Presymptomatic cognitive and neuroanatomical changes in genetic frontotemporal dementia in the Genetic Frontotemporal dementia Initiative (GENFI) study: a cross-sectional analysis

Findings Between Jan 30, 2012, and Sept 15, 2013, we recruited participants from 11 research sites in the UK, Italy, the Netherlands, Sweden, and Canada. We analysed data from 220 participants: 118 mutation carriers (40 symptomatic and 78 asymptomatic) and 102 non-carriers. For neuropsychology measures, we noted the earliest signiﬁ cant diﬀ erences between mutation carriers and non-carriers 5 years before expected onset, when diﬀ erences were signiﬁ cant for all measures except for tests of immediate recall and verbal ﬂ uency. We noted the largest Z score diﬀ erences between carriers and non-carriers 5 years before expected onset in tests of naming (Boston Naming Test –0·7; SE 0·3) and executive function (Trail Making Test Part B, Digit Span backwards, and Digit Symbol Task, all –0·5, SE 0·2). For imaging measures, we noted diﬀ erences earliest for the insula (at 10 years before expected symptom onset, mean volume as a percentage of total intracranial volume was 0·80% in mutation carriers and 0·84% in non-carriers; diﬀ erence –0·04, SE 0·02) followed by the temporal lobe (at 10 years before expected symptom onset, mean volume as a percentage of total intracranial volume 8·1% in mutation carriers and 8·3% in non-carriers; diﬀ erence –0·2, SE 0·1). in ﬁ ndings ne biomarkers that can stage presymptomatic disease track disease for future therapeutic trials. onset, and cognitive impairment around 5 years before expected onset, when the genetic group is investigated as a whole. Exploratory analyses suggest that diﬀ erent cortical and subcortical areas are aﬀ ected earliest in each of the MAPT , GRN , and C9orf72 groups, and that structural imaging changes can be seen 15 years or more before symptoms onset. Our results provide an insight into the early neuroanatomical changes in genetic frontotemporal dementia and suggest the potential for use of structural imaging measures as biomarkers in future therapeutic trials.


Introduction
Frontotemporal dementia is a neurodegenerative disorder characterised by focal neuronal loss in the frontal and temporal lobes. 1 It is a common cause of early-onset dementia, but can also present in old age and has an estimated prevalence of between 15 and 22 per 100 000 individuals in the population. 2 It presents clinically with either behavioural symptoms (behavioural variant frontotemporal dementia) or language disturbance (primary progressive aphasia), but patients can also develop symptoms of motor neuron disease, progressive supranuclear palsy, or corticobasal syndrome. 1 It is highly heritable, with an autosomal dominant family history reported in around a third of people with the disease. 3 Mutations in three genes are proven major causes of genetic frontotemporal dementia: microtubule-associated protein tau (MAPT), progranulin (GRN), and chromosome 9 open reading frame 72 (C9orf72). 4 Frequencies of mutations of these three genes vary by geography, but together they account for 10-20% of all cases of frontotemporal dementia. 4 The study of autosomal dominant frontotemporal dementia in its presymptomatic period provides a window into the earliest stages of the disease process. 5 Evidence from familial Alzheimer's disease and Huntington's disease shows that changes in some biomarkers occur many years before symptom onset, [6][7][8] suggesting that the ideal time to treat neurodegenerative disease could be before clinical presentation, at a point when the minimum of irreversible neuronal loss has occurred and cognitive function is still preserved. To optimise therapeutic opportunities, biomarkers of frontotemporal dementia are therefore needed that signify disease onset and can measure changes in disease trajectory in the presymptomatic period. Furthermore, biomarkers that allow accurate staging of the disease process will be important to identify individuals most suitable for particular trials, to reduce heterogeneity, and increase the statistical power.

