The Genetics of TDP-43 Type C Neurodegeneration

Abstract

Background and Objectives

Frontotemporal lobar degeneration TDP43 type C (TDP-C) is a rare and unique neurodegenerative disease that attacks the anterior temporal lobe. Recently, it was shown that Annexin-A11 and TDP-43 coaggregate specifically in TDP-C. Current literature on the genetic associations with TDP-C, reviewed here, lacks a discernible corpus of robust or replicated findings. In this study, using blood tissue, we completed whole genome sequencing to investigate ANXA11 and TARDBP genetic variants for their association with TDP-C. Then, we completed genome-wide hypothesis-free analyses using artificial intelligence to identify rare pathogenic variants associated with TDP-C.

Methods

(1) We tested common variants in ANXA11 and TARDBP for their association with 37 TDP-C cases vs 290 controls. We attempted to replicate our findings in a different cohort of 467 TDP-C cases vs 3,153 controls and contrasted them with cohorts of TDP-A and TDP-B. (2) AI-guided analyses were completed to prioritize pathogenic rare variants associated with TDP-C in our cohort.

Results

(1) Four common variants in ANXA11 (rs113772135, rs2789686, rs1079242, rs61860017) were significantly associated with TDP-C in the discovery cohort and replicated in the other cohort of TDP-C but not in TDP-A or TDP-B, providing evidence for ANXA11 specific association with TDP-C. Rs1079242-A showed the most robust replication (p = 7.35 × 10⁻⁰⁵) and correlates with higher ANXA11 level in CSF (p = 4 × 10⁻¹¹). No associations were found between TARDBP and TDP-C (p > 0.05). Using AI-guided rare variant analyses, we identified a pathogenic variant in FIG4, a gene that has been implicated in amyotrophic lateral sclerosis (ALS). Because of the observed potential genetic overlap between some ALS genes and TDP-C, we leveraged mendelian randomization and found that ALS genetic load is associated with TDP-C risk (p = 0.0046).

Discussion

This study provides replicated evidence for the association between common variants in ANXA11 with TDP-C. Knowing rs1079242-A affects ANXA11 level in CSF, future studies may aim to investigate ANXA11 level as potential CSF biomarker for TDP-C. Moreover, FIG4 and ANXA11 have been implicated in the inositol pathway. Our results provide novel insights into the genetic risk of TDP-C and offer new clues about its underpinning mechanisms.

Introduction

Neurodegenerative disorders are a major cause of morbidity and mortality with no available treatments that stop the disease progression, which is partly due to the limited understanding of the underlying pathogenesis.^1,2 Frontotemporal lobar degeneration type C (TDP-C) is a neurodegenerative disease that consistently starts by attacking the anterior temporal lobe (ATL).³ The resultant syndromes depend on the hemispheric asymmetry of atrophy, in which typically left ATL variant leads to semantic variant primary progressive aphasia (svPPA) while right ATL variant leads to associative agnosia or behavioral variant FTD syndrome (bvFTD).^4-8 TDP-A and TDP-B neurodegenerative syndromes although associated with the same proteinopathy differ from TDP-C in which they do not have a consistent predilection for ATL and have neuronal cytoplasmic inclusions preponderance while TDP-C is characterized by long dystrophic neurites concentrated mostly in upper layers of cortex.^4-6,9

TDP-43 (transactive response DNA binding protein of 43 kDa) is an intranuclear protein encoded by the TARDBP gene that is involved in RNA splicing, stabilization, and regulation of genes expression.^9,10 The phosphorylated form of TDP-43 is a hallmark of amyotrophic lateral sclerosis (ALS) and a subset of FTD disorders.¹⁰ Most recently, annexin A11 (encoded by ANXA11 gene) was found to coassemble with TDP-43 in aggregates in all TDP-C cases but much more rarely, if at all, in TDP-A or B.^11,12 The unique characteristics of TDP-C makes this disease a promising target for disease-modifying interventions once its pathogenesis is identified.

Arguably, one of the first steps to understand this disease pathogenesis is through identifying its associated genes, which can shed light to the implicated biological pathways and potential pharmacologic targets. TDP-C is the least studied neurodegenerative syndrome from the molecular perspective, in which no known pathogenic genetic variants have been robustly identified.^13-15 Current evidence suggests that both rare and common genetic variants may play a role in TDP-C risk.¹⁶ Some familial presumptive TDP-C cases have been reported,¹⁷ with existing evidence suggesting rare genetic variants are likely to play a role in TDP-C risk.¹⁸ Although common genetic variants have been proposed to contribute to the disease pathogenesis,¹⁹ the small sample size (even in international consortia) of this rare disease has been a limitation against identifying robust common variants.

Given the paucity of this literature and its rather meager content of pathologically proven TDP-C cases, we review the reported genetic associations in Table 1 to provide context for our investigations. The prior studies were limited either by lack of pathologic confirmation of the diagnosis (for example, none of the reported TARDBP associations were based on pathologically confirmed TDP-C). Most—if not all—studies were limited by reduced power for traditional whole genome search strategies and lack of replication. In addition, the unique coaggregation of TDP-43 and annexin A11 has not been known to previous genetic work, and thus, hypothesis-driven studies did not focus on ANXA11.

Table 1.

