characterization of functional connectivity along various clinical stages of C9orf72 would provide the foundational next step toward studies that anticipate symptom onset and monitor disease progression. Most longitudinal studies have included limited sample sizes that focus on a single clinical stage, and to date, no previous studies have examined longitudinal network connectivity in prodromal C9orf72 or C9orf72‐bvFTD. Longitudinal studies of C9orf72‐ALS have reported a lack of connectivity changes compared to controls, ¹³ , ²⁴ possibly due to limited follow‐up time and sample sizes. For asymptomatic carriers, two tf‐fMRI studies involving the same 15 participants were conducted over a period of up to 18 months. One study identified regions of declining and increasing connectivity within a motor network and increasing connectivity within a speech production network in asymptomatic carriers compared to controls. ⁹ The other study found that asymptomatic carriers had increasing intra‐network homogeneity in the somatomotor network. ¹¹ Overall, these studies suggest that functional network abnormalities and their progressive alterations precede detectable structural degeneration, but the extent to which longitudinal connectivity changes appear in each clinical stage of C9orf72 remains unclear.

RESEARCH IN CONTEXT

1. Systematic review: The authors conducted a systematic review using major databases (e.g., PubMed and Google Scholar). While cross‐sectional studies have investigated functional connectivity patterns in C9orf72 expansion carriers (C9orf72), longitudinal studies are scarce. Notably, no study has examined longitudinal functional connectivity changes during the prodromal stage, a critical transition stage from the asymptomatic to symptomatic phases, which was included in this study.

2. Interpretation: Leveraging a large longitudinal neuroimaging cohort of C9orf72, we comprehensively characterized stage‐specific connectivity changes in asymptomatic, prodromal and symptomatic carriers. Associations between longitudinal connectivity, and neurodegeneration and symptom severity highlight the potential of task‐free fMRI for tracking disease progression during each clinical stage.

3. Future directions: This study lays important groundwork for tracking disease progression as early as the asymptomatic stage. Future research should establish phenotype‐ and stage‐specific connectivity trajectories in carriers who convert to advanced stages or develop distinct C9orf72‐associated syndromes.

In the present study, we analyzed longitudinal tf‐fMRI and structural MRI changes in a cohort of 82 C9 (36 asymptomatic [aSxC9], 17 prodromal [proC9], and 29 symptomatic [SxC9]). We aimed to examine longitudinal functional and structural changes for each clinical stage and to determine how imaging measures relate to symptom severity and plasma neurofilament light chain (NfL) concentrations, a non‐specific biomarker of axonal injury that increases around symptom onset in genetic FTD. ³¹ Using a seed‐based connectivity approach, we examined the SN, DMN, SMN and a medial pulvinar thalamus network (MPN), networks that have shown alterations in cross‐sectional studies. ² , ⁸ We hypothesized that all three clinical stages would show longitudinal connectivity changes within these ICNs, but that longitudinal GM loss would be detectable only during the prodromal and symptomatic stages. We also hypothesized that longitudinal changes in connectivity or GM would correlate with baseline NfL and symptom severity in those participants with available data.

2. METHODS

2.1. Participants

Participants were recruited from the University of California, San Francisco (UCSF) Memory and Aging Center and the Advancing Research and Treatment for Frontotemporal Lobar Degeneration (ARTFL) / Longitudinal Evaluation of Familial Frontotemporal Dementia Subjects (LEFFTDS) Longitudinal Frontotemporal Lobar Degeneration (ALLFTD) study, a multisite genetic frontotemporal dementia study. ³² Clinical diagnoses were rendered at each study site.

We identified 82 C9 with usable neuroimaging data (Table 1). A pathological expansion was defined as having more than 40–50 repeats following previous methods, ³³ and all individuals tested negative for microtubule‐associated protein tau (MAPT) ³⁴ and progranulin (GRN) ³⁵ variants. C9 were grouped based on the clinical diagnosis at the time of the first visit.

TABLE 1.

Baseline demographics and clinical measures of C9orf72 expansion carriers and healthy controls.

	aSxC9 (n = 36)	proC9 (n = 17)	SxC9 (n = 29)	HC (n = 107)	Test statistic, p	Post hoc, p < 0.05
Age, years	42.7 (13.0)	55.4 (8.7)	58.4 (7.1)	52.3 (16.4)	H(3) = 19.00, p < 0.001*	aSxC9 < proC9, SxC9, and HC
Education, years	15.6 (1.8)	14.4 (2.5)	15.8 (3.0)	16.0 (2.3)	H(3) = 8.20, p = 0.04*	proC9 < HC
Sex, F:M	22:14	10:7	12:17	70:37	χ2(3) = 5.54, p = 0.14
Handedness, R/L/A	31/4/1	14/2/1	27/2/0	96/9/2	p = 0.76
CDR plus NACC FTLD, global score, median [range]	0 [0]	0.5 [0.5, 1]	2 [0.5, 3]	0 [0]	H(3) = 175.96, p < 0.001*	aSxC9 and HC < proC9 and SxC9
CDR plus NACC FTLD, sum of boxes	0 (0)	1.8 (1.3)	8.3 (4.0)	0 (0)	H(3) = 175.84, p < 0.001*	aSxC9 and HC < proC9 and SxC9
MMSE, total score	29.1 (1.0)	26.8 (2.5)	25.3 (3.5)	29.1 (0.9)	H(3) = 43.49, p < 0.001*	aSxC9 and HC > proC9 and SxC9
MoCA, total score	27.3 (2.5)	25.8 (3.2)	19.7 (7.1)	27.5 (2.1)	H(3) = 26.12, p < 0.001*	aSxC9 and HC > SxC9
NfL, pg/mL	7.6 (3.6)	18.4 (20.4)	43.4 (42.4)	5.8 (2.9)	F(5,100) = 37.13, p < 0.001*	proC9 > HC; SxC9 > aSxC9, proC9, and HC
Longitudinal visits, n, 1/2/3/4	36/18/6/0	17/10/5/0	29/8/5/1	107/52/16/2
Longitudinal follow‐up duration, years	1.8 (0.8)	2.1 (0.7)	2.2 (1.6)	2.4 (1.4)	H(3) = 2.90, p = 0.40

Note: This table presents group differences in the mean (SD) values (continuous variables) or proportions (categorical variables), unless otherwise specified. Original NfL concentration values are shown, whereas statistical analyses are based on linear regression models using log‐transformed NfL concentrations, which regress baseline age and sex as nuisance covariates. Mean follow‐up duration is calculated based on participants with longitudinal visits only. Abbreviations: aSxC9, asymptomatic C9orf72 expansion carriers; CDR plus NACC FTLD, CDR Dementia Staging Instrument plus Behavior and Language domains from the National Alzheimer's Coordinating Center (NACC) Frontotemporal Lobar Degeneration Module; F/M, female/male; HC, healthy control; MMSE, Mini‐Mental State Examination; MoCA, Montreal Cognitive Assessment; NfL, plasma neurofilament light chain; proC9, prodromal C9orf72 expansion carriers; R/L/A, right/left/ambidextrous; SD, standard deviation; SxC9, symptomatic C9orf72 expansion carriers.

^*Significance at p < 0.05.

