Apathy in presymptomatic genetic frontotemporal dementia predicts cognitive decline and is driven by structural brain changes

Abstract Introduction Apathy adversely affects prognosis and survival of patients with frontotemporal dementia (FTD). We test whether apathy develops in presymptomatic genetic FTD, and is associated with cognitive decline and brain atrophy. Methods Presymptomatic carriers of MAPT, GRN or C9orf72 mutations (N = 304), and relatives without mutations (N = 296) underwent clinical assessments and MRI at baseline, and annually for 2 years. Longitudinal changes in apathy, cognition, gray matter volumes, and their relationships were analyzed with latent growth curve modeling. Results Apathy severity increased over time in presymptomatic carriers, but not in non‐carriers. In presymptomatic carriers, baseline apathy predicted cognitive decline over two years, but not vice versa. Apathy progression was associated with baseline low gray matter volume in frontal and cingulate regions. Discussion Apathy is an early marker of FTD‐related changes and predicts a subsequent subclinical deterioration of cognition before dementia onset. Apathy may be a modifiable factor in those at risk of FTD.


INTRODUCTION
Apathy is a common and disabling feature of frontotemporal dementia (FTD). It is part of the diagnostic criteria for behavioral variant of FTD (bvFTD), 1 and frequently occurs across all FTD variants. 2,3 Apathy is a multifaceted construct that describes dysfunctional goal-directed behavior, arising from affective, behavioral, and cognitive impairments.
FTD has been associated with concurrent affective, behavioral, and cognitive apathy symptoms, 4 which worsen the prognosis in terms of survival, 5 disability 6-9 and functional independence. Better understanding of the causes and consequences of apathy and its role in the clinical progression of FTD is vital to develop effective treatment strategies, including preventive strategies in the context of genetic risk of FTD.
Previous imaging studies have identified structural correlates and changes associated with apathy in FTD. The severity of apathy correlates with widespread atrophy in frontotemporal areas, including the dorsolateral, ventromedial and orbital prefrontal cortex, anterior cingulate cortex, and insula and basal ganglia 3,10-12 (see 13,14 ). In people with symptomatic FTD, apathy is associated with the severity of executive function impairment, 12,15 including deficits in working memory, decision making, selective/sustained attention, planning, processing speed, inhibitory processes and mental/cognitive flexibility. 12,[15][16][17][18] Deficits in executive function occur in both behavioral and aphasic syndromes of FTD, with subtler impairments in the presymptomatic phase. [19][20][21] Indeed, executive dysfunction, like apathy, is a diagnostic criterion for bvFTD 1 and shares several anatomical correlates with apathy (see 13 29, 30 We therefore examined longitudinal changes in apathy and their association with subclinical cognitive decline in presymptomatic gene carriers, in the international Genetic FTD Initiative (GENFI). 20 We first tested the hypothesis that apathy increases over time in presymptomatic carriers of FTD mutations, and is more severe in those closer to symptom onset. We used latent growth curve modeling of longitudinal data to test the predictive value of apathy for subclinical deterioration of cognitive performance in the Digit Symbol test in gene carriers versus non-carriers. To understand the relationship between apathy and FTD-related brain changes, we tested whether baseline and longitudinal changes in apathy were a function of atrophy in the presymptomatic gene carriers. Previous studies suggest a detrimental effect of apathy on clinical progression and survival of FTD patients, [5][6][7][8][9] Highlights • Apathy progresses in presymptomatic genetic frontotemporal dementia • Apathy predicts prospective cognitive decline, and not vice versa • Structural changes in frontal and cingulate regions predict apathy progression • Apathy is an early marker in frontotemporal dementia, even before dementia onset and have highlighted frontal lobe and cingulate cortex atrophy as neural correlates of apathy in FTD. 13,14 Based on this, we predicted: (1) that baseline apathy predicts future cognitive deterioration; and (2) an association between apathy and structural brain change, in the frontal lobe and cingulate cortex.