Review of the Literature of Previously Implicated Genes in Confirmed and Probable TDP-C Neurodegeneration

Genes	Approacha	Study
C19orf52 (common) TRPC4 (common)a TMEM135 (common)a LRP1B (common)a COL22A1 (common)a RBPJL (rare) L3MBTL1 (rare) ANO9 (rare)	WGS of blood- or brain-derived DNA from combined 199 pathologically confirmed TDP-C cases and 268 svPPA cases (no pathologic confirmation) vs 3,153 controls.a In subgroup analysis of the pathologically confirmed cases, associations were found with TRPC4, TMEM135, LRP1B, COL22A1. All common variants were not tested for replication	16
791 differentially expressed genes in the FTLD-TDP-C group compared with controls	Bulk RNA-sequencing on the frontal cortex from pathologically confirmed TDP-C cases (n = 22) and controls (n = 24)	e1
HNF4A, NR5A1, TAL1, SLC2A4, PSEN1, KRT81, MYBL2, UBE2I, EBI3, BATF, ARFRP1, NR6A1, PACS1, PELP1, and TEF	Bio-informatic search for common genetic variants supported by gene expression data. Cases in first stage: Publicly available 361 with semantic dementia vs 4,308 controls. Cases in s stage: Publicly available data set FTLD with ubiquitinated inclusions (n = 10) vs controls (n = 11)a	19
ANXA11 (rare)	Candidate gene study. One out 34 pathologically confirmed TDP-C cases had possible pathogenic ANXA11 rare variant	12
ANXA11 (rare)	Case report of anomia and prosopagnosia with predominant right-ATL syndrome, no pathologic diagnosis	e2
ANXA11 (rare)	Candidate gene study for possible pathogenic variants in ANXA11 in 2 FTD cohorts in Korea, reported ANXA11, c.119A > G (p.D40G) in 6 out of 109 clinically diagnosed svPPA cases. Only 1 patient later developed ALS symptoms	e3
TARDBP (rare)	Candidate gene study of 122 svPPA/SD cases, no pathologic diagnosis	18
TARDBP (rare)	Deep whole exome sequencing searching for brain somatic variants from the middle temporal gyrus, in which 2/16 cases with TDP-C confirmed pathology were found to have somatic TARDBP variants. This study excluded germline variants	e4
TARDBP (rare)	Case series of FTD syndromes with TARDBP pathogenic variants, 5 of 29 patients with TARDBP variants have svPPA/SD, no pathologic diagnosis	e5
TARDBP (rare)	Case report of a patient with svPPA/SD and a sibling with ALS, no pathologic diagnosis	e6
TARDBP (rare)	Case report of family with TARDBP variant, in which three members of the family had different age at onset and clinical presentations (1 with ALS, 1 with SD, 1 with bvFTD-MND). They all carried c.1147A!G (p.I383V) missense pathogenic variant in the TARDBP gene coding sequence, no pathologic diagnosis	e7
TARDBP (rare)	A cohort of 149 FTLD-MND cases and 400 controls screened for TARDBP pathogenic variants. One patient with svPPA/SD presentation followed by MND had TARDBP p.Gly295Ser variant. One sibling with the same variant had ALS presentation, father with ALS presentation and unknown genotype, no pathologic diagnosis	e8
TARDBP (rare)	Candidate gene analysis of TARDBP in 391 Chinese patients with sporadic ALS. One patient with TARDBP variant had ALS with sPPA presentation	e9
TARDBP (rare)	Candidate genes study in FTD Chinese cohort, including 41 clinically diagnosis svPPA. One svPPA case had rare variant in TARDBP. Other genes were linked with other svPPA cases, but those genes were known not to be associated with TDP-C	e10
TARDBP (rare)	Screening for TARDBP I383V variant in 3 Netherlands cohorts with FTD. Thirteen patients with FTD with TARDBP I383V variant presented with a combination of behavioral changes and semantic deficits and predominant and severe atrophy of the temporal lobes. Two cases had neuropathology data showing scarcity of TDP-43 reactivity in the cortical areas, not classifiable as any TDP subtype	e11
TARDBP (rare)	Candidate genetic panel screen in 118 unrelated Greek FTD cohort. Two clinical svPPA cases out of 14 had pathogenic TARDBP genetic variants	e12
TUBA4A (rare)	Whole exome sequencing of a patient with SD diagnosis and family history of parkinsonism; pathologic diagnosis confirmed TDP-C. Patient was found to have novel frameshift variant in the TUBA4A gene leading to a decrease in total TUBA4A mRNA and protein levels	e13
ANXA11 (common, REPLICATED) UBQLN2 (rare) FIG4 (rare)	This study leveraged WGS on a cohort of autopsy confirmed (n = 22) or probable (n = 15) TDP-C cases with svPPA (total n = 37). Common variants in ANXA11 were associated with TDP-C based on hypothesis-driven analyses and replicated in a larger cohort. Rare variants in UBQLN and FIG4 were identified in AI-guided genome wide analyses, mendelian randomization analysis showed association between ALS genetic load and TDP-C	This study

Abbreviations: ALS = amyotrophic lateral sclerosis; svPPA = semantic variant primary progressive aphasia; TDP-C = TDP43 type C; WGS = whole genome sequencing.

^aThe source of DNA for the included studies is blood tissue unless otherwise specified. Blood tissue was assigned as a source of DNA when explicitly mentioned or no explicit source of tissue is mentioned in the referenced studies.

In this study, we leveraged a well-characterized (clinically and pathologically) cohort of TDP-C from Northwestern University.⁵ Using blood samples from pathologically confirmed or probable TDP-C cases and using whole genome sequencing (WGS), we ran 2 sets of analyses. We first completed hypothesis-driven analyses focusing on TARDBP and ANXA11 genes, knowing the recent discovery of their coassembly specifically in TDP-C cases. Then, we completed genome wide hypothesis-free analyses guided by artificial intelligence (AI) to identify and prioritize rare pathogenic variants associated with TDP-C.

