GRN rs5848 variant associates with TDP-43 pathology and cancer in opposite directions Authors: Khine Zin Aung, Peter T Nelson, Xiaotong Ning, Xian Wu, Inori Tsuchiya, Shama Karanth, Erin L Abner, David W Fardo, Yuriko Katsumata Published: 2026 ABSTRACT AbstractEpidemiologic studies have reported that cancer survivors have a relatively low risk of developing dementia, but the mechanisms underlying that inverse relationship are mostly unknown. The Granulin (GRN) gene single nucleotide variant rs5848 T allele is associated with increased risk of limbic-predominant age-related TDP-43 encephalopathy neuropathologic change (LATE-NC) and hippocampal sclerosis of aging (HS-Aging). The T allele is also associated with lower expression of the cognate protein progranulin (PGRN), which is a mitogen implicated in neoplasia. We examined whether the rs5848 variant associated with LATE-NC/HS-Aging pathology and cancer in the same cohort. This study leveraged genotype data from the Alzheimer’s Disease Genomics Consortium (n = 8121) and the Alzheimer’s Disease Sequencing Project (n = 3231), with cancer history and neuropathology data drawn from the National Alzheimer’s Coordinating Center. The rs5848 T allele was associated with higher odds of LATE-NC (p < 0.001) and was also associated with lower odds of cancer (p = 0.012). Established TMEM106B, APOE, and BIN1 risk alleles for Alzheimer’s disease showed no associations with cancer, implying that the GRN-related associations could not be completely explained by selection bias in the study sample. The finding of a specific allele with opposite correlative impact on cancer risk and dementia-related pathology has potential therapeutic implications. Introduction Epidemiologic and observational studies have reported an inverse relationship between cancer and dementia such that individuals with a history of cancer have a relatively low risk of developing dementia, while those with dementia are comparatively unlikely to be diagnosed with cancer.1–6 Methodological factors may contribute to the detected phenomena, with potential underlying confounders including survival and surveillance biases as patients with cancer may die before developing dementia (and vice versa). Moreover, study inclusion criteria, health-care utilization and volunteer participation may differ across observational cohorts. In addition to study design and related sources of bias, biological mechanisms may underlie this relationship. Treatments commonly used in cancer care, such as chemotherapies that affect neuroinflammation and endocrine function, may modify dementia risk.7–9 In addition, constitutive biological pathways that drive neurodegeneration or tumorigenesis may suppress the other. Neurodegenerative diseases and cancer are influenced by overlapping biological pathways involving protein homeostasis, inflammatory regulation, and cell growth.1,4,7,10 Among genes implicated in human diseases, granulin (GRN) is distinctive in being directly linked to both dementia and cancer. GRN plays key roles in normal cellular function and brain health and its encoded protein progranulin (PGRN), is involved in lysosomal activity, inflammation, and cell proliferation.11,12 Increased levels of PGRN are mitogenic and PGRN is recognized as an oncogenic marker in several cancers.13 Conversely, PGRN deficiency promotes the accumulation of TDP-43 pathology, a hallmark of frontotemporal lobar degeneration with TDP-43 inclusions (FTLD-TDP) and limbic-predominant age-related TDP-43 encephalopathy neuropathologic change (LATE-NC); it exacerbates lysosomal dysfunction, neuroinflammation and microglial hyperactivation thereby apparently contributing to neuronal damage and cognitive decline.14–16 The GRN rs5848 single-nucleotide polymorphism (SNP) is a notable variant located in the 3'-untranslated region (UTR) of GRN. The T allele of rs5848 is associated with reduced PGRN expression in both brain and blood.17–20 In numerous studies, this SNP has been implicated in neurodegenerative conditions such as LATE-NC and hippocampal sclerosis of aging (HS-Aging), and the T allele has also been reported as a risk factor for younger age-at-onset in FTLD-TDP.17,21–26 Although existing studies have not directly examined the association of rs5848 with cancer (with the exception of