Participants
From the GENFI study, 20

Latent growth curve model
Univariate latent growth curve models (LGCMs) were fitted to the combined data from three time points of longitudinal behavioral/cognitive and imaging assessments, to test the relationships between apathy, cognition, and brain volumes. The LGCM provides insight into baseline scores, change, and individual differences by estimating (1) an intercept, which represents the initial level of the outcome measures; (2) a slope, quantifying the rate of change; (3) a variance of the intercept and slope, capturing individual differences in baseline and change over time; and (4) the relation between intercept and slope, that is, how the initial level is associated with the rate of change over time. Predictors can be added to the model to assess their effects (as an interaction) with intercept and/or slope. The LGCM estimation has two main steps: (1) a linear or curvilinear regression is conducted to fit across the repeated measures of each subject, eliciting a growth curve shape which describes the change over time; and (2) the potential predictors of individual differences in intercepts/slopes are then evaluated. In this way the growth model, as a collection of individual trajectories, describes the individual differences in the changes over time, and the change at group level. 34 LGCM is a powerful and flexible tool well suited to specifying and testing hypotheses of changes, predictors of change and clinical progression, 34,35 and can be estimated using open source software such as R (R Core Team). Compared to simpler longitudinal analysis methods, LGCM is preferred for complex models with more than one dependent variable and/or more than one predictor, with complex variance functions, or multigroup model estimation with partial constraints, to assess global model fit, and to deal with random missing data. 35 LGCM guidelines recommend ≥3 time points and ≥5 cases per parameter. 35  We also report the model chisquare test (χ 2 ), noting this index is sensitive to the sample size and is liable to reject models of large cohorts (good fit: low values and P > 0.05). 37 We also report the ratio between chi-square and degrees of freedom (χ 2 /df) as an alternative model fit index (acceptable fit: <2, good fit: <3). 37 To test group differences on parameters of interest in LGCMs, we compared each model to a model that constrained the relevant parameters (eg, the slope) to be equal between the two groups. For model comparisons, we used Akaike Information Criteria (AIC), penalizing model complexity.

LGCM of apathy and cognitive decline
In all models, the intercept was centered at baseline and a linear slope was tested. CBI-R apathy scores and Digit Symbol scores at follow-up visits were annualized and recomputed at one and 2 years to adjust for small differences in intervals. EYO was included as a predictor of both intercept and slope, and the genetic status used to define groups.
We applied four different LGCMs to behavioral and cognitive data to test our main hypothesis: (1) a LGCM on the longitudinal CBI-R apathy subscale scores; (2) the same as the previous item, but with baseline Digit Symbol as predictor; (3) a LGCM on the longitudinal Digit Symbol scores; and (4) the same as the previous item, but with baseline CBI-R apathy subscale scores as predictor. These four models allowed us to test whether apathy progresses over time in presymptomatic carriers, and predicts a subclinical cognitive deterioration, or vice versa.
First, an LGCM was fitted on the CBI-R apathy z-scores, estimating the parameters freely in a multigroup model defined by genetic diagnosis. This model was compared to one that was fitted by constraining the slope estimation to be equal in the two groups, in order to test the difference in fit of the group equality constrained model with the one accounting for differences between presymptomatic carriers and noncarriers on the annual rate of change (slope). Second, baseline Digit Symbol scores were added to the model as predictor of both intercept and slope of apathy, to test the predictive value of baseline cognitive performance on longitudinal change in apathy. An analogous approach was applied to the longitudinal and annualized Digit Symbol z-scores: first, the initial LGCM with EYO as predictor of the intercept and slope was fitted in a multigroup model by freely estimating all parameters; second, we compared this free model with a model where we constrain key parameters to test for between-group differences; and lastly, baseline CBI-R apathy scores were added to the model as a predictor variable on intercept and slope.

