Distinct Subtypes of Behavioral Variant Frontotemporal Dementia Based on Patterns of Network Degeneration

Importance  Clearer delineation of the phenotypic heterogeneity within behavioral variant frontotemporal dementia (bvFTD) will help uncover underlying biological mechanisms and improve clinicians’ ability to predict disease course and to design targeted management strategies.

Objective  To identify subtypes of bvFTD syndrome based on distinctive patterns of atrophy defined by selective vulnerability of specific functional networks targeted in bvFTD using statistical classification approaches.

Design, Setting and Participants  In this retrospective observational study, 90 patients meeting the Frontotemporal Dementia Consortium consensus criteria for bvFTD underwent evaluation at the Memory and Aging Center of the Department of Neurology at University of California, San Francisco. Patients underwent a multidisciplinary clinical evaluation, including clinical demographics, genetic testing, symptom evaluation, neurologic examination, neuropsychological bedside testing, and socioemotional assessments. All patients underwent structural magnetic resonance imaging at their earliest evaluation at the memory clinic. From each patient’s structural imaging scans, the mean volumes of 18 regions of interest (ROI) constituting the functional networks specifically vulnerable in bvFTD, including the salience network (SN), with key nodes in the frontoinsula and pregenual anterior cingulate, and the semantic appraisal network (SAN), anchored in the anterior temporal lobe and subgenual cingulate, were estimated. Principal component and cluster analyses of ROI volumes were used to identify patient clusters with anatomically distinct atrophy patterns. Data were collected from from June 19, 2002, to January 13, 2015.

Main Outcomes and Measures  Evaluation of brain morphology and other clinical features, including presenting symptoms, neurologic examination signs, neuropsychological performance, rate of dementia progression, and socioemotional function, in each patient cluster.

Results  Ninety patients (54 men [60%]; 36 women [40%]; mean [SD] age at evaluation, 55.1 [9.7] years) were included in the analysis. Four subgroups of patients with bvFTD with distinct anatomic patterns of network degeneration were identified, including 2 salience network–predominant subgroups (frontal/temporal [SN-FT] and frontal [SN-F]), a semantic appraisal network–predominant group (SAN), and a subcortical-predominant group. Subgroups demonstrated distinct patterns of cognitive, socioemotional, and motor symptoms, as well as genetic compositions and estimated rates of disease progression.

Conclusions and Relevance  Divergent patterns of vulnerability in specific functional network components make an important contribution to the clinical heterogeneity of bvFTD. The data-driven anatomic classification identifies biologically meaningful anatomic phenotypes and provides a replicable approach to disambiguate the bvFTD syndrome.

Behavioral variant frontotemporal dementia (bvFTD) is a clinical syndrome associated with frontal-predominant neurodegeneration.^1-3 Despite numerous detailed accounts, no systematic understanding of the nature, frequency, and mechanisms of bvFTD clinical symptoms has arisen, in part owing to tremendous diversity across patients.

The recent realization that unique functional neural networks are selectively vulnerable in bvFTD has fueled new insights into the syndrome’s clinical and pathologic diversity.^4-7 Evidence suggests that along with other networks, bvFTD targets intrinsic networks responsible for social-emotional-autonomic processing (the salience network [SN]) and semantically driven personal evaluation (the limbic or semantic appraisal network [SAN]).^8-10 The SN includes the frontoinsula and pregenual anterior cingulate structures and is closely allied with the SAN, which includes the temporal pole, ventral striatum, subgenual cingulate, and basolateral amygdala. We hypothesized that patterns of regional vulnerability of brain structures that undergo network-specific degeneration may reveal distinct subtypes within the bvFTD syndrome. Using networks as an organizational schema may reveal clinical phenotypes generated by differentially disrupted functional architecture and may provide specific therapeutic targets for pharmacologic and behavioral treatments.

