Cognitive reserve and TMEM106B genotype modulate brain damage in presymptomatic frontotemporal dementia: a GENFI study

Frontotemporal dementia (FTD) shows substantial phenotypic variability. In a multicentre study, Premi et al. explore the effect of cognitive reserve and TMEM106B genotype in modulating grey matter volume in presymptomatic FTD. Environmental as well as genetic factors affect rates of brain atrophy, suggesting a possible strategy for delaying disease onset.


Introduction
Frontotemporal dementia (FTD) is a neurodegenerative disorder characterized by neuronal loss in the frontal and temporal lobes (Hodges et al., 2004;Rohrer et al., 2011;Warren et al., 2013). It presents clinically with behavioural symptoms, deficits of executive functions and language impairment, and in some cases, with motor neuron disease, progressive supranuclear palsy or corticobasal syndrome (Seelaar et al., 2011). Up to 40% of cases have a family history of dementia, with an autosomal dominant inheritance in around a third of patients (Stevens et al., 1998). Mutations within microtubule-associated protein tau (MAPT) (Hutton et al., 1998), granulin (GRN) (Baker et al., 2006;Cruts et al., 2006), and chromosome 9 open reading frame 72 (C9orf72) (DeJesus-Hernandez et al., 2011;Renton et al., 2011) are proven major causes of genetic FTD, accounting for 10-20% of all FTD cases. MAPT mutations lead to FTD with neuronal tau inclusions, while GRN and C9orf72 are associated with intraneuronal TAR DNA-binding protein 43  inclusions (Baborie et al., 2011).
Recently, it has been demonstrated in the Genetic Frontotemporal Dementia Initiative (GENFI) study that grey matter and cognitive changes can be identified 5-10 years before the expected onset of symptoms in adults at risk of genetic FTD (Rohrer et al., 2015), and even earlier for those with C9orf72 expansions. However, there is wide variation in the age at onset within families, and possible modifiers of disease progression (including genetic and environmental factors) have yet to be investigated. Such modifiers will be important for several reasons: to properly define biomarkers that can stage presymptomatic disease and track disease progression, to correctly identify individuals most suitable for clinical trials, and to reduce heterogeneity and increase the statistical power of analyses of such trials.
Cognitive reserve and genetic factors have both been proposed as moderators of the onset of disease. Cognitive reserve is a theoretical concept proposing that certain lifetime experiences, including education, individual intelligence quotient, degree of literacy, and occupational attainment, increase the flexibility, efficiency, and capacity of brain networks, thereby allowing individuals with higher cognitive reserve to sustain greater levels of brain pathology before showing clinical impairment (for a review, see Stern, 2009). In healthy individuals, higher educational attainment (Arenaza-Urquijo et al., 2013) as well as cognitive enrichment (Sun et al., 2016) have been related to greater volume and greater metabolism in frontotemporal regions, thus likely enhancing brain performance (Barulli and Stern, 2013).
Genetic modifiers of disease expression also exist, affecting the phenotype and prognosis. TMEM106B has been identified as a genetic modifier in FTD, modulating the age at disease onset in frontotemporal lobar degeneration-TDP-43 disease (Cruchaga et al., 2011;Gallagher et al., 2014;van Blitterswijk et al., 2014). The TMEM106B rs1990622 TT genotype is detrimental and associated with earlier age at disease onset (Cruchaga et al., 2011) and greater functional impairment in frontal regions in presymptomatic GRN mutation carriers (Premi et al., 2014). Conversely, the role of this polymorphism in C9orf72 mutation carriers is still unclear, as it has been suggested a detrimental effect of TMEM106B rs1990622 CC genotype on disease onset and death (Deming and Cruchaga, 2014;Gallagher et al., 2014;van Blitterswijk et al., 2014).
In this study, we aimed to evaluate modifiers of structural brain changes in presymptomatic mutation carriers from a large international cohort of subjects at risk for genetic FTD, investigating the effect of (i) pathogenetic mutation, i.e. MAPT, GRN and C9orf72 carriers versus non-carriers; (ii) cognitive reserve, as measured by years of formal schooling; and (iii) TMEM106B rs1990622 genotype, and their interaction, on grey matter volume.
We hypothesized that individually and together, these three factors will modulate the degree of structural atrophy.