one study in a cohort of persons with Parkinson disease27) the potential effects of PGRN expression decreasing rs5848 can be inferred from the role of GRN in cancer biology, as elevated PGRN levels promote tumor growth and function as a mitogenic factor.11 These observations suggest that the T allele of rs5848, which is associated with increased dementia risk may theoretically exert a protective influence against cancer development or progression. Therefore, we investigated whether rs5848 was associated with cancer status as well as TDP-43–related neuropathologic outcomes, focusing on autopsy-confirmed LATE-NC and HS-Aging. We also conducted parallel analyses of rs1990622, located in TMEM106B, which is a known genetic risk SNP for LATE-NC and HS-Aging, as well as established Alzheimer disease (ad) risk alleles in Apolipoprotein E (APOE; ε4), and two bridging integrator 1 (BIN1) risk SNPs (rs744373 and rs6733839). These comparisons allowed us to test the hypothesis that the inverse relationship between cancer and neurodegenerative disease was partly attributable to the GRN rs5848 variant. We therefore hypothesize a biological trade-off model whereby GRN rs5848 influences neurodegeneration and proliferative pathways through differential progranulin expression (Figure 1). llustrations were adapted using BioArt resources provided by the National Institutes of Health (NIH) (https://bioart.niaid.nih.gov). Biological trade-off model of GRN rs5848. PGRN: progranulin, LATE-NC: limbic-predominant age-related TDP-43 encephalopathy neuropathologic change, HS-aging: hippocampal sclerosis of aging. Methods Study population and data sources As all study participants were de-identified and the data were available to the public for secondary analyses after an application process, the University of Kentucky Institutional Review Board (IRB) classified this project as not Human Subjects Research under National Institute on Health (NIH) Exemption #4 (https://grants.nih.gov/policy-and-compliance/policy-topics/human-subjects/research). Clinical and neuropathological data were drawn from the National Alzheimer’s Coordinating Center (NACC) Uniform Data Set (UDS) and Neuropathology (NP) Data Set from the June 2025 data freeze (https://naccdata.org).28,29 Data were collected from 46 Alzheimer’s Disease Research Centers (ADRCs) across the U.S. National Institute on Aging (NIA)–funded ADRC network. All participating centers received IRB approval at their local site and obtained written informed consent from participants. Cancer history was ascertained from the NACC UDS version 3.0 (March 2015), using Form D2 (Clinician-Assessed Medical Conditions), in which cancer status in the 12 months prior to the UDS visit was clinician-determined.30 Participants were considered positive for cancer history based on at least one instance of clinician-reported cancer (excluding non-melanoma skin cancer) across UDS follow-up. For analyses involving NP outcomes, participants diagnosed with any of 27 rare brain diseases (Supplementary Table 1, available at Journal of Neuropathology & Experimental Neurology online) were excluded as previously described;31 this mitigated biases inherent to clinic-based cohorts that are enriched for uncommon neurodegenerative disorders relatively affecting younger (middle-aged) individuals.32 Cancer and neuropathological outcomes Cancer status was coded as: 1 for individuals with a cancer history (as defined above) and 0 for individuals with no reported cancer history.30 TDP-43-related pathologies were coded as 0 = absent or 1 = present in each of the amygdala, hippocampus, and neocortex. Pathologically diagnosed HS-Aging, including cases recorded as having mesial temporal sclerosis (MTS), was coded as 0 = absent or 1 = present. ad neuropathologic change (ADNC) was defined according to the National Institute on Aging-Alzheimer’s Association (NIA-AA) ABC scoring system.33 Braak neurofibrillary tangle (NFT) stage (NIA-AA B score) was grouped into four ordered categories reflecting increasing severity (B0 = 0; B1 = I–II; B2 = III–IV; B3 = V–VI). Neocortical neuritic plaque density was assessed using the Consortium to Establish a Registry for ad (CERAD) score (NIA-AA C score). Genetic data processing SNP data were obtained from the Alzheimer’s Disease Sequencing