Methods

Review of Literature

We created Table 1 to review the genetic associations reported in the literature with TDP-C. We used PubMed to review the relevant literature and included original, review, and case report papers that were published prior to the initial submission of this work on May 1, 2025. We used the following terms of the search: “semantic variant primary progressive aphasia,” “semantic dementia,” “TDP43 type C,” “genetic,” and “gene.” We also reviewed the references included in selected publications for further review. We only reported associations specific to TDP-C and did not include associations with FTLD-TDP in general (knowing the different clinical and neuropathologic characteristics between TDP-C and another FTLD-TDP pathologies and to reduce heterogeneity). Thus, we intentionally did not include clinical svPPA cases that were associated with genes known to be associated with TDP-A or TDP-B (such as GRN, C9orf72, TBK1 etc.). We also did not include genes associated with svPPA due to tauopathy (such as MAPT). We included large cohorts, family studies, and case reports in Table 1. We explained if the pathology was confirmed by autopsy or not. We also added our findings at the end of the table for the reader to be able to compare them with other studies.

WGS Study in the Northwestern University TDP-C Cohort

Northwestern TDP-C Cohort

Premortem blood samples from TDP-C cases (n = 37) were included in this study, which were selected from the clinically and pathologically well-characterized PPA cohort of the Mesulam Center at Northwestern University. In brief, 22 cases had confirmed TDP-C pathology, and 15 cases were probable TDP-C cases [characterized with a clinical syndrome most likely (ML)—with about 90% correlation—to be because of TDP43-C (svPPA/semantic dementia (SD), temporal pole as the most predominant region of atrophy, and when available biomarkers not consistent with Alzheimer disease)].⁵ The 15 probable TDP-C cases were either still alive at the time of conducting the analyses or did not consent for brain autopsy at death. Clinical diagnosis was based on the clinical evaluation of a single behavioral neurologist with expertise in PPA (M-Marsel Mesulam), using criteria described elsewhere.⁶ In brief, the svPPA diagnosis is based on prominent impairment of single word comprehension, and severe anomia (reflecting word comprehension as well as retrieval failures). Grammar and repetition are typically preserved, while speech can be vague and contains usually circumlocutions and semantic paraphasia.⁶ MRI of enrolled patients with svPPA showed peak atrophy in dominant anterior temporal pole.⁶ Demographic and clinical characteristics for the included cases are summarized in Table 2. The pathologic characterization of the first 10 of the autopsied cases has been discussed elsewhere (for review, please refer to Kawles A., et al).⁹

Table 2.

The Demographic and Clinical Characteristics of the TDP-C Cohort

Cases	N	Females (%)	Mean age onset (range)	Mean age at death (range)	APOE2 carriers (%)a	APOE4 carriers (%)a	Phenotypes other than svPPA
Total	37	20 (54)	59.3 (50–79)	NA	5 (14)	5 (14)	—
Autopsy confirmed	22	14 (64)	—	72 (61–93)	—	—	1 m-PPA 1 SD 1 bvFTD
Clinically diagnosed	15	6 (40)	—	NA	—	—	1 SD 1 isolated anomia

Abbreviation: svPPA = semantic variant primary progressive aphasia.

^aNo cases with homozygous APOE 2,2 or APOE 4,4, and no cases with APOE 2,4 genotype.

WGS

WGS was performed using PCR-free library preps from Illumina to a cohort average coverage of × 35. Sequencing was performed using paired-end 151bp chemistry on the Illumina NovaSeq-X. Alignment and preprocessing were conducted using NVIDIA parabricks germline pipeline v4.1.0. The workflow includes alignment with BWA-mem, BAM sorting, marking duplicates, and base-quality score recalibration. gVCFs were combined for variant joint call using GLnexus v1.2.6.

Hypothesis-Driven Analyses Focusing on Common and Rare Variants in TARDBP, and ANXA11 vs Controls

We selected the regions of interest for the candidate genes analyses from a genome association search completed on the 37 TDP-C cases vs 290 controls (242 after quality control; further details available in eMethods and eTable 1). We investigated common and rare variants in genes that have sufficient pathologic evidence in the literature for being likely pathogenic for TDP-C based on forming histopathologic protein aggregates in TDP-C cases (TARDBP, ANXA11). We used ANNOVAR to annotate variants as upstream/downstream when at least 1-kb from the transcription start/end site. A total of 30 common ANXA11 variants (MAF >5%), pruned at linkage disequilibrium threshold of R² 0.6 (PLINK option: --indep-pairwise 50 5 0.6), were included in the analysis. Total of 247 rare ANXA11 variants (MAF <1%) were included in the analysis. A total of 2 common TARDBP variants (MAF >5%), pruned at linkage disequilibrium threshold of R² 0.6, were included in the analysis. A total of 60 rare TARDBP variants (MAF <1%) were included in the analysis. p values for rare variants analyses were adjusted using the Benjamini and Hochberg method.