We examined the frequency of early clinical symptoms in a large sample of patients with bvFTD (n = 90) and report the key features of their initial clinical presentation. We used statistical classification approaches to identify subgroups of patients with different patterns of gray matter loss in specific regions of interest (ROIs) in the SN and SAN and report distinct symptom profiles of each subgroup.

Box Section Ref ID

Ninety patients met the International Behavioral Variant FTD Criteria Consortium (FTDC) revised guidelines for the diagnosis of bvFTD¹¹ and underwent evaluation at the University of California, San Francisco (UCSF), Memory and Aging Center (MAC). All patients underwent a complete clinical history, physical examination, neuropsychological evaluation, and structural brain imaging. The diagnosis was made at a multidisciplinary consensus meeting for each patient individually. A retrospective medical record review fulfilled the FTDC criteria for a subset of patients (n = 13) who received an initial diagnosis using the Neary criteria.¹ Eligibility criteria for age-matched healthy participants (n = 44) who contributed as imaging controls included normal cognitive performance, normal structural brain imaging results, and absence of neurologic, psychiatric, or other major medical illnesses. Written informed consent was obtained from all participants or their assigned surrogate decision makers. The study was approved by the UCSF institutional review board for human research.

Data were collected from June 19, 2002, to January 13, 2015. The Clinical Dementia Rating (CDR) (range, 0-3, with higher scores indicating more disability), CDR Sum of Boxes (CDR-SOB) (range, 1 to 18, with higher scores indicating more disability),¹² and frontotemporal lobar degeneration (FTLD)–modified CDR-SOB (range, 0 to 24, with higher scores indicating more disability)¹³ scales were completed via a structured caregiver interview, and the Mini Mental State Examination (MMSE) score (range, 0-30, with higher scores indicating better functioning)¹⁴ was recorded for each patient at the first presentation to the MAC. Neuropsychological, socioemotional, and neurologic examinations were completed within 90 days of structural brain imaging (eMethods in the Supplement). Socioemotional tests included The Awareness of Social Inference Test,¹⁵ UCSF Cognitive Theory of Mind,¹⁶ Interpersonal Reactivity Index,^17,18 and Revised Interpersonal Adjective Scales.^19,20 The National Alzheimer Coordinating Center symptom checklist (https://www.alz.washington.edu/WEB/nacc_handbook.html) to identify specific behavioral deficits was completed for 81 patients within 12 months of the first presentation to the MAC.

Genetic analysis identified the presence of FTD-causing mutations in the C9orf72 (NCBI Entrez Gene 203228), microtubule-associated protein tau (MAPT [NCBI Entrez Gene 4137]), and progranulin (GRN [NCBI Entrez Gene 2896]) genes.²¹ Six patients chose not to undergo genetic testing. Among the patients with no FTD-causing mutation, the frequency of GRN and MAPT allele distributions were compared with normal population data from the National Center for Biotechnology database. Twenty-four patients underwent autopsy and neuropathologic assessment (eMethods in the Supplement).

To examine the effect of disease severity on symptoms, we used the data from follow-up evaluations at the MAC to generate a cross-sectional model to categorize patients according to their FTLD-modified CDR-SOB score.¹³ Based on previous evidence of an approximately 3.5-point annual gain in the CDR-SOB score among patients with bvFTD,²² we identified the following 4 levels of disease severity for the FTLD-modified CDR-SOB: 4.0 to 7.5, 8.0 to 11.5, 12.0 to 15.5, and 16.0 or higher. We semirandomly assigned patients into roughly equal sets of 27, 25, 23, and 15 patients for these respective sets.