Project (ADSP)34,35 and the Alzheimer’s Disease Genomics Consortium (ADGC).36 For these two resources, we constructed two independent sample groups: participants in ADSP and NACC (hereafter referred to as “ADSP-NACC”) and participants in ADGC and NACC excluding those in ADSP-NACC (hereafter referred to as “ADGC-NACC”). ADSP whole genome sequencing (WGS) data in release 5,34,35 were downloaded from the National Institute on Aging Genetics of Alzheimer’s Disease Data Storage Site (NIAGADS), which comprised 58 507 samples from 57 study cohorts.34,35 SNP assay data from ADRC1 to 17 batches, (which are a portion of Alzheimer’s Disease Genetics Consortium (ADGC) genotype data), were quality-controlled, removing variants with high missing rate (>5%) and rare variants (minor allele frequency <5%) and then imputed using the TOPMed Imputation Server (https://imputation.biodatacatalyst.nhlbi.nih.gov/)37 based on the Genome Reference Consortium Human Build 38 (GRCh38) reference panel. These genetic data were linked to clinical and neuropathological outcome data from the NACC dataset, as previously described.38 We then extracted SNP data for rs5848 located in GRN (coded as 0 = C/C, 1 = C/T, and 2 = T/T); rs1990622 located in TMEM106B (0 = G/G, 1 = G/A, and 2 = A/A); and rs744373 and rs6733839 located in BIN1 (0 = A/A, 1 = A/G, and 2 = G/G in rs744373 and 0 = C/C, 1 = C/T, and 2 = T/T in rs6733839) from each data set. APOE ε4 dosage was derived from NACC UDS and coded as 0 = no ε4 allele, 1 = one ε4 allele, and 2 = two ε4 alleles. We extracted phenotype-genotype data from participants with European ancestry. Individual genetic ancestry was inferred using principal component analysis (PCA) based on linkage disequilibrium (LD) pruned SNP assay data implemented in PLINK39 (—indep-pairwise 1500 50 0.2; window size = 1500 SNPs, sliding window increment = 50 SNPs, r2 threshold = 0.2), filtered by minor allele frequency (MAF ≥0.05) and genotype missingness (≤0.05), and excluding strand-ambiguous (A/T and C/G) variants. We performed PCA separately for the ADSP-NACC and ADGC-NACC datasets after merging each with genotype data from the 1000 Genomes Project Phase 3 (1000G),40,41 followed by uniform manifold approximation and projection (UMAP) for the first 20 principal components (PCs).42 A clustering-based approach was then applied to two-dimensional UMAP embeddings of study participants using k-means clustering with a pre-specified number of clusters (k = 15; random_state = 42) to capture fine-scale clustering patterns. Clusters were assigned to 1000G superpopulations based on proximity to the corresponding 1000G samples in the UMAP space, and participants assigned to the European reference cluster were extracted (Supplementary Figure 1, available at Journal of Neuropathology & Experimental Neurology online).43,44 To further adjust for residual European ancestry effects, PCA was rerun in PLINK using the “—pca” option, and a scree plot of eigenvalues was generated to determine the optimal number of PCs to include as covariates. Statistical analysis Participants were included if NP data and/or cancer history were available, with missing data handled through an outcome-specific stratification, and analyses conducted within phenotype-specific subsets (Figure 2). Sensitivity analyses were restricted to individuals of European ancestry with complete data on both cancer status and hippocampal TDP-43 pathology for the rs5848 variant. Additionally, established Alzheimer disease risk variants of TMEM106B, APOE ε4, and BIN1 were also evaluated for associations with cancer status to assess the specificity of the observed findings. Study workflow and sample selection. n*: Number of samples with at least one neuropathology data. WGS: Whole Genome Sequencing, NP: Neuropathology, ADNC: NIA-AA Alzheimer’s disease neuropathologic change, ADSP: the Alzheimer’s Disease Sequencing Project, ADGC: Alzheimer’s Disease Genetics Consortium, NACC: National Alzheimer’s Coordinating Center, HS aging: Hippocampal sclerosis of Aging, including mesial temporal sclerosis. For binary outcomes, association testing was performed using the Scalable and Accurate Implementation of Generalized Mixed Model (SAIGE) with a binary outcome specification.45 SAIGE accounts for participant relatedness and addresses