Underlying genetic variation in familial frontotemporal dementia: sequencing of 198 patients

Frontotemporal dementia (FTD) presents with a wide variability in clinical syndromes, genetic etiologies, and underlying pathologies. Despite the discovery of pathogenic variants in several genes, many familial cases remain unsolved. In a large FTD cohort of 198 familial patients, we aimed to determine the types and frequencies of variants in genes related to FTD. Pathogenic or likely pathogenic variants were revealed in 74 (37%) patients, including 4 novel variants. The repeat expansion in C9orf72 was most common (21%), followed by variants in MAPT (6%), GRN (4.5%), and TARDBP (3.5%). Other pathogenic variants were found in VCP, TBK1, PSEN1, and a novel homozygous variant in OPTN. Furthermore, we identified 15 variants of uncertain significance, including a promising variant in TUBA4A and a frameshift in VCP, for which additional research is needed to confirm pathogenicity. The patients without identified genetic cause demonstrated a wide clinical and pathological variety. Our study contributes to the clinical characterization of the genetic subtypes and confirms the value of whole-exome sequencing in identifying novel genetic variants.

In this study, we systematically assessed a broad set of dementia-related genes in our large cohort of patients with FTD and a positive family history using WES, C9orf72 repeat-primed PCR, and copy number variation analysis. Our objectives were to investigate the frequencies of pathogenic variants in the Netherlands and to identify potential novel variants, which might ultimately provide new pathophysiological insights.

Clinical data collection
Patients were selected from our large FTD cohort in the Netherlands (Erasmus Medical Center, Rotterdam) (Seelaar et al., 2008), which currently includes 656 patients with a clinical diagnosis of either the behavioral variant of FTD (bvFTD) or primary progressive aphasia (PPA), classified into 3 different forms (semantic variant [svPPA], nonfluent variant (nfvPPA), and logopenic variant [lvPPA]). We excluded patients and relatives with a pathological diagnosis other than frontotemporal lobar degeneration (FTLD). The family history was considered positive with the presence of at least one firstor second-degree relative affected by an FTLD spectrum disorder (besides bvFTD and PPA, this includes FTD with motor neuron disease, amyotrophic lateral sclerosis [ALS], progressive supranuclear palsy, and corticobasal syndrome [CBS]) or another type of dementia or Parkinson's disease [PD]). Family history was further classified into one of the following adjusted Goldman categories (Goldman et al., 2005). Psychiatric disorders were not considered in this classification as these were not known for all patients.
(1) Autosomal dominant: 2 relatives with either an FTLD spectrum disorder at any age or another type of dementia or PD < 65 years, occurring in at least 2 generations with one person being a first-degree relative of both other 2; (2) Familial aggregation: 3 relatives (first, second, or third degree) with an FTLD spectrum disorder, another type of dementia or PD at any age, not meeting criteria for autosomal dominant inheritance; (3) Possible familial: 1 firstor second-degree relative with an FTLD spectrum disorder at any age or another type of dementia or PD < 65 years; 4) Possible familial late-onset: 1 first-degree relative with any type of dementia or PD > 65 years; (5) Negative family history: none of the above.
From the total cohort (n ¼ 656), we selected 198 unrelated patients with a positive family history (Goldman 1e4) and DNA availability ( Supplementary Fig. A.1). For 41 familial patients, DNA was not available.

