The shared genetic architecture of the TDP-43 spectrum: Bridging clinical FTLD and neuropathological endophenotypes.

The shared genetic architecture of the TDP‐43 spectrum: Bridging clinical FTLD and neuropathological endophenotypes PATEL et al.
THE SHARED GENETIC ARCHITECTURE OF THE TDP‐43 SPECTRUM

Abstract

INTRODUCTION

TAR DNA‐binding protein 43 (TDP‐43) proteinopathies encompass a spectrum of conditions including frontotemporal lobar degeneration with TDP‐43 inclusions (FTLD‐TDP), limbic‐predominant age‐related TDP‐43 encephalopathy (LATE‐NC), and hippocampal sclerosis (HS). FTLD‐TDP is further classified into several pathological subtypes (A, B, C, D, and E), which differ in the distribution and morphology of TDP‐43 inclusions. Clinical and neuropathological variability across FTLD, FTLD‐TDP, LATE‐NC, and HS has hindered the translatability of genetic discoveries between clinical and pathological disease categories. By leveraging their genetic convergence, we aimed to refine effect size estimates and uncover variants that influence the TDP‐43 spectrum.

METHODS

We conducted an inverse variance weighted fixed‐effects meta‐analysis using METAL, combining neuropathological datasets (FTLD‐TDP subtypes: A [N
cases = 193], B [N
cases = 288], C [N
cases = 467], and controls [N
controls = 3153]), HS (N
cases = 751; N
controls = 6413), or LATE‐NC (N
cases = 1325; N
controls = 1787). Additionally, we performed a random‐effects meta‐analysis that included clinically diagnosed FTLD (N
cases = 4685; N
controls = 15,308) and neuropathological datasets. Moreover, we performed colocalization, fine‐mapping, and functional annotation analyses.

RESULTS

Three loci reached genome‐wide significance. The GRN (rs5848) and TMEM106B (rs3807865) loci were associated with FTLD‐TDP (A and B), LATE‐NC, and HS.  Colocalization and fine‐mapping analyses supported genetic signals at GRN and TMEM106B loci, while functional annotation suggested that rs3807865 and rs3807866 influence regulatory activity and TMEM106B expression in glial cells. MAPT demonstrated association across all combinations of TDP‐43 proteinopathies (except FTLD‐TDP B and C) and clinical FTLD, specifically the H1c haplotype.

DISCUSSION

Our results extend the known role of GRN and TMEM106B beyond FTLD‐TDP subtype A, while MAPT involvement suggests a role beyond traditional tauopathies.

HIGHLIGHTS

Meta‐analysis across clinical frontotemporal lobar degeneration (FTLD) and TAR DNA‐binding protein 43 (TDP‐43) proteinopathies identifies MAPT, GRN, and TMEM106B as key shared risk loci.

GRN and TMEM106B associations extend beyond limbic‐predominant age‐related TDP‐43 encephalopathy, hippocampal sclerosis, and FTLD‐TDP subtype A to include FTLD‐TDP B.

MAPT H1c is associated with clinical FTLD and FTLD‐TDP subtype A, suggesting subtype‐specific genetic effects across the TDP‐43 spectrum.

1BACKGROUND

Neuropathologies featuring TAR DNA‐binding protein 43 (TDP‐43) inclusions in neuronal and glial cells have been widely reported, with frontotemporal lobar degeneration with TDP‐43 inclusions (FTLD‐TDP) being the most extensively studied form. FTLD‐TDP is classified into at least five subtypes—A, B, C, D and E1—based on the distribution and structure of TDP‐43 aggregates in the cortical layers of the brain. FTLD‐TDP manifests clinically into the FTLD syndrome characterized by gradual deterioration of the frontal and temporal regions of the brain, resulting in alterations in personality, changes in behavior, and language impairment.2 TDP‐43 positive cytoplasmic inclusions are present in up to 50% of FTLD patients.3, 4, 5 More recently, a new neurodegenerative entity called limbic‐predominant age‐related TDP‐43 encephalopathy neuropathologic change (LATE‐NC) has been identified, and involves distinct TDP‐43 neuropathological changes affecting the amygdala and hippocampus.6 The diagnostic criteria for LATE‐NC are based on the density of TDP‐43 immunoreactive neuronal cytoplasmic inclusions, with advanced stages indicating more widespread involvement in neocortical regions. Another condition associated with TDP‐43 proteinopathy is hippocampal sclerosis (HS), which commonly affects individuals over the age of 85 (≈ 10%) and for which up to 90% of cases are TDP‐43 positive.7, 8

Genetic studies, including genome‐wide association studies (GWASs) and whole‐genome sequencing (WGS), have identified loci associated with increased risk for FTLD‐TDP and related conditions, enhancing our understanding of these TDP‐43 proteinopathies’ etiology and progression. A large WGS study recently reported the GRN and TMEM106B loci associated with FTLD‐TDP,9 while Shade et al. showed that both of these loci were associated with LATE‐NC and HS.10 However, it did not always reached genome‐wide significance and additional studies are necessary to validate these associations.

The neuropathological and clinical variability of FTLD has limited the generalizability/biological relevance of genetic findings from clinically diagnosed FTLD and neuropathologically defined FTLD‐TDP, even though most clinically diagnosed FTLD patients will present with TDP‐43 pathology at autopsy.11, 12 For instance, TNIP1 and UNC13A, associated with FTLD‐TDP, were not reported in the large clinical FTLD study,12 and similarly MAPT and RPSA/MOBP, associated with clinical FTLD, were not associated with FTLD‐TDP. On the other hand, the apolipoprotein E (APOE) locus was recently associated with clinical FTLD, which overlapped with findings in LATE‐NC.10, 12 Therefore, leveraging identified genetic loci across clinical FTLD, FTLD‐TDP, and related TDP‐43 proteinopathies such as LATE‐NC and HS provides an opportunity to bridge this gap. We hypothesized that this strategy could increase the power to detect TDP‐43–driven genetic risk factors, refine estimates of their effect sizes across endophenotypes, and highlight variants whose effects are consistent across both clinical and neuropathologically defined TDP‐43 proteinopathies. In this study, we therefore conducted a meta‐analysis integrating data from different FTLD and TDP‐43 proteinopathy studies. We comprehensively mapped the genetic landscape and enhanced our understanding of FTLD and TDP‐43 proteinopathies.