Replication Analyses for Common Variants

To assess the robustness of our common variants significantly associated with TDP-C in the discovery cohort from Northwestern University, we attempted to replicate the results in a different cohort (the international FTLD-TDP Whole-Genome Sequencing Consortium).¹⁶ The replication cohort used whole blood– or brain-derived DNA from pathologically confirmed and clinical frontotemporal lobar degeneration cases from the 3 main pathologic subtypes TDP-A (all pathologically confirmed n = 193), TDP-B (total n = 288, pathologically confirmed 222, clinical FTD-MND 66), and TDP-C (total n = 467, pathologically confirmed 199, clinical svPPA 268) as well as 3,153 controls.¹⁶ Of note, the Northwestern cohort (discovery cohort in this study) contributed a very small sample size to the replication cohort. Fourteen out of 467 cases in the replication TDP-C cohort were also in the Northwestern discovery cohort of 37 cases. Removing those 14 cases was not possible because only the summary statistics of the replication cohort were available at the time of completing this study. The replication summary statistics were obtained from the GWAS Catalog Database under accessions codes GCP001210 (GCST90558311, GCST90558312, GCST90558313, GCST9055831).

AI-Guided Analyses to Identify Rare Pathogenic Variants

Using our generated WGS data, variants were prioritized with Illumina's Emedgene AI-based genomic analysis and interpretation platform (Illumina; Emedgene; 2022) and individually inspected using Integrated Genome Viewer browser to verify their quality.^20,21 Emedgene is an AI model and interpretation platform. Emedgene generates a shortlist of potential causative variants along with supporting evidence from the literature and curated databases, providing a proprietary relevance score to classify the variants, significantly reducing the time to interpret a case.²⁰ The automated interpretation machine learning algorithms have been trained on thousands of patient samples and incorporate dozens of features typically used by geneticists to interpret cases.²⁰ The algorithms use a proprietary knowledge graph of variants, genes, diseases, phenotypes, and their connections, which include information extracted from literature with natural language processing.²⁰ Overall, the system automates time-intensive aspects of variant analysis and variant research and enhances laboratory analysis capabilities. For this analysis, we provided a list of Human Phenotype Ontology terms related to TDP-C (olivopontocerebellar atrophy, dysarthria, hypotension, urinary urgency, neurodegeneration, bradykinesia, parkinsonism, postural instability, urinary incontinence, orthostatic hypotension, and rigidity). The cases were analyzed with the AI tier 2 setting, which divides the high score variants into 4 main categories: ML, ML genes with unknown significance (GUS) GUS, Candidates, and Candidate GUS, with ML variants having higher scores than Candidate Variants. The algorithm also divides variants that are in genes with a known disease connection and within GUS and restricts the ML list to the ML variants.

In Silico Functional Characterization of the Identified Likely Pathogenic Variants

Protein structures were downloaded from AlphaFold Protein Structure Database in Protein Data Bank format.²² The target residue was changed using PyMOL v3.0, and the protein stability in both the wild-type and mutated protein was estimated using Foldx v4.²³ Specifically, the protein stability refers to the Gibbs free energy change (ΔG) associated with the protein folding process, which measure how energetically favorable or unfavorable the folded state of the protein is compared with the unfolded state.

Mendelian Randomization Analyses

Knowing that TARDBP, ANXA11, and other reported associations with TDP-C (including our findings) have been implicated in ALS, we sought to evaluate whether the genetic load for ALS is associated with TDP-C risk via leveraging two-sample mendelian randomization (MR) approach. We selected 16 common SNPs associated at a genome wide level of significance with ALS in the largest ALS GWAS to date to be genetic proxy of ALS genetic load. We removed C9orf72 SNP (rs2453555) from the ALS genetic proxy because C9orf72 is known to be associated with TDP43 type A or B, but not TDP-C.²⁴ We then ran a two-sample MR analysis testing if ALS genetic load can be associated with TDP-C risk using the inverse-variance weighted method.²⁵ To assess the robustness of the findings, we then ran MR model-based sensitivity analyses, in addition, to heterogeneity, pleiotropy and outliers analyses, and created scatter plot.^25,26 Genetic associations with exposure and outcome were harmonized by aligning beta coefficients to the same effect allele. The MR analyses were conducted using the R-studio v2024.04.1 + 748 and the TwoSampleMR v 0.4.26 package.

Standard Protocol Approvals, Registrations, and Patient Consents

All participants have provided written consents to enroll in the cohort and for their tissue samples and clinical data to be used and published in research studies. This study was approved by Northwestern University Institutional Review Board.

Data Availability

The summary statistics for the common variants association tests in our discovery cohort are openly available in Synapse database; accession syn64465917.²⁷ The replication summary statistics were obtained from the GWAS Catalog Database under accessions codes GCP001210 (GCST90558311, GCST90558312, GCST90558313, GCST9055831).²⁸ Investigators may request anonymized individual patient data, which will be shared in accordance with the specific study protocol on approval.

Results

TDP-C Genetics Literature Review

A comprehensive review of the literature for genetic variants found to be associated with TDP-C is included in Table 1. This was completed to help compare our results with the existing literature and to guide our hypothesis-driven analyses. We illustrated the reported previous associations based on whether the TDP-C diagnosis was pathologically confirmed as TDP-C or clinically diagnosed. The TDP-C literature in multiple studies hinted to genes that have been also implicated or linked to ALS (TARDBP, ANXA11, L3MBTL1, TUBA4A) (Table 1). TARDBP and ANXA11 have been implicated in TDP-C at the genetic and neuropathologic levels, which was the reason we focused on them in our study. None of the previously reported cases that had a germline genetic association with TARDBP were pathologically confirmed. Most the reported associations in the literature require further studies, validation, or replication. No single gene to date has been robustly associated with TDP-C cases in an autosomal-dominant mode of inheritance (in contrast with TDP-A association with GRN and TDP-B with C9orf72).²⁹ To our knowledge, no previously associated common variants with TDP-C have been replicated.