The 36 aSxC9 were diagnosed as clinically normal and had a global score of 0 on the Clinical Dementia Rating (CDR) plus Behavior and Language domains from the National Alzheimer's Coordinating Center (NACC) FTLD module (CDR plus NACC FTLD). ³⁶ The 17 proC9 included 15 diagnosed with mild cognitive impairment (MCI), ³⁷ of whom two had a global score of 1 on the CDR plus NACC FTLD: one with a mood disorder and one with alcohol use. The 29 SxC9 included 17 bvFTD, 7 bvFTD‐MND, one bvFTD with primary progressive aphasia of unspecified subtype, one ALS‐MCI, and three ALS. To enable comparisons of connectivity abnormalities between each of the three C9 cohorts and a common reference group, we included a single group of 107 healthy control participants (HCs). HCs consisted of non‐carrier members of GRN, MAPT, and C9orf72 families with unrelated controls who were included as necessary to ensure an age distribution comparable to the C9. All HCs were required to have a Mini‐Mental State Examination (MMSE) score ≥27, and if unavailable, a corresponding score ≥21 on the Montreal Cognitive Assessment (MoCA), ³⁸ a global score of 0 on the CDR plus NACC FTLD, and no significant white matter lesions or history of neurological disease. Baseline MRI scanning was required to be within 180 days of clinical assessment.

2.2. Statistical analyses of demographic, clinical, and neuropsychological measures

Demographic, clinical, and neuropsychological measures were compared between groups using one‐way analysis of variance (ANOVA) or Kruskal–Wallis tests if the assumptions required for ANOVA were not met, implemented in R (version 4.3.1). Significant differences were followed by post hoc multiple comparisons: Tukey's Honestly Significant Difference (HSD) after ANOVA or Dunn's test (FSA package, v0.9.5) with Bonferroni adjustment after Kruskal–Wallis tests, with a significance threshold of p < 0.05. For categorical variables, chi‐square tests were used to compare groups when the cell frequencies were 5 or more, and Fisher's exact tests when the cell frequencies were less than 5. A linear regression model was used to examine the effects of group on log‐transformed NfL concentrations at baseline while controlling for baseline age and sex. Significant differences were followed by Tukey's HSD post hoc tests (p < 0.05). Linear mixed‐effects (LME; lme4 v1.1.34) models were used to compare changes in clinical and neuropsychological measures over time between groups, controlling for baseline age, sex and education. HC was set as the reference group. The fixed effects included group, time since baseline, the interaction between group and time, and baseline age, sex, and education. Random intercepts for each subject were included to account for repeated measures. For measures for which a group had fewer than seven subjects with longitudinal data, that group was excluded from the LME analyses. To aid the interpretation of significant longitudinal group differences, raw annualized change for each outcome measure was calculated. For each subject, the difference between the last and first observed score was divided by the time (in years) between those visits to obtain an annualized rate of change. Group‐level summaries were then computed as the mean and standard deviation (SD) of these per‐subject annualized changes. In addition, because familial age at onset was not available in the ALLFTD data cohort, we calculated the estimated years to onset for the aSxC9 group by subtracting the mean reported age at onset from a previous study of 1433 C9 (58.2 years) ³⁹ from the baseline age of each aSxC9.

2.3. Image acquisition

Each participant underwent a T1‐weighted (T1w) structural MRI scan and a T2*‐weighted echo‐planar imaging tf‐fMRI scan on a 3T scanner at each site. The tf‐fMRI session acquired either 240 (8:08 min) or 197 volumes (10 min), following previous protocols at the UCSF Memory and Aging Center and ALLFTD. ³²

A standard imaging protocol for ALLFTD scans was used and reviewed for quality control at the Mayo Clinic, Rochester, Minnesota. ³² Following initial quality assessment, six C9 and two HCs were excluded because none of their visits passed quality control, and individual timepoints from 1 aSxC9, 1 SxC9, and 4 HCs were excluded. The proportion of participants with excluded imaging sessions did not differ between groups (χ ²(3) = 0.69, p = 0.87). We applied additional quantitative fMRI motion criteria requiring a mean framewise displacement (FD) below 0.5 mm and at least 4 min of data meeting this threshold per scan. All included participants met these motion criteria. Voxel‐based morphometry analyses and functional imaging were conducted using T1w and tf‐fMRI scans acquired at the same visit.

2.4. Voxel‐based morphometry analysis

2.4.1. Preprocessing

T1w images were processed with the standard longitudinal voxel‐based morphometry pipeline in Statistical Parametric Mapping (SPM12), (Wellcome Trust Centre for Neuroimaging, UCL, London, UK) embedded in MATLAB (version R2021a). Structural T1w images of each participant at each visit were bias field corrected and segmented into tissue probability maps using the “New Segment” routine. ⁴⁰ The tissue maps were normalized to standard Montreal Neuroimaging Institute (MNI) space using serial longitudinal registration, via a subject‐specific average template created using non‐linear diffeomorphic and rigid‐body registration. ⁴¹ Next, the maps were modulated by multiplying the timepoints’ Jacobians with the corresponding tissue maps, and smoothed using an 8 mm full width at half maximum (FWHM) isotropic Gaussian kernel. Finally, these GM maps were parcellated into 246 regions of interest (ROIs) based on the Brainnetome atlas, ⁴² and GM volume (GMV) within each ROI was obtained for subsequent statistical analyses. Total intracranial volume (TIV) was estimated by the sum of GM, white matter, and cerebrospinal fluid volumes.

2.4.2. Harmonization

To harmonize multisite structural data for subsequent statistical analyses, a customized ComBat approach ⁴ was applied to individual GM maps to remove batch effects of no interest (different scanners), while preserving biological variability of interest (i.e., age, sex, education, and handedness). ComBat parameters were estimated in a reference group of 190 HCs (ComBat HC, including the 107 study HC; Table S1) whose demographic characteristics spanned the age ranges and scanner sites of the study cohorts (both carriers and study HC) and were used to harmonize study cohort data. Details on evaluating ComBat harmonization performance are included in the Supporting Information.

2.4.3. Statistics

Longitudinal changes in GMV were examined using LME models (lme4 v1.1.34, lmerTest v3.1.3), with fixed effects for group, time between baseline and follow‐up, group‐by‐time interaction, age, sex, education, and TIV. Random effects included individual intercepts, while random slopes were omitted because the number of repeated observations per participant was insufficient to reliably estimate subject‐specific slope variability. Bonferroni correction (p < 0.05) was applied for the 246 ROIs.

2.5. Functional imaging analysis

2.5.1. Preprocessing

Data were preprocessed using the fMRIPrep pipeline. ⁴³ Each T1w image was corrected for intensity non‐uniformity (N4BiasFieldCorrection, ANTs v2.2.0), and used as the T1w‐reference throughout the workflow. The T1w‐reference was then skull‐stripped, and segmented to GM, white matter, and cerebrospinal fluid (fast, FSL v5.0.9). Volume‐based spatial normalization to one standard space was performed through first creating an unbiased subject template by averaging T1w acquired across the visits included in the analysis, and then normalized to the template (FSL's MNI ICBM 152 non‐linear 6th Generation Asymmetric Average Brain Stereotaxic Registration Model; antsRegistration, ANTs 2.2.0), using brain‐extracted versions of both T1w‐reference and the template.