Participants
The Genetic Frontotemporal dementia Initiative (GENFI) consists of 11 research sites, in the UK, Italy, the Netherlands, Sweden, and Canada. We recruited participants who were either known carriers of a pathogenic mutation in MAPT, GRN, or C9orf72, or at risk of carrying a mutation because a fi rst-degree relative was a known symptomatic carrier. We genotyped all participants at their local site, with a pathogenic expansion in C9orf72 being defi ned as the presence of greater than 30 repeats. We enrolled 220 participants between Jan 30, 2012, and Sept 15, 2013. Local ethics committees at each site approved the study and all participants provided written informed consent at enrolment.

Procedures
Participants underwent a standardised clinical assessment consisting of a medical history, family history, and physical examination. We based symptomatic status on this assessment, which included a collateral history from a family member or close friend. We measured functional status using the Frontotemporal Dementia Rating Scale 27 and assessed behavioural symptoms using the Cambridge Behavioural Inventory Revised version (CBI-R). 28 Patients underwent a neuropsychological battery consisting of tests from the Uniform Data Set: 29 the Logical Memory subtest of the Wechsler Memory Scale-Revised with Immediate and Delayed Recall scores, Digit Span forwards and backwards from the Wechsler Memory Scale-Revised, a Digit Symbol Task, Parts A and B of the Trail Making Test, the short version of the Boston Naming Test, and Category Fluency (animals). We also tested Letter Fluency and did the Wechsler Abbreviated Scale of Intelligence Block Design task, and the Mini-Mental State Examination (MMSE). For each test, apart from the MMSE and CBI-R, we calculated Z scores based on language-specifi c norms. Most at-risk participants (158 [88%] of 180) had not undergone presymptomatic genetic testing and were therefore not aware of their mutation status, and for these participants the clinicians and neuropsychologists who did the assessments were masked to mutation status.
We did volumetric T1-weighted MRI on 3T and 1·5T scanners at sites where 3T scanning was not available. We designed scan protocols at the outset of the study to match across scanners as much as possible. For the volumetric analysis, we did a cortical parcellation using a multiatlas segmentation propagation approach following the brainCOLOR protocol, 30,31 combining regions of interest to calculate grey matter volumes of the entire cortex, separated into the frontal, temporal, parietal, occipital, cingulate, and insula cortices. We also did a subcortical parcellation using the Neuromorphometrics protocol 32,33 for the hippocampus, amygdala, striatum, and thalamus, and a parcellation of the cerebellum using the Diedrichsen cerebellar atlas, 33,34 producing a measure for the entire cerebellum by combining regions of interest. We measured whole-brain volumes using a semi-automated segmentation method. 35 We expressed all measures as a percentage of total intracranial volume (measured with SPM12 with a combination of grey matter, white matter, and CSF segmentations). In view of previous evidence for asymmetrical atrophy in GRN mutation carriers compared with MAPT and C9orf72 carriers, 4,5 we also assessed diff erences between left and right hemisphere volumes using a laterality index, calculated as the absolute diff erence between left and right cortical volumes divided by total cortical volume.
Findings from individual case series of individuals with dementia with a known genetic cause suggest that variability of age at symptom onset exists within families. However, authors of a large study of familial Alzheimer's disease 36 suggest that a strong relation exists between individual age at symptoms onset and both parental age at onset and mean age at onset within the family. To our knowledge, no similar studies have been done in frontotemporal dementia. We therefore did an initial analysis on the basis of the symptomatic carriers within our cohort, investigating the relation between their age at symptoms onset and parental age at onset, their age at onset and mean age at onset for other members of the same family, and their age at onset and median age at onset for other members of the same family (excluding the symptomatic individual from mean and median calculations). Parental age at onset did not show a signifi cant correlation with age at symptoms onset of the symptomatic carriers (Pearson correlation coeffi cient 0·39; p=0·0685), but we found both mean and median ages at onset within the family to be signifi cantly correlated with the symptomatic carriers' age at onset (Pearson correlation coeffi cient 0·53, p=0·0019 for the mean and 0·50, p=0·0036 for the median). Furthermore, in addition to being correlated with mean age of onset within their families, age at symptoms onset of symptomatic carriers did not signifi cantly diff er from mean age at onset within their families (p=0·3216 Wilcoxon signed rank). On the basis of this analysis, we decided to use mean familial age at onset to estimate time to expected symptom onset-ie, someone aged 50 years old at the time of assessment with a mean age at onset of 55 years old in their family would be given an expected time from symptoms onset of -5 years. Data were available for this calculation from one family member in 35 families, from two in 15 families, from three in ten families, from four in four families, from fi ve in fi ve families, from six in two families, and from seven in two families; 12, 16, and 30 family members were available in a further three families.