Hypothesis-Driven Analyses Focusing on TARDBP and ANXA11

We aimed to select all significantly associated common variants in TARDBP and ANXA11 in our discovery cohort and test their association in the replication cohort. In the discovery cohort, we found significant association between 4 common variants in ANXA11 and TDP-C vs controls (Table 3, eTable 2). No common variants in TARDBP were found to be associated with TDP-C vs controls (p > 0.05) (eTable 3). We then selected those 4 common variants from the replication cohort and tested their association with TDP-C vs controls and found that all 4 SNPs were associated with TDP-C vs controls in the replication cohort and had the same direction of effect (rs113772135 was nominally significant at p = 0.07) (Table 3). The most robust replication was for rs1079242 (p = 7.35 × 10⁻⁰⁵), which has been previously associated with an increased ANXA11 protein level in the CSF.³⁰ Then to assess the specificity of the genetic association to TDP-C, we tested the 4 genetic variants for their association with TDP-A and TDP-B. We found that none of the 4 genetic variants were associated with TDP-A or TDP-B (Table 3).

Table 3.

ANXA11 Common Variants Association With TDP-C vs Controls

ANXA11 common variants associated with TDP-C vs controls in our discovery cohort
SNPa	Allele1	Allele2	Minor allele frequencyb	Allele1 carriers in TDP-c/controls	OR	Beta	SE	p Value	Location/function	gnomAD v4.1.0 Allele1 frequency	gnomAD v4.1.0 Total sample
rs113772135	T	C	0.052	9/19	1.3	0.23	0.1	0.00036	UTR3	0.076	76,199
rs2789686	T	C	0.099	13/40	1.2	0.15	0.1	0.002	UTR3	0.154	78,009
rs1079242	G	A	0.498	24/184	0.9	−0.1	0	0.03	Intronic	0.4996	806,174
rs61860017	A	T	0.065	1/33	0.9	−0.1	0.1	0.046	Intronic	0.0401	179,684

ANXA11 variants replication analyses in a larger cohort of TDP-C as well as TDP-A and TDP-B
SNP	Allele1	Allele2	Minor allele frequency	OR	Beta	SE	p Value	gnomAD v4.1.0 Allele 1 frequency	gnomAD v4.1.0 Total sample
TDP-C (n = 467 casesc vs controls 3,153)
rs113772135	T	C	0.06	1.3	0.26	0.2	0.075	0.076	76,199
rs2789686	T	C	0.12	1.4	0.35	0.1	0.001	0.154	78,009
rs1079242d	A	G	0.499	1.3	0.29	0.1	7.35 × 10−05	0.5004	806,174
rs61860017	A	T	0.07	0.6	−0.5	0.2	0.01	0.0401	179,684
TDP-A (n = 193 cases vs controls 3,153)
rs113772135	T	C	0.05	0.7	−0.4	0.3	0.13	0.0764	76,199
rs2789686	T	C	0.11	0.8	−0.3	0.2	0.18	0.1540	78,009
rs1079242d	A	G	0.49	0.9	−0.1	0.1	0.23	0.5004	806,174
rs61860017	A	T	0.06	1.1	0.06	0.2	0.77	0.040	179,684
TDP-B (n = 288 casesc vs controls 3,153)
rs113772135	T	C	0.05	1	−0	0.2	0.85	0.0764	76,199
rs2789686	T	C	0.11	1.1	0.11	0.1	0.42	0.154	78,009
rs1079242d	A	G	0.49	0.9	−0.1	0.1	0.45	0.500	806,174
rs61860017	A	T	0.06	0.8	−0.2	0.2	0.39	0.0401	179,684

Abbreviation: TDP-C = TDP43-Type-C.

^aSNPs were annotated using Annovar software, referencing dbSNP build 151 (GrCh38 reference genome).

^bMinor Allele Frequency (Allele 1) refers to the total sample size of the discovery cohort (both cases and controls).

^cIncludes both pathologically confirmed and probable cases (clinical diagnosis).

^dThe replication analysis reported the other allele as the effect allele; thus, the direction of effect is consistent between the original and replication sample.

We then completed candidate gene analyses comparing the frequency of rare variants in TARDBP and ANXA11 in TDP-C cases vs controls. We found 19 rare variants nominally associated with TDP-C vs controls; however, only single variant rs144114856 in ANXA11 remained statistically significant after correcting for multiple testing (BH adjusted p = 0.003); Table 4 presents the top 5 rare variants associated with TDP-C vs controls (complete results presented in eTable 4). Six rare variants in TARDBP were found to be associated with TDP-C vs controls at nominal level, but none passed multiple testing correction (BH adjusted p > 0.05) (eTable 5).

Table 4.

Top 5 Rare Variants in ANXA11 Associated With TDP-C vs Controls in Our Cohort

SNP	Allele1	Allele2	Minor allele frequencya	OR	Beta	SE	p Value	Location/function	gnomAD v4.1.0 Allele 1 frequency	gnomAD v4.1.0 Total sample
rs144114856	A	G	0.009	2	0.7	0.2	9.43 × 10−06b	Intronic	0.0075	76,149
rs184651165	C	T	0.005	1.7	0.6	0.2	0.005	Intronic	0.0039	76,176
rs1180489738	A	C	0.002	2.5	0.9	0.3	0.006	Intronic	—	—
rs1262499074	T	C	0.002	2.5	0.9	0.3	0.008	Intronic	0.0000066	76,067
rs753662122	A	G	0.002	2.5	0.9	0.3	0.008	Intronic	0.000046	76,089

^aMinor Allele Frequency (Allele 1) refers to the total sample size of the discovery cohort (both cases and controls).