The first five fMRI volumes were dropped for magnetic field stabilization, and the following preprocessing was performed. First, a reference volume and its skull‐stripped version were generated using a custom methodology of fMRIPrep. The blood‐oxygen level–dependent (BOLD) reference was then co‐registered to the T1w‐reference using boundary‐based registration (bbregister, FreeSurfer v6.0.0) with six degrees of freedom. Head‐motion parameters with respect to the BOLD reference (transformation matrices, and six corresponding rotation and translation parameters) were estimated before spatiotemporal filtering (mcflirt, FSL v5.0.9). BOLD data were slice‐time corrected (3dTshift, AFNI v16.02.07). BOLD time series were resampled onto their original, native space by applying a single, composite transform to correct for head‐motion and susceptibility distortions. To normalize BOLD time‐series into MNI152 standard space, motion correcting transformations, BOLD‐to‐T1w transformation, and T1w‐to‐MNI warp were concatenated and applied in a single step (antsApplyTransforms, ANTs v2.1.0; Lanczos interpolation). Data were smoothed with a 6 mm FWHM Gaussian kernel (SUSAN). ⁴⁴ Following these steps, we applied “non‐aggressive” ICA‐AROMA ⁴⁵ to identify independent components representing motion‐related artifacts based on spatial and temporal discriminative features, and regressed them from the time series. Finally, data underwent linear detrending, band‐pass filtering (0.008 ∼ 0.08 Hz), and nuisance regression (the 6 head motion parameters, 6 head motion parameters one time point before, and the 12 corresponding squared items, mean signals from the white matter and cerebrospinal fluid, and global signal).

2.5.2. Seed‐based connectivity maps

We used 4 mm radius spheres (i.e., seed regions) around peak coordinates from the literature (Table S2) ² , ⁸ to derive ICNs of interest. These seeds were selected to anchor large‐scale ICNs known a priori to be altered in C9, which included the SN, SMN, DMN, and MPN. These networks have shown cross‐sectional alterations in both asymptomatic and symptomatic C9. We conducted separate bivariate regression analyses between the average time series within each seed region and time series in the remaining voxels across the whole brain for each ICN of interest. This resulted in four ICN maps for each participant.

2.5.3. Harmonization

Similar to GM maps, ICN maps were harmonized with ComBat based on the same Combat HC group. Covariates included age, sex, education, and handedness.

2.5.4. Statistics

Voxelwise analyses were carried out in R (version 4.3.3) using the neuropointillist toolbox (0.0.0.9, http://ibic.github.io/neuropointillist) ⁴⁶ on each ICN map separately. To assess cross‐sectional differences between groups (aSxC9, proC9, SxC9, HC), we analyzed participants’ first visits using GLM. These models evaluated group effects with nuisance covariates for age, sex, education, handedness, and eyes‐open status during scan acquisition. Linear contrasts were used to compare mean group levels, and interaction contrast tests were conducted to compare: (1) aSxC9 and HC, (2) proC9 and HC, and (3) SxC9 and HC.

Longitudinal change patterns were assessed by LME models (lme4 v1.1‐35.1, lmerTest v3.1‐3) with fixed‐effects terms for group (aSxC9, proC9, SxC9, HC), age at baseline, time since baseline, sex, education, handedness, and eyes‐open status during scan acquisition. A random intercept was included for subject identity. Random slopes were not included because the number of repeated observations per participant was insufficient to reliably estimate subject‐specific slope variability. LME models were fitted using restricted maximum likelihood (ReML). All available sessions were included to maximize the estimation accuracy of between‐subject variability. For further analysis, results were converted to Z scores using Satterthwaite approximation to estimate the effective degrees of freedom. Based on the omnibus model with all four subgroups, the marginal means of linear trends (group ∙ time since baseline interaction; emmeans, v1.9.0) associated with each subgroup were compared using planned linear contrasts and interaction contrasts. These tests compared (1) aSxC9 and HC, (2) proC9 and HC, and (3) sxC9 and HC.

A follow‐up analysis explored the relationship between baseline CDR plus NACC FTLD sum of boxes ³⁶ and longitudinal connectivity changes in proC9 (n = 10) and SxC9 (n = 8) separately using LME models, with fixed‐effects terms for CDR plus NACC FTLD sum of boxes at baseline, baseline age, time since baseline, sex, education, handedness, and eyes‐open status. A random intercept was included for subject identity.

All connectivity analyses were masked to the relevant network (Figure S1). Network masks were created by using 1‐sample t‐tests of seed‐based ICN maps in an independent group of 103 HC (Seed mask HC; Table S1), thresholded at t ≥ 4. Results were family‐wise error (FWE) corrected based on Monte‐Carlo simulations using AFNI's 3dClustSim program (NN1). ⁴⁷ Spatial smoothness was estimated for each subject and network using 3dFWHMx ⁴⁸ based on data that were smoothed and detrended but not denoised by ICA‐AROMA, regression, or temporal filtering. The spherical autocorrelation function parameters were averaged across all scans in the analysis when simulating the noise distribution. Significance was set at voxelwise thresholds of p < 0.05, with a cluster‐level FWE‐correction applied at p < 0.05 using extent thresholding. We used one‐tailed inference for linear contrasts for consistency with our previous findings. ² , ³ , ⁴

2.6. Longitudinal changes associated with NfL concentrations

Plasma NfL concentrations were assayed utilizing the Simoa NF‐light Advantage Kit from Quanterix on a SiMoA HD‐X Analyzer instrument following previous protocols ⁴⁹ , ⁵⁰ , ⁵¹ at two processing labs.

The relationships between baseline log(NfL) concentrations and longitudinal change patterns in GMV in 246 ROIs and voxel‐wise functional connectivity were assessed using LME models within each group. LMEs were run for aSxC9 (n = 16) and a combined cohort of proC9 (n = 8) and SxC9 (n = 6), due to limited baseline NfL data for proC9 and SxC9. The models included fixed‐effects terms for log(NfL) at baseline, age at baseline, time since baseline, sex, education, handedness, TIV (GMV only) and eyes‐open status (functional connectivity only). A random intercept was included for subject identity. Results were masked to the relevant network and considered significant at voxelwise thresholds of p < 0.05, with cluster‐level FWE‐correction at p < 0.05.

3. RESULTS

3.1. Demographic and clinical features, longitudinal GM comparisons

Demographic, clinical, and neuropsychological measures were compared between groups at baseline (Tables 1, 2) and longitudinally (Table 3). Groups differed in age and education, with aSxC9 being younger than all other groups, and proC9 having fewer years of education than HC. The mean estimated years to onset for aSxC9 was −15.5 years (SD = 13.0; range = −39.2 to 10.8), based on a large study estimating the mean age at onset in C9 to be 58.2 years. ³⁹ As expected, aSxC9 showed no differences in any clinical or neuropsychological measures compared to HC. SxC9 (vs HC) had lower scores in most measures, except for the Multilingual Naming Test, and for this test, Sx had lower scores compared to aSxC9. ProC9 had impairments in Trail Making Tests, Semantic Fluency Test, and Benson Figure Recall Test compared to HC. Detailed group comparison results are reported in Table 2.

TABLE 2.

Baseline neuropsychological measures of C9orf72 expansion carriers and healthy controls.