Neuropsychological (Z score)
Logical Memory-Immediate Recall Digit Span backwards

Statistical analysis
We used linear mixed-eff ects models to examine whether diff erences existed between non-carriers and mutation carriers in the association between each clinical, behavioural, or structural imaging measure and the time to expected onset of symptoms (we combined all genes because of low numbers in each individual genetic group). This modelling framework allows estimation of fi xed and random eff ects of predictor variables, including the intercept. Fixed eff ects represent non-random sources of variation, where the predictor variable has the same relation with the outcome in all observations. Random eff ects estimate the variance in the eff ect of a predictor between diff erent clusters in the data and this estimation allows for correlation in the outcome between members of the same cluster. 37,38 For analysis of each measure, a random intercept for family allowed values of the marker to be correlated between family members. The fi xed eff ect predictor variables of interest were mutation carrier status, time to expected onset, and terms for the interaction between mutation carrier status and time to expected onset. We expected a non-linear change in each measure over time, so models also included a quadratic term for time to expected onset and the interaction between this term and mutation carrier status. We included a more complex cubic relation association between the measure and time to expected onset only when signifi cant (p<0·05) evidence existed that addition of a cubic term and the interaction between the cubic term and mutation carrier status improved model fi t. An example of the mixed eff ect model is given in the appendix for analysis of whole-brain volume to show the modelling framework that we used for analysis.
We also did exploratory analyses to assess whether diff erences between non-carriers and MAPT, GRN, and C9orf72 mutation carriers existed in the association between values of each measure and time to expected onset of symptoms. Because of the small number of participants in each gene group, we considered only linear changes in markers over time in this analysis.
We did a Wald test for each model to assess whether the mean value of the measure diff ered between mutation carriers and non-carriers. We predicted average values from the mixed eff ects model for each group and diff erences between mutation carriers and non-carriers every 5 years between 25 years before expected onset and 10 years after expected onset. All analyses were adjusted for study site and sex. Model diagnostics for both MMSE and CBI-R suggested nonconstant variance, so we used robust standard errors for these analyses.
In addition to the prespecifi ed analysis of markers of disease progression, we did a post-hoc analysis to examine whether diff erences existed between noncarriers and MAPT, GRN, and C9orf72 mutation carriers in the association between laterality of brain volume and time to expected onset of symptoms. Because of strong skew in laterality, we used a log transformation for this analysis, and results are presented as ratios of laterality between mutation carriers and non-carriers for ease of interpretation. We did all analyses with STATA (version 12.1 or later).
Boston Naming Test Diff erences calculated from unrounded values.