Sequencing and variant filtering
In 38 patients, targeted Sanger sequencing of MAPT or GRN, or C9orf72 repeat-primed PCR had previously revealed a pathogenic variant. WES was performed in 151 patients, and 9 were wholegenome sequenced at the Mayo Clinic Genome Analysis Core as part of another study. As the data were collected from various sources, different capture kits were used (see Appendix A for bioinformatics details). The presence of a C9orf72 repeat expansion was tested using either repeat-primed PCR or a commercial kit (AmplideX PCR/CE, Asuragen), with a repeat length 30 considered pathogenic variants (Renton et al., 2011).
We analyzed 26 prespecified genes, based on an extensive literature search of genes associated with FTD, FTD-ALS, and Alzheimer's disease (AD), as AD may clinically resemble FTD (Supplementary Table A.1). Variants were selected based on the following criteria: (1) affecting coding (missense, nonsense, frameshift) or splicing regions; (2) with a minor allele frequency of <0.1% in the Genome Aggregation Database (gnomAD); and (3) with a quality by depth score 5. The untranslated regions (UTRs) of the genes GRN, MAPT, and TARDBP were investigated for the presence of known pathogenic regulatory variants. Variants reported as pathogenic in the AD&FTD Mutation Database (http:// www.molgen.ua.ac.be/ADMutations) were classified accordingly. We classified novel variants as pathogenic, likely pathogenic, or as VUS in a conservative and systematic approach according to the recently refined guidelines by The American College of Medical Genetics and Genomics (ACMG) (Nykamp et al., 2017;Richards et al., 2015). The following criteria were jointly considered to obtain evidence of pathogenicity: (1) bioinformatic in silico prediction scores: SIFT, PolyPhen2, MutationTaster, FATHMM, combined annotation dependent depletion; score 10), Human Splicing Finder, and MaxEnt; (2) presence in other online genetic databases [OMIM, HGMD, ClinVar, AlzGene, Healthy Exomes (HEX) (Guerreiro et al., 2018)]; (3) existing literature on the variant or a different variant in the same position; (4) segregation analysis if available; (5) functional biomarker if available (blood progranulin levels for GRN); and (6) pathological confirmation of disease if available. Variants reported in the previously mentioned genetic databases as likely benign were only discarded if these reports were consistent and in concordance with in silico prediction tools. Pathogenic and likely pathogenic variants were confirmed by Sanger sequencing.

SNP array and CNV detection
We performed copy number variant (CNV) analysis of the same 26 genes using single nucleotide polymorphism (SNP) array data to identify deletions or duplications in subjects without a pathogenic variant (including those with a VUS). The SNP array platform used was Illumina GSA BeadChip GSA MD, v2 (Illumina GSA Arrays "Infinium iSelect 24x1 HTS Custom BeadChip Kit"). Samples were processed using the Illumina manufacturer's recommended protocol. CNV calling was performed using Nexus Copy Number software (v.4.1, BioDiscovery, Inc, El Segundo, CA, USA) with default parameters.

Neuropathology
Neuropathological examination was available in 76 subjects (46 probands and 30 affected relatives). Immunohistochemistry was performed as previously described (Seelaar et al., 2008), and FTLD diagnosis was based on the criteria by Cairns et al. (2007). The pattern of FTLD with TDP-43 or FET pathology was classified into different subtypes according to the morphology and distribution of neuronal inclusions as proposed by Neumann and Mackenzie (2019).

Frequencies of known pathogenic variants
We detected a pathogenic or likely pathogenic genetic variant in 74 of 198 (37%) patients (Table 1). The most common cause was the C9orf72 repeat expansion identified in 21% (42/198), followed by pathogenic variants in MAPT in 6% (11/198; 6 unique variants), GRN in 4.5% (9/198; 8 unique variants, 3 of which were not reported previously), and TARDBP in 3.5% (7/198, 2 unique variants). Clinical and pathological characteristics of patients carrying genetic variants in these 4 genes are shown in Fig. 1 and Supplementary  Table A.2. Furthermore, we identified 2 different pathogenic missense variants in VCP (1%), one nonsense variant in TBK1 (0.5%), one missense variant in PSEN1 (0.5%), and one novel homozygous variant in OPTN (0.5%). Subsequent CNV analysis performed in all remaining cases (n ¼ 124) did not reveal any deletions or duplications. No cases were identified with a double pathogenic variant, although this could not be excluded in 38 cases tested for single genes.