RESEARCH IN CONTEXT

Systematic review: The authors reviewed the most recent literature using traditional sources such as PubMed and Google Scholar. Clinical frontotemporal lobar degeneration (FTLD) and FTLD with TAR DNA‐binding protein 43 (TDP‐43) proteinopathies represent a complex overlapping spectrum of neurodegenerative conditions. Genetic studies including clinical FTLD and TDP‐43 proteinopathies have identified key risk genes such as GRN, TMEM106B, MAPT, and MOBP. The findings have enhanced our understanding of the genetic architecture and progression of these conditions.

Interpretation: The meta‐analyses presented in this paper integrate genetic data from clinical FTLD and TDP‐43 proteinopathies. This comprehensive approach has led to the identification of significant genetic associations, including the MAPT, GRN, and TMEM106B loci. The GRN and TMEM106B loci showed significant association across various TDP‐43 proteinopathies including FTLD‐TDP A and B subtypes. Moreover, this study highlights the potential role of MAPT as a modifier of TDP‐43 proteinopathies.

Future directions: Future research should focus on conducting further neuropathological studies to confirm the impacts of these genetic variants on disease progression and to validate these associations across diverse populations. These steps will deepen our understanding of genetic mechanisms underlying TDP‐43 proteinopathies and pave the way for therapeutic interventions.

2METHODS

2.1Study design

To identify potential studies, we searched the recent literature in PubMed and Google Scholar. We included any GWAS of clinical FTLD or neuropathological TDP‐43–related phenotypes for which genome‐wide summary statistics were available for harmonization and meta‐analysis. Studies were excluded if summary statistics were unavailable or if the phenotype was not relevant to the FTLD/TDP‐43 spectrum. In this study, we included both TDP‐43 proteinopathy studies and clinical FTLD (Table S1 in supporting information). The Pottier et al.9 study specifically targeted participants diagnosed with FTLD‐TDP (n
cases = 948; n
controls = 3153; N = 4101) and its subtypes, including FTLD‐TDP A (n
cases = 193), FTLD‐TDP B (n
cases = 288), and FTLD‐TDP C (n
cases = 467). From the Shade et al.10 study, we included pathological endophenotypes such as LATE‐NC (n
cases = 1325; n
controls = 1787; N = 3112) and HS (n
cases = 751; n
controls = 6413; N = 7164). In addition, as the majority of clinical cases of FTLD present with TDP‐43 neuropathology at autopsy, we included clinical FTLD from Manzoni et al.12 (n
cases = 4685; n
controls = 15308; N = 19993). The summary statistics from Pottier et al. were downloaded from the GWAS catalog (FTLD‐TDP: GCST90558311; FTLD‐TDP A: GCST90558312; FTLD‐TDP B: GCST90558313; FTLD‐TDP C: GCST90558314). Summary statistics from Shade et al. were obtained upon request to the corresponding author. Summary statistics from Manzoni et al. were downloaded from the University College London research data repository.13 The studies we used in our analysis include Pottier et al., which conducted WGS, and Manzoni et al. and Shade et al., which both performed single nucleotide polymorphism array–based GWASs.

2.2Quality control and preprocessing

Rigorous quality control on raw data were previously performed as described.9, 10, 12 All GWAS summary statistics included in the meta‐analysis were derived from European ancestry–focused studies. In each GWAS study, population genetic ancestry was assessed using ancestry‐anchored principal components or admixture analyses. Accordingly, the meta‐analysis was conducted within a relatively homogeneous European ancestry population, reducing the likelihood of confounding due to population stratification. Genome‐wide summary statistics across all cohorts were harmonized and quality controlled using a standardized pipeline prior to downstream analyses. Variant coordinates were unified to the GRCh38 human reference genome build using chromosome, genomic position, and rsIDs. We used the Broad Institute liftover tool to convert variants originally based on GRCh37 to the GRCh38 reference genome to ensure uniformity.14 For bi‐allelic variants for which the effect and non‐effect alleles were harmonized between study datasets, and effect sizes (beta) were inverted to align effect estimates to a consistent reference allele. Variants with non‐complementary or mismatched alleles across studies (e.g., A/G vs. A/C) were excluded. To eliminate potential strand‐alignment errors, strand‐ambiguous variants (palindromic: A/T or C/G) with a minor or effect allele frequency between 0.45 and 0.55 were removed. Furthermore, lead variants identified in top loci were manually verified to confirm non‐palindromic status (Table S1).

2.3Meta‐analysis approach

Inverse variance weighted meta‐analyses were conducted using METAL software,15 integrating both neuropathological and clinical datasets. The analyses were conducted in two stages. In the first stage, neuropathological datasets were combined across various combinations of FTLD‐TDP subtypes, LATE‐NC, and HS. Meta‐analysis models were selected based on the degree of phenotypic and methodological similarity across studies. For the neuropathological analysis, the two available cohorts were both based on brain autopsy and therefore showed high methodological homogeneity; a fixed‐effect model was used assuming a common underlying effect. In the second stage, clinical FTLD cohorts were combined with the neuropathological studies. Because these underlying diagnostic frameworks differ between these studies, we applied a random effect model to account for between‐study heterogeneity using Plink (v2.0). In addition to this unbiased genome‐wide approach, we specifically assessed sentinel variants of genome‐wide significant loci from clinical FTLD and neuropathological studies (P values [P] < 5×10−08). Note that the APOE variants failed quality controls in the Pottier et al.9 summary statistics and therefore could not be assessed.

2.4MAPT sub‐haplotype analysis

To assess genetic signal for known MAPT haplotypes, we evaluated six variants tagging the MAPT haplotypes including rs1467967, rs242557 (a partial tag for the H1c sub‐haplotype), rs3785883, rs2471738, rs8070723 (a tag for the H2 haplotype), and rs7521.16, 17, 18 The tag variants used to define haplotypes were selected according to linkage disequilibrium (LD) measures including R
2 and D′ (Table S2 in supporting information). Using summary statistics of FTLD‐TDP subtypes, HS, LATE‐NC, and clinical FTLD, we examined the association evidence at these six markers to infer whether the observed MAPT signals were consistent with specific haplotype backgrounds.

2.5Colocalization analysis

Colocalization analysis between FTLD‐TDP (type A and type B) and LATE‐NC was performed at the TMEM106B and GRN loci using ± 250 kb regions centered on sentinel variants from the significant meta‐analyses. Colocalization was assessed using the coloc R package (v2.1).19 The posterior probability of hypothesis 4 (PP.H4) was used to assess evidence that the two traits share a common causal variant at the same locus, with higher PP.H4 values indicating stronger evidence for colocalization.