All patients underwent a unified structural magnetic resonance imaging acquisition protocol (1.5-, 3-, or 4-T) and were included in gray matter volume assessment (eMethods in the Supplement) using the voxel-based morphometry toolbox of Statistical Parametric Mapping (SPM8; http://www.fil.ion.ucl.ac.uk/spm/software/spm8/). We used the Neuromorphometrics, Inc, brain atlas (http://www.neuromorphometrics.com) to define ROIs uniquely contributing to 4 potentially bvFTD-affected networks, including the SN, SAN, default mode,^23,24 and frontoparietal network.^25,26 Initial analyses showed no statistical benefit of default-mode and frontoparietal structures; hence, the final model included only the 18 frontal and temporal ROIs constituting the SN and SAN (eMethods in the Supplement). These ROIs included right and left ROIs from the temporal poles, gyrus recti, subcallosal areas, anterior cingulate gyri, anterior insulae, basal forebrains, frontal operculae, posterior orbital gyri, and amygdalae. We estimated the mean gray matter volume of each ROI for each patient and used a cluster analysis based on Euclidean distance to derive 4 clusters of patients with bvFTD (eMethods in the Supplement). Using the same ROIs, we performed a principal component analysis (PCA) to demonstrate the association between patient clusters and SN and SAN volume loss. We mapped the ROIs for each patient onto the Euclidean space based on the first and second components of the PCA, keeping their cluster identity. We modeled the analysis to identify the greatest possible number of distinct clusters of patients with bvFTD that remained meaningfully separated when mapped onto dimensional space, with the result that the 4-cluster solution was the best fit (eFigure in the Supplement). We used a polytomous logistic regression model to determine the stability of the clusters (eMethods in the Supplement). The voxel-based morphometry–derived gray matter volumes of the 4 bvFTD clusters were compared with the age-matched control population (mean [SD] age, 62.7 [1.5] years). The PCA, cluster, and logistic regression analyses used the MatLab statistical toolbox (MathWorks). Statistical comparisons of clinical-behavioral measures between subgroups were performed using SAS (version 9.4; SAS Institute Inc).

Ninety patients (54 men [60%]; 36 women [40%]; mean [SD] age at evaluation, 55.1 [9.7] years) were included in the analysis. Patients with bvFTD presented to the MAC at a mean (SD) age of 61.1 (8.2) years (eTable in the Supplement). Initial clinical assessment within 12 months of the first presentation revealed an MMSE score of 23, a CDR-SOB score of 6.9, and an FTLD-modified CDR-SOB score of 9.2 (eTable in the Supplement). Of the 84 patients who underwent genetic assessment, 24 (29%) showed a genetic mutation. C9orf72 accounted for 14 mutations (58%). The 60 patients with mutation-negative bvFTD showed a significantly higher frequency of the GRN T allele (rs5848; P = .02).

Each core symptom of FTDC criteria showed frequency of more than 70%, although no symptom had a frequency of 100%, highlighting the clinical heterogeneity of bvFTD (Figure 1A). Sixty-seven patients with bvFTD (74%) presented with behavioral abnormalities as their first symptom (Figure 1B), whereas the rest presented with executive (16 [18%]), memory (3 [3%]), language (3 [3%]), or motor (1 [1%]) deficits as their first symptom. Visuospatial, sensory, or constitutional deficits were not reported as first symptoms. Based on National Alzheimer Coordinating Center symptom categories, personality change was the most common early behavioral symptom (Figure 1C). Apathy, emotional blunting, disinhibition, obsessive behavior, and altered eating habits were the next most common behavioral symptoms, respectively (Figure 1C). Violent behaviors and hallucinations were the least prevalent symptoms. Frequency of symptoms in all domains increased as the disease progressed (Figure 1D). All 90 patients reported behavioral and executive deficits by the point of moderate severity (ie, FTLD-modified CDR-SOB scores 12.0-15.5). Although never reported as first symptoms, visuospatial and constitutional symptoms were reported in a subset of patients at all stages of severity (Figure 1D).

