Genetic predictors of survival in behavioral variant frontotemporal degeneration

Abstract

Objective

To determine autosomal dominant genetic predictors of survival in individuals with behavioral variant frontotemporal degeneration (bvFTD).

Methods

A retrospective chart review of 174 cases with a clinical phenotype of bvFTD but no associated elementary neurologic features was performed, with diagnosis either autopsy-confirmed (n = 57) or supported by CSF evidence of non-Alzheimer pathology (n = 117). Genetic analysis of the 3 most common genes with pathogenic autosomal dominant mutations associated with frontotemporal degeneration was performed in all patients, which identified cases with C9orf72 expansion (n = 28), progranulin (GRN) mutation (n = 12), and microtubule-associated protein tau (MAPT) mutation (n = 10). Cox proportional hazards regressions were used to test for associations between survival and mutation status, sex, age at symptom onset, and education.

Results

Across all patients with bvFTD, the presence of a disease-associated pathogenic mutation was associated with shortened survival (hazard ratio [HR] 2.164, 95% confidence interval [CI] 1.391, 3.368). In separate models, a GRN mutation (HR 2.423, 95% CI 1.237, 4.744), MAPT mutation (HR 8.056, 95% CI 2.938, 22.092), and C9orf72 expansion (HR 1.832, 95% CI 1.034, 3.244) were each individually associated with shorter survival relative to sporadic bvFTD. A mutation on the MAPT gene results in an earlier age at onset than a C9orf72 expansion or mutation on the GRN gene (p = 0.016).

Conclusions

Our findings suggest that autosomal dominantly inherited mutations, modulated by age at symptom onset, associate with shorter survival among patients with bvFTD. We suggest that clinical trials and clinical management should consider mutation status and age at onset when evaluating disease progression.

Frontotemporal degeneration (FTD) is a progressive neurodegenerative condition that affects the frontal and temporal lobes of the brain and results in progressive deterioration in executive functioning, language, and social comportment that ultimately leads to death. There is a wide range of variability in survival in FTD, from 3 to 14 years (for review, see reference 1). Studies of survival in FTD spectrum disorders suggest there are multiple factors that contribute to this variability, including clinical phenotype and genetic background, but these studies have been mostly limited to phenotypically diverse cohorts of FTD.² In this present study, we sought to examine genetic factors associated with survival in a well-characterized cohort of individuals with behavioral variant frontotemporal degeneration (bvFTD). bvFTD is the most common FTD syndrome and is genetically heterogeneous, therefore predictors of survival in this FTD phenotype are of high clinical importance.

bvFTD is associated with an autosomal dominant pattern of inheritance in about 15%–20% of cases.^3,4 Mutations in the microtubule-associated protein tau gene (MAPT), progranulin gene (GRN), or a hexanucleotide repeat expansion of C9orf72 are the most common types of pathogenic mutations in bvFTD and are highly penetrant.^5–7 The presence of any pathogenic mutation may be associated with a shortened survival in FTD compared to sporadic cases²; however, individual effects on survival for each disease-associated gene in bvFTD cases have not yet been reported. Hereditary causes of FTD may be associated with distinct neuroanatomical patterns of disease compared to sporadic cases, and thus may have differing effects on survival.⁸ Therefore, it is important to consider each gene individually. The aim of this article is to examine survival in the most common genetic causes of hereditary FTD in well-characterized bvFTD cases. We used a statistical analysis method that appropriately accounts for both censoring and truncation in the data,⁹ and allows us to rigorously evaluate survival in a cohort that includes individuals who have not yet died.

Methods

Participants

We performed a retrospective evaluation of 174 patients with a clinical diagnosis of bvFTD. All patients were initially evaluated at the University of Pennsylvania Medical Center by experienced cognitive neurologists (M.G., D.J.I.) using published consensus criteria^10,11 subsequently confirmed by a multidisciplinary consensus committee based on history and neurologic examination including a detailed mental status evaluation. Further inclusion criteria included a postmortem neuropathologic diagnosis of tau or TDP-43 inclusions consistent with frontotemporal lobar degeneration (FTLD)¹² performed by a board-certified neuropathologist (E.B.L., J.Q.T.) using a previously reported procedure.^13,14 For those patients who met criteria for probable bvFTD¹⁰ where we did not have neuropathologic confirmation of FTLD, we used an autopsy-validated CSF tau: β-amyloid (Aβ) cutoff of <0.34, which has >95% accuracy for identifying non-Alzheimer pathology.¹³ Patients with a concurrent clinical diagnosis of motor neuron disease, corticobasal degeneration, or progressive supranuclear palsy were excluded because of their known shorter lifespan associated with motor factors.^1,8,15 Medical and psychiatric causes of dementia were excluded by clinical examination and blood and brain imaging tests. Years of education was recorded for each patient. Table 1 summarizes the demographic features of these patients.