Novel pathogenic and likely pathogenic variants
The novel OPTN variant is a homozygous splice-site variant (c.1242þ1G>A) in a patient with lvPPA, decreased frontotemporal FDG uptake on positron emission tomography-computed tomography, and a normal profile in cerebrospinal fluid of ptau and amyloid-b, which is incompatible with AD. Family history revealed a sibling diagnosed with nfvPPA and consanguinity between parents (Goldman 3). No other relatives were known to have dementia, PD, or ALS. We considered the variant likely pathogenic  for the following reasons: (1) it is extremely rare in gnomAD (minor allele frequency, 8.8ee06) and has not been reported in the homozygous state; (2) it is predicted to change the canonical splice donor site resulting in skipping of exon 12 (MaxEnt, NNSplice, HSF), leading to a shift of the open reading frame; (3) the variant segregates with the disease as the sibling with nfvPPA carried the same homozygous variant. Three variants in GRN have not been reported previously, including 2 truncating (p.C388LfsX26 and p.C315X) and 1 missense variant (p.W7G). The truncating variants were found in 2 patients with bvFTD leading to death within 5 years. Family history revealed an autosomal dominant pattern in the patient with the C388LfsX26 variant (Goldman 1), whereas the patient with the C315X variant only had 2 relatives with dementia at old age (Goldman 4). Segregation analysis could not be performed because of lack of DNA from family members and serum was not available to measure progranulin levels. However, all truncating variants in GRN are currently considered as likely pathogenic.
The GRN missense variant was identified in a patient who presented with apathy, severe visual hallucinations, fluctuations in cognitive functioning, and a mild asymmetrical hypokinetic rigid syndrome, leading to a differential diagnosis of dementia with Lewy bodies, bvFTD, and CBS. Neuroimaging showed severe left frontal atrophy, suggestive of underlying FTLD. The patient's brother was clinically diagnosed with CBS and also suffered from prominent visual hallucinations. Their father had died at the age of 59 years with severe behavioral and memory disturbances. DNA of these affected relatives was not available for testing. Its pathogenicity is supported by absence in gnomAD, reduced serum progranulin levels (13.4 ng/mL) in the carrier, and the previous report of a different amino acid change in the same codon in 2 other families (p.W7R) (Saracino et al., 2019).

Variants of uncertain significance
We found 15 different VUS (Table 2 and Supplementary  Table A.3). The variant we identified in TUBA4A (p.R105C) seems most relevant, as it was found in a proband with an autosomal dominant inheritance pattern, and segregation analysis revealed the same variant in 4 additional affected relatives (2 with bvFTD and 2 with unspecified dementia), whereas it was absent in an unaffected relative (aged>70 years). Its pathogenicity is further supported by its absence in gnomAD, and in silico tools predict a deleterious effect. FTLD-TDP pathology was confirmed in the proband, with features fitting subtype A. Based on the ACMG guidelines, without supporting functional data thus far, we interpreted the variant as VUS.
Three other variants (K389Rfs*23 in VCP, p.W541C in GRN, and p.P1084S in DCTN1) in patients with familial aggregation (Goldman 2) are potential candidates, but DNA of family members was not available for segregation analysis. The frameshift variant in VCP, due to an insertion resulting in a truncated protein, was found in a patient with bvFTD. Family history was positive for dementia and PD. Its pathogenicity is unknown as frameshift or nonsense variants have not been previously reported in VCP. Therefore, this variant was classified as VUS. The missense variant in GRN (p.W541C), predicted to be damaging, was found in a patient with nfvPPA, but plasma progranulin levels were not available. The p.P1084S variant in DCTN1 was found in a patient with bvFTD and additional semantic deficits, without parkinsonism or motor neuron disease.
For the remaining 11 variants, pathogenicity remains questionable either because of benign or contradictory in silico predictions or because DNA from other family members was not available for segregation analyses. Of note, the VUS in SQSTM1 (p.A33V) was detected in 2 unrelated patients. This variant was also found in the Healthy Exomes database (minor allele frequency, 0.004).

Patients with unknown genetic cause
We did not identify any pathogenic variant, likely pathogenic variant, or VUS in the 26 screened genes in the remaining 108 (55%) patients. Although >75% had Goldman scores 3e4, this group also included 6 (6%) patients with Goldman 1 and 18 (17%) with Goldman 2. The majority (65%) was diagnosed with bvFTD; a relatively large proportion (21%) in this group had svPPA. Other diagnoses included nfvPPA (13%) and lvPPA (1%). Concomitant parkinsonism was present in 14 patients and 6 suffered from ALS. Seventeen patients underwent pathological examination and showed a variety of FTLD pathologies (Supplementary Tables A.3

Discussion
In the present study of a large cohort of familial FTD, we revealed pathogenic variants in 8 FTD-related genes, with the C9orf72 repeat expansion as most common, followed by variants in MAPT and GRN. Furthermore, we identified an unexpected high frequency of the p.I383V variant in TARDBP, a novel homozygous OPTN variant, and 3 novel GRN variants. Finally, we found 15 VUS, including a promising variant in TUBA4A that cosegregated with the disease. The overall frequency of pathogenic variants sums up to 37%. Confining the analysis to patients with a strong family history (Goldman 1e2;n¼70) raises this to 57%. Nonetheless, it indicates that still a substantial proportion of familial cases remains genetically unresolved.