	aSxC9 (n = 36)	proC9 (n = 17)	SxC9 (n = 29)	HC (n = 107)	Test statistic, p	Post hoc, p < 0.05
NPI‐Q, total score	1.0 (2.6)	8.5 (7.4)	10.2 (7.4)	0.6 (1.0)	H(3) = 38.92, p <0.001*	aSxC9 and HC < proC9 and SxC9
GDS‐15, total score	1.1 (1.8)	2.8 (3.2)	3.6 (3.1)	1.3 (1.8)	H(3) = 15.73, p =0.001*	aSxC9 and HC < SxC9
Trails A, correct lines per minute	61.3 (18.0)	47.0 (13.3)	38.8 (16.2)	79.7 (28.3)	H(3) = 27.6, p < 0.001*	SxC9 < aSxC9 and HC; proC9 < HC
Trails B, correct lines per minute	26.2 (6.7)	19.2 (5.0)	17.9 (10.2)	30.5 (9.8)	F(3, 87) = 8.56, p < 0.001*	proC9 and SxC9 < HC
Digit span forward	6.5 (1.3)	6.5 (1.6)	5 (1.1)	7 (1.1)	H(3) = 33.80, p < 0.001*	aSxC9, proC9, and HC > SxC9
Digit span backward	5.3 (1.0)	5.0 (1.7)	3.2 (1.3)	5.4 (1.2)	H(3) = 38.20, p < 0.001*	aSxC9, proC9, and HC > SxC9
Semantic fluency, animals in 1 min	22.7 (4.8)	18.5 (5.5)	11.4 (6.1)	23.7 (5.2)	F(3, 158) = 35.59, p < 0.001*	aSxC9, proC9, and HC > SxC9; proC9 < HC
Lexical fluency, words in 1 min	27.3 (8.2)	24.7 (7.3)	15.9 (11.6)	29.2 (8.0)	F(3,89)=7.44, p < 0.001*	aSxC9 and HC > SxC9
Delayed free recall (California Verbal Learning Test‐short form, 10 min recall)	7.4 (1.6)	6.6 (2.2)	4.5 (2.7)	7.9 (1.4)	H(3) = 33.36, p < 0.002	aSxC9 and HC > SxC9
Benson figure 10 min recall, total score	12.6 (2.3)	10.6 (2.7)	9.2 (4.0)	13.1 (2.3)	H(3) = 27.00, p < 0.001*	aSxC9 and HC > SxC9; proC9 < HC
Multilingual Naming Test, total score	30.4 (1.8)	28.3 (3.3)	27.4 (3.5)	30.2 (2.0)	H(3) = 11.71, p = 0.008*	aSxC9 > SxC9
Benson figure copy, total score	15.4 (2.9)	14.6 (4.2)	13.4 (3.2)	15.1 (2.6)	H(3) = 11.46, p = 0.01*	aSxC9 and HC > SxC9

Note: This table presents group differences in the mean (SD) values.

Abbreviations: aSxC9, asymptomatic C9orf72 expansion carriers; GDS‐15, Geriatric Depression Scale 15‐item version; HC, healthy control; NPI‐Q, Neuropsychiatric Inventory Questionnaire; proC9, prodromal C9orf72 expansion carriers; SxC9, symptomatic C9orf72 expansion carriers.

^**Significance at p < 0.05.

TABLE 3.

Longitudinal comparisons of clinical and neuropsychological measures in C9orf72 expansion carriers and healthy controls.

	aSxC9 vs HC		proC9 vs HC			SxC9 vs HC		Mean annualized change
	β, 95% CI	Test statistic, p	β, 95% CI	Test statistic, p	β, 95% CI	Test statistic, p	aSxC9	proC9	SxC9	HC
CDR plus NACC FTLD, global score,	0.0003, [−0.07, 0.07]	t(152) = 0.01, p = 0.99	0.06, [−0.02, 0.1]	t(138) = 1.49, p = 0.14	0.05, [−0.02, 0.1]	t(176) = 1.51, p = 0.13	0 (0)	−0.03 (0.34)	0.11 (0.39)	0 (0)
CDR plus NACC FTLD, sum of boxes	−0.002, [−0.3, 0.3]	t(122) = −0.01, p = 0.99	0.46, [0.09, 0.83]	t(115) = 2.39, p = 0.02 *	0.33, [−0.005, 0.6]	t(131) = 2.03, p = 0.04 *	0 (0)	0.11 (1.41)	1.13 (2.13)	0 (0)
MMSE, total score	−0.53, [−1.3, 0.2]	t(111) = −1.39, p = 0.17	0.12, [−0.7, 0.9]	t(97) = 0.29, p = 0.78	−0.14, [‐0.8, 0.6]	t(142) = −0.35, p = 0.73	−0.46 (1.25)	−0.08 (0.58)	−1.95 (5.31)	0.01 (0.32)
MoCA, total score	−−0.20, [−1.1, 0.7]	t(88) = −0.44, p = 0.66	−0.81, [−1.9, 0.4]	t(84) = −1.37, p = 0.17	−0.25, [−1.4, 1.0]	t(182 = −0.41, p = 0.68	0.08 (1.28)	−0.61 (1.37)	−3.35 (7.69)	0.27 (1.05)
NPI‐Q, total score	−0.10, [−0.9, 0.7]	t(68) = −0.25, p = 0.80	−−0.77, [−1.7, 0.07]	t(64) = −1.75, p = 0.09			0.17 (0.72)	−0.81 (2.94)		0.11 (1.32)
GDS‐15, total score	0.02, [−0.5, 0.5]	t(137) = 0.07, p = 0.94	0.13, [−0.5, 0.7]	t(127) = 0.43, p = 0.67	−0.31, [−‐0.98, 0.37]	t(160)= −0.89, p = 0.38	−0.23 (0.69)	−0.43 (2.0)	−−0.58 (0.79)	−0.297 (0.88)
Trails A, correct lines per minute	−1.99, [−8.0,4.1]	t(89) = −0.64, p = 0.52	1.69, [−‐5.2, 8.6]	t(84) = 0.48, p = 0.64			−1.04 (10.9)	0.85 (4.79)		0.63 (11.60)
Trails B, correct lines per minute	−1.41, [−3.6, 0.8]	t(89) = −1.25, p = 0.21	−−2.19, [−4.7, 0.3]	t(89) = −1.71, p = 0.09			−0.12 (3.68)	−2.61 (5.68)		2.49 (4.47)
Digit span forward	0.02, [−0.3, 0.3]	t(126) = 0.13, p = 0.89	0.09, [−0.3, 0.4]	t(108) = 0.48, p = 0.63	0.07, [−0.3, 0.5]	t(137) = 0.35, p = 0.73	0.28 (0.68)	−0.11 (0.88)	−0.22 (0.44)	0.15 (0.53)
Digit span backward	0.05, [−0.3, 0.4]	t(126) = 0.32, p = 0.75	−0.14, [−0.5, 0.2]	t(111) = −0.76, p = 0.45	0.01, [−0.4, 0.4]	t(138)= −0.07, p = 0.95	0.21 (0.59)	−0.22 (0.51)	−0.03 (0.44)	0.22 (0.68)
Semantic fluency, animals in 1 min	0.58, [−0.7, 1.8]	t(129) = 0.92, p = 0.36	−0.2, [−1.8, 1.4]	t(118)= −0.26, p = 0.80	0.29, [−1.5, 2.1]	t(166) = 0.31, p = 0.76	0.07 (2.58)	−1.11 (2.87)	−−1.83 (3.68)	−0.52 (2.56)
Lexical fluency, words in 1 min	0.008, [−1.8, 1.8]	t(82) = 0.01, p = 0.99	−0.61, [−2.7, 1.6]	t(78) = −0.55, p = 0.58			0.83 (2.85)	−0.74 (2.41)		−0.05 (3.73)
Delayed free recall (California Verbal Learning Test‐short form, 10 min recall)	−0.03, [−0.4, 0.4]	t(116) = −0.14, p = 0.89	0.06, [−0.4, 0.5]	t(90) = 0.24, p = 0.81	−0.32 [−0.9, 0.3]	t(107)= −1.07, p = 0.29	−0.04 (0.98)	0.26 (0.51)	−0.94 (1.95)	0.25 (0.77)
Benson figure 10 min recall, total score	0.75, [0.1, 1.4]	t(129) = 2.18, p = 0.03 *	0.15, [−0.7, 1.0]	t(113) = 0.35, p = 0.72	0.09, [−‐0.9, 1.1]	t(154) = 0.19, p = 0.85	0.65 (1.30)	0.20 (1.76)	−1.22 (2.57)	0.13 (1.44)
Multilingual Naming Test, total score	0.06, [−0.3, 0.5]	t(75) = 0.30, p = 0.76	−0.26, [−0.7, 0.2]	t(72) = −1.17, p = 0.25			0.29 (0.99)	0.04 (1.10)		0.22 (0.43)
Benson figure copy, total score	−0.12, [−0.6, 0.3]	t(86) = −0.51, p = 0.61	−0.10, [−0.7, 0.5]	t(81) = −0.36, p = 0.72	−0.27, [−0.9, 0.4]	t(93) = −0.79, p = 0.43	−0.21 (0.71)	−0.14 (0.88)	−0.61 (1.08)	0.01 (0.80)

Note: This table presents longitudinal group differences and mean annualized change scores in clinical and neuropsychological measures between each carrier group and HC. Longitudinal group differences were estimated using linear mixed‐effects models (LMEs). For measures where a group had fewer than seven subjects with longitudinal data, that group was excluded from the LME analyses. Annualized change was calculated for each subject as (last score − first score) ÷ years between visits. Values represent the group means (SD). Positive values indicate increasing scores per year and negative values indicate decreasing scores in original test units per year.