Table 1.

Patient demographics by mutation status

Genetic screening

All 174 patients with bvFTD were evaluated for a pathogenic GRN mutation, MAPT mutation, or C9orf72 repeat expansion. Briefly, DNA was extracted from peripheral blood or frozen brain tissue following the manufacturer’s protocols (Flexigene [Qiagen] or QuickGene DNA whole blood kit [Autogen] for blood, and QIAsymphony DNA Mini Kit [Qiagen] for brain tissue).

All patients were tested for C9orf72 hexanucleotide repeat expansions using a modified repeat-primed PCR as previously described.¹⁶ Evaluation for mutations in neurodegenerative disease–associated genes, including GRN and MAPT, was performed using a custom-targeted next-generation sequencing panel (MiND-Seq)¹⁷ as previously described. A pathogenic mutation was identified in 50 of the 174 cases (28.7%) with 12 (6.9%) in GRN, 10 (5.7%) in MAPT, and 28 (16.1%) in C9orf72 (see table 2 for a summary of mutations). The remaining 124 bvFTD cases were negative for mutations in all 3 genes and therefore defined as sporadic bvFTD. Of these sporadic bvFTD cases, the risk of having a genetic mutation was evaluated using a validated pedigree system⁴ in 72 cases with pedigrees available for classification; 16 cases were classified as medium risk (30% mutation detection rate), 20 as low risk (12.5% detection rate), and 52 as unknown significance (4.6% detection rate).

Table 2.

Summary of mutation details in analysis cohort

Standard protocol approvals, registrations, and patient consents

All participants completed a written informed consent procedure in accordance with the Declaration of Helsinki and approved by an institutional review board convened at the University of Pennsylvania.

Statistical analysis methods

Continuous variables are summarized with mean and SD, and a 2-sample t test was performed for each to assess differences in demographics between mutation and sporadic cases (table 1). Categorical variables are summarized with percentages, and a χ² test for each determined the presence of an association between each categorical variable and diagnosis method or mutation status. All statistical analyses were 2-sided. Statistical significance was set at <0.05. Kaplan-Meier curves¹⁸ were used to summarize postsymptom onset survival by mutation status (figure).

Figure. Survival among the 3 gene mutation types and sporadic cases.

(A) Survival among C9orf72, GRN, and MAPT cases vs sporadic cases. (B) Survival among C9orf72 cases vs sporadic cases. (C) Survival among GRN cases vs sporadic cases. (D) Survival among MAPT cases vs sporadic cases. *Number at risk in sporadic group at specific time on horizontal axis. **Number at risk in mutation group at specified time on horizontal axis.

To examine the association between mutation status and survival, Cox¹⁹ proportional hazards models were used. In order to reduce bias in survival models, it is important to include censored data (i.e., patients who are not yet deceased) because exclusion of these patients will skew the resulting hazard ratio (HR) estimates by ignoring patients with longer survival.⁹ We also acknowledge the possibility that patients with very short survival may not live long enough to enter the study. To account for this, we considered the data to be left-truncated at the date of initial visit, with survival measured from date of symptom onset. Age at disease onset was estimated based on family report of the earliest persistently abnormal clinical feature in the domain of social function, personality change, or executive functioning. We adjusted for this left truncation in the Cox model.⁹ Ninety-two of the 174 patients were deceased, and the remaining 82 patients were considered to be right-censored in the model. For each of these patients, the censoring date was taken to be the latest date at which the patient was confirmed not deceased.