We next tested the hypothesis that subsets of patients within the bvFTD syndrome would show distinct patterns of atrophy^27-29 based on vulnerability of brain structures that undergo network-specific degeneration in bvFTD. More specifically, we predicted that unique subgroups of patients would be identified based on volumetric differences in SN and SAN structures.^7-9,30 We identified 4 distinct patient clusters (based on 18 ROIs). A polytomous regression model predicted the assigned cluster with more than 0.80 probability for 99% of the patients and confirmed cluster stability. Keeping their cluster identity, we mapped each patient’s data onto PCA-derived dimensional space based on the same ROIs (Figure 2A). Three of the 4 subgroups showed a distinct mapping along the first dimension (Figure 2A–top panel, orange, blue, and green subgroups). Both SN and SAN volumes contributed strongly to first dimension of the PCA (Figure 2A–bottom panel [gray-scale gradient along x-axes]), indicating that the first dimension primarily predicts the overall degree of cortical atrophy. The second dimension of the PCA more distinctly predicted the SAN volume (Figure 2A–bottom panel), and the cluster analysis identified a subset of patients with high SAN and intermediate SN volume loss (pink subgroup). Both the o and blue subgroups showed a higher degree of SN volume loss (Figure 2A).

The voxel-based morphometry analysis of each subgroup compared with an age-matched control group demonstrated distinct patterns of atrophy (Figure 2B). Three of the 4 clusters showed significant cortical atrophy (Figure 2A-B). One group showed both frontal and temporal gray matter loss with significant subcortical involvement (orange dots, labeled salience network frontal/temporal predominant [SN-FT]), whereas the second group showed predominantly frontal and subcortical involvement (blue group, labeled salience network frontal predominant [SN-F]). The third cluster, which included patients with a high degree of atrophy to SAN structures, showed atrophy localized to the right temporal lobe, with only modest subcortical involvement (pink group, labeled semantic appraisal network [SAN]). Finally, the fourth subset of patients showed minimal cortical and predominantly subcortical atrophy (green dots, labeled subcortical).

We found no differences between subgroups in age at evaluation, disease duration, and depression (Table). The SN-FT subgroup had significantly higher dementia rating scores (ie, CDR and FTLD-modified CDR-SOB) and poorer MMSE scores compared with the subcortical group, whereas all other pairwise comparisons on symptom severity were equivalent.

The subgroups showed differences in their distribution of mutation carriers and of primary neuropathologic features. C9orf72 mutations were found in all subgroups, constituting most of the mutations (8 of 9 [88%]) found in the subcortical group, 2 of 5 mutations (40%) in the SN-FT subgroup, 3 of 7 mutations (43%) in the SN-F subgroup, and 1 of 3 mutations (33%) in the SAN subgroup (Figure 2C). The remaining 3 patients (60%) in the SN-FT group had exclusively MAPT mutations, whereas the remaining 4 (57%) in the SN-F group had exclusively GRN mutations (Figure 2C). All subgroups (n = 24) showed both tau and transactive response DNA-binding protein 43 (TDP) neuropathologic features (Figure 2D). Five of 6 patients (83%) in the subcortical group showed TDP, and the remaining patient had argyrophilic grain disease. All 3 patients with corticobasal degeneration belonged to the SN-F group. Pick disease was found in 2 patients, 1 in the SN-FT and 1 in the SAN subgroups. These results suggest that patients with bvFTD who have predominantly subcortical presentations have a higher likelihood of TDP pathologic features. Patients in the subcortical subgroup, compared with the other subgroups, also showed a trend toward slower disease progression (Figure 3) (P = .07).

High frequencies of core FTDC symptoms were found in patients in all 4 subgroups. The overall heterogeneity of symptom profiles across all patients with bvFTD was partly explained by specific within-subgroup patterns; for example, the frequency of disinhibition was 100% in the SAN subgroup, with lower rates in other subgroups, and the frequency of hyperorality was 100% in the SN-FT subgroup, with less than 80% in the other subgroups (Figure 4A). Patients in the SN-F subgroup had the highest frequency of altered neuropsychological performance (Figure 4A) and the highest rate of dyexecutive first symptoms (Figure 4B). Five of 9 patients in the SAN subgroup (60%) showed early obsessive behaviors, although this proportion was less than 25% in other groups (Figure 4C).