^bStatistically significant after p value multiple testing correction (corrected p = 0.003).

Hypothesis-Free AI-Guided Search for Rare Pathogenic Variants in the Northwestern TDP-C Cohort

In our cohort, 37 confirmed and probable TDP43-C cases were included in the WGS analyses. One case was found to have a pathogenic variant in FIG4, and another a variant of unknown significance (VUS) in UBQLN2 (Table 5, eFigures 1 and 2); both genes were previously implicated in ALS. Further bioinformatic functional analyses have been completed to predict the molecular pathways implicated by those 2 genes. The FIG4 rare variant is a splice donor variant that likely leads to a truncated transcript. The UBQLN2 rare variant causes a moderate decrease in the stability of the mutated protein ((ΔG): −2.899 kcal/mol). Such moderate instability is likely to cause biologically significant changes to related protein functions; thus, we believe that this variant may in the future be reclassified as likely pathogenic.

Table 5.

TDP-C Cases in Our Cohort With Rare Genetic Variants of Interest

Case	Clinical diagnosis	Pathologic diagnosis	APOE	Gene	Variant	Genotype	Allele frequency in gnomAD	Combined annotation dependent depletion (CADD)b	ACMG/AMP criteria	Evidence codes	Note
1	svPPA	Probable TDP-C, with negative clinical amyloid PET scan	3,3	FIG4	NM_014845.6:c.2459 + 1G>A; splice donor variant (truncated)	Heterozygous	0.000026	34	Pathogenic	PVS1, PS3, PM2, PP3c	Gene was implicated in ALSd
2	svPPA	Confirmed TDP-Ca	2,3	UBQLN2	NM_013444.4:c.431G>A NP_038472.2 p.Gly144Glu	Heterozygous, missense variant	0.000018	23.2	Variant of uncertain significance (VUS)	—	Gene was implicated in ALS

Abbreviations: ALS = amyotrophic lateral sclerosis; svPPA = semantic variant primary progressive aphasia; TDP-C = TDP43 type C.

^aCause of death: motor vehicle accident.

^bScore >20 placing the variant in the top 1% of all possible pathogenic variants^e14.

^cPVS1 (very strong): predicted loss-of-function variant in a gene where loss of function is a known disease mechanism; PS3 (strong): well-established functional studies support a damaging effect on gene or protein; PM2 (moderate): variant absent or extremely rare in population databases; PP3 (supporting): multiple computational prediction tools support a deleterious effect on the gene or gene product.

^dA homozygous genotype of the same variant has been reported in an autosomal recessive form for child onset leukoencephalopathy due to demyelination.³⁴

We then sought to further evaluate those 2 cases from a clinical and neuropathologic aspect. The first case with a variant in FIG4 is a patient with svPPA presentation with predominant left temporal pole atrophy (Figure1A), and negative clinical Amyloid PET scan. The svPPA diagnosis was based on expert assessment (M-Marsel Mesulam) and neuropsychological data showing that the most predominant cognitive impairments were in naming (using Boston Naming Test) and single word comprehension (using Peabody Picture Vocabulary Test-Form B). This patient had no signs of motor neuron disease. Family history was remarkable for a son formally diagnosed with a learning disability with a need for accommodations throughout school. The parents of this patient did not have dementia per report. The second case with UBQLN2 variant is a patient with svPPA syndrome and TDP-C confirmed pathology (Figure1B). There was also predominant left temporal pole atrophy at presentation, with a similar pattern to the one seen in Figure 1A. There were no signs of motor neuron disease. The family history was positive for a brother with either learning or behavioral disability and was held back in school for a year. The parents of the patient did not have dementia per report.

Figure 1. Imaging and Pathological Data for the TDP-C Cases With Rare Genetic Variants of Interest.

(A) Neuroimaging for the probable TDP-C case of svPPA with FIG4 variant. These MRI-based reconstructed images show predominant left anterior temporal lobe atrophy. (B) Pathologic data for the confirmed TDP-C case with UBQLN2 variant. Histopathologic images showing TDP-43 aggregates of type C pattern. (A) The atrophy map was generated using voxel-based morphometry (VBM) in CAT12 (running in SPM12), by performing voxel-level, nonparametric comparisons of gray matter volume between the participant and a group of healthy, age-matched participants (N = 46). Total intracranial volume (TIV) was included as a covariate in the VBM model to account for differences in head size. Threshold-free cluster enhancement (TFCE) statistics was used to detect statistically significant differences between groups at each voxel, and a family-wise (FWE) correction at p < 0.1 was applied to account for multiple comparisons. The map was then displayed on the FSaverage brain template available within CAT12. The color bar indicates the strength of p values at each voxel, where yellow indicates areas of greater atrophy and darker red areas of lesser atrophy. (B) Sections of paraffin-embedded left inferior parietal (right) and left posterior hippocampus/dentate gyrus (right) were immunostained with antiphosphorylated TDP antibody (pTDP, pSer409/410; CosmoBio, polyclonal rabbit) to visualize TDP-C pathology. Image of hippocampus specifically depicts granule cells of dentate gyrus. Arrows indicate characteristic TDP-C pathologic inclusions, including long dystrophic neurites (red arrows) and cytoplasmic dense round bodies (black arrows). svPPA = semantic variant primary progressive aphasia; TDP-C = TDP43 type C.