Abbreviations: aSxC9, asymptomatic C9orf72 expansion carriers; CDR plus NACC FTLD, CDR Dementia Staging Instrument plus Behavior and Language domains from the National Alzheimer's Coordinating Center (NACC) Frontotemporal Lobar Degeneration Module; GDS‐15, Geriatric Depression Scale 15‐item version; HC, healthy control; MMSE, Mini‐Mental State Examination; MoCA, Montreal Cognitive Assessment; NPI‐Q, Neuropsychiatric Inventory Questionnaire; proC9, prodromal C9orf72 expansion carriers; SD, standard deviation; SxC9, symptomatic C9orf72 expansion carriers.

^* Significance at p < 0.05.

Longitudinally, aSxC9 (vs HC) showed slower decline in a visual memory task (Benson Figure Recall) compared to HC. Both proC9 and SxC9 demonstrated a faster rate of clinical decline (CDR plus NACC FTLD sum of boxes) compared to HC. There were no differences in longitudinal follow‐up duration between the groups.

Both aSxC9 and proC9 lacked detectable longitudinal GMV decline, whereas SxC9 had decline in the precuneus and medial prefrontal cortex (Figure S2).

3.2. Asymptomatic carriers had baseline hypoconnectivity and longitudinal connectivity changes within C9orf72‐relevant networks

At baseline, aSxC9 showed connectivity disruptions (i.e., lower connectivity) within networks previously reported to show alterations in C9 (Figure 1A). SN hypoconnectivity (vs HC) was detected in the bilateral anterior cingulate and medial prefrontal cortex, left anterior insula and inferior frontal gyrus, left angular gyrus, and right middle temporal gyrus (Cohen's d = 0.75). Within the SMN, decreased connectivity was seen bilaterally in the occipital lobe (d = 0.95). MPN hypoconnectivity (vs HC) was concentrated in bilateral posterior thalamus, precuneus, and postcentral gyrus (d = 0.9). No significant alterations in DMN connectivity were found.

FIGURE 1.

Asymptomatic carriers exhibit baseline hypoconnectivity and longitudinal connectivity changes within C9orf72‐relevant networks. (A) At baseline, aSxC9 show regions of connectivity disruptions including in key regions of the SN (bilateral MPFC and left anterior INS), SMN (bilateral Occ), and MPN (bilateral posterior THA) compared to HC. (B) Longitudinally, aSxC9 shows regions of connectivity decline in the SN, increasing connectivity in the DMN, and regions of both connectivity decline and increase in the MPN. Notably, longitudinal SN connectivity decline is observed in the right anterior INS, which shows baseline connectivity disruption contralaterally. (C) Mean connectivity versus time since the baseline visit within the significant voxels shown in (B) is plotted for visualization purposes only. Results are thresholded at p < 0.05 voxelwise, with cluster‐level FWE correction at p < 0.05, masked to the relevant network mask. Color bars represent z‐scores, where warm colors indicate baseline hyperconnectivity or longitudinal increases in aSxC9 compared to HC, whereas cool colors indicate hypoconnectivity or longitudinal decreases. Result maps are superimposed on the MNI brain template. ACC, anterior cingulate cortex; ANG, angular gyrus; aSxC9, asymptomatic C9orf72 expansion carriers; DMN, default mode network; FC, functional connectivity; HC, healthy control; INS, insula; L, left; MPFC, medial prefrontal cortex; MPN, medial pulvinar network; ns, not significant; Occ, occipital lobe; PCC, posterior cingulate cortex; PCu, precuneus; PostCG, postcentral gyrus; SMN, sensorimotor network; SN, salience network; STG/Op, superior temporal gyrus extending to operculum; THA, thalamus; VLPFC, ventrolateral prefrontal cortex.

Longitudinally, aSxC9 showed declining connectivity in the SN compared to HC (Figure 1B,C). SN connectivity declines were observed in the right middle frontal cortex, ventrolateral prefrontal cortex, anterior insula, and angular gyrus. Of note, longitudinal SN connectivity decline arose in the right anterior insula, a region showing connectivity disruption contralaterally at baseline. MPN connectivity declines emerged in bilateral superior temporal gyrus extending to the operculum, along with MPN connectivity increases appearing in bilateral precuneus and posterior cingulate cortex. For the DMN, aSxC9 showed increasing connectivity principally in bilateral medial prefrontal cortex and left angular gyrus. These changes reflected absolute connectivity increases or decreases, rather than solely differences in rates of change relative to HC (Figure 1C). No significant longitudinal changes in SMN connectivity were observed. Given the earlier finding of slower visual memory decline in aSxC9, we performed a post hoc LME analysis to test whether this increasing DMN connectivity was associated with Benson Figure Recall changes. Mean DMN connectivity extracted from significant voxels was separated into each participant's average connectivity across visits and visit‐to‐visit deviations from that participant's own average, allowing stable between‐subject differences to be distinguished from longitudinal within‐subject changes. The model included time, within‐subject connectivity, and their interaction (time × connectivity), along with the participant mean connectivity term, covariates (age, sex, education, and handedness), and random intercepts for subject. We did not find evidence that increasing DMN connectivity was associated with slower visual memory decline in aSxC9 (β = −2.90, p = 0.44).

3.3. Prodromal carriers showed baseline connectivity alterations but lacked detectable longitudinal connectivity changes compared to HC

At baseline, proC9 had regions of hypoconnectivity in the SN and SMN, and hyperconnectivity in the DMN compared to HC (Figure 2). Specifically, SN hypoconnectivity was seen in bilateral medial prefrontal cortex (d = 0.83). SMN hypoconnectivity was observed in the bilateral occipital regions (d = 0.78). DMN hyperconnectivity was detected in bilateral precuneus and posterior cingulate cortex (d = 1.1). ProC9 showed no detectable alterations in the MPN, or longitudinal connectivity changes in any networks compared to HC.

FIGURE 2.

Group comparison of prodromal carriers versus HCs identifies baseline alterations in network connectivity, but a lack of detectable longitudinal connectivity changes. At baseline, proC9 display regions of hypoconnectivity in SN (MPFC) and SMN (Occ), and DMN hyperconnectivity, which include bilateral PCu and PCC, compared to HCs. No significant longitudinal connectivity changes are observed. Results are thresholded at p < 0.05 voxelwise, with cluster‐level FWE correction at p < 0.05, masked to the relevant network mask. Color bars represent z‐scores, where warm colors indicate cross‐sectional hyperconnectivity in proC9 compared to HCs, whereas cool colors indicate hypoconnectivity. Result maps are superimposed on the MNI brain template. DMN, default mode network; HC, healthy control; L, left; MPFC, medial prefrontal cortex; MPN, medial pulvinar network; ns, not significant; Occ, occipital lobe; PCC, posterior cingulate cortex; PCu, precuneus; proC9, prodromal C9orf72 expansion carriers; SMN, sensorimotor network; SN, salience network.