2.6Fine‐mapping analysis

We performed fine mapping for each genome‐wide significant sentinel variant using SuSiE RSS, a Bayesian fine‐mapping method that integrates GWAS summary statistics with a LD correlation matrix to identify credible sets and estimates the posterior inclusion probabilities to identify likely causal variants. For the LD reference panel, we used 1000 Genomes project variant call data formatted as a Plink binary file and aligned to the GRCh38 genome build.20 We defined ± 500 kb genomic regions around the lead variant for LD calculation and 95% credible sets (i.e., each credible set has at least 95% probability of containing a true causal variant). Variants were prioritized based on their posterior inclusion probability (PIP), with higher PIP values indicating greater support for a variant being causal within a fine‐mapped credible set.

2.7Functional analysis

AlphaGenome was used to predict the functional consequences of TMEM106B variants.21 Predictions were generated using a ± 500 kb sequence centered on the lead variant at the locus. Alzheimer's disease– and FTLD‐relevant brain tracks were identified directly from AlphaGenome output metadata using the keywords brain, cortex, cortical, frontal, temporal, hippocampus, hippocampal, entorhinal, prefrontal, microglia, microglial, neuron, neuronal, astrocyte, astrocytic, oligodendrocyte, oligodendroglia, glia, glial, cerebrum, cerebral, amygdala, substantia nigra, spinal cord. The evaluated modalities included RNA‐seq, CAGE, ATAC‐seq, DNase‐seq, histone‐modification ChIP‐seq, splice‐site probability, splice‐site usage, splice junction, and different histone‐track usage. GENCODE v46 annotations were used to display the variant relative to TMEM106B. The significant variants at the TMEM106B locus were queried in the SingleBrain portal to assess regulatory associations across different cell types.22

2.8Generation of forest plots

To visually represent the effect sizes and confidence intervals of identified genetic associations, forest plots were generated using the forestplot R package (v3.1.7).

3RESULTS

3.1Study design

In this study, we aimed to dissect genetic risk loci associated with TDP‐43 proteinopathies using a two‐step approach (Figure 1). First, we focused on neuropathologically defined TDP‐43 proteinopathies and performed comprehensive fixed‐effect meta‐analyses of FTLD‐TDP and its subtypes with LATE‐NC and HS. Second, we leveraged the high expected proportion of TDP‐43 positive cases in clinical FTLD cohorts and performed random‐effects meta‐analyses to evaluate genetic associations across clinical FTLD and TDP‐43 proteinopathies.

3.2Meta‐analyses of TDP‐43 proteinopathies

We performed a meta‐analysis including datasets enriched in FTLD‐TDP patients from the study by Pottier et al. (FTLD‐TDP subtype A, FTLD‐TDP subtype B, and FTLD‐TDP subtype C) together with LATE‐NC data from Shade et al. (Table S1). The most significant locus was identified on chromosome 17, led by the rs5848 (P
meta:FTLD‐TDP+LATE‐NC = 5.96×10−12; beta (β)meta:FTLD‐TDP+LATE‐NC = 0.280; standard error (SE) meta:FTLD‐TDP+LATE‐NC = 0.041; effect allele = T; Figure 2A and Table 1; and Table S3 and Figure S1A in supporting information), which is located in the 3’UTR of the GRN gene. This variant was not previously reported as genome‐wide significant in the overall FTLD‐TDP group in the Pottier et al. study. However, because rs5848 previously reached genome‐wide significance in the FTLD‐TDP A subtype, we further meta‐analyzed each FTLD‐TDP subtype with LATE‐NC summary statistics. As expected, rs5848 was genome‐wide significant in FTLD‐TDP A and LATE‐NC meta‐analysis (P
meta:FTLD‐TDPA+LATE‐NC = 2.50 × 10−12, βmeta:FTLD‐TDP A+LATE‐NC = 0.356; SEmeta:FTLD‐TDP A+LATE‐NC = 0.050; effect allele = T; Figure 2A and Table 1; and Table S3 and Figure S1B). In line with this, colocalization analysis provided strong evidence for shared association signal with a 99.3% posterior probability of a shared causal variant (PP.H4). Interestingly, rs5848 was genome‐wide significant in FTLD‐TDP B and LATE‐NC meta‐analysis (P
meta:FTLD‐TDPB+LATE‐NC = 4.47×10−9, βmeta:FTLD‐TDPB+LATE‐NC = 0.288; SEmeta:FTLD‐TDPB+LATE‐NC = 0.049; effect allele = T; Figure 2A and Table 1; and Table S3 and Figure S1C). Colocalization analysis provided modest evidence for a shared causal signal between FTLD‐TDP B and LATE‐NC (PP.H4 = 65.2%). We identified a second genome‐wide significant signal at the TMEM106B locus driven by rs3807865 (P
meta:FTLD‐TDP+LATE‐NC = 4.31×10−12; βmeta:FTLD‐TDP+LATE‐NC = −0.267; SEmeta:FTLD‐TDP+LATE‐NC = 0.038; effect allele = A; reference allele = G; Figure 2A and Table 1; and Table S3 and Figure S1D). rs3807865 and the previously known protective variant rs1990622 at the TMEM106B locus are in strong LD (D′ = 0.96, R2
 = 0.90).23 Interestingly, rs3807865 also reached genome‐wide significance in the meta‐analysis of FTLD‐TDP B and LATE‐NC (P
meta:FTLD‐TDPB+LATE‐NC = 3.69×10−12; effect allele = A; Figure 2A; and Table S3 and Figure S1E). In the FTLD‐TDP A and LATE‐NC meta‐analysis, the TMEM106B locus was driven by rs6460900 (P
meta:FTLD‐TDPA+LATE‐NC = 8.45 × 10−16, βmeta:FTLD‐TDPA+LATE‐NC = −0.393; SEmeta:FTLD‐TDPA+LATE‐NC = 0.048; effect allele = A; Figure 2A and Table 1; and Table S3 and Figure S1F), while rs3807865 also reached genome‐wide significance (P
meta:FTLD‐TDPA+LATE‐NC = 1.84×10−15, βmeta:FTLD‐TDPA+LATE‐NC = −0.387; SEmeta:FTLD‐TDPA+LATE‐NC = 0.048; effect allele = A). rs6460900 and rs3807865 were in strong LD, with an R2
of 0.94. Colocalization analysis provided strong evidence for shared association signal between FTLD‐TDP A and LATE‐NC (PP.H4 = 93.6%), whereas evidence for colocalization between FTLD‐TDP B and LATE‐NC was weak (PP.H4 = 26.9%), possibly due to limited statistical power. No signal at this locus was detected in FTLD‐TDP C.