Neurologic examination showed different rates of signs across subgroups; however, only abnormal gait reached statistical significance (Table) (SN-FT subgroup, 0; SN-F subgroup, 10 of 27 [37%]; SAN subgroup, 2 of 8 [25%]; and subcortical subgroup, 5 of 30 [17%]; P = .01 between the SF-FT and SN-F subgroups). In neuropsychological bedside testing, all subgroups showed clinical impairments in aspects of executive, visuoconstruction, memory, and language testing. The SN-F subgroup showed worse executive function with significantly poorer digit-span-backward and lexical fluency abilities in pairwise comparisons (Table). The SN-FT subgroup showed the most predominant memory dysfunction and significantly better syntax comprehension in pairwise comparisons (Table).

Socioemotional testing also revealed distinct clinical impairments across subgroups. In the SAN subgroup, 4 of 4 (100%) were impaired at sarcasm detection, while only 8 of 29 (28%) were impaired in the subcortical subgroup (Figure 4D). The SAN subgroup showed significantly low scores in sarcasm detection compared with the subcortical subgroup in pairwise comparisons (Table). As a whole, the SAN subgroup was in the normal range for empathy and interpersonal warmth (Table), whereas the SN-FT and subcortical subgroups showed diminished empathic perspective taking, empathic concern, and warmth. Only 10 of 17 patients in the subcortical subgroup (59%) were impaired on complex social cognition, whereas 100% of patients in all other subgroups were impaired (Figure 4D).

We characterized a large cohort of patients with bvFTD with detailed clinical and genetic evaluations and structural brain imaging. When patients were clustered into subgroups based on quantification of atrophy patterns, distinct anatomic profiles emerged, each with substantial overlap within the realm of bvFTD symptoms, but also with notably divergent patterns of neuropsychological, socioemotional, motor, and other clinical symptoms. Beginning with anatomic subgroups provided greater precision in understanding each patient’s clinical features in the context of striking heterogeneity seen across the bvFTD group as a whole.

Our results underscored the diversity of the bvFTD syndrome. None of the FTDC criteria reached 100% frequency, indicating that no single symptom is perfectly sensitive and specific to bvFTD. Nearly one-quarter of our sample presented with first symptoms that were not behavioral (ie, executive, memory, language, or motor), suggesting that a nonbehavioral first symptom cannot be taken as evidence that bvFTD will not emerge as the dominant clinical syndrome. Motor, visuospatial, and constitutional symptoms, although considered atypical for bvFTD, were observed in up to one-half of patients. Despite the heterogeneity, however, personality and socioemotional changes dominated throughout the disease, highlighting the importance of standardized tools to quantify these deficits in bvFTD.

The discovery of intrinsic brain connectivity patterns has led to delineation of specific functional networks distributed in frontal and temporal lobes that are selectively vulnerable in bvFTD. Previous attempts have been made to identify subtypes of bvFTD based on clinical symptoms.^2,43-45 Whitwell and colleagues²⁷ performed a large study using statistical classification approaches based on regional anatomy, specifically 26 volumetric ROIs representing all structures in the frontal, temporal, and parietal lobes, and suggested that bvFTD could be subdivided into different anatomic subtypes. Herein, we extend these findings by demonstrating that volume loss in specific vulnerable functional networks alone accounts for most of the meaningful anatomic variance in bvFTD. Although in our initial analytic approach we included additional network structures from frontoparietal and default-mode networks, we found that 18 volumetric ROIs from SN and SAN were adequate to classify patients into 4 subgroups. The SN-F and SN-FT subgroups paralleled the frontal-predominant and frontal/temporal-predominant subgroups, respectively, identified by Whitwell et al.²⁷ The SAN subgroup was comparable to the temporal-dominant group of Whitwell et al, although, as opposed to the largely bitemporal atrophy distribution they identified, our network-based approach identified a subset of patients with predominantly right temporal disease.