Frequencies of known pathogenic variants
We found relatively high frequencies of variants in MAPT (6%) and TARDBP (3.5%) compared with other cohorts (Supplementary Table A.5). Variants in TARDBP have been reported in around 4% of familial ALS (Zou et al., 2017) but much less often in FTD (Blauwendraat et al., 2018;Ramos et al., 2019Ramos et al., , 2020. Surprisingly, 5 unrelated TARDBP carriers harbored the same variant (p.I383V), suggestive of a possible founder effect. The same variant was found in other FTD cohort screens across the world (Caroppo et al., 2016;Ramos et al., 2019Ramos et al., , 2020. Family history of our patients was not consistent with autosomal dominant transmission (i.e., high Goldman scores), possibly indicating reduced penetrance of this variant, as also suggested by others (Caroppo et al., 2016).
The repeat expansion in C9orf72 is the most common genetic cause of familial FTD in our cohort, accounting for 21% of cases. This is in line with previous studies revealing it as the major genetic cause of familial and sporadic FTD and ALS (Majounie et al., 2012). However, there is substantial geographical variation with frequencies up to 40% in Scandinavian countries (Fostinelli et al., 2018;Oijerstedt et al., 2019;Ramos et al., 2019), contrasting with its absence in Asian cohorts (Kim et al., 2018;Tang et al., 2016). We found a GRN variant in 4.5%, which is less than in other cohorts, especially compared with an Italian study that reported a remarkably high frequency (Fostinelli et al., 2018) (Supplementary  Table A.5). The pathogenic variant in TBK1 (p.W445X) identified in a proband and an affected sibling is the first FTD kindred caused by a variant in TBK1 in the Netherlands. In contrast, other studies have reported variants in TBK1 as the fourth most common genetic cause in FTD (Greaves and Rohrer, 2019). Studies of French and Belgian cohorts found frequencies between 1% and 2% in bvFTD and even higher frequencies in FTD-ALS (Gijselinck et al., 2015;Le Ber et al., 2015;van der Zee et al., 2017).

Novel likely pathogenic variant in OPTN
The presence of a novel homozygous splice-site variant in OPTN (c.1242þ1G>A) in a patient with lvPPA extends the clinical spectrum of OPTN variants, as it has never been associated with this phenotype. OPTN variants are extremely rare in FTD; only a few cases have been described with variants in compound heterozygous state or in combination with a TBK1 variant, which are functionally related genes (Pottier et al., 2015. Homozygous nonsense/ missense OPTN variants were first described to cause autosomal recessive ALS (Maruyama et al., 2010). Subsequently, numerous heterozygous variants were reported in ALS as either disease causing or as risk factor (Markovinovic et al., 2017). Thus far, the proband and sibling with nfvPPA are the first FTD cases without motor neuron disease caused by a homozygous OPTN variant. The parents of our patient were unaffected. A heterozygous variant in the same position was reported in a patient with familial ALS (c.1242þ1delGinsAA) (Belzil et al., 2011). In this case, a second defectdpossibly intronic or a copy number variationdin either OPTN or TBK1 cannot be ruled out because the authors performed targeted sequencing of OPTN only. Others have also suggested a complex mode of inheritance regarding OPTN with an oligogenic basis (Pottier et al., 2015). A recent study on patients with dementia identified heterozygous missense variants in OPTN, but functional or segregation analyses were not available (Bartoletti-Stella et al., 2018).