imbalance in the distribution of the modeled event outcomes through Firth’s Bias-Reduced Logistic Regression.46 For ordinal neuropathologic outcomes, we applied Proportional Odds Logistic Mixed Models (POLMM) using the Genome-wide Regression Analysis under Bias control (GRAB) framework,47 which extends cumulative logit models by incorporating random effects derived from a genetic relationship matrix to account for participant-relatedness and population structure.47,48 Ordinal outcomes included Braak NFT stages and CERAD neocortical neuritic plaque density.49,50 The proportional odds assumption was evaluated using the Brant test.51 Due to evidence that the proportional odds assumption was violated, the ADNC and CERAD neocortical neuritic plaque density outcomes were recategorized. Specifically, ADNC was dichotomized (ADNC: 0 = not ad, low ADNC, or intermediate ADNC vs 1 = high ADNC), and CERAD neocortical neuritic plaque density was trichotomized (0 = none, 1 = sparse, vs 2 = moderate or frequent). Each model was analyzed assuming an additive mode of inheritance and adjusted for sex, age at last visit for cancer history, age at death for neuropathological outcomes, and top 2-3 PCs as determined by inspection of scree plots of PCA eigenvalues derived from genotype data. Study-specific estimates of regression coefficients in ADSP-NACC and ADGC-NACC were meta-analyzed using METAL.52 Results The overall study workflow is summarized in Figure 2. Demographic and neuropathological characteristics of the ADSP-NACC and ADGC-NACC cohorts are presented in Table 1. The frequencies of rs5848 T-allele carriers across major cancer types are shown in Supplementary Table 2, available at Journal of Neuropathology & Experimental Neurology online. In the ADSP-NACC cohort, T-allele carrier frequencies ranged from 24.0% in lung cancer to 51.6% in breast cancer, whereas in the ADGC-NACC cohort, the range was from 41.4% in breast cancer to 55.4% in lung cancer. Table 1Demographic and neuropathological characteristics of ADGC and ADSP participants stratified by cancer status.Characteristic, n (%) or mean ± SDADSP-NACCADGC-NACCNP available*No cancerCancer casesNP available*No cancerCancer cases(n = 1853)(n = 2336)(n = 895)(n = 2583)(n = 5945)(n = 2296)Age at death79.0 ± 10.878.5 ± 11.184.0 ± 8.676.9 ± 10.084.0 ± 10.685.5 ± 8.8Hippocampal sclerosis of aging†216 (13.0)84(15.9)19 (11.4)297 (14.7)100 (14.6)39 (13.5)TDP-43 pathologyAmygdala252 (15.2)158 (37.4)48 (36.1)331 (16.3)185 (33.0)79 (31.7)Hippocampus222 (13.4)124 (27.7)37 (26.2)297 (14.7)157 (27.2)59 (23.0)Neocortex39 (2.4)25 (6.3)5 (3.9)70 (3.5)34 (6.5)12 (5.1)Braak NFT stage015 (0.8)1 (0.2)2 (1.2)97 (3.9)22 (3.1)6 (2.1)I-II196 (10.6)58 (10.8)32 (18.8)485 (19.7)131 (18.7)60 (20.8)III-IV377 (20.4)91 (16.9)39 (22.9)655 (26.6)170 (24.2)89 (30.8)V-VI1261 (68.2)389 (72.2)97 (57.1)1229 (49.8)379 (54.0)134 (46.4)CERAD neuritic plaque scoreNo534 (21.6)70 (13.0)38 (22.4)224 (12.1)152(21.6)72 (24.9)Sparse352 (14.2)56 (10.4)25 (14.7)191 (10.3)77(10.9)35 (12.1)Moderate + Severe1588 (64.2)413(76.6)107 (62.9)1436 (77.6)475 (67.5)182 (63.0)ADNCHigh ADNC738 (44.5)360 (67.0)80 (47.1)690 (34.0)335 (48.1)107 (37.2)*Participants with available neuropathologic assessments used for genetic analysis, with or without cancer history. †: Hippocampal sclerosis of Aging, including mesial temporal sclerosis, NP: Neuropathology, ADNC: NIA-AA Alzheimer’s disease neuropathologic change, ADSP: the Alzheimer’s Disease Sequencing Project, ADGC: Alzheimer’s Disease Genetics Consortium, NACC: National Alzheimer’s Coordinating Center, SD: Standard Deviation, NFT: Neurofibrillary Tangles, CERAD: Consortium to Establish a Registry for Alzheimer’s Disease. Demographic and neuropathological characteristics of ADGC and ADSP participants stratified by cancer status. Participants with available neuropathologic assessments used for genetic analysis, with or without cancer history. †: Hippocampal sclerosis of Aging, including mesial temporal sclerosis, NP: Neuropathology, ADNC: NIA-AA Alzheimer’s disease neuropathologic change, ADSP: the Alzheimer’s Disease Sequencing Project, ADGC: Alzheimer’s Disease Genetics Consortium, NACC: National Alzheimer’s Coordinating Center, SD: Standard Deviation, NFT: Neurofibrillary Tangles, CERAD: Consortium to Establish a