We also identified a novel subgroup, representing 32 patients (36% of our cohort), who have minimal cortical atrophy (ie, subcortical). This finding is relevant to the debate around slowly progressive bvFTD or bvFTD phenocopy.^45-47 Previous studies^45-47 have reported a significant fraction of patients with minimal cortical atrophy and slow progression, yet otherwise having a clear bvFTD syndrome. Some investigators⁴⁵ have suggested that these patients may not have an underlying neurodegenerative disease. Our subcortical subgroup showed at least 70% frequency in each core diagnostic symptom, whereas nearly 30% carried an FTD-related mutation, and the 6 autopsy diagnoses were 100% FTLD. This subgroup also included patients with the slowest progression. These results confirm that the subcortical subtype is a true neurodegenerative phenotype and the most likely degenerative pattern in patients with slowly progressive bvFTD. The minimal cortical atrophy suggests that, in these patients, the typical bvFTD syndrome may result from disconnection of major network hubs from their striatal or thalamic partners.⁴⁸

Identifying distinct subgroups within the bvFTD syndrome provides a neural substrate explaining much of the diversity reported in previous studies. The statistical division of patients with clear SN atrophy into 2 separate subgroups (SN-F and SN-FT) emphasizes that the presence of subgenual cingulate and anterior temporal atrophy in addition to SN damage provides a meaningful anatomic differentiation of patients with bvFTD. Although one might initially presume that the SN-FT subgroup simply represents the SN-F at a more advanced stage of disease severity, this hypothesis is contradicted by the clinical data, where patients in the SN-F subgroup showed worse executive abilities and more abnormalities on the neurologic examination, particularly in gait and coordination, compared with patients with SN-FT subgroup. The genetic and pathologic probabilities were also distinct between the SN-F and SN-FT subgroups. This divergence likely results from differences in multifocal onset seeding and progression of neurodegeneration in each anatomic subgroup, reflecting the complex patterns of selective vulnerability known to characterize FTD. The resulting patterns of unique functional disruptions may provide neural substrates for distinctive outcomes.

The SAN subgroup, although consisting of fewer patients, showed very distinct behavioral characteristics. All patients in the SAN group showed behavioral disinhibition, highlighting the centrality of right SAN–temporal lobe contributions to socioemotional sensitivity and maintenance of social decorum. The high frequency of obsessive behaviors may reflect disruption in patients’ ability to value objects of interest, which is consistent with the dysfunctional rostral caudate and medial orbitofrontal nodes of the SAN.^49,50 Consistent with the central role of right temporal structures in person perception, the SAN subgroup also showed striking impairments in paralinguistic sarcasm detection (ie, voice prosody, facial expression).⁵¹ This finding suggests that SAN disruption may play a specific role in the difficulty of patients with bvFTD in translating auditory and visual signals into meaningful social interactions. Preserved interpersonal warmth and empathic concern among patients in the SAN subgroup indicates the critical role of their relatively intact frontal brain structures in these behaviors.^52,53 Despite the focality of atrophy in the SAN subgroup, they were neuropathologically diverse, consisting of patients with TDP-B, Pick disease, and FTDP-17 tau pathologic features and included patients with C9orf72, GRN, and MAPT mutations.

Although the UCSF MAC has a high diagnostic accuracy rate, with clinical diagnosis of bvFTD predicting FTLD pathology in more than 95% of cases,¹¹ future studies must corroborate these results in a clinicopathologically diagnosed cohort and in longitudinal studies with prospective follow-up. The differences among cluster groups were subtle, and substantial symptom overlap across subgroups remained that reflected the core clinical features of the bvFTD syndrome as a whole. However, this study suggests that classification of patients with bvFTD by their patterns of volume loss in key networks may yield incremental increases in clarity in predicting patients’ specific cognitive, behavioral, and motor symptoms and rate of disease progression. The identification of a subset of patients with bvFTD with primarily subcortical atrophy who are more likely to have TDP neuropathologic features and C9orf72 mutations suggests this approach also has modest utility for prediction of pathological molecules.

