In this study, we included 57 FTD index patients from the memory clinic, in Centro Hospitalar e Universitário de Coimbra, Portugal, a tertiary reference hospital from the central region of Portugal. The diagnosis was performed according to the most widely accepted criteria (12, 13), and supported by extensive characterization. Our clinical protocol included (1) a complete and systematic clinical and neuropsychological evaluation; (2) structural neuroimaging [computed tomography and/or magnetic resonance imaging]; (3) lumbar puncture to determine Alzheimer's Disease (AD) cerebrospinal fluid biomarkers and/or amyloid PET, to strengthen the diagnosis, by excluding AD.

Patients are from a convenience sample recruited between 2012 and 2016. We included 57 patients, 31 of which were female (53.4%). Mean age of onset was 54.5 (34 to 73) years. All patients were from Portuguese ancestry. Of the whole cohort, twenty patients (34.4%) had positive family history (a first-degree relative with a neurodegenerative condition).

The study was approved by the ethics committee of Coimbra University Hospital and biological samples were obtained following written informed from the patients and/or the patients legal representatives.

Exome sequencing was performed for all patients on a HiSeq2500 with 75-100 bp pair-end reads after library preparation with the SureSelect Exome Capture Kit v4 (Agilent). Exome data processing was done following GenomeAnalysisTK best practices. Alignment was done with the Burrows-Wheeler Aligner (bwa-mem) v0.7.12 against the hg19 genome assembly. Duplicates were identified with samblaster v0.1.21 and bases were recalibrated with GenomeAnalysisTK v3.8-1 (14). Variant Quality Score Recalibration (15), and annotation with snpEff v4.2 and dbNSFP v2.9 were applied to all variants (16, 17). Variants were filtered according to their quality metrics as described by Patel and collaborators (18).

We explored variants in genes reported to cause FTD (10): FUS, TBK1, TARDBP, SQSTM1, CHMP2B, VCP, CHCHD10, UBQLN2, CCNF, OPTN, DCTN1, and TUBA4A. We also explored other genes previously linked to FTD: TREM2, TYROBP, hnRNPA1, hnRNPA2B1 and TMEM106B. We further analyzed variants in genes reported to cause ALS (19), namely: SOD1, ANG, MATR3, FIG4, VAPB, PFN1, SETX, ERBB4, ANXA11, KIF5A, PRPH, GLT8D1, TIA1, DAO, ELP3, TAF15, EWSR1, SS18L1, NEK1, C21orf2, NEFH, CHRNA3, CHRNA4, CHRNB4, PON1, PON2, and PON3. Genes known to cause ALS in an autosomal recessive pattern and genes associated with copy number variations were not examined. All patients were previously screened for C9ORF72 expansions, GRN and MAPT mutations, and were found to be negative. Variants identified in CYLD in this cohort have been reported elsewhere (20).

Only rare variants were included in the study. Variants with a minor allele frequency (MAF) > 0.001 in gnomAD overall sample, or in any of the detailed subpopulations were excluded. In silico pathogenicity prediction was assessed using SIFT (Scale-Invariant Feature Transform) (21), Polyphen-2 (22), PROVEAN (Protein Variation Effect Analyzer) (23), Mutation Taster (24) and CADD (Combined Annotation Dependent Depletion) (25). Scores and cut-offs suggested by the software's authors were considered. In CADD, a score higher than 15 was considered as deleterious. The significant findings were comprehensively assessed by considering MAF, predicted pathogenicity, disease association, and family history.

An adapted version of the Exomiser software (v12.1.0) was used to identify variants that could potentially cause the disease in the individual FTD cases by prioritizing variants related to FTD (HP:0002145) and ALS (HP: 0007354), using an autosomal dominant inheritance model.

The results from the individual runs of Exomiser analyses were filtered by excluding variants without alternative alleles, not passing the Exomiser filter (inheritance), synonymous, with low coverage, and high frequency. Variants that had heterozygous calls in more than three controls and more than 100 cases were filtered out. Variants that were not present in the available databases were kept and were sorted by the Exomiser Gene Combined score in a descending order. A list was compiled with the three top ranked candidate variants separately for FTD and ALS.

An adapted version of the InterVar software (v2.2.1) was used to classify variants based on ACMG guidelines (26). Out of the 28 ACMG criteria, InterVar can automatically apply 18 criteria. The variants in Table 1 were analyzed using the InterVar classification followed by a manual analysis to determine if any of the ten manual criteria were present.