Evaluating the ALS Genetic Load Association With TDP-C

The MR analyses included 11 SNPs from the ALS proxy that were available in the TDP-C vs control GWAS summary statistics (eTable 6). We found a significant association between ALS genetic load and TDP-C risk, which was supported by multiple lines of MR sensitivity analyses (Table 6, eFigure 3).

Table 6.

MR Primary and Sensitivity Analyses Testing the Association Between ALS Genetic Load and TDP-C in Our Cohort

MR method	n-SNP	Beta	se	p Value
Inverse variance weighted	11	0.31	0.11	0.0046
MR-Egger	11	0.69	0.48	0.18
Weighted median	11	0.40	0.14	0.005
Weighted mode	11	0.41	0.18	0.04
MR heterogeneity inverse variance weighted				0.54a
MR pleiotropy Egger intercept				0.44b
Outliers testing MR-Global PRESSO				0.57c

Abbreviations: ALS = amyotrophic lateral sclerosis; MR = mendelian randomization.

^aIndicating no heterogeneity.

^bIndicating no pleiotropy.

^cIndicating no outliers.

Discussion

In this study, we report evidence for the association of ANXA11 common variants with TDP-C. To our knowledge, no prior common variants in any gene have been replicated for their association with TDP-C. In our study, 4 common variants in ANXA11 were significantly associated with TDP-C and replicated in a different and larger cohort of TDP-C cases. Those 4 variants did not replicate in TDP-A or TDP-B, showing the specificity of our ANXA11 genetic associations with TDP-C only. The specific genetic association between ANXA11 and TDP-C is in line with the neuropathologic evidence showing that annexin-A11 was found to coassemble with TDP-43 in aggregates in all TDP-C cases and not in TDP-A or TDP-B.¹¹ It is worth mentioning that rs1079242-A, which showed the most robust association with TDP-C in our replication analysis, was previously associated higher ANXA11 protein level in CSF. This observation motivates future studies to investigate ANXA11 level as potential spinal fluid biomarker for TDP-C. Although a very small overlap exists between the discovery and replication TDP-C cohorts, we believe that the more robust association found in the replication analyses points to the fact that increasing the sample size further supports the true association rather than dilute the results. Nonetheless, future genetic and functional studies of ANXA11 will help accumulate more evidence to support the robustness of the reported associations in our cohort.

As listed in Table 1, a number of cases of TDP-C have been previously associated with likely pathogenic rare variants in ANXA11. Although our AI-guided approach did not identify likely pathogenic rare variants in ANXA11 in our cohort, our rare variants candidate gene analysis showed multiple rare variants associated with ANXA11 vs control with high effect size (OR about 2), with most robust association in rs144114856, which is based on RegulomeDB it has 60% probability to be a regulatory variant. Our overall findings provide evidence that in addition, to rare variants, common variants in ANXA11 are also associated with TDP-C risk. The lack of association between TARDBP and TDP-C in our cohort could be due speculatively to the notion that ANXA11 is the main driver of the neuropathology in TDP-C and that the accumulation of TDP-43 is a secondary consequence in the disease cascade. Knowing the small sample size of the cohort, the ANXA11 rare variant results should be considered preliminary and require further validation in future studies.

In our cohort and using comprehensive WGS and AI-guided approach, we identified 2 novel rare genetic variants that are likely to contribute to the pathogenesis of TDP-C. Our results support the hypothesis that TDP-C can be caused by rare genetic variants even in the absence of family history, which could be explained by de novo variants or pathogenic variants with low penetrance. Another potential explanation is the presence of an interaction with other genetic factors that can increase or decrease the risk of developing the TDP-C phenotype among carriers. One such example is the TMEM106B protective variant that protect GRN carriers from developing the FTD phenotype,³¹ or the Christchurch variant that protects APOE4 carriers from developing the AD phenotypes.³² Thus, the family members of genetic TDP-C cases could carry the risk variant but not have the phenotype due to protective genetic variants for example. On the other hand, environmental or epigenetic factors might contribute to the development of the TDP-C phenotype among the genetic variant carriers in a double hit model, in which the genetic variant is necessary but not sufficient to develop the disease. This proof-of-concept work supports further research using WGS to identify rare pathogenic variants in patients with TDP-C even with the lack of family history.

Notably, both identified rare genetic variants in our cohort are in genes that have been associated with ALS, which is also a TDP43 neurodegenerative disease in most cases. FIG4 encodes a phosphoinositide 5-phosphatase, which is implicated in trafficking of endosomal vesicles to the trans-Golgi network.³³ Loss of function of FIG4 in mice results in degeneration of neurons.³³ Compound heterozygous pathogenic variants in FIG4 in humans (indicating severe loss of function) lead to the recessively inherited severe form of Charcot-Marie-Tooth disease with early onset and involvement of both sensory and motor neurons.³³ On the other hand, heterozygous autosomal-dominant nonsynonymous variants of FIG4 (with less severe loss of function phenotype) have been associated with ALS and primary lateral sclerosis.³³ Our reported variant in FIG4 linked to TDP-C here is a heterozygous genotype. A homozygous genotype of the same variant has been reported in an autosomal recessive form for child onset leukoencephalopathy due to demyelination.³⁴ Our finding is in line with literature that homozygous pathogenic variants in FIG4 lead to early onset and severe neurologic disease, while a heterozygous pathogenic variant leads to a late onset neurodegenerative disease. One example of this phenomena can be seen among GRN pathogenic variants carriers, in which heterozygous carriers have late onset neurodegenerative FTD phenotype, while homozygous carriers have severe early onset neuronal ceroid lipofuscinosis-11 syndrome.^35,36