3.4. Symptomatic carriers had predominantly baseline hypoconnectivity and longitudinal increases within networks examined

SxC9 showed SN and SMN connectivity disruptions compared to HC (Figure 3A). Specifically, SN hypoconnectivity was identified in bilateral anterior cingulate and medial prefrontal cortex, insula, and posterior cingulate cortex (d = 1.07). SMN hypoconnectivity was found in bilateral perirolandic cortex and supplementary motor area, as well as the bilateral occipital lobe (d = 0.84). For the DMN, divergent connectivity alterations included hypoconnectivity in regions of bilateral medial prefrontal cortex and right superior frontal cortex (d = 0.87), accompanied by hyperconnectivity in bilateral posterior cingulate cortex, precuneus, and middle temporal gyrus (d = 0.74). Of note, the precuneus and posterior cingulate cortex regions exhibiting DMN hyperconnectivity in SxC9 showed substantial spatial overlap with the patterns observed in proC9. No significant connectivity alterations were detected in the MPN at baseline.

FIGURE 3.

Symptomatic carriers have baseline connectivity alterations and longitudinal connectivity increases in the SN and SMN. (A) At baseline, SxC9 (vs HCs) show connectivity disruptions in the SN and SMN, and divergent connectivity alterations in the DMN. (B) At follow‐up, SxC9 exhibit connectivity increases over time within the SMN in bilateral PRC. These changes overlap well with regions showing baseline connectivity disruptions in the same networks. (C) Mean connectivity versus time since the baseline visit within the significant voxels shown in (B) is plotted for visualization purposes only. Results are thresholded at p < 0.05 voxelwise, with cluster‐level FWE correction at p < 0.05, masked to the relevant network mask. Color bars represent z‐scores, where warm colors indicate baseline hyperconnectivity or longitudinal increases in SxC9 compared to HC, whereas cool colors indicate hypoconnectivity or longitudinal decreases. Result maps are superimposed on the MNI brain template. Overlapping regions between cross‐sectional and longitudinal results are highlighted with cyan outlines. ACC, anterior cingulate cortex; DMN, default mode network; FC, functional connectivity; HC, healthy control; INS, insula; L, left; MPFC, medial prefrontal cortex; MPN, medial pulvinar network; ns, not significant; Occ, occipital lobe; PCC, posterior cingulate cortex; PCu, precuneus; PRC, perirolandic cortex; SFG, superior frontal gyrus; SMN, sensorimotor network; SN, salience network; SxC9, symptomatic C9orf72 expansion carriers.

SxC9 demonstrated longitudinal increases in the SMN (Figure 3B,C). Increasing SMN connectivity emerged in bilateral perirolandic regions. Notably, these SMN regions showing that longitudinal increases overlapped well with regions showing connectivity disruptions at baseline. These changes reflected absolute connectivity increases or decreases, rather than solely differences in rates of change relative to HC. No significant longitudinal changes were detected in the SN, DMN, and MPN. To determine whether the increasing SMN connectivity was associated with faster increases in symptom severity in SxC9, we performed an exploratory LME analysis analogous to that used for longitudinal DMN increase in aSxC9 (Section 3.2), using CDR plus NACC FTLD sum of boxes as the clinical outcome measure. Within‐subject changes in SMN connectivity were not associated with CDR plus NACC FTLD sum of boxes progression over time in SxC9 participants (β = −0.23, p = 0.74).

Cross‐sectional and longitudinal connectivity profiles across the C9orf72 spectrum are summarized in Figure S3.

3.5. Higher baseline NfL was associated with longitudinal connectivity changes in aSxC9, and longitudinal DMN increases and GMV decline in prodromal/symptomatic carriers

For aSxC9, baseline NfL concentrations showed an association with longitudinal connectivity changes in the SMN and MPN (Figure 4A). Higher NfL concentrations were associated with longitudinal SMN connectivity decreases within the right postcentral gyrus. In addition, higher NfL concentrations were associated with longitudinal MPN connectivity increases in the right anterior thalamus as well as longitudinal decreases in this network in the bilateral posterior cingulate, precuneus, and left middle temporal gyrus. No significant associations were found between baseline NfL concentrations and longitudinal GMV changes, or connectivity changes in the SN or DMN for aSxC9.

FIGURE 4.

Baseline NfL concentrations are associated with longitudinal GMV declines and functional connectivity changes in C9orf72 expansion carriers. (A) In aSxC9, greater NfL concentrations at baseline are associated with longitudinal SMN connectivity decline in a right PostCG region, and both connectivity decreases (bilateral PCu and left temporal regions) and increases (right anterior THA) in the MPN. (B) A combined group of proC9 (n = 8) and SxC9 (n = 6) show that higher baseline NfL concentrations correlate with longitudinal DMN increases in the left ANG and longitudinal atrophy in the left PreCG and right MCC. Connectivity results are thresholded at p < 0.05 voxelwise, with cluster‐level FWE correction at p < 0.05, masked to the relevant network mask. Color bars represent z‐scores for association analyses with longitudinal connectivity (warm colors: longitudinal increases; cool colors: longitudinal decreases), or indicate regions with NfL‐associated GMV declines surviving multiple comparisons correction. Result maps are superimposed on the MNI brain template. ANG, angular gyrus; aSxC9, asymptomatic C9orf72 expansion carriers; DMN, default mode network; FC, functional connectivity; GMV, gray matter volume; L, left; MCC, midcingulate cortex; MPN, medial pulvinar network; MTG, middle temporal gyrus; NfL, plasma neurofilament light chain; PCu, precuneus; PostCG, postcentral gyrus; PreCG, precentral gyrus; proC9, prodromal C9orf72 expansion carriers; SMN, sensorimotor network; SxC9, symptomatic C9orf72 expansion carriers; THA, thalamus.

Given the limited NfL data for proC9 (n = 8) and SxC9 (n = 6) participants at baseline, we combined these groups for our analyses. In this proC9 and SxC9 combined group, higher baseline NfL concentrations were associated with longitudinal DMN connectivity increases within the left angular gyrus (Figure 4B). Higher baseline NfL concentrations were also associated with longitudinal GMV atrophy in the left precentral gyrus and right midcingulate cortex (Figure 4C).  No significant associations were observed between baseline NfL concentrations and longitudinal connectivity changes in the SN, SMN, or MPN for this group.

3.6. In prodromal and symptomatic carriers, longitudinal connectivity changes in the SN and SMN were associated with greater baseline symptom severity

Although proC9 lacked longitudinal connectivity changes (Figure 2) over the relatively limited average follow up of 2.1 years, we identified longitudinal connectivity changes in the SN and SMN that were associated with a measure of symptom severity (Figure 5A). Greater symptom severity (higher baseline CDR plus NACC FTLD sum of boxes scores) correlated with longitudinal SN connectivity declines in the dorsomedial prefrontal cortex and the supplementary motor area, and with longitudinal SMN increases in bilateral medial perirolandic regions.

FIGURE 5.