Variants of uncertain significance
The segregation of a TUBA4A variant (p.R105C)da gene mostly associated with ALSdin several affected family members seems promising. Neuropathologic findings in the proband resembled FTLD-TDP pathology type A. Other groups have also reported likely pathogenic TUBA4A variants in clinical ALS and FTD cases, yet without neuropathologic confirmation, suggesting a plausible role for this gene (Perrone et al., 2017;Smith et al., 2014). Functional studies investigating the pathogenicity of the p.R105C variant are currently ongoing.
A novel frameshift variant in VCP (p.K389Rfs*23) is also a plausible candidate. This variant was found in a patient with bvFTD and familial aggregation, without any symptoms of motor neuron disease or myopathy. Variants in VCP are associated with the classical phenotype of inclusion body myopathy with Paget's disease of bone and frontotemporal dementia (Watts et al., 2004), but cases with pure FTD or ALS have also been described, including 2 other patients in our cohort (Johnson et al., 2010;Wong et al., 2018). Some of the previously reported variants are located in the same D1 domain as this frameshift variant (Abrahao et al., 2016;Watts et al., 2004), which is predicted to lead to a truncated protein. A loss of function mechanism has not been described for VCP. Therefore, segregation and/or neuropathological findings consistent with previous VCP cases are needed to confirm its pathogenicity.
For the other identified VUS in our cohort (Table 2), genetic screens in additional cohorts, segregation analyses, and functional studies should provide further insight. Of note, the p.A33V variant in SQSTM1 has been considered as pathogenic despite the lack of functional evidence (Dols-Icardo et al., 2018;Fecto et al., 2011;Le Ber et al., 2013), and it was detected in controls in another study (van der Zee et al., 2014).

Patients with unknown genetic cause
The wide variety of clinical syndromes and pathologies in the patients with FTD and without an identified genetic cause likely fit various underlying molecular mechanisms. The tau pathology in 4 patients may suggest the presence of unknown causal variants in genes related to MAPT, which may have an impact in its transcription or on the physiology of the tau protein. The strong family history in 2 svPPA cases with confirmed TDP type C was remarkable, as svPPA is nearly always sporadic (Convery et al., 2019). In addition, the presence of FUS pathology in a patient with a family history of dementia, PD, and psychiatric disorders contrasts with the sporadic occurrence of FUS cases in the literature . As FUS is part of the FET protein family, an undefined variant in 1 of the other FET genes, TAF15 or EWSR1, could be considered. Variants in these genes have been reported in a small number of patients, although these were not confirmed to have FUS pathology (Ramos et al., 2019). In our patient, we did not identify any potential causal variants in these genes.
We have not identified variants in patients with bvFTD and concomitant parkinsonism or motor neuron disease, but could not exclude variants in all genes related to these disorders. It might be worthwhile to extend genetic screening to a larger set of genes, as a recent study on sporadic FTD has shown potential variants in genes associated with a variety of disorders (Ciani et al., 2019). Such genetic pleiotropy alludes to an important issue in all nextgeneration sequencing studies: the list of genes associated with neurodegeneration continuously grows, and the phenotypical spectra of different subtypes coincide. In this study, we confined to genes associated with FTD, FTD-ALS, and AD to avoid large numbers of VUS and report those that justify further investigation. Fortunately, as the sequencing data permit constant reanalysis of novel genes, we expect that more and more cases will be resolved over time.

Limitations of the study
As our objective was to give an overview of familial FTD, we focused on cases with a positive family history, representing 43% of our total cohort. Despite our interpretation of a positive family history being rather unconstrained, we might have missed de novo variants or variants with incomplete penetrance in sporadic cases. Several previous studies have revealed GRN variants and C9orf72 repeat expansions in sporadic patients (Blauwendraat et al., 2018;Oijerstedt et al., 2019;Ramos et al., 2019). Nonetheless, we believe that this work reflects clinical practice, where generally familial patients are selected for genetic assessment. As we have not included patients with exclusively psychiatric disorders in the family history, we may have missed the presence of several GRN or C9orf72 carriers (Lanata and Miller, 2016). Finally, as a substantial number of pathogenic variants was identified by targeted singlegene testing, we could not exclude the coexistence of pathogenic variants in other genes (e.g., in C9orf72 carriers) (Giannoccaro et al., 2017;van Blitterswijk et al., 2013).

Conclusions
We present the genetic screen of a large cohort of familial FTD in which we identified a genetic cause in 37% of the patients, including novel pathogenic variants in OPTN and GRN. A large proportion of carriers of the p.I383V variant in TARDBP was found, suggestive of a common founder. We found several VUS, of which the novel variants in TUBA4A and VCP seem most promising. Future studies are needed to confirm their potential pathogenicity. As a whole, our study contributes to the disentanglement of the wide genetic landscape of FTD.

Disclosure statement
The authors declare no conflict of interest.