UBQLN2 is an intron-less gene encoding ubiquilin-2 protein, which is a member of the ubiquitin-like protein family (ubiquilins). In humans, there are 4 ubiquilin genes.³⁷ Ubiquilins promote protein degradation through the ubiquitin-proteasome system and macroautophagy.^37,38 UBQLN2 pathogenic variants can lead to loss of function and have been found to impair misfolded proteins degradation and clearance pathways.^37,39-41 In mouse models, pathogenic Ubqln2 was also found to cause neuron death through a toxic gain of function properties.⁴² In a familial ALS cohort, UBQLN2 has been causally associated with ALS and FTD-MND (with behavioral executive dysfunction presentation) in an x-linked manner (no male-to-male transmission).³⁷ Patients with ALS and FTD-MND (with or without UBQLN2 pathogenic variants) have been found ubiquilin-2 brain inclusions.³⁷ In a subsequent genetic screening study of patients with nonfamilial ALS or FTD, 3 novel UBQLN2 pathogenic variants were reported including a patient with pure bvFTD without MND.⁴³

From a mechanistic standpoint, it is of interest that we found an association between TDP-C with ANXA11 and FIG4, both of which are implicated in the inositol metabolism pathway. Dysfunction in this pathway has been linked with excessive and toxic levels of intracellular Ca²⁺ that may contribute to the neurodegeneration process.^44-46 ANXA11 is one of the annexin groups, which are Ca2+-regulated phospholipid binding proteins that are implicated in the cell life cycle, apoptosis, and exocytosis.⁴⁷ ANXA11 encodes a protein that leads RNA granules to lysosomes as they traffic in axons, with an N-terminus that contains a low-complexity domain, and C-terminus with calcium binding domains.⁴⁸ ANXA11 has been shown to bind to phosphatidylinositol 3-phosphate (PI (3) P)-containing giant unilamellar vesicles (composed of lysosomal membrane lipids) in a Ca2+-dependent manner and is dependent on the structured C-terminal annexin repeat domain.^49,50 Taken together, our findings suggest that TDP-C neurodegeneration might emerge because of dysregulation in the inositol-calcium pathway leading to RNA to lysosomes trafficking abnormalities.

Our post-hoc analyses provided further supportive evidence of the association between ALS genes with TDP-C. The MR analyses showed robust association between ALS genetic load and risk for TDP-C, as supported by multiple lines of sensitivity analyses. Despite the relatively small sample size, those analyses further support that the accumulation of common variants associated with TDP-43 related pathways (in this case ALS common variants) might be associated with increased risk for TDP-C neurodegeneration.

In this cohort, we reported the association between common variants in ANXA11 and TDP-C. We also reported 2 novel variants, one a pathogenic variant in FIG4 and the other a VUS in UBQLN2 that are associated with TDP-C neurodegeneration. Our overall results point to the possible overlap between pathways implicating ALS genetic risk and TDP-C and involve the inositol-calcium pathway as a potential underpinning mechanism for the RNA to lysosomes trafficking abnormalities in TDP-C neurodegeneration. Further functional studies of the identified genes could further illuminate the pathophysiologic mechanism of this enigmatic disease and pave the way for targeted and personalized interventions. In addition, specialized cognitive genetic clinics may consider TDP-C genetic panel (once available) or whole exome/genome sequencing in suspected patients with TDP-C even in the absence of family history. However, genetic counseling and a discussion of benefit vs risk and cost effectiveness and the actionability of the discovered genetic data should take place in the clinical setting as indicated. Our data support further studies of both rare and common genetic variants as risk factors for TDP-C neurodegeneration.

Glossary

AI

artificial intelligence

ALS

amyotrophic lateral sclerosis

ATL

anterior temporal lobe

bvFTD

behavioral variant FTD syndrome

ML

most likely

MR

mendelian randomization

svPPA

semantic variant primary progressive aphasia

WGS

whole genome sequencing

Author Contributions

M. Nassan: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; analysis or interpretation of data. I. Ayala: drafting/revision of the manuscript for content, including medical writing for content. J. Sloan: major role in the acquisition of data; analysis or interpretation of data. A. Bonfitto: major role in the acquisition of data; analysis or interpretation of data. B. Stark: major role in the acquisition of data; analysis or interpretation of data. S. Song: major role in the acquisition of data; analysis or interpretation of data. M. Naymik: major role in the acquisition of data; analysis or interpretation of data. C. Geula: drafting/revision of the manuscript for content, including medical writing for content. T. Gefen: drafting/revision of the manuscript for content, including medical writing for content. E. Barbieri: drafting/revision of the manuscript for content, including medical writing for content. I. Piras: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; analysis or interpretation of data. M-M. Mesulam: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; analysis or interpretation of data. M. Huentelman: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; analysis or interpretation of data.

Study Funding

This study was supported by the NIH awards: R01AG077444, R01AG091388, and R01NS085770 and Alzheimer's Association and National Alzheimer's Coordinating Center (NACC) New Investigator Award NIAP24-1277,067.

Disclosure

The authors report no competing interests specific to this work. Go to Neurology.org/NG for all disclosures.