Corresponding Author: Katherine P. Rankin, PhD, Memory and Aging Center, Department of Neurology, University of California, San Francisco, 675 Nelson Rising Ln, Ste 190, San Francisco, CA 94158 (krankin@memory.ucsf.edu).

Accepted for Publication: May 4, 2016.

Published Online: July 18, 2016. doi:10.1001/jamaneurol.2016.2016.

Author Contributions: Drs Ranasinghe and Rankin had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Ranasinghe, Rankin, Kramer, B. Miller.

Acquisition, analysis, or interpretation of data: Ranasinghe, Rankin, Pressman, Perry, Lobach, Seeley, Coppola, Karydas, Grinberg, Shany-Ur, Lee, Rabinovici, Rosen, Gorno-Tempini, Boxer, Z. Miller, Chiong, DeMay, Possin, Sturm, Bettcher, Neylan, Zackey, Nguyen, Ketelle, Block, Wu, Dallich, Russek, Caplan, Geschwind, Vossel.

Drafting of the manuscript: Ranasinghe, Rankin.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Ranasinghe, Rankin, Lobach.

Obtained funding: Ranasinghe, Rankin, Seeley, Vossel, B. Miller.

Administrative, technical, or material support: Seeley, Karydas, Grinberg, Shany-Ur, Boxer, Z. Miller, Possin, Bettcher, Neylan, Zackey, Nguyen, Ketelle, Block, Wu, Dallich, Russek, Caplan, Geschwind.

Study supervision: Rankin, Lobach, Coppola, Gorno-Tempini, Kramer, Possin, B. Miller.

Conflict of Interest Disclosures: Dr Rabinovici reports receiving research support from Avid Radiopharamceuticals and speaking honoraria from GE Healthcare and Piramal Imaging. Dr B. Miller reports receiving grant support from the National Institute on Aging, National Institutes of Health (NIH), and the Centers for Medicare & Medicaid Services as grants for the Memory and Aging Center and serving as medical director for the John Douglas French Foundation, scientific director for the Tau Consortium, director of the medical advisory board of the Larry L. Hillblom Foundation, and scientific advisory board member for the National Institute for Health Research Cambridge Biomedical Research Centre and its subunit, the Biomedical Research Unit in Dementia. No other disclosures were reported.

Funding/Support: This study was supported by grants P01AG019724 (Dr B. Miller), P50AG02350 (Dr B. Miller), R01AG029577 (Dr Rankin), K23AG021606 (Dr Rankin), K23 AG038357 (Dr Vossel), K23AG040127 (Dr Sturm), K23 AG042492-01 (Dr Bettcher), K23 AG039414 (Dr Lee), and K23 AG045289 (Dr Perry) from the NIH; grant PCTRB-13-288476 from the Alzheimer’s Association made possible by Part the Cloud^TM (Dr Vossel); grants 2013-A-029-SUP, 2007/2I, 2015-A-034-FEL (Dr Ranasinghe), and 2002/2J (Dr Rankin) from the Larry L. Hillblom Foundation, Inc; and grants from the J. D. French Alzheimer’s Foundation (Dr Vossel), the Rainwater Charitable Foundation (Dr Seeley), and the Bluefield Project (Dr Seeley).

Role of the Funder/Sponsor: The funding sources had no role in the design and conduct of the study; collection, management, analysis, or interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Additional Contributions: We thank the patients, caregivers, and healthy controls for participation in this research. Babu Adhimoolam, MBBS, consulted on the voxel-based morphometry analysis. Jennifer Yokoyama, PhD, consulted on the statistical analysis of genetic data. Neither received compensation for their contributions.