Survival models were performed using mutation status, education, sex, and age at symptom onset. An overall model assessed the effect of the presence of any mutation on survival. Following this, a separate model evaluated survival in each of the 3 mutations. To account for the possibility that mutation affects survival differently by education, sex, or age at onset, interaction terms between mutation status and these 3 predictors were assessed for statistical significance and removed from the model if no significance was found. We tested the assumption of proportional hazards,²⁰ and found no violations.

Data availability

De-identified data necessary to replicate the proposed analyses are available to other parties upon an approved request to the corresponding author.

Results

There was no difference between sex and education between mutation and sporadic groups. Age at onset and disease duration differed between mutation and sporadic groups. Mean age at onset was 56.0 years in mutation cases and 59.3 years in sporadic cases (t = 2.500, df = 84, p = 0.014). Mean disease duration, which, unlike survival, does not account for whether a participant died or was lost to follow-up, was 7.3 years in mutation cases and 8.6 years in sporadic cases (t = 2.127, df = 111, p = 0.036). These results are summarized in table 1. Median survival, which is the time at which survival is 50% and distinguishes between deceased and censored cases, among C9orf72-positive patients was 7.0 years (95% confidence interval 3.1, 11.2); median survival with a GRN mutation was 6.2 years (one-sided lower 95% confidence limit 4.0); median survival with a MAPT mutation was 6.9 years (1-sided lower confidence limit 4.0). Finally, sporadic patients had a median survival of 10.9 years (95% confidence interval 9.4, 12.1). The figure summarizes survival by the Kaplan-Meier method.

In Cox models, the presence of having any mutation was found to have a significant association with reduced survival (HR 2.164, p < 0.001). Survival in GRN mutation cases was significantly reduced compared to sporadic cases (HR 2.423, p = 0.010); the same was found for C9orf72 (HR 1.832, p = 0.038). These results are summarized in table 3. Finally, due to the significant interaction term between MAPT status and age at onset (p < 0.001), the effect of MAPT on survival is dependent on age at onset. Therefore, the HR for MAPT must be reported for a specific age at onset (table 3)⁹; MAPT patients with a younger age at onset had a shorter survival (HR 0.892, p = 0.013). There were no significant differences in survival between mutation types.

Table 3.

Results of Cox proportional hazards models for mutation status, C9orf72, and GRN

Among the mutation cases, we found that mean age at onset differs by mutation (F = 4.548, df = 2, 47, p = 0.016), with the youngest age at onset in the MAPT cases. Pairwise comparison tests with a Holm adjustment revealed that the mean age at onset in MAPT is significantly different from both C9orf72 (p = 0.025) and GRN (p = 0.020). C9orf72 and GRN do not differ from each other (p = 0.465). Later age at onset reduced survival in GRN and C9orf72 (GRN: HR 1.078, p < 0.001; C9orf72: HR 1.056, p = 0.004), compared to sporadic cases. There is a significant interaction between MAPT and age at onset (interaction term p < 0.001, main effect of MAPT p < 0.001, main effect of age at onset p < 0.001). For patients with a MAPT mutation, a younger age at onset was associated with shorter survival (HR 0.892, p = 0.013), while in sporadic cases, a later age at onset was associated with shorter survival (HR 1.095, p < 0.001). We did not observe a difference in survival in TDP-43 (n = 39) and tau pathology (n = 18) cases.

Exploratory analysis

While our CSF exclusion criteria for removing cases with AD pathology have previously been reported to have >95% sensitivity and specificity,¹³ it is possible that pathologic heterogeneity in our sporadic cases may affect survival rates. Therefore, we performed an exploratory analysis to evaluate whether having a mutation (C9orf72, GRN, or MAPT) was associated with reduced survival compared to autopsy-confirmed or likely TDP-43 or tau pathology. To define likely TDP-43 or tau pathology cases, we used a previously reported 2-step classification procedure²¹ in 92 cases with unknown pathology to (1) exclude cases with potential comorbid AD based on p-tau/Aβ CSF ratio (n = 10) and (2) sort the remaining cases with available CSF p-tau into likely pathologic groups applying the previously reported regression model that integrates CSF p-tau adjusted for age and disease duration. This resulted in a total of 22 TDP cases (19 autopsy-confirmed, 3 likely) and 41 tau cases (13 autopsy-confirmed, 28 likely). The remaining cases (n = 51) did not have CSF p-tau data available and sample sizes were too small to evaluate individual genes relative to likely or definite TDP-43 or tau groups. A Cox model analysis revealed that mutation cases significantly differed from FTLD-tau (HR 3.169, p = 0.010) but not FTLD-TDP (HR 1.191, p = 0.640). Thus, our overall observation of mutation carriers differing in survival from sporadic cases may in part be due to the relatively increased survival rate of bvFTD due to FTLD-tau relative to FTLD-TDP. Future studies with larger pathology-confirmed cohorts will be necessary to confirm this exploratory finding.