We identified a novel OPTN variant (p.Gln79Arg), predicted to be pathogenic in silico, in a male patient with progressive primary aphasia, starting at 62 years old (patient 5). This variant is absent from gnomAD, has a high CADD score (24.7), and leads to a change in the NEMO-like domain of the protein, thought to interact with TBK1 (32). So far most variants reported in ALS are located at the end of this domain (p.Ala93Pro, p.Lys95Asn, p.Arg96Leu), but there are some variants, namely p.Glu50Lys, which have been shown to be pathogenic, causing normal tension glaucoma (32).

Another variant absent from gnomAD was identified in CCNF (p.Glu125Lys) in a female patient diagnosed with bvFTD at age 48 (patient 57). CCNF encodes the Cyclin F protein that has been shown to cause abnormal ubiquination and accumulation of ubiquinated proteins, including TDP-43 when mutated (33). This variant is predicted to be pathogenic by SIFT, Polyphen 2 and Mutation Taster and is located between the F-box and the cyclin D regions of the protein (33). Other variants have been described in this region, such as p.Lys97Arg, p.Thr181Ile and p.Ser195Arg (33). These variants have been shown to cause the mislocalization of Cyclin F to the cytoplasm and increased association of Cyclin F with VCP, resulting in activation of VCP ATPase, which plays a positive role in TDP-43 aggregation (34).

For both variants, in OPTN and CCNF the in silico analyses performed pointed to potential deleterious effects. In addition, the absence of these variants from gnomAD supports a potential pathogenic role in disease. Still, these variants had uncertain significance when using the ACMG guidelines and further evidence would need to be found to support their pathogenicity.

Other rare variants were identified in DCTN1, FUS and TREM2, but contrary to the ones described above, these are all reported in gnomAD.

The DCTN1 variant (p.Arg409Trp), although rare and predicted to be pathogenic in silico, is located outside the cytoskeleton-associated protein-glycine-rich domain of the protein, where all the variants reported to date and considered to be pathogenic are located (35). This was found in another female patient with bvFTD, starting at the age of 48 (patient 51). This variant was also determined to be of uncertain significance by the ACMG guidelines and further evidence would be needed to support its pathogenicity.

FUS mutations in pure FTD are rare (36). Here, one female bvFTD patient with an age at onset of 53 years (patient 34) carried a missense variant (Pro432Leu), predicted to be pathogenic in silico. This variant has been previously associated with familial essential tremor (37). The proline at position 432 is highly conserved and located in the zinc finger domain. It may affect splicing as it affects the last nucleotide of exon 12. However, no other variants in this domain have been associated with ALS or FTD (36) and this was found to be of uncertain significance by the ACMG guidelines. This patient also carried a rare SETX variant (Table 1).

In TREM2, we identified an heterozygous missense variant (p.Asp39Glu), previously described and of unclear pathogenicity in FTD (38). This variant was found in a female patient with a primary progressive aphasia variant, starting at the age of 61 (patient 10).

One of the main aims of this study was to search for rare, potentially pathogenic, variants in genes associated with ALS with no previous association with FTD. Specifically looking into these genes, we identified rare variants in ERBB4, ANG, SETX, CHRNA4, CHRNB4 and GLT8D1, the latter being a frameshift variant present in one patient.

Recently, a variant in ERBB4 has been described in one FTD-ALS patient (39), supporting the idea that genes associated with ALS may be also implicated in FTD pathogenesis. However, the assessment of cognitive impairment in that patient was complicated by dysarthria and no clear frontal impairment was documented in the study. We report a variant (p.Arg1112His, in patient 23), which is predicted to be pathogenic in silico by SIFT and Mutation Taster. This, together with the previously described ALS-FTD case, gives some support to the potential for this gene to be involved in FTD. This variant was classified as likely pathogenic by the ACMG guidelines.