Similarly, both GRN and TMEM106B loci reached genome‐wide significance in the meta‐analysis of FTLD‐TDP and HS. The signal at the GRN locus was driven by rs5848 (P
meta:FTLD‐TDP+HS = 1.75×10−13; βmeta:FTLD‐TDP+HS = 0.305; SEmeta:FTLD‐TDP+HS = 0.041; effect allele = T; Figure 2B and Table 1; and Table S3 and Figure S2A in supporting information), and rs5848 reached genome‐wide significance in the meta‐analysis of FTLD‐TDP A and HS (P
meta:FTLD‐TDPA+HS = 1.88×10−14; βmeta:FTLD‐TDPA+HS = 0.404; SEmeta:FTLD‐TDPA+HS = 0.052; effect allele = T; Figure 2B and Table 1; and Table S3 and Figure S2B). The meta‐analysis including FTLD‐TDP B and HS also revealed rs5848 as genome‐wide significant (P
meta:FTLD‐TDPB+HS = 1.09×10−10; βmeta:FTLD‐TDPB+HS = 0.328; SEmeta:FTLD‐TDPB+HS = 0.051; effect allele = T; Figure 2B and Table 1; and Table S3 and Figure S2C).

In addition, we identified a genome‐wide significant signal at the TMEM106B locus driven by rs3807866 (P
meta:FTLD‐TDP+HS = 5.17×10−13; βmeta:FTLD‐TDP+HS = −0.294; SEmeta:FTLD‐TDP+HS = 0.040; effect allele = A; reference allele = G; Figure 2B and Table 1; and Table S3 and Figure S2D) for FTLD‐TDP. This variant reached genome‐wide significance in the meta‐analysis of FTLD‐TDP B and HS (P
meta:FTLD‐TDPB+HS = 9.42×10−14; c
meta:FTLD‐TDPB+HS = −0.376; SEmeta:FTLD‐TDPB+HS = 0.051; effect allele = A; Figure 2B and Table 1; and Table S3 and Figure S2E) and is in strong LD with the previously identified rs1990622 variant (D′ = 0.97, R2
 = 0.91). Moreover, the TMEM106B locus signal was driven by rs7805419 in the FTLD‐TDP A and HS meta‐analysis (P
meta:FTLD‐TDPA+HS = 1.33×10−18; βmeta:FTLD‐TDPA+HS = −0.470; SEmeta:FTLD‐TDPA+HS = 0.053; effect allele = C; Figure 2B and Table 1; and Table S3 and Figure S2F), with rs7805419 being in strong LD with rs1990622 (D′ = 1.0, R2
 = 0.98). The TMEM106B index variants identified across the FTLD‐TDP subtypes map to the same LD block, consistent with a shared association signal at this locus. No signal was detected from the FTLD‐TDP C subtype. Overall, our data suggests that the GRN and TMEM106B are associated with FTLD‐TDP A as well as FTLD‐TDP B.9

Because deciphering the causal variant at the TMEM106B is still an area of intense research, we performed fine mapping in this region. In the FTLD‐TDP and LATE‐NC analysis, the strongest fine‐mapping evidence was observed at the genome‐wide significant sentinel variant, rs3807865, which showed a PIP of 23%. In the FTLD‐TDP A and LATE‐NC analysis, the highest PIP was observed for rs2043539 at 9%, whereas in the FTLD‐TDP B and LATE‐NC analysis, rs3807865 showed the highest PIP at 10%. In the FTLD‐TDP and HS fine‐mapping analysis, the highest PIP was 20% for rs3807866. For FTLD‐TDP A and HS, the highest PIP was 7% for rs10281425, while for FTLD‐TDP B and HS, it was 8% for rs3807866. These genetic variants at the TMEM106B locus were in strong LD with the sentinel variant (R2
 > 0.80). These findings suggest that the lead genome‐wide significant variants remained among the most likely causal candidates within their respective loci, although the relatively low posterior probabilities likely reflect the high LD among correlated variants, which limits the ability to resolve a single causal variant in this region.

Furthermore, we identified several high‐ranking brain‐relevant regulatory predictions for candidate causal variants, including rs3807865 and rs3807866 at the TMEM106B locus. Even though modest, both rs3807865 and rs3807866 alternative alleles showed increased predicted chromatin accessibility in motor‐neuron and brain vascular endothelial cells. Despite this, H3K27ac predictions in astrocytes and spinal cord suggested that rs3807865 and rs3807866 alternative alleles weaken enhancer activity, potentially leading to reduced downstream TMEM106B expression (Figure 3 and Figure S3 in supporting information). Similarly, the alternative alleles of rs3807865 and rs3807866 reduced the promoter‐associated chromatin activity marker H3K4me3 in spinal‐cord astrocytes and brain microvascular endothelial cells. In line with this, single‐cell quantitative trait loci data showed that rs3807865 and rs3807866 alternative alleles were trending toward a reduced TMEM106B expression in astrocytes (P = 5.66×10−2; β = −0.039; P = 9.26×10−2; β = −0.034; respectively). Similarly, rs3807865 (P = 1.51×10−2; β = −0.051), rs3807866 (P = 1.74×10−2; β = −0.050), and rs6460900 (P = 1.55×10−2; β = −0.05) showed nominal associations with TMEM106B expression in oligodendrocytes, whereas rs7805419 showed a stronger association (P = 7.45×10−5; β = −0.06), supporting a potential cell type–specific regulatory role for TMEM106B variants.