Role of the funding source
The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. All authors had full access to all the data in the study except for the results of genetic mutation screening in presymptomatic participants. Only JMN and DMC had access to all of the genetic results to avoid risk of disclosure of genetic status to at-risk participants who were unaware of whether they carried a mutation. All authors had fi nal responsibility for the decision to submit for publication.
In the symptomatic cohort, most participants had a diagnosis of behavioural variant frontotemporal dementia (meeting the Rascovsky diagnostic criteria), 39 except for six participants with GRN mutations who had diagnoses of the non-fl uent variant of primary progressive aphasia (Gorno-Tempini diagnostic criteria) 40 and four participants with C9orf72 mutations (one with the non-fl uent variant of primary progressive aphasia, two with frontotemporal dementia with motor neuron disease, and one with a dementia syndrome not otherwise specifi ed). Functionally, one participant (with a MAPT mutation) in the symptomatic cohort was very mildly aff ected (according to the Frontotemporal Dementia Rating Scale), three (one GRN and two C9orf72) were mildly aff ected, 16 (four MAPT, fi ve GRN, and seven C9orf72) were moderately aff ected, 13 (four MAPT, four GRN, and fi ve C9orf72) were severely aff ected, and seven (two MAPT, three GRN, and two C9orf72) were very severely aff ected.
MMSE, CBI-R, and all neuropsychology measures showed signifi cant mean diff erences between mutation   GRN, and 33 C9orf72). Whole-brain volume showed a signifi cant diff erence between mutation carriers as a whole group and non-carriers (p<0·0001), with strong evidence for a diff erence in all cortical and subcortical volumes (p≤0·0030), except for the occipital lobe, which was not signifi cant (p=0·0598). The cerebellum had a less signifi cant diff erence than the cortical and subcortical volumes (p=0·0211). We noted diff erences in group means between mutation carriers and noncarriers at the earliest timepoint for the insula (10 years before expected symptom onset) followed by the temporal lobe (also 10 years before expected symptom onset, but with a less signifi cant diff erence; table 3 and fi gure). We noted diff erences in the frontal lobe, all subcortical volumes, and whole-brain volume between carriers and non-carriers at 5 years before expected onset, whereas we noted diff erences in the parietal lobe and cingulate only just before expected time of onset (table 3, fi gure, and appendix). Although we noted only weak evidence for a diff erence between mutation carriers and non-carriers, the results suggest that signifi cant diff erences might exist in the occipital lobe at 5 years after symptoms onset and in the cerebellum at 10 years after symptoms onset.
When we analysed the individual genetic groups separately, we noted a diff erent ordering of cortical and subcortical involvement in each group (appendix): in the MAPT group, we noted diff erences between mutation carriers and non-carriers in the hippocampus and amygdala at 15 years before expected onset, followed by the temporal lobe at 10 years before expected onset, and the insula at 5 years before expected onset; in the GRN group, we noted diff erences between carriers and non-carriers in the insula at 15 years before expected onset, then in the temporal and parietal lobes at 10 years before expected onset, with the earliest subcortical area aff ected being the striatum at 5 years before expected onset; and in the C9orf72 group, subcortical areas including the thalamus, the insula, and posterior cortical areas diff ered between carriers and controls at 25 years before expected onset, followed by the frontal and temporal lobes at 20 years before expected onset. We noted signifi cant diff erences in the cerebellum presymptomatically in the C9orf72 group at 10 years before expected onset. Examination of the laterality index showed evidence for asymmetry between left and right cortical volumes in the GRN mutation carriers (p=0·0001 vs non-carriers), but not in the MAPT carriers (p=0·3283 vs non-carriers) or C9orf72 carriers (p=0·2018 vs non-carriers). GRN mutation carriers showed signifi cantly greater asymmetry than non-carriers at 5 years before expected onset (appendix).