Registry for Alzheimer’s Disease. Among participants without a reported history of cancer who had complete data on both Braak NFT stages and TDP-43 pathology in the hippocampus (Figure 3), 40.3% had Braak NFT stages V or VI, 9.2% had TDP-43 pathology in the hippocampus, 18.2% had both pathologies, and 32.3% had neither (Figure 3A). In contrast, among participants with cancer history, a larger proportion had neither pathology (47.6%), while 28.2% had Braak NFT stages V or VI, 10.3% had TDP43 pathology in the hippocampus, and 13.9% had both pathologies (Figure 3B). (A, B) Distribution of Braak NFT stages V or VI and TDP-43 pathology in hippocampus among pooled ADGC and ADSP participants stratified by cancer history. Analyses were restricted to participants with complete data on of Braak NFT stages V or VI and TDP-43 pathology in the hippocampus, and cancer status. In the meta-analysis across the ADSP-NACC and ADGC-NACC cohorts, the T allele of rs5848 was associated with increased odds of TDP-43 pathology in hippocampus (OR = 1.42, 95% CI = 1.20-1.69, p = 5.9 × 10−5) and neocortex (OR = 1.97, 95% CI = 1.46-2.68, p = 1.2 × 10−5), but not in amygdala (OR = 1.16, CI = 0.99-1.35, p = 0.075). The association with HS-Aging was significant in both cohorts and in the meta-analysis (OR = 1.50, CI = 1.30-1.74, p = 4.9 × 10−8). In contrast, a significant protective association with cancer history was observed in the meta-analysis (OR = 0.91, 95% CI = 0.84-0.98, p = 0.012) (Figure 4). Forest plots for the association of the T allele rs5848 in GRN with TDP-43 pathology, HS Aging and cancer status across cohorts. The red dashed vertical line corresponds to the null hypothesis of no association (odds ratio = 1). Genetic association analyses were performed using SAIGE logistic mixed models, adjusting for ancestry principal components, sex, and age. Age at death was used as an age covariate for neuropathologic outcomes in genetic analyses, while age at last clinical assessment was used for cancer outcomes, with cancer defined as having been reported at least once during follow-up. ADSP: the Alzheimer’s Disease Sequencing Project, ADGC: Alzheimer’s Disease Genetics Consortium, NACC: National Alzheimer’s Coordinating Center, HS aging: Hippocampal sclerosis of Aging, including mesial temporal sclerosis. In the parallel analysis of the TMEM106B rs1990622, a well-established TDP-43–related SNP, the A allele was strongly associated with an increased odds of TDP-43 pathology in amygdala and hippocampus and HS-Aging in both the ADSP-NACC and ADGC-NACC cohorts. The meta-analysis demonstrated robust pooled effects on HS-Aging (OR = 1.77, 95% CI =1.53-2.04, p = 1.0 × 10−14) and on TDP-43 pathology in amygdala and hippocampus (OR point estimate range was 1.47 to 1.67 and both p-values were < 1 × 10−6), indicating that the A allele confers increased susceptibility to HS-Aging as well as TDP-43–related neuropathology in amygdala and hippocampus. In contrast, the meta-analysis showed no evidence of an association with cancer status (OR = 0.99, 95% CI = 0.93-1.05, p = 0.66) (Figure 5). Forest plots of the association between the TMEM106B rs1990622 A allele and TDP-43 pathology, HS Aging, and cancer status across cohorts. The red dashed vertical line corresponds to the null hypothesis of no association (odds ratio = 1). Genetic association analyses were performed using Scalable and Accurate Implementation of Generalized Linear Mixed Models (SAIGE) logistic mixed models, adjusting for ancestry principal components, sex, and age. Age at death was used as an age covariate for neuropathologic outcomes in genetic analyses, while age at last clinical assessment was used for cancer outcomes, with cancer defined as having been reported at least once during follow-up. ADSP: the Alzheimer’s Disease Sequencing Project, ADGC: Alzheimer’s Disease Genetics Consortium, NACC: National Alzheimer’s Coordinating Center, HS aging: Hippocampal sclerosis of Aging, including mesial temporal sclerosis. We also assessed whether similar protective associations of other ad-related SNPs (analogous to those observed for rs5848) could be detected. As shown in Figure 6, APOE ε4 was not associated with cancer in either cohort or in the meta-analysis (OR = 0.95, 95% CI 0.89-1.02; p = 0.15). Additionally, two