3.3Meta‐analyses of clinical FTLD datasets and TDP‐43 proteinopathies

Manzoni et al.12 conducted a comprehensive GWAS that incorporated a large number of clinically diagnosed FTLD participants. We first performed a meta‐analysis that included the FTLD‐TDP, LATE‐NC, and clinical FTLD datasets. We identified a single locus within the MAPT genomic region that achieved genome‐wide significance. rs2696557 was the most significant variant (P
meta:FTLD‐TDP+LATE‐NC+clinical‐FTLD = 2.82×10−11, βmeta:FTLD‐TDP+LATE‐NC+clinical‐FTLD = 0.123, SEmeta:FTLD‐TDP+LATE‐NC+clinical‐FTLD = 0.019, I2
 = 0.0; effect allele = C; Figure 4A and Table 1; and Table S4 and Figure S4A in supporting information). To assess whether the FTLD‐TDP signal was driven by a subtype, we performed meta‐analyses including the clinical FTLD, LATE‐NC, and FTLD‐TDP subtype studies. Interestingly, rs2696557 was nominally significant only in FTLD‐TDP A (P
FTLD‐TDPA = 1.55×10−2; βFTLD‐TDPA = 0.350; SEFTLD‐TDPA = 0.144; effect allele = C; Figure 4A and Table 1; and Table S4 and Figure S4B) as opposed to the other FTLD‐TDP subtypes, suggesting that the overall FTLD‐TDP signal is primarily driven by FTLD‐TDP A. The rs2696557 signal was also nominally significant in LATE‐NC (Table S4).

Given the well‐established haplotypic structure of the MAPT locus, we further evaluated the association of H1 sub‐haplotypes with clinical FTLD and TDP‐43 proteinopathies. We observed a significant association signal at rs242557 (P
meta:FTLD‐TDPA+HS+clinical‐FTLD = 8.29×10−4; βmeta:FTLD‐TDPA+HS+clinical‐FTLD = 0.053, SEmeta:FTLD‐TDPA+HS+clinical‐FTLD = 0.016), a partial tag for the MAPT H1c sub‐haplotype, with FTLD‐TDP subtype A (P
FTLD‐TDPA = 8.95×10−3), clinical FTLD (P
clinical‐FTLD = 7.73×10−3), and LATE‐NC (P
LATE‐NC = 1.25×10−1).

Similar to the meta‐analysis with LATE‐NC, only the MAPT locus reached genome‐wide significance when including the HS study. rs199499 was the most significant variant (P
meta:clinical‐FTLD+FTLD‐TDP+HS = 1.56×10−9, βmeta:clinical‐FTLD+FTLD‐TDP+HS = −0.116, SEmeta:clinical‐FTLD+FTLD‐TDP+HS = 0.019, I2
 = 0.0; effect allele = T; Figure 4B and Table 1; and Table S4 and Figure S4C). The FTLD‐TDP signal was nominally significant in the FTLD‐TDP A subtype (P
FTLD‐TDPA = 3.70×10−2; βFTLD‐TDPA = −0.306, SEFTLD‐TDPA = 0.147). The rs2696557 variant was significant in LATE‐NC and showed a consistent direction of effect in HS. In addition, the H1c rs242557 tagging variant was also significant (P
meta:clinical‐FTLD+FTLD‐TDPA+HS = 4.99×10−2, βmeta:clinical‐FTLD+FTLD‐TDPA+HS = 0.104, SEmeta:clinical‐FTLD+FTLD‐TDPA+HS = 0.053; P
FTLD‐TDPA = 8.95×10−3, βFTLD‐TDPA = 0.29, SEFTLD‐TDPA = 1.110; P
FTLD = 7.73×10−3, βclinical‐FTLD = 0.045, SEclinical‐FTLD = 0.016; and P
HS = 7.34×10−2; βHS = 0.105; SEHS = 0.058).

Overall, our findings suggest that the MAPT locus is associated with multiple TDP‐43 proteinopathies and its H1c sub‐haplotype is primarily associated with clinical FTLD and the FTLD‐TDP A subtype.

3.4Previously identified variants and meta‐analyses validation

Finally, we specifically assessed the other previously identified genome‐wide significant variants from all three studies at the TNIP1, UNC13A, and RPSA/MOBP loci. None of these variants and loci reached genome‐wide significance at the meta‐analysis stage (Table S5 in supporting information), suggesting that these variants are associated with a specific endophenotype.

4DISCUSSION

In this study, we conducted an extensive meta‐analysis focusing on the genetic landscape of the TDP‐43 spectrum, leveraging data from both clinical FTLD and neuropathologically defined TDP‐43 proteinopathies. By integrating clinical FTLD data with TDP‐43 proteinopathies, we further delineated genetic association signals, particularly at the MAPT locus, which suggests that MAPT is implicated not only in clinically diagnosed FTLD but also in TDP‐43 proteinopathies. Our analyses confirmed and refined previously known association signals at the GRN and TMEM106B loci, further strengthening their roles in TDP‐43 proteinopathies. These findings bridge the gap between clinical presentations of FTLD and neuropathologically defined TDP‐43 proteinopathies and underscore the value of integrating genetic data from neuropathologically and clinically defined cohorts to enhance our understanding of these complex conditions.

We identified a strong association at the MAPT locus across our combined clinical and TDP‐43 proteinopathy datasets highlighting the critical role of this genomic region. Importantly, we revealed that the MAPT signal in FTLD‐TDP was driven by the FTLD‐TDP A subtype. This signal aligns with the previously reported association of the MAPT H1 haplotype with an earlier age of onset in GRN mutation carriers, which consistently results with FTLD‐TDP A at autopsy.24
MAPT encodes the tau protein, which is known to aggregate in the brains of FTLD patients,25 including in a subset of patients with MAPT mutations.26 Accumulating neuropathological evidence reveals the presence of comorbid TDP‐43 pathology in some tauopathy patients, and vice versa, with distinct morphological characteristics and propagation patterns within the brain.27, 28, 29, 30 Of note, the TMEM106B locus has previously been linked to Braak neurofibrillary tangle stages.10 While further neuropathology studies are necessary, our findings strongly suggest that variants in the MAPT region may act as modifiers of TDP‐43 proteinopathy etiology.