Angiogenin (ANG) is a pro-angiogenic and neurotrophic factor with an important role in stress-induced injury, that acts by promoting neovascularization and neuronal survival (40). Pathogenic variants in this gene have been implicated in ALS (41). Interestingly, in an ALS-FTD cohort, the levels of angiogenin in CSF were only elevated in patients with FTD (40) and at least one patient reported in the literature seemed to have ALS-FTD (42). The variant we identified (p.Gly123Ala) is located close to the previously reported variants p.Arg121His and p.Arg122His (43, 44). The patient carrying this variant is part of a family with 3 of 4 siblings affected, and this variant was only present in two of them. It was also present in one descendant, currently not affected. This, together with the in silico data, argues against the pathogenicity of this variant, or, less likely, the presence of a phenocopy in the family.

Senataxin (SETX) is associated with juvenile ALS in an autosomal dominant inheritance pattern and with ataxia with oculomotor apraxia type 2 in a recessive inheritance pattern. Senataxin is a protein associated with R-loop regulation. An R-loop forms when an RNA transcript displaces homologous DNA sequence of one of the DNA strands of chromosomes and hybridizes to the complementary sequence of the other strand, generating a short region of RNA-DNA hybrid and single stranded DNA. R-loop dysfunction is increasingly recognized as the cause of several disorders, such as C9orf72 associated ALS (45). In our cohort, we identified three rare SETX variants. One patient carried a variant (p.Thr1558Ala, in patient 16) close to a previously reported pathogenic variant (p.Thr1863Ala) (46). Other rare variant identified in this Portuguese cohort was p.Thr372Pro (patient 34), located in the helicase domain. The third variant was p.Ser213Phe, the only one predicted to be deleterious. This variant is also not reported in gnomAD. This variant was found in a male patient with bvFTD starting at the age of 55 (patient 13).

CHRNA3, CHRNA4 and CHRNB4 have been associated with ALS (47). These genes encode neuronal nicotinic acetylcholine receptors and regulate Ca2⁺ influx to the neurons. Rare missense variants in the intracellular domains of these proteins have been shown to be more prevalent in sporadic ALS patients (48). They also have been shown to be associated with cognitive performance (49). In particular, CHRNB4 has been linked to the modulation of attention in attention deficit/hyperactivity disorder (50). Together, these data point to the possible role of variability in FTD risk and/or pathogenesis. We identified three rare variants in this gene cluster, with two variants located in CHRNB4 (p.Leu60Phe, in patient 29, and p.Val220Met, in patient 23) being particularly relevant, as they are very rare or absent in gnomAD and predicted to be pathogenic in silico by five of five (p.Leu60Phe) and three of five (p.Val220Met) tools (Table 1). However, all three of these variants were classified with uncertain significance by the ACMG guidelines.

Finally, we found one female patient (patient 42) with an age at onset of FTD at 66 years carrying a frameshift variant in GLT8D1, a gene recently associated with ALS (51). It encodes a glycosyltransferase enzyme of unknown function, possibly linked to ganglioside synthesis. GLT8D1 has also been linked with schizophrenia (52, 53). So far, the vast majority of variants reported to cause ALS are missense and located in exon 4 (51). One variant in exon 10 has also been described in a Chinese cohort (47). The variant identified in our cohort (which is not present in gnomAD) leads to a change in the same glycosyltransferase domain of the protein but is predicted to cause a premature stop codon. It is unclear if variants causing a truncated protein will have the same functional effect as the missense variants previously identified in ALS. However, given that in this case we were able to assess the sister of the proband that also presented with FTD-ALS and did not carry the variant, the likelihood of pathogenicity for this variant is very low.

This study has some limitations, including the small size of the cohort, the lack of pathological diagnoses and the unavailability of samples from family members to test co-segregation of variants with disease. Only one of the FTD patients had a pathological confirmation of diagnosis, although all patients in the cohort were clinically evaluated and thoroughly studied in an experienced dementia focused centre. Due to the absence of family history or the unavailability of samples from family members we did not perform segregation analyses to further study the pathogenicity of most of the reported variants. However, the frequency of the variants, together with the use of in silico prediction tools and functional annotation allowed us to weight the possible role of the identified variants.

This study described for the first time the genetic variability beyond GRN, MAPT and C9ORF72, in a clinically well characterized cohort of FTD cases from Portugal. The identification of rare variants in FUS, SETX, CHRNB4, CHRNA4 and ERBB4 in FTD patients provides additional support for a role of ALS-associated genes in pure FTD cases. This expands the spectrum between FTD and ALS on a genetic level. Replication and extension of these findings can have a significant impact on the genetic counseling of affected patients and families and on the understanding of FTD mechanisms.