Discussion

We examined survival in a well-characterized cohort of patients with bvFTD. We found that individuals with inherited forms of bvFTD involving mutations of MAPT, GRN, and C9orf72 are at risk of shorter survival, with each mutation independently predicting a poor prognosis. Furthermore, we found that individuals with a MAPT mutation with younger age at onset are at risk for shorter survival. Key features of this study include the examination of a single FTD phenotype (e.g., bvFTD) that was confirmed with CSF or autopsy data, availability of genetic data on all participants in the study, which allowed us to extrapolate the effect of each genetic mutation on survival probability, and use of Cox proportional hazards model to examine potential independent factors associated with survival.

Previous studies have reported that the presence of a genetic mutation reduces survival in a heterogeneous cohort of FTD.² Our study tested whether mutation status influences survival in the most common clinical phenotype of FTD and investigated prognosis of individual mutations associated with the 3 most common forms of hereditary FTD. While we did not observe a difference in survival by individual mutation, the examination of individual mutations is nonetheless important because there are specific molecular etiologies associated with FTD mutations.²² Thus, we sought to examine the individual effects of common causes of hereditary FTD as each mutation contributes to distinct pathology and therefore could contribute to differences in survival. For example, tau-positive pathology has been associated with shorter survival in autopsy-confirmed FTLD.⁸ Irwin et al.²³ reported C9orf27-positive cases contributed to shorter survival in amyotrophic lateral sclerosis but not in FTLD. This could be a result of the inclusion of mixed phenotypes of bvFTD and primary progressive aphasia (PPA) and the small sample size. While other studies²⁴ observed that the presence of language impairment affects survival, we did not observe that having a secondary diagnosis of PPA contributes to differences in survival in our cohort. In a study of FTLD-TDP with 43 cases, patients with GRN mutations had a longer survival than sporadic cases, but this cohort included individuals with comorbid AD pathology.²⁵ Here we focus on a large cohort of well-characterized patients with bvFTD, excluding cases with AD pathology to identify an accurate selection of patients with underlying FTLD. In addition, our study adheres to a high standard of statistical rigor by accounting for left truncation in our survival analysis. If differences in survival by mutation type were to be found, their discovery would be made easier by our discerning inclusion criteria for our cohort, along with a statistical analysis that addresses key characteristics of the dataset. While we did not find differences in survival by mutation type, we may have been limited by our sample size (see table 2 for individual mutation types). An important area for future study will be to evaluate differences in survival by mutation type since, for example, the locus of a specific mutation is known to modify neuropathologic features of disease due to MAPT.²⁶ Future natural history studies will require a larger series of genetic cases with sufficient power in mutation type to detect group differences.

Evidence suggests there are distinct differences in anatomic distribution of disease in genetic cases, and this may contribute to differences in survival. For example, symptomatic mutation carriers showed significantly reduced gray matter (GM) in orbitofrontal and inferior frontal regions, anterior temporal regions, insula, cingulate, parietal lobe, and cerebellum in comparison with sporadic cases, suggesting more severe widespread neurodegeneration in genetic cases.²⁷ Others have also suggested that there are distinct neuroanatomical patterns of disease in different mutation types. In a study examining GM differences in FTD-associated genetic mutations, temporal lobe atrophy was most evident in MAPT, temporoparietal loss was observed in GRN, and frontal and anterior temporal atrophy was most evident in C9orf72.²⁸

There have been very few longitudinal studies evaluating rate of progression in genetic bvFTD cases. In a longitudinal cohort study of bvFTD, a positive family history was associated with a faster rate of progression on the Addenbrooke’s Cognitive Examination–Revised, a global measure of cognitive impairment.²⁹ In a longitudinal evaluation of GM volume in genetic and sporadic FTD, faster rates of atrophy were observed in GRN cases followed by sporadic, C9orf72, and MAPT.³⁰ These results suggest that some mutation carriers may be at a higher risk for worse prognosis, and question whether this is true for all mutations. However, these studies included small cohorts. Our data point to worse prognosis in survival for any mutation carrier.