Additionally, we reported the previously known TMEM106B‐associated locus with TDP‐43‐related proteinopathies. The association of variants at the TMEM106B locus with FTLD‐TDP type A was recently reported when collapsing common variants at the gene level (Pottier et al.9). We are now reporting on a genome‐wide significant association at the single variant level when combining FTLD‐TDP type A and LATE‐NC cases. Even though FTLD‐TDP type A and LATE‐NC are distinct entities in terms of clinical and regional disease burden,31, 32, 33 they share some histological features. Indeed, advanced LATE‐NC and FTLD‐TDP type A were recently reported as displaying similar densities of neuronal cytoplasmic inclusions and dystrophic neurites in amygdala and hippocampus.33, 34 In addition, both present with dense mesh‐like pattern in the hippocampus. On the other hand, we also report a genome‐wide significant signal when combining LATE‐NC and FTLD‐TDP type B. While FTLD‐TDP type A cases, like LATE‐NC, are associated with GRN and TMEM106B variants, FTLD‐TDP B cases have not been reported as such so far. The latter group of patients is usually associated with C9ORF72 repeat expansions and more recently with UNC13A and TNIP1 variants.9 We cannot, however, rule out that the FTLD‐TDP B signal from Pottier et al.9 originates from mixed FTLD‐TDP A and B pathologies.35, 36 Similarly, we also report that rs5848 at the GRN locus, which was previously thought to be FTLD‐TDP type A specific, reaches genome‐wide significance when combining FTLD‐TDP B subtype with LATE‐NC or HS. These findings suggest that the TMEM106B and GRN loci are not restricted to cases with FTLD‐TDP A and LATE‐NC pathology. In addition, this supports that TMEM106B‐ and GRN‐mediated pathways, such as endolysosomal regulation,37 may represent a broader mechanism across diverse TDP‐43 pathologies, rather than a subtype‐restricted risk factor.

To resolve whether a specific functional signal drives this broad association across FTLD‐TDP subtypes, we performed fine mapping at the TMEM106B locus and revealed multiple variants in strong LD with our lead signals, pointing to a complex underlying genetic mechanism. Historically, the TMEM106B locus has been characterized by a single LD block with as many as 104 genetic variants in strong LD (R2
 > 0.80) with the lead protective variant, rs1990622.38 However, a recent fine‐mapping effort has identified a finder structural composition of the locus.39 Salazar et al.39 reported that the TMEM106B locus contains multiple haplotypes associated with different neurodegeneration risks including Alzheimer's disease and FTLD. In our dataset, while our lead variant showed strong LD with the primary haplotypic background, it was not in LD with rs13237415, marker of the secondary independent LD block. Thus, our association signal did not capture the LD block2 signal described by Salazar et al. This finding adds to the growing body of evidence supporting the existence of a complex haplotypic structure at the TMEM106B locus.

Functional validations are, however, essential to identify the TMEM106B variant that directly influences gene regulation or protein function. In our study, rs3807865 and rs3807866 were associated with reduced H3K27ac‐marked regulatory activity at the TMEM106B locus in astrocytes, suggesting altered enhancer activity in glial cells at this locus. This finding is consistent with prior work showing that TMEM106B loss impairs lysosomal organization and remyelination in oligodendrocytes, and disrupts autophagosome maturation in astrocytes.40, 41 Moreover, rs7805419 is in near‐perfect LD with rs1990620, suggesting that it likely tags the same regulatory haplotype identified in previous studies, where the rs1990620 risk allele was shown to preferentially recruit CTCF transcription factor, strengthen long‐range chromatin interactions with the TMEM106B promoter, and increase TMEM106B expression in association with lysosomal abnormalities.38 Together, these findings suggest that TMEM106B regulatory variation as a plausible glial disease mechanism, linking non‐coding risk haplotypes to altered chromatin regulation, cell type–specific gene expression, and lysosomal dysfunction.

This work has several limitations. First, we relied on summary statistics rather than raw genotype data, which limited our ability to apply fully harmonized quality control and prevented complete assessment of some loci (e.g., APOE). Second, heterogeneity in case definitions and phenotyping across clinical FTLD and neuropathologically defined TDP‐43 proteinopathies may have diluted endophenotype‐specific signals and contributed to the absence of genome‐wide significance at some previously reported loci (e.g., UNC13A, MOBP). Finally, the majority of the participants from all studies were of European ancestry, restricting the generalizability of our findings to other populations.

Overall, our comprehensive meta‐analysis identified significant association signals at the MAPT, GRN, and TMEM106B loci, shedding light on their critical roles in the etiology and modulation of TDP‐43 proteinopathies. Importantly, our findings underscore the intricate genetic underpinnings linking tauopathies and TDP‐43 proteinopathies themselves. These insights not only enhance our understanding of the genetic architecture underlying FTLD and TDP‐43 proteinopathies but also highlight the importance of leveraging genetic data to bridge the gap between clinical and neuropathological diagnoses. Ultimately, our findings pave the way for improved genetic risk assessments and hold promise for the development of targeted therapeutic strategies to mitigate the burden of these devastating diseases.

AUTHOR CONTRIBUTIONS

Cyril Pottier and Maulikkumar Patel designed and oversaw the study. Maulikkumar Patel and Cyril Pottier were responsible for acquisition of data. Maulikkumar Patel, Pawan Kumar, Matthew Johnson, John Budde, Menghan Liu, and Cyril Pottier did the analysis or interpretation of the data. Maulikkumar Patel, Cyril Pottier, and Carlos Cruchaga did the statistical analysis of the data. Maulikkumar Patel, Carlos Cruchaga, and Cyril Pottier drafted the manuscript content. Maulikkumar Patel, Pawan Kumar, Matthew Johnson, Menghan Liu, John Budde, Carlos Cruchaga, and Cyril Pottier revised the manuscript for content. Cyril Pottier did the study supervision and coordination. Maulikkumar Patel, Pawan Kumar, Matthew Johnson, John Budde, Daniel Western, Menghan Liu, Carlos Cruchaga, and Cyril Pottier provided intellectual input. All authors contributed to and reviewed the final version of the manuscript.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest. Author disclosures are available in the Supporting Information.

CONSENT STATEMENT

Informed consent was not necessary for this study.