Discussion
We have shown that imaging changes can be identifi ed at least 10 years before expected onset of symptoms in genetic frontotemporal dementia. Structural neuroimaging identifi es a sequence of change in atrophy through cortical and subcortical regions, with the insular and temporal cortices aff ected initially (around 10 years before expected symptoms onset), followed by the frontal cortex and subcortical areas (around 5 years before expected onset), parietal and cingulate cortices (around time of expected onset), and, lastly, the occipital cortex (5 years after expected onset) and cerebellum (10 years after expected onset). We noted that neuropsychological measures were fi rst diff erent between carriers and noncarriers later than initial imaging measures, up to 5 years before expected symptoms onset. These fi ndings suggest that the disease process signifi cantly precedes onset of symptoms in genetic frontotemporal dementia. Whereas previous studies have shown inconsistent fi ndings (panel), the value of investigation of a large cohort of presymptomatic participants is confi rmed in this study, consistent with similar approaches previously done in patients with familial Alzheimer's disease 8 and patients with Huntington's disease. 7 The fi ndings from this study are consistent with our understanding of the earliest structural changes in frontotemporal dementia. The insula is thought to act as a crucial hub in many key networks that become aff ected (particularly the so-called salience network connecting the insula, frontal lobe, and anterior cingulate, and frontoparietal networks). 25,41,42 Here, we noted that the insula was the fi rst cortical area to show evidence of atrophy in the mutation group as a whole, and was one of the earliest areas aff ected in the analyses of each individual genetic group, suggesting that it might be an early focus of pathology followed by connectivity-based spread of disease.
Our primary analysis focused on genetic fronto temporal dementia as a single group. The rationale for this decision lies in the shared clinical features and overlapping disease mechanisms seen in genetic frontotemporal dementia. However, diff erences have been shown between genetic subgroups in previous neuroimaging studies, 43,44 and signatures of network disintegration with particular genetic proteinopathies are predicted on both empirical and theoretical grounds. 45 Our exploratory analyses are consistent with and extend this previous work. In the MAPT group, temporal lobe and medial temporal structures (the hippocampus and amygdala) were aff ected initially, consistent with previous fi ndings suggesting that the disease is a temporal-predominant disorder. 18,43,46 However, this study shows that signifi cant changes can be seen in these areas much earlier than previously suggested. In the GRN group, the insula was the fi rst area aff ected (around 15 years before expected onset), followed by the temporal and parietal lobes. Consistent with previous neuroimaging studies of symptomatic carriers showing early temporal and parietal involvement in patients with GRN mutations, 11,43,46 fi ndings from this study identify the insula as the key region aff ected signifi cantly earlier than other areas. Distinct from the other groups, the earliest subcortical involvement in the GRN group was in the striatum (around 5 years before expected onset), an area known to be involved in symptomatic GRN mutation cases, but not previously shown presymptomatically. 47 In the C9orf72 group, the thalamus and more posterior cortical areas were aff ected early. No previous presymptomatic studies of this group have been done, but previous imaging analyses of symptomatic carriers suggest that the thalamus is a key area aff ected in people with C9orf72 expansions and that posterior areas are more involved than in the other two genetic groups. 43,44 Similarly, the cerebellum has been identifi ed as an area aff ected in symptomatic C9orf72 expansion carriers, and here we show evidence for presymptomatic involvement. The exploratory analysis suggested very early detectable structural imaging changes, particularly in the C9orf72 group, more than 20 years before expected symptoms onset. The timing of Figure: Standardised diff erence between all mutation carriers and non-carriers in cortical grey matter volumetric imaging measures versus estimated years from expected symptoms onset Individual datapoints not plotted to prevent disclosure of genetic status. The time at which the upper 95% CI for each curve crosses zero on the y-axis (ie, the point at which a signifi cant diff erence exists between mutation carriers and non-carriers) is shown on the x-axis. Individual curves with 95% CIs are shown in the appendix. Subcortical and cerebellar volumes are also shown in the appendix. presymptomatic involvement before expected symptoms onset might, to some extent, result from limitations of the simple linear association used in modelling, but this intriguing fi nding needs further investigation and could be consistent with the very slow progressive change in symptoms seen in some patients with C9orf72-related frontotemporal dementia. [48][49][50] Another possibility is that some of the very early diff erences between mutation carriers and non-carriers in the C9orf72 group represent diff erences in brain volume that are, in fact, developmental and longstanding, with superimposed atrophy only late in the disease process.
A key strength of this study is its ability to show robust presymptomatic diff erences in clinical and imaging biomarkers in genetic frontotemporal dementia. However, we analysed only cross-sectional diff erences between carriers and controls at diff erent times from expected symptoms onset. Whether the apparent progression of atrophy through a sequence of cortical and subcortical regions is followed within individuals remains to be shown in a longitudinal study. A further limitation of the study is the method used for estimation of age at onset in presymptomatic mutation carriers. Despite our initial analysis showing a signifi cant correlation between actual age at onset in symptomatic carriers and mean familial age at onset, this measure is imperfect, with variability in age at onset within a family in all frontotemporal dementia mutations. This variability is greater for C9orf72 and GRN mutations than for MAPT mutations, which could lead to greater error in estimated time to onset in these subtypes than in the MAPT subtype (and could therefore suggest that changes can be seen earlier than actually occur). Another limitation of the study is its ability to detect subtle neuropsychiatric or neuropsychological abnormalities. The behavioural and cognitive battery used in the study includes a series of standard validated tests, but these tests might not have suffi cient sensitivity for diagnosis of subtle cognitive or neuropsychiatric dysfunction identifi ed with experimental tests.
In further studies, imaging, genetic, biochemical, and cognitive measures might be able to be combined to identify changes even earlier than noted here. Findings from initial studies [19][20][21][22][23][24][25] suggest that presymptomatic diff erences between carriers and non-carriers of mutations associated with frontotemporal dementia might be seen with other imaging methods, such as diff usion tensor imaging and resting-state functional MRI. Findings from presymptomatic studies of Alzheimer's disease 8 also suggest earlier changes in ¹¹C Pittsburgh compound B PET and CSF measures than diff usion tensor imaging and resting-state functional MRI. Although no fl uid biomarkers have been identifi ed for frontotemporal dementia, tau PET scanning is now available 51 and will be important to examine`` in this cohort as the GENFI study progresses. Our fi ndings suggest that some readily measurable markers can show rates of decline before symptom onset in frontotemporal dementia; if confi rmed in the longitudinal stages of the GENFI study, these measures could be suitable for use in clinical trials and, we hope, contribute to development of preventive strategies.