In prodromal and symptomatic carriers, greater symptom severity is associated with longitudinal connectivity changes. (A) In proC9, greater symptom severity (higher FTLD‐CDR‐SB scores) is associated with longitudinal SN decreases in the bilateral MPFC extending to SMA, and SMN increases in bilateral medial PRC. (B) In SxC9, higher scores correlate with longitudinal declines in the SMN (bilateral MCC and SMA), DMN (right MTG), and MPN (left PreCG and PostCG). Results are thresholded at p < 0.05 voxelwise, with cluster‐level FWE correction at p < 0.05, masked to the relevant network mask. Color bars represent z‐scores for association analyses with longitudinal connectivity, where warm colors indicate an association with longitudinal increases, whereas cool colors indicate an association with longitudinal decreases. Result maps are superimposed on the MNI brain template. DMN, default mode network; FTLD‐CDR‐SB, CDR Dementia Staging Instrument plus Behavior and Language domains from the National Alzheimer's Coordinating Center (NACC) Frontotemporal Lobar Degeneration Module, sum of boxes; GMV, gray matter volumes; L, left; MCC, midcingulate cortex; MPFC, medial prefrontal cortex; MPN, medial pulvinar network; MTG, middle temporal gyrus; PostCG, postcentral gyrus; PRC, perirolandic cortex; PreCG, precentral gyrus; proC9, prodromal C9orf72 expansion carriers; SMA, supplementary motor area; SMN, sensorimotor network; SN, salience network; SxC9, symptomatic C9orf72 expansion carriers.

For SxC9, greater symptom severity at baseline (higher CDR plus NACC FTLD sum of boxes scores) correlated with longitudinal connectivity decreases (Figure 5B). These included longitudinal decreases in the SMN in bilateral midcingulate cortex and supplementary motor area, in the DMN in the right middle temporal gyrus, and in the MPN in the left pre‐ and post‐central gyrus.

4. DISCUSSION

Longitudinal connectivity changes were evident as early as the asymptomatic stage despite a lack of detectable GM decline in aSxC9 and proC9, and limited atrophy change in SxC9. ProC9 showed regions of reduced baseline SN and SMN connectivity, and DMN enhancements. Although proC9 lacked group‐level longitudinal connectivity changes (vs HC), greater baseline symptom severity correlated with longitudinal SN and SMN connectivity changes. In addition, we identified associations between longitudinal connectivity changes and measures of neurodegeneration and symptomatic decline unique to each cohort. Higher baseline NfL was associated with longitudinal connectivity changes in the SMN and MPN in aSxC9, and GM decline and DMN connectivity increases in proC9/SxC9. Greater baseline symptom severity was associated with longitudinal connectivity changes in distinct ICNs for proC9 and for SxC9. These findings reveal stage‐specific longitudinal functional connectivity changes despite minimal GM atrophy declines, offering novel insights into network‐level changes underlying C9orf72‐related disease progression.

4.1. Despite lacking GM decline, asymptomatic carriers demonstrate longitudinal connectivity changes in C9orf72‐relevant networks that show associations with NfL

At baseline, aSxC9 showed SN, SMN, and MPN connectivity disruptions, consistent with previous findings in asymptomatic carriers. ² These early disruptions suggest vulnerability of networks involved in emotional salience processing (SN), ¹⁶ motor functioning (SMN), ¹⁹ and connectivity to the medial pulvinar, which participates in evaluating visual salience and in social cognition. ²² These alterations may represent a vulnerable substrate for the development of future behavioral and motor symptoms. Individuals with C9orf72‐bvFTD and asymptomatic carriers also show disruptions in these networks. ² , ²⁰ , ⁵² , ⁵³ , ⁵⁴ For the DMN, we found no differences in aSxC9 (vs HC), although our previous study identified DMN hypoconnectivity in asymptomatic carriers, ² but no differences in C9orf72‐bvFTD. ⁸ A possible explanation is that the aSxC9 in the present study may have included more participants closer to symptom onset, resulting in the absence of group differences in DMN connectivity.

Longitudinally, SN and MPN connectivity decreases were observed, aligning with a study of asymptomatic carriers showing regions of longitudinal SN and thalamic network connectivity decline. ⁹ SN disruptions in the anterior insula, a key network hub, were present at baseline on the left side and emerged longitudinally in the right insula, a region linked to lower memory performance in asymptomatic carriers. ⁵⁵ Longitudinal MPN connectivity increases were also detected, suggesting early and dynamic vulnerability in this network. We identified regions of DMN connectivity increases, in line with previous work showing DMN increases in asymptomatic carriers. ¹¹ Hyperconnectivity has been noted as an early response in various neurological diseases, including neurodegeneration, followed by diminishing connectivity as networks degrade with atrophy progression. ⁵⁶ , ⁵⁷ , ⁵⁸ In the present study, DMN hyperconnectivity persisted across proC9 and SxC9 stages, and higher DMN connectivity has been associated with greater behavioral symptoms and symptom severity in bvFTD. ⁸ , ⁵² These findings raise the possibility that increased DMN connectivity represents a compensatory response to disintegration of the remaining networks.

In aSxC9, connectivity changes were linked to markers of neurodegeneration. For example, higher baseline NfL correlated with longitudinal SMN and MPN connectivity changes. Together, these findings suggest that NfL‐associated network connectivity changes reflect network vulnerability and compensatory reorganization accompanying emerging subclinical neurodegeneration during the asymptomatic stage.

4.2. In prodromal carriers, SN and SMN changes point toward risk of further decline

Substantial challenges remain in identifying C9 at greatest risk of symptomatic conversion. The age at symptom onset and rate of progression are highly variable, even within families, ⁷ , ⁵⁹ and C9orf72‐bvFTD presents with insidious behavioral changes, ⁶⁰ challenging to quantify. Nevertheless, identifying prodromal carriers at risk for symptomatic progression is crucial for early treatment.

To our knowledge, this study is the first to assess network connectivity profiles during the prodromal stage. As in aSxC9, proC9 showed SN and SMN disruptions at baseline, suggesting ongoing vulnerability during this transitional stage. ProC9 lacked MPN alterations, contrasting with the regions of MPN hypoconnectivity in aSxC9. DMN hyperconnectivity in proC9 may reflect the continuation of asymptomatic‐stage DMN increases and a compensatory response to early neurodegeneration during the transition to symptomatic disease. Alternatively, this DMN hyperconnectivity may reflect pathological processes, such as a loss of network segregation, ⁶¹ aberrant excitatory signaling due to impaired inhibition–excitation balance, ⁶² or alterations in the physiological signal‐to‐noise ratio. ⁶³ Taken together, these patterns support the characterization of proC9 as an intermediate stage, characterized by the persistence of asymptomatic patterns alongside the emerging network breakdown of the symptomatic stage.

Although group‐level longitudinal connectivity changes were not observed in proC9, this may reflect limited statistical power due to the limited longitudinal sample size and follow‐up time. In contrast, we found that greater baseline symptom severity in proC9 correlated with longitudinal SN declines, suggesting that proC9 have heterogeneity with respect to symptom severity that may be obscured by group‐level comparisons. The SN declines appear driven by proC9 individuals with greater symptom severity, suggesting that declining SN reflects greater risk for clinical progression. Given SN's role in emotional processing, this connectivity decline may underlie subtle symptom emergence. Similarly, increases in SMN connectivity in proC9 may indicate an elevated risk of progression. Our data suggest a non‐linear SMN trajectory, with reduced connectivity across all stages, but increasing longitudinal SMN connectivity in SxC9, which may emerge as early as the proC9 stage in individuals with greater symptom severity. Longer‐term longitudinal studies spanning the prodromal to symptomatic transition are needed to elucidate these trajectories.