Supporting information

REFERENCES

1Meneses A , Koga S , O'Leary J , Dickson DW , Bu G , Zhao N . TDP‐43 Pathology in Alzheimer's Disease . Mol Neurodegener . 2021 ;16 (1 ):84 . doi:10.1186/s13024-021-00503-x PMC8691026 34930382

2Dattola S , Ielo A , Varone G , Cacciola A , Quartarone A , Bonanno L . Frontotemporal dementia: a systematic review of artificial intelligence approaches in differential diagnosis . Front Aging Neurosci . 2025 ;17 :1547727. doi:10.3389/fnagi.2025.1547727 PMC12018464 40276595

3Jo M , Lee S , Jeon YM , Kim S , Kwon Y , Kim HJ . The role of TDP‐43 propagation in neurodegenerative diseases: integrating insights from clinical and experimental studies . Exp Mol Med . 2020 ;52 (10 ):1652 ‐1662 . doi:10.1038/s12276-020-00513-7
PMC8080625 33051572

4Hasegawa M , Arai T , Nonaka T , et al. Phosphorylated TDP‐43 in frontotemporal lobar degeneration and amyotrophic lateral sclerosis . Ann Neurol . 2008 ;64 (1 ):60 ‐70 . doi:10.1002/ana.21425
PMC2674108 18546284

5Weihl CC , Temiz P , Miller SE , et al. TDP‐43 accumulation in inclusion body myopathy muscle suggests a common pathogenic mechanism with frontotemporal dementia . J Neurol Neurosurg Psychiatry . 2008 ;79 (10 ):1186 ‐1189 . doi:10.1136/jnnp.2007.131334
PMC2586594 18796596

6Robinson JL , Porta S , Garrett FG , et al. Limbic‐predominant age‐related TDP‐43 encephalopathy differs from frontotemporal lobar degeneration . Brain . 2020 ;143 (9 ):2844 ‐2857 . doi:10.1093/brain/awaa219
PMC7526723 32830216

7Nelson PT , Schmitt FA , Lin Y , et al. Hippocampal sclerosis in advanced age: clinical and pathological features . Brain . 2011 ;134 (Pt 5 ):1506 ‐1518 . doi:10.1093/brain/awr053
PMC3097889 21596774

8Sieben A , Van Langenhove T , Vermeiren Y , et al. Hippocampal Sclerosis in Frontotemporal Dementia: when Vascular Pathology Meets Neurodegeneration . J Neuropathol Exp Neurol . 2021 ;80 (4 ):313 ‐324 . doi:10.1093/jnen/nlab010
33638350

9Pottier C , Kucukali F , Baker M , et al. Deciphering distinct genetic risk factors for FTLD‐TDP pathological subtypes via whole‐genome sequencing . Nat Commun . 2025 ;16 (1 ):3914 . doi:10.1038/s41467-025-59216-0
PMC12032271 40280976

10Shade LMP , Katsumata Y , Abner EL , et al. GWAS of multiple neuropathology endophenotypes identifies new risk loci and provides insights into the genetic risk of dementia . Nat Genet . 2024 ;56 (11 ):2407 ‐2421 . doi:10.1038/s41588-024-01939-9
PMC11549054 39379761

11Dickson DW , Kouri N , Murray ME , Josephs KA . Neuropathology of frontotemporal lobar degeneration‐tau (FTLD‐tau) . J Mol Neurosci . 2011 ;45 (3 ):384 ‐389 . doi:10.1007/s12031-011-9589-0
PMC3208128 21720721

12Manzoni C , Kia DA , Ferrari R , et al. Genome‐wide analyses reveal a potential role for the MAPT, MOBP, and APOE loci in sporadic frontotemporal dementia . Am J Hum Genet . 2024 ;111 (7 ):1316 ‐1329 . doi:10.1016/j.ajhg.2024.05.017
PMC11267522 38889728

13Manzoni C , Genome‐wide analysis—Summary Statistics of sporadic frontotemporal dementia cohort . https://rdr.ucl.ac.uk/articles/dataset/Genome‐wide_analysis_‐_Summary_Statistics_of_sporadic_frontotemporal_dementia_cohort/25600692

14Park KJ , Yoon YA , Park JH . Evaluation of Liftover Tools for the Conversion of Genome Reference Consortium Human Build 37 to Build 38 Using ClinVar Variants . Genes (Basel) . 2023 ;14 (10 ):1875 . doi:10.3390/genes1410187510.3390/genes14101875 PMC10606611 37895222

15Willer CJ , Li Y , Abecasis GR . METAL: fast and efficient meta‐analysis of genomewide association scans . Bioinformatics . 2010 ;26 (17 ):2190 ‐2191 . doi:10.1093/bioinformatics/btq340
10.1093/bioinformatics/btq340 PMC2922887 20616382

16Pittman AM , Myers AJ , Abou‐Sleiman P , et al. Linkage disequilibrium fine mapping and haplotype association analysis of the tau gene in progressive supranuclear palsy and corticobasal degeneration . J Med Genet . 2005 ;42 (11 ):837 ‐846 . doi:10.1136/jmg.2005.031377
10.1136/jmg.2005.031377 PMC1735957 15792962

17Valentino RR , Koga S , Walton RL , et al. MAPT subhaplotypes in corticobasal degeneration: assessing associations with disease risk, severity of tau pathology, and clinical features . Acta Neuropathol Commun . 2020 ;8 (1 ):218 . doi:10.1186/s40478‐020‐01097‐z
10.1186/s40478-020-01097-z PMC7720600 33287913

18Gagliardi M , Procopio R , Felicetti A , et al. MAPT Subhaplotypes in Different Progressive Supranuclear Palsy Phenotypes . Biomedicines . 2025 ;13 (6 ):1405 . doi:10.3390/biomedicines13061405
10.3390/biomedicines13061405 PMC12190731 40564124

19Rasooly D , Peloso GM , Giambartolomei C . Bayesian Genetic Colocalization Test of Two Traits Using coloc . Curr Protoc . 2022 ;2 (12 ):e627. doi:10.1002/cpz1.62710.1002/cpz1.627 36515558

201000 Genomes Project Consortium
, Auton A , Brooks LD , et al, 1000 Genomes Project Consortium
. A global reference for human genetic variation . Nature . 2015 ;526 (7571 ):68 ‐74 . doi:10.1038/nature15393
10.1038/nature15393 PMC4750478 26432245

21Avsec Z , Latysheva N , Cheng J , et al. Advancing regulatory variant effect prediction with AlphaGenome . Nature . 2026 ;649 (8099 ):1206 ‐1218 . doi:10.1038/s41586‐025‐10014‐0
10.1038/s41586-025-10014-0 PMC12851941 41606153

22Jang B , Bp K , Tokolyi A , et al. SingleBrain: a Meta‐Analysis of Single‐Nucleus eQTLs Linking Genetic Risk to Brain Disorders . medRxiv . 2025 :2025.03.06.25323424. doi:10.1101/2025.03.06.25323424. Preprint.