LGCM for structural brain changes
We applied eight independent univariate LGCMs to estimate longitudinal changes in gray matter volumes of frontal, temporal, parietal and occipital lobes, insular cortex, cingulate cortex, subcortical central structures, and brainstem. As for the behavioral and cognitive scores, all gray matter values at follow-up visits were computed at 1 and 2 years to adjust for small differences in retest interval. In all models, the intercept was centered at baseline and a linear slope was tested.
EYO and TIV were included as predictors of both intercept and slope.
Genetic status (presymptomatic carrier versus non-carrier) defined the groups. When change is homogeneous, or modeled in smaller subgroups, LGCM estimation may occasionally yield improper solutions (ie, impossible values such as negative variances) which necessitate imposing constraints to achieve plausible solutions, which will be noted when necessary. In presymptomatic carriers, we applied a bivariate LGCM model on longitudinal apathy scores and longitudinal gray matter volumes in each of the brain regions that changed over time. With the bivariate LGCM it is possible to investigate the association between the annual rates of change (slopes) in the two variables considered, as well as the associations between initial scores (intercepts) and the longitudinal changes. Thus, we tested our hypothesis on the association between atrophy in fronto-cingulate brain regions and apathy severity in presymptomatic FTD. For these longitudinal analyses of imaging data, we used an ROI-based approach that included the regional volumes in the bivariate LGCM with apathy. We considered bilateral lobar values rather than single subregions or lateralized lobar values, to simplify analyses and constrain the parameter-to-subject ratio. Although individuals may have asymmetric atrophy, the group pattern is typically bilateral and symmetric. To assess the degree of symmetry, we tested for brain volume differences between left and right hemispheres at baseline using a laterality index (absolute difference between left and right volumes divided by total volume). We then applied t tests on this index between presymptomatic carriers and non-carries, across the whole population and by genetic mutation.
The parameters in each of these models are estimated independently from the other region-specific models. We correct for multiple comparisons for slope estimates and group comparisons across regionspecific tests, although it would not be appropriate to apply corrections

Descriptive statistics
Demographic and clinical characteristics at baseline, and descriptive statistics are summarized in Table 1 However, there was no significant effect of genetic status on this association (presymptomatic carriers vs. non-carriers; P > 0.05).

LGCM on longitudinal apathy scores
The LGCM on longitudinal CBI-R apathy scores fit the data well In summary, presymptomatic carriers showed a longitudinal increase in apathy severity over a 2-year period, which was greater than non-carriers. This change was not predicted by Digit Symbol test performance at baseline.

LGCM on longitudinal cognition
The LGCM on longitudinal Digit Symbol scores fit the data adequately In summary, presymptomatic carriers showed a progressive cognitive decline over 2 years, which was greater than non-carriers. This subclinical cognitive deterioration was faster when approaching the estimated age of onset, and was predicted by apathy severity at the baseline.

LGCM on longitudinal gray matter brain volumes
Model fit indices for LGCM on z-scored brain volumes in cortical and subcortical regions, and the estimated slope for both presymptomatic carrier and non-carrier groups, are reported in Table 2. In summary, for non-carriers there were no significant structural changes in the regions of interest. In contrast, presymptomatic carriers showed progressive atrophy, which was significantly different from the non-carrier group, in the frontal, temporal, and parietal lobes, cingulate cortex and in subcortical central structures (but not in the occipital lobe and brainstem). Insular cortex showed longitudinal decline in the presymptomatic group, but this did not significantly differ from non-carriers' rate of change. In the model on parietal lobe values, the slope variance term was constrained to zero in non-carriers to make the model converge correctly. In Appendix B (Table B. and MAPT groups showed a significant annual rate of atrophy, but the GRN group did not. However, interpretation of this result requires caution given the unbalanced sample size of the three gene mutation groups. The group comparisons on the laterality index for each lobar value did not identify asymmetry between left and right volumes in presymptomatic carriers as compared to non-carrier family members (P > 0.05).
Considering the gene-specific groups, only insula showed an effect of laterality (left > right volume, t = 2.00; P = 0.048) in MAPT carriers as compared to the family non-carrier members.

Bivariate LGCMs on longitudinal apathy scores and gray matter brain volumes
In the previous models of brain changes, significant longitudinal changes in apathy and atrophy were identified in the presymptomatic group only. We therefore applied five new, bivariate, LGCMs of longitudinal apathy and apathy of frontal, temporal, and parietal lobes, cingulate cortex and the subcortical structures, constraining the covariance term between apathy intercept and slope to zero in all models to ensure proper solutions. We report model fit indices and estimated covariance parameters for all brain regions in Table 3. In summary, the annual progression of apathy severity was associated with baseline gray matter volumes in frontal lobe (est = -0.208, SE = 0.100, z = -2.077, std est = -0.348, P = 0.038) and cingulate cortex (est = -0.139, SE = 0.058, z = -2.085, std est = -0.237, P = 0.037; Figure 3). Comparing the bivariate LGCMs with and without constraining the estimation of covariance between brain volume intercept and progressive apathy to zero, freely estimating the association between brain structure and apathy change improved model fit for both frontal lobe (∆χ 2 = 5.056, ∆df = 1, P = 0.025) and cingulate cortex (∆χ 2 = 7.206, ∆df = 1, P = 0.007) gray matter volumes. With reduced sample sizes in gene specific subgroups, the LGCM method is not suitable for gene-specific analysis in this dataset. Larger future datasets in GENFI, or merged datasets between genetic FTD cohort studies, may enable gene-specific modeling.