Participants
Data for this study were drawn from the GENFI multicentre cohort study (Rohrer et al., 2015), which consists of 13 research centres in the UK, Italy, The Netherlands, Sweden, and Canada. Inclusion and exclusion criteria have been previously described (Rohrer et al., 2015). Local ethics committees approved the study at each site and all participants provided written informed consent according to the Declaration of Helsinki. For the aim of the present work, we considered participants at 50% risk of carrying a GRN, C9orf72 or MAPT mutation based on having a first-degree relative who was a known symptomatic mutation carrier. Between January 2012 and April 2015, 365 participants were recruited into GENFI, of which 294 were at risk and 71 symptomatic. Of the 294 atrisk participants, 22 did not have a T 1 -weighted MRI scan suitable for volumetric analysis. Included at-risk subjects underwent a careful recording of demographic data, including years of formal schooling (education), past medical history, and a standardized clinical and neuropsychological assessment, as previously published (Rohrer et al., 2015). Genotyping was then performed for the TMEM106B rs1990622 (C/T) single nucleotide polymorphism according to standard procedures (Premi et al., 2014) (at the individual sites in 70.6% of cases, and at the University of Brescia, Italy in the remaining 29.4%). Genotype was not available for 41 participants, and so the final analysis was performed on 231 participants: 108 presymptomatic mutation carriers [genetic status (GS) = 1], 61 with GRN, 33 with C9orf72 and 14 with MAPT mutations) and 123 non-carriers (GS = 0)]. Participants (GS = 0 and GS = 1) came from 77 families (15 with MAPT, 33 with GRN, and 29 with C9orf72 mutations). TMEM106B genotype distribution was comparable between groups (GS = 0 versus GS = 1, Pearson 2 test, P = 0.958), as well as among GS = 1 subgroups (i.e. GRN, C9orf72 and MAPT mutation carriers, P = 0.419). Demographic characteristics of GS = 1, subgrouped on the basis of mutation type, and GS = 0 are reported in Table 1. Mean age of GS = 1 was 45.9 years (range 20.5-70.5 years) and of GS = 0 was 48.3 years (range 19.4-85.7 years). No significant differences were found in age, gender, years of education, and neuropsychological tests between the groups.

Statistical analysis
We fitted a linear mixed effect interaction model (Galecki and Burzykowski, 2013). We assessed the main effect of three factors on grey matter: (i) the presence of pathogenetic mutation (GS, coded as GS = 1 for GRN, MAPT or C9orf72 mutation carriers and GS = 0 for mutation non-carriers); (ii) the role of cognitive reserve as measured by years of formal education; and (iii) the TMEM106B rs1990622 C/T genotype (coded as CC, CT or TT). The relationship between each factor and grey matter volume was labelled as b1, b2, and b3, respectively ( Fig. 1, dark blue lines). Furthermore, we considered the two-way interaction effect of each factor (i.e. GS and education, labelled as b4 (red line), GS and TMEM106B genotype, labelled as b5 (orange line), and education and TMEM106B genotype, labelled as b6 (green line) (Fig. 1). Finally, we considered the three-way interaction effect, i.e. GS, education, and TMEM106B genotype, on grey matter (b7) (Fig. 1). These main and interaction effects were adjusted by fixed covariates, namely age and gender. Moreover, we considered two random effect factors, study site and pedigree, which permitted analysis of the correlations of subjects in the same cluster (centres of subjects' enrolment or individual families).
To overcome the complexity of multiple comparison corrections, we first carried out data reduction of grey matter parcellation data. We proposed a graph-Laplacian Principal Component Analysis (gLPCA) to obtain a low dimensional representation of grey matter parcellation, which incorporated graph structure (Jiang et al., 2013). We did not apply principal component analysis (PCA), widely used to obtain a low-dimensional representation, as the imaging data presented spatial distribution and high left-right correlation (Belkin and Niyogi, 2001). Graph-Laplacian PCA (gLPCA) has several advantages: (i) it is modelled on the representation of the data; (ii) it can be easily calculated, presenting a compact closedform solution; and (iii) it allows noise removal. Once we obtained data reduction, bivariate correlations between principal component (PC) scores and each grey matter measure were computed. Finally, we fitted the mixed-effect interaction models with grey matter, summarized by the first PC scores, as outcome variable. Statistical analysis was performed via R packages (www.r-project.org) and in-house R scripts.

Results
By gLPCA using the skeleton graph between grey matter measures ( Supplementary Fig. 1), the first PC (PC1) was selected to summarize the grey matter volume data. Frontal, parietal and temporal regions were the areas that contributed most to graph construction and PC1 scores, based on correlations between PC1 scores and grey matter measures (Supplementary Table 1).
Fitting the linear mixed-interaction model with fixed covariates (age and gender) and random effects (study site and pedigree), a significant direct effect of GS and years of  education on grey matter outcome (PC1 scores) was observed (P = 0.002 and P = 0.02, respectively), while no effect of TMEM106B genotype on grey matter was detected. We did not find any significant two-way interaction between the considered variables, but did find a three-way interaction on grey matter (P = 0.007) ( Table 2). The data are summarized in Fig. 2. On the x-axis, years of education (i.e. cognitive reserve) are reported, and on the y-axis grey matter volume (PC1). Years of education had a significant direct effect on grey matter volume, independently of GS. We found that the greater the years of education, the greater the grey matter volume (in both GS = 0 and GS = 1), suggesting that cognitive reserve was able to exert an effect in presymptomatic at-risk participants carrying pathogenetic mutations as well as in non-carriers, by increasing grey matter volume.