23Van Deerlin VM , Sleiman PM , Martinez‐Lage M , et al. Common variants at 7p21 are associated with frontotemporal lobar degeneration with TDP‐43 inclusions . Nat Genet . 2010 ;42 (3 ):234 ‐239 . doi:10.1038/ng.536
10.1038/ng.536 PMC2828525 20154673

24Wauters E , Van Mossevelde S , Sleegers K , et al. Clinical variability and onset age modifiers in an extended Belgian GRN founder family . Neurobiol Aging . 2018 ;67 :84 ‐94 . doi:10.1016/j.neurobiolaging.2018.03.007
10.1016/j.neurobiolaging.2018.03.007 29653316

25Irwin DJ , Cairns NJ , Grossman M , et al. Frontotemporal lobar degeneration: defining phenotypic diversity through personalized medicine . Acta Neuropathol . 2015 ;129 (4 ):469 ‐491 . doi:10.1007/s00401‐014‐1380‐1
10.1007/s00401-014-1380-1 PMC4369168 25549971

26Goedert M , Jakes R . Mutations causing neurodegenerative tauopathies . Biochim Biophys Acta . 2005 ;1739 (2‐3 ):240 ‐250 . doi:10.1016/j.bbadis.2004.08.007
10.1016/j.bbadis.2004.08.007 15615642

27Riku Y , Yoshida M , Iwasaki Y , Sobue G , Katsuno M , Ishigaki S . TDP‐43 Proteinopathy and Tauopathy: do They Have Pathomechanistic Links? Int J Mol Sci . 2022 ;23 (24 ):15755. doi:10.3390/ijms23241575510.3390/ijms232415755 PMC9779029 36555399

28Latimer CS , Liachko NF . Tau and TDP‐43 synergy: a novel therapeutic target for sporadic late‐onset Alzheimer's disease . Geroscience . 2021 ;43 (4 ):1627 ‐1634 . doi:10.1007/s11357‐021‐00407‐0
10.1007/s11357-021-00407-0 PMC8492812 34185246

29de Boer EMJ , Orie VK , Williams T , et al. TDP‐43 proteinopathies: a new wave of neurodegenerative diseases . J Neurol Neurosurg Psychiatry . 2020 ;92 (1 ):86 ‐95 . doi:10.1136/jnnp‐2020‐322983
10.1136/jnnp-2020-322983 PMC7803890 33177049

30Koga S , Zhou X , Murakami A , et al. Concurrent tau pathologies in frontotemporal lobar degeneration with TDP‐43 pathology . Neuropathol Appl Neurobiol . 2022 ;48 (2 ):e12778. doi:10.1111/nan.12778
10.1111/nan.12778 PMC9300011 34823271

31Nelson PT , Lee EB , Cykowski MD , et al. LATE‐NC staging in routine neuropathologic diagnosis: an update . Acta Neuropathol . 2023 ;145 (2 ):159 ‐173 . doi:10.1007/s00401‐022‐02524‐2
10.1007/s00401-022-02524-2 PMC9849315 36512061

32Shahidehpour RK , Katsumata Y , Dickson DW , et al. LATE‐NC Stage 3: a diagnostic rubric to differentiate severe LATE‐NC from FTLD‐TDP . Acta Neuropathol . 2025 ;149 (1 ):38 . doi:10.1007/s00401‐025‐02876‐5
10.1007/s00401-025-02876-5 PMC12037668 40293530

33Tome SO , Gawor K , Ospitalieri S , et al. Differences and overlaps in TDP‐43 pathology of ‘pure’ LATE‐NC compared to LATE‐NC coexisting with Alzheimer's disease . Acta Neuropathol . 2025 ;150 (1 ):28 . doi:10.1007/s00401‐025‐02929‐9
10.1007/s00401-025-02929-9 40952471

34Fischer DL , Spina S , Miller BL , Seeley WW , Grinberg LT . FTLD‐TDP versus LATE‐NC: the UCSF Neurodegenerative Disease Brain Bank Experience . Alzheimer's & Dementia . 2025 ;21 (S1 ):e106279. doi:10.1002/alz70855_10627910.1002/dad2.70411 PMC13319421 42388895

35Alafuzoff I , Pikkarainen M , Neumann M , et al. Neuropathological assessments of the pathology in frontotemporal lobar degeneration with TDP43‐positive inclusions: an inter‐laboratory study by the BrainNet Europe consortium . J Neural Transm (Vienna) . 2015 ;122 (7 ):957 ‐972 . doi:10.1007/s00702‐014‐1304‐1
10.1007/s00702-014-1304-1 25239189

36Neumann M , Lee EB , Mackenzie IR . Frontotemporal Lobar Degeneration TDP‐43‐Immunoreactive Pathological Subtypes: clinical and Mechanistic Significance . Adv Exp Med Biol . 2021 ;1281 :201 ‐217 . doi:10.1007/978‐3‐030‐51140‐1_13
10.1007/978-3-030-51140-1_13 PMC8183578 33433877

37Takahashi H , Strittmatter SM . The role of endolysosomal progranulin and TMEM106B in neurodegenerative diseases . Mol Neurodegener . 2025 ;20 (1 ):86 . doi:10.1186/s13024‐025‐00873‐610.1186/s13024-025-00873-6 PMC12297828 40713630

38Gallagher MD , Posavi M , Huang P , et al. A Dementia‐Associated Risk Variant near TMEM106B Alters Chromatin Architecture and Gene Expression . Am J Hum Genet . 2017 ;101 (5 ):643 ‐663 . doi:10.1016/j.ajhg.2017.09.004
10.1016/j.ajhg.2017.09.004 PMC5673619 29056226

39Salazar A , Tesi N , Knoop L , et al. Fine‐mapping of the TMEM106Blocus identifies four haplotypes with differential associations to neurodegeneration . medRxiv . 2026 :2025.02.02.25321330. doi:10.1101/2025.02.02.25321330. Preprint.

40Zhou X , Nicholson AM , Ren Y , et al. Loss of TMEM106B leads to myelination deficits: implications for frontotemporal dementia treatment strategies . Brain . 2020 ;143 (6 ):1905 ‐1919 . doi:10.1093/brain/awaa141
10.1093/brain/awaa141 PMC7296855 32504082

41Bauer CS , Webster CP , Shaw AC , et al. Loss of TMEM106B exacerbates C9ALS/FTD DPR pathology by disrupting autophagosome maturation . Front Cell Neurosci . 2022 ;16 :1061559. doi:10.3389/fncel.2022.1061559
10.3389/fncel.2022.1061559 PMC9812496 36619668