DISCUSSION
In this study we found that apathy progresses significantly in presymptomatic carriers of mutations associated with FTD, and that individual differences in apathy at baseline predict the severity of progressive deterioration of performance on the Digit Symbol test over time.
In presymptomatic carriers, the progression of apathy over 2 years is associated with atrophy of the frontal lobe and cingulate gyrus at baseline. In contrast, subclinical cognitive impairments do not predict the worsening of apathy.
Apathy is one of the most prevalent symptoms in patients with FTD syndromes, 38  apathy, often as a presenting symptom (see review 42 ). Apathy has been reported in ∼69% of patients with GRN mutations, 43 but is less common with MAPT mutations. 41,44,45 Although apathy is sometimes reported as more common than disinhibition, 45   Abbreviations: Ap, apathy; Br, brain; CFI, comparative fit index; est, estimate; RMSEA, root-mean-square error of approximation; SRMR, standardized root mean-square residual; std est, standard estimate; z, z-value; χ 2 , chi-square test.
FTD, and are associated with lesions or dysfunction involving the fronto-subcortical networks. 13,47 We quantified apathy from the subscale of CBI-R, as it was the principal measure for apathy available in presymptomatic cases from the GENFI study. Although this has been successfully employed in previous studies on FTD, and more recently also in presymptomatic FTD, 46  Another challenge in the quantification of apathy and executive function is the potential overlap with other symptoms, such as depression and akinesia. 38 In particular, depression might be a confounding symptom for apathy. The wider spectrum of neuropsychiatric symptoms has been described in GENFI, at least in terms of cross-sectional prevalence of symptoms/signs and their neural correlates. 46,48 Our study and hypothesis focused on apathy, but we verified that depression and apathy measures were not significantly associated. This suggests that the CBI-R apathy subscale is not simply measuring, or confounded by, depression symptoms. This aligns with previous evidence that supports the dissociation between apathy and depression in FTD and other neurodegenerative diseases. 15,[49][50][51][52][53] While akinesia is common in symptomatic genetic FTD, 54 it is not common in presymptomatic cases and does not correlate with apathy measures in other cohorts. 10 Finally, an open longitudinal study like GENFI will have incomplete longitudinal data. We therefore included only the first three waves of assessment, fulfilling the minimum requirement in the LGCM guidelines. 35 In years to come, it will be possible to examine a larger data sample and/or a longer follow-up period, including the role of apathy in the transition from presymptomatic to symptomatic phases of FTD, and the relationship between apathy, cognitive, and brain changes by gene mutation groups.
To conclude, our results demonstrate that apathy occurs early in disease progression of genetic FTD, reflecting early brain changes and predicting individual future clinical trajectories of cognitive and executive function deterioration. The assessment of apathy could help with cohorts' stratification, according to their prognosis, and improve the power and design of future therapeutic trials. Apathy may also be a modifiable factor in its own right, by pharmacological 55 or nonpharmacological interventions. 56 As such, it becomes a potential target not only for symptomatic treatment but also interventions to slow down or delay clinical decline in people at risk of FTD.

ACKNOWLEDGMENTS
We thank our participant volunteers and their families for their partic-

F I G U R E A 1
Cross-sectional and longitudinal changes in apathy severity as assessed by the Cambridge Behavioural Inventory (CBI-apathy) and Digit Symbol performance in presymptomatic carriers (red) and non-carriers (blue). (A) Graphs represent the relationships of the estimated initial scores ("intercept"; left graph) and the annual rate of change ("slope"; right graph) in apathy scores with the estimated years from onset (EYO). (B) Graphs represent the relationships of the estimated intercept (left graph) and slope (right graph) in Digit Symbol test scores with EYO. Individuals' data are not plotted, to protect anonymity.

TA B L E B 1
Model fit indices and estimated slopes of Latent Growth Curve Models on longitudinal z-scored brain volumes in presymptomatic carriers by gene groups   Abbreviations: CFI, comparative fit index; est, estimate; FDR, false discovery rate correction; Non-Car, non-carriers; Pres-Car, presymptomatic carriers; RMSEA, root-mean-square error of approximation;