Contributors
JDR drafted the initial version of the report and the fi gures. JMN did the statistical analysis. RvM, SM, ER, HT, LB, and BN did genetic analyses. All authors recruited patients, collected data, and contributed to reviewing and editing of the report.

Systematic review
We searched PubMed for articles on presymptomatic studies in genetic frontotemporal dementia up to Nov 16, 2014, using the following terms: "frontotemporal dementia AND genetics" and "frontotemporal dementia AND presymptomatic". We identifi ed one review article of presymptomatic studies in genetic frontotemporal dementia, 5 and 18 original research studies that had investigated neuropsychology or neuroimaging, or both, in presymptomatic genetic frontotemporal dementia (appendix) 9-21 A few case studies 9,11,16,17 and two other studies 15,22 have shown evidence of presymptomatic abnormalities on neuropsychometry in asymptomatic mutation carriers, usually with tests of executive dysfunction. However, fi ndings from some other studies have not shown any abnormalities before onset. 13,19,21,[23][24][25] In two single case studies 9,11 and two small case series 12,13 of presymptomatic GRN mutation carriers, focal brain atrophy has been shown a few years before symptoms onset using volumetric T1 MRI, with the prefrontal cortex being predominantly involved, often in an asymmetric pattern. MAPT carriers have been studied less than GRN carriers, with a single case study 17 and a small case series 18 showing presymptomatic atrophy, with hippocampal involvement predominating. We identifi ed no presymptomatic studies of C9orf72 mutation carriers. Some studies have focused on other types of MRI in GRN and MAPT carriers, particularly diff usion tensor imaging and resting-state functional MRI; [19][20][21][22][23][24][25][26] however, Borroni and colleagues, 19,21 Whitwell and colleagues, 20 and Dopper and colleagues 22 also did voxel-based morphometry analyses using volumetric T1 imaging in their studies and did not fi nd any diff erences between asymptomatic carriers and controls.

Interpretation
This work is the fi rst multicentre study of presymptomatic genetic frontotemporal dementia and identifi es structural imaging changes around 10 years before expected onset, and cognitive impairment around 5 years before expected onset, when the genetic group is investigated as a whole. Exploratory analyses suggest that diff erent cortical and subcortical areas are aff ected earliest in each of the MAPT, GRN, and C9orf72 groups, and that structural imaging changes can be seen 15 years or more before symptoms onset. Our results provide an insight into the early neuroanatomical changes in genetic frontotemporal dementia and suggest the potential for use of structural imaging measures as biomarkers in future therapeutic trials.