We observed that patients with mutations in MAPT have a significantly younger age at onset than those with mutations in C9orf72 and GRN. Previous studies examined the relationship between age at onset and mutation. In one study of C9orf72 and GRN, the presence of a mutation in C9orf72 strongly influenced age at onset but the effect was not significant in GRN cases,³¹ while another study found significantly earlier age at onset in C9orf72 cases compared to GRN cases.³² MAPT cases were not examined in these aforementioned studies. It may be that variability in age at onset may be related to mutation type and other genetic modifiers such as single nucleotide polymorphisms (SNPs), and possibly clinical phenotype.^33,34 For example, in a heterogeneous cohort of FTD cases, SNPs associated with a mutation on the GRN gene were associated with a delayed age at onset.³⁵ In future studies, it would be useful to evaluate the relationship between age at onset and particular types of mutations in MAPT, GRN, and C9orf72 as well as other genetic modifiers like the presence of SNPs in a cohort of individuals with a single clinical phenotype. We also found that age at symptom onset was a significant predictor of mortality. Our data suggest that having an older age at onset in GRN, C9orf72, and sporadic cohorts is associated with a shorter survival. Conversely, we found that a younger age at onset was associated with shorter survival in MAPT patients. The effect of age at onset on survival in bvFTD has been mixed in previous reports; some reports suggest later age at onset is associated with shorter survival,^2,36 while others do not find a link between age and survival.²⁴ Future research will be necessary to clarify associations between age and survival within the bvFTD phenotype in well-characterized samples.¹

Although these findings are novel and informative, our study is subject to several limitations. First, we did not have autopsy confirmation of FTLD in all of our cases; however, our cohort was well-characterized with CSF consistent with underlying FTLD pathology, which increases our confidence regarding diagnostic accuracy. Since the timing of symptom onset cannot be precisely ascertained and is obtained from a historical estimate in most cases, the actual survival times may differ slightly from the recorded survival times, introducing variability in the analysis. We recognize that this is a common challenge in dementia research where onset of disease symptoms is insidious. Although FTD is a relatively rare disorder and mutations causing FTD are even rarer, our sample was comparatively large; yet additional study of larger cohorts is needed to confirm our findings. While we observed a strong association between mutation status and survival, we did not ascertain supporting biological evidence. It will be important to consider the mechanistic evidence of shorter survival induced by each individual gene mutation. For example, future studies may require an integration of neuroimaging data to elucidate possible neural mechanisms associated with reduced survival. Next, our study was cross-sectional in nature. Future longitudinal studies should be performed to ascertain how mutation status affects the rate of cognitive decline. While gene-specific modifiers have been associated with heterogeneity within each mutation type (e.g., C9orf72 promoter methylation with C9orf72 expansions, TMEM106B with GRN mutations),^37–39 our current study focused on comparisons of inherited relative to sporadic disease. Future larger-scale studies should evaluate these factors within each mutation group. Similarly, our exploratory analysis revealed that patients with sporadic bvFTD with underlying FTLD-TDP did not differ in survival from mutation cases, a finding that will need to be confirmed with larger pathology-confirmed cohorts. With these caveats in mind, our work suggests that the presence of hereditary mutations and age at symptom onset affect survival among patients with bvFTD. This has important implications for counseling of patients and families, prognosis, and stratification of future clinical trials.

Glossary

Aβ

β-amyloid

bvFTD

behavioral variant frontotemporal degeneration

FTD

frontotemporal degeneration

FTLD

frontotemporal lobar degeneration

GM

gray matter

HR

hazard ratio

PPA

primary progressive aphasia

SNP

single nucleotide polymorphism

Appendix. Authors

Study funding

This work was supported by NIH grants AG000255, AG056054, AG17586, AG043503, AG10124, NINDS NS 088341, the Penn Institute on Aging, and the Dana Foundation.

Disclosure

The authors report no disclosures relevant to the manuscript. Go to Neurology.org/N for full disclosures.