ad-associated variants, rs744373 (G allele) and rs6733839 (T allele), located in BIN1 had robust significant associations with neocortical neuritic plaques and Braak NFT stage but not with cancer diagnosis (Figure 7). All detailed cohort-specific and meta-analyzed results are provided in Supplementary Table 3, available at Journal of Neuropathology & Experimental Neurology online. Analyses were subsequently focused on two major cancer types, breast and prostate cancers, to assess associations of the rs5848 T allele (with the implicit caveat of smaller sample sizes). Although a significant protective association of T allele with breast cancer was observed in the ADGC-NACC cohort, this was not replicated in the ADSP-NACC cohort and meta-analysis (Supplementary Table 4, available at Journal of Neuropathology & Experimental Neurology online). Forest plots showing the association of APOE ε4 with Alzheimer disease–related neuropathologic changes and reported cancer history across two cohorts (ADSP-NACC and ADGC-NACC), followed by fixed-effect meta-analysis. ADNC scores and cancer status were analyzed as binary traits. The red dashed vertical line corresponds to the null hypothesis of no association (odds ratio = 1). Genetic association analyses were performed using Scalable and Accurate Implementation of Generalized Linear Mixed Models (SAIGE) logistic mixed models, adjusting for ancestry principal components, sex, and age. Age at death was used as an age covariate for neuropathologic outcomes in genetic analyses, while age at last clinical assessment was used for cancer outcomes, with cancer defined as having been reported at least once during follow-up. ADNC: NIA-AA Alzheimer’s disease neuropathologic change, ADSP: the Alzheimer’s Disease Sequencing Project, ADGC: Alzheimer’s Disease Genetics Consortium, NACC: National Alzheimer’s Coordinating Center. Forest plots showing the association of two BIN1 variants—rs6733839 (G allele), rs744373(T allele) with Alzheimer’s disease–related neuropathologic changes and reported cancer history across two cohorts (ADSP and ADGC), followed by fixed-effect meta-analysis. Neuropathologic outcomes (CERAD score, Braak B score) were analyzed as ordinal traits using proportional odds logistic mixed models, whereas cancer status was analyzed as a binary outcome using Scalable and Accurate Implementation of Generalized Linear Mixed Models (SAIGE) logistic mixed models. All models were adjusted for sex, principal components of ancestry, and age. Age at death was used as an age covariate for neuropathologic outcomes in genetic analyses, while age at last clinical assessment was used for cancer outcomes, with cancer defined as having been reported at least once during follow-up. ADSP: the Alzheimer’s Disease Sequencing Project, ADGC: Alzheimer’s Disease Genetics Consortium, NACC: National Alzheimer’s Coordinating Center, CERAD score: Consortium to Establish a Registry for Alzheimer’s Disease neuritic plaque score, Braak stage: Braak neurofibrillary tangle stage. In sensitivity analyses restricted to the subsample of included participants with available data on both history of cancer and TDP-43 pathology in hippocampus (n = 1,423; demographic and neuropathological characteristics shown in Supplementary Table 5, available at Journal of Neuropathology & Experimental Neurology online), the rs5848 T allele was not associated with cancer history in the meta-analysis (OR = 0.97, 95% CI = 0.80-1.17), whereas its association with TDP-43 pathology in hippocampus remained significant (OR = 1.52, 95% CI 1.25-1.84). We additionally evaluated the interaction effect between cancer history and rs5848 T allele on LATE-NC stages (0 to 3) using an ordinal regression analysis, showing that rs5848 T allele did not modify the relationship between cancer history and LATE-NC severity (Supplementary Table 6, available at Journal of Neuropathology & Experimental Neurology online). Discussion Analyzing data from two cohorts, we found, as expected, that the T allele of rs5848 in GRN was associated with increased burden of LATE-NC and HS-Aging pathologies. In contrast, the same allele was associated with lower odds of documented cancer history. The findings of a negative association between the GRN dementia-related risk allele with cancer history were