In comparison to non-carriers (GS = 0, red line), mutation carriers (GS = 1, blue line) showed a significant decrease of grey matter volume, confirming the effect of pathogenetic mutations in shaping progressive atrophy before the onset of symptoms (Rohrer et al., 2015).
The TMEM106B genotype did not exert a direct effect on grey matter volume, and it did not affect grey matter (Fig. 2, red line, GS = 0 and TMEM106B CC or CT or TT). However, in those individuals carrying pathogenetic mutations (GS = 1), TMEM106B polymorphism modulated the slope of the relationship between education and grey matter volume (GS * Education * TMEM106B): a steeper slope was found in TMEM106B TT carriers compared with CT carriers, which in turn was greater than that of CC carriers, with a dose-dependent effect (Fig. 2, green and purple lines). Considering the contribution of each mutation separately, the effect of TMEM106B genotype (GS * Education * TMEM106B) was mainly driven by C9orf72 mutation carriers, the subgroup of patients with the greatest atrophy (Supplementary Table 2 and Supplementary Fig. 2).

Discussion
Autosomal dominant FTD presents with significant interand intra-familial variability among individuals bearing the same pathogenetic mutation. This suggests the presence of environmental, genetic and/or epigenetic modifiers, influencing the age at disease onset and clinical phenotype (Borroni and Padovani, 2013). The effect of genetic modifiers and environmental factors that might trigger the onset of neurodegeneration in carriers of the pathogenetic mutation (which is present at birth but only manifests symptoms in mid-late adulthood) are of extreme interest. In this view, the pathogenesis of inherited FTD may be a model of  'Latent Early-Life Associated Regulation' (LEARn), in which latent expression of associated genes is triggered by environmental and non-environmental factors (Maloney et al., 2012;Maloney and Lahiri, 2016), with neurodegeneration being modulated by lifetime exposure to one or more environmental factors as well as genetic background.
In the present study, we aimed at identifying modulating factors of neuronal loss in presymptomatic subjects bearing pathogenetic mutations within GRN, MAPT and C9orf72 genes through the surrogate marker of volumetric MRI.
We analysed the effect of (i) pathogenetic mutations; (ii) cognitive reserve as measured by years of schooling; and (iii) TMEM106B genotype on grey matter volume in the large GENFI cohort. Indeed, we chose grey matter volume as an endpoint measure as it correlates well with indexes of disease severity (Premi et al., 2016) and progression (Brambati et al., 2007;Lam et al., 2014).
First, we confirmed, as previously reported (Rohrer et al., 2015), that pathogenetic mutations are detrimental to grey matter volume years before expected age at disease onset, being associated with smaller volumes i.e. greater atrophy as compared to siblings who did not inherit the mutation. C9orf72 repeat expansion carriers had a greater degree of atrophy (Rohrer et al., 2015), as compared to GRN mutation carriers and MAPT mutation carriers, with the latter being the smallest group in our sample.
Furthermore, we demonstrated that cognitive reserve is associated with brain atrophy and also modulate neuronal loss years before the onset of symptoms. TMEM106B polymorphism, on the other hand, only modulated grey matter volume in those with an autosomal dominant mutation and with the lowest education.
The duration of formal schooling, as a proxy of cognitive reserve, was associated with greater grey matter volume in both non-carriers and in mutation carriers. This might suggest that those subjects with higher educational attainment were able to better counteract the detrimental effect of a pathogenetic mutation than their counterparts with lower education. However, this effect was also found in the group that did not carry mutations: those with higher education attainment had greater grey matter volume then those with low education. Hence the finding is not specific to mutation carriers suggesting a broader effect of cognitive reserve on maintaining and ameliorating brain functioning. Interestingly, even if presymptomatic mutation carriers already had mild structural changes (Rohrer et al., 2015), the relationship between years of education and structural changes was comparable to that observed in non-carriers (direct correlation, the higher the education the greater the grey matter volume) (Sole-Padulles et al., 2009;Rzezak et al., 2015), rather than that reported in symptomatic FTD (inverse correlation, the higher the education the lower grey matter volume) (Borroni et al., 2009). The concept of cognitive reserve was originally proposed to explain the lack of a direct relationship between the degree of brain pathology and the severity of the clinical manifestations that should supposedly result from such damage in neurodegenerative disorders such as Alzheimer's disease (Stern et al., 1992(Stern et al., , 1994 and FTD (Borroni et al., 2009;Premi et al., 2012Premi et al., , 2013. Cognitive reserve represents the hypothesized capacity of the adult brain to compensate for the effects of a disease or injury that would be sufficient to cause clinical dementia in an individual with less cognitive reserve (Stern, 2002). Herein, we propose that high education might postpone the onset of dementia in those subjects at risk of developing FTD (Akbaraly et al., 2009;Craik et al., 2010;Pettigrew et al., 2013). These findings extend previous results obtained in healthy subjects (Sole-Padulles et al., 2009;Rzezak et al., 2015), and might represent a possible strategy to delay onset of inherited FTD.