4.3. Symptomatic carriers predominantly have baseline hypoconnectivity with longitudinal connectivity increases

Consistent with our prior study, SxC9 had baseline hypoconnectivity within key SN and SMN regions. ⁸ Because greater symptom severity has been associated with lower SN connectivity ⁸ and longitudinal SMN decline in SxC9 in the present study, these baseline network alterations may reflect underlying clinical impairments. Notably, SxC9 had longitudinal SMN connectivity increases in perirolandic regions showing disrupted baseline connectivity. These findings contrast with a previous study reporting no such changes in carriers with predominantly ALS or FTD‐ALS, ¹¹ a discrepancy that may reflect the predominance of the bvFTD phenotype and longer follow‐up in our SxC9 cohort.

In SxC9, greater baseline symptom severity was associated with longitudinal SMN declines in bilateral midcingulate and supplementary motor areas, suggesting progressive SMN disruption in more advanced clinical stages. By contrast, SxC9 exhibited longitudinal SMN increases in perirolandic regions that minimally overlapped with areas showing SMN declines associated with greater baseline symptom severity. One potential explanation is that the group‐level SMN increases are driven by less impaired individuals, whereas those with greater baseline symptoms undergo subsequent SMN breakdown.

At baseline, SxC9 showed regions of DMN hyperconnectivity and hypoconnectivity. Although a previous study of C9orf72‐bvFTD found no DMN alterations, single‐subject analyses of early‐stage C9orf72‐bvFTD identified DMN hyperconnectivity, ⁸ suggesting that individual heterogeneity may contribute to discrepancies across group‐level studies. Baseline DMN hyperconnectivity in SxC9 and proC9 largely overlapped within the precuneus/posterior cingulate, key DMN hubs involved in memory and self‐referential processing. In contrast, greater baseline symptom severity was associated with longitudinal DMN decline in the right middle temporal gyrus. This parietal DMN hyperconnectivity may reflect a reactive, potentially compensatory response to neurodegeneration that becomes unsustainable with disease progression, consistent with the absence of longitudinal DMN increases.

The medial pulvinar represents a key site of degeneration in C9orf72‐bvFTD and ‐ALS. ¹⁴ , ²⁸ , ⁶⁴ , ⁶⁵ Reduced connectivity involving this region has been reported in C9orf72‐bvFTD and ‐ALS, ⁸ , ¹⁵ and asymptomatic carriers. ² Although SxC9 lacked baseline and longitudinal MPN connectivity differences, greater baseline symptom severity was associated with longitudinal MPN decline, suggesting that MPN disruption accompanies symptom progression.

In a combined proC9 and SxC9 group, higher baseline NfL correlated with precentral gyrus and midcingulate cortex decline, consistent with early degeneration in these regions. ⁶⁶ Moreover, higher baseline NfL was associated with longitudinal DMN connectivity increase in the angular gyrus, a potential reactionary response.

Consistent with previous studies, ¹¹ , ²⁵ , ²⁹ longitudinal GM declines were not detected in either aSxC9 or proC9. Slow atrophy progression has been reported in symptomatic carriers, ⁶⁷ , ⁶⁸ which is in line with the limited regions of atrophy decline we observed in SxC9. Although studies with longer follow‐up times are needed, the absence of extensive GMV decline in C9 suggests that these longitudinal functional connectivity changes are not secondary to atrophy. Instead, they likely represent earlier, more sensitive markers of progression before substantial atrophy emerges.

4.4. Baseline and longitudinal functional connectivity alterations across the C9orf72 clinical spectrum

Our findings support a model of stage‐dependent functional network alterations (Figure S3), where network‐level connectivity alterations precede GM changes and represent early disease markers. Asymptomatic carriers exhibit baseline disruptions alongside longitudinal connectivity increases and decreases, with some increases potentially reflecting adaptive or compensatory responses. During the prodromal stage, these disruptions persist alongside baseline connectivity increases, and longitudinal connectivity changes become associated with symptom severity. In the symptomatic stage, widespread network disruption predominates, accompanied by region‐specific increases, that may reflect reactive processes. These patterns broadly correspond to the functions of the affected networks, with early alterations in salience and sensorimotor systems (targeted in bvFTD and MND), and increasing DMN involvement across disease stages, likely representing interacting rather than discrete disease processes.

5. CONCLUSIONS AND LIMITATIONS

Our large cohort of C9 carriers enabled connectivity profiling unique to the asymptomatic, prodromal, and symptomatic stages. We identified stage‐specific associations between ICN changes and clinical measures of neurodegeneration and symptom severity, even in the absence of detectable longitudinal GM loss. These results demonstrate that functional connectivity profiling detects change across C9orf72 clinical stages, even with limited timepoints. Limitations include limited NfL data and a small C9orf72‐ALS subgroup, which precluded subgroup comparisons with C9orf72‐bvFTD. Future studies with larger cohorts and longer follow‐up are needed to define connectivity trajectories throughout the lifespan and in distinct clinical phenotypes.

CONFLICT OF INTEREST STATEMENT

C.U.O. has grant funds from Alector and Denali, and has served as consultant to Alector, Otsuka, Eli Lilly, Sanofi, Reata, and Neuvivo. J.C.R. reported serving as a site PI for clinical trials sponsored by Eli Lilly, Eisai, and Amylyx; and received consulting fees from Ferrer International, Adept Field Solutions, and REACH. J.S.Y. serves on the scientific advisory board for the Epstein Family Alzheimer's Research Collaboration and the Charleston Conference on Alzheimer's Disease and is the editor‐in‐chief of npj Dementia. B.F.B. has received honoraria for SAB activities for the Tau Consortium, funded by the Rainwater Charitable Foundation; and institutional research grant support for clinical trials from Alector, Transposon, Cognition Therapeutics, and EIP Pharma/Cervomed. A.L.B. has served as a paid consultant to Alector, Alexion, Arrowhead, Arvinas, Eli Lilly, Janssen, Merck, Neurocrine, Novartis, Oligomerix, Ono, Oscotec, Switch, and Transposon. He is a scientific cofounder of Neurovanda. His institution received research support from Biogen and Eisai for serving as a site investigator for clinical trials, as well as from Regeneron. K.K. consults for Biogen Inc., Eisai Inc., and BioArctic Inc. with no personal compensation; she received research material support form Eli Lilly. H.J.R. has received grant funding from the National Institutes of Health (NIH) and the CA department of public health, and is a consultant for Eli Lilly and Alector. B.L.M. serves on the Scientific Advisory Board of the Bluefield Project to Cure FTD, the John Douglas French Alzheimer's Foundation, Fundación Centro de Investigación Enfermedades Neurológicas, Madrid, Spain, Genworth, Inc., the Kissick Family Foundation, the Larry L. Hillblom Foundation, and the Tau Consortium; serves as a scientific advisor for the Arizona Alzheimer's Consortium and the Stanford University Alzheimer's Disease Research Center; receives royalties from Cambridge University Press, Elsevier, Inc., Guilford Publications, Inc., Johns Hopkins Press, Oxford University Press, and the Taylor & Francis Group; serves as editor for Neurocase and section editor for Frontiers in Neurology; and receives grants from the Bluefield Project to Cure FTD and NIH: P0544014, R01AG057234. L.Z., S.H., Y.J., M.L.M., D.L., J.L., M.L.G.‐T., V.E.S., J.K., L.K.F., H.W.H., E.M.R., W.W.S., T.M.F., and S.E.L. do not have any conflicts to disclose. Author disclosures are available in the Supporting Information.

CONSENT STATEMENT

This study was approved by institutional review boards and was performed in accordance with the Declaration of Helsinki. All participants or their surrogates provided informed consent.

Supporting information

Supporting Information

Supporting Information

Zhang L, Häkkinen S, Jung Y, et al. Longitudinal functional network connectivity changes across the clinical stages of C9orf72 hexanucleotide repeat expansion carriers. Alzheimer's Dement. 2026;22:e71744. 10.1002/alz.71744