Conversely to education attainment, TMEM106B genotype did not have any effect on mutation free individuals, but this genetic trait might represent an additional nonmodifiable risk factor in mutation carriers. Literature data have widely proven that TMEM106B variants are genetically associated to frontotemporal lobar degeneration-TDP-43 pathology and are considered a major risk factor for this disease (Chen-Plotkin et al., 2012;Busch et al., 2013;Nicholson and Rademakers, 2016). It has been suggested that the TMEM106B polymorphism might modulate progranulin plasma levels, thus affecting age at onset of symptoms in GRN mutation carriers and explaining in part the reported variability (Cruchaga et al., 2011). Furthermore, presymptomatic GRN mutation carriers bearing the TMEM106B TT genotype showed greater functional brain damage than those with CT/CC TMEM106B genotypes (Premi et al., 2014). In frontotemporal lobar degeneration-TDP-43 due to C9orf72 mutations, the relationship is less clear, and it has been suggested that TMEM106B might be able to affect disease pathology, but with an opposite association (Gallagher et al., 2014;van Blitterswijk et al., 2014): two independent groups analysed the association of TMEM106B variants with disease risk, age at onset, and age at death in C9orf72 expansion carriers with the CC genotype (protective in GRN carriers) found to be associated with earlier onset and earlier death in C9orf72 expansion carriers (Deming and Cruchaga, 2014;Gallagher et al., 2014;van Blitterswijk et al., 2014). This effect may be an example of the general phenomenon of epistasis, in which a genetic variant is beneficial on some genetic backgrounds but deleterious in others (Gallagher et al., 2014;Busch et al., 2016). In particular, as hypothesized, if in GRN-related TDP-43 pathology TMEM106B is related to endosomal-lysosomal dysfunction and to the perturbation of the progranulin pathway, in C9orf72 knockdown mice TMEM106B over-expression may produce a phenotypic rescue effect (Busch et al., 2016). However, further studies are needed to elucidate its mechanism of action. Another possibility is that TMEM106B is simply in linkage disequilibrium with the actual associated variant and when different populations are examined, the allele associated with disease modulation is different.
In the present work, a moderating, dose dependent effect of TMEM106B rs1990622 genotype together with education attainment was observed on grey matter volume in presymptomatic subjects carrying pathogenetic mutations. This finding supports the idea that epigenetic modifications in TMEM106B might occur. Epigenetic mechanisms, mostly mediated by DNA methylation, have been shown to be important in other neurodegenerative disorders (Piaceri et al., 2015) and to be influenced by socioeconomic status, which is strongly associated with cognitive reserve (Tehranifar et al., 2013). Thus, it could be hypothesized that TMEM106B, a gene containing a number of methylation sites (http://genome. ucsc.edu), might exert its effect on structural changes in atrisk subjects via cognitive reserve. Future studies, however, need to be performed to confirm this hypothesis.
We indeed found a detrimental effect of the CC genotype, but the present results were mainly driven by subjects carrying C9orf72 mutations, in which CC is the risk genotype (Supplementary Table 2).
We acknowledge that there are limitations with this work. First, the correlation between the factors herein considered and age at disease onset would benefit from longitudinal follow-up and independent studies to confirm the present results. The lack of assessment of leisure activities prevents us from characterizing the entire spectrum of cognitive reserve proxies (Nucci et al., 2012), and we are aware of possible biases determined by different school systems across the involved countries. Finally, the role of TMEM106b genotype should be further evaluated in the different genetic groups alone.
In conclusion, our findings indicate that even several years before the onset of symptoms, brain changes in inherited FTD may be modulated by environmental and genetic factors. In the absence of effective pharmacotherapeutic treatments for counteracting the onset of symptoms in pathogenetic mutation carriers, high education may represent a large-scale strategy to be considered by national health system policies.
TMEM106b genotype needs to be considered as an extra non-modifiable trait affecting brain pathology, and each FTD mutation should be analysed individually. Future clinical trials in genetic FTD should take into account both education level and TMEM106b genotype to define subjects with greater brain damage, thus representing those at higher risk of developing FTD at an earlier age.