different than for other dementia-related risk alleles. Specifically, established TMEM106B, APOE, and BIN1 risk alleles had no significant associations with cancer history in the same sample. Although the T allele of rs5848 has been extensively studied as a risk factor for HS-Aging, LATE-NC, and other TDP-43-related proteinopathies in older adults, its relationship with cancer has not previously been examined, despite long-standing evidence that PGRN has potential roles in tumor biology.13 One exception to this was the work of Rosas et al that showed in a cohort of 753 Parkinson disease patients, the rs5848 T allele was associated with lower cancer likelihood.27 Otherwise, our findings provide novel evidence that rs5848 T allele confers a protective effect against cancer in later life. Mechanistically, a single genetic variant with opposite impacts across different clinical syndromes is biologically plausible. Previous studies showed that rs5848, located in the 3'-UTR of GRN, was functional. The T allele reduced GRN translation (ie, rs5848 acts as an expression quantitative trait locus [eQTL]), leading to lower PGRN levels in blood24 and was linked to increased expression of inflammatory mediators central nervous system (CNS) such as AXL and CLU, reflecting PGRN’s role in regulating inflammation and lysosomal function.11,53–55 Remarkably, rs5848 has been associated a diverse set of conditions linked to inflammation such as Parkinson disease, ad, and multiple sclerosis.22,56,57 Given that reduced GRN/PGRN levels, particularly associated with the rs5848 T allele, have been consistently linked to neurodegeneration (Supplementary Table 7, available at Journal of Neuropathology & Experimental Neurology online),14,16,17,20,22,24,25,53,55–65 PGRN deficiency appears to promote neuroinflammation and TDP-43 pathology. By contrast, increased GRN/PGRN expression may have the opposite consequence in oncogenesis. Prior studies reported that GRN/PGRN plays an important role across multiple cancers, where its expression/overexpression promotes tumor cell growth, migration, and invasivity (Supplementary Table 8, available at Journal of Neuropathology & Experimental Neurology online).13,27,66–83 Increased PGRN expression in blood and/or in CNS tissue may attenuate inflammation while promoting tumor proliferation. Together, these observations support a mechanistic model in which GRN/PGRN-mediated regulation of inflammatory, lysosomal, and mitogenic pathways creates a biological tradeoff, offering a plausible mechanistic basis for the opposing associations between rs5848 and dementia versus cancer-related outcomes. Our interaction analysis indicated that rs5848 does not function as an effect modifier in the association between cancer and TDP-43 pathology suggesting that the presence of the rs5848 T allele does not alter the strength or direction of this association. Instead, these results support a model in which the rs5848 T allele independently influences the risk for both disease domains. This in turn implies that the mechanisms linking rs5848 to cancer and to TDP-43 pathology may operate through concurrent but not necessarily identical biological pathways, providing a remarkable example of an impactful pleiotropic mechanism. These results have clinical implications. Decreased blood and organ PGRN levels (as with the rs5848 T allele, and also in persons with homozygous or heterozygous deletions of the functional GRN gene), are associated with several human diseases, including FTLD-TDP, LATE-NC, and neuronal ceroid lipofuscinosis.62–65,84–88 The possibility of augmenting GRN/PGRN expression in vivo has catalyzed exciting work aimed at clinical therapies that are designed to augment PGRN levels in blood and/or brain.58–61 However, among gliomas and other subtypes of cancer, PGRN has been shown to stimulate cell growth and malignancy.66–68,89 The present study underscores that efforts to increase PGRN expression should be undertaken with caution due to the potential increased risk of neoplasia. We also note that our results shed light on the reciprocal relationship between LATE-NC/HS-Aging and cancer but not that between ADNC and cancer. However, the distributions of autopsy-confirmed dementia-associated pathologies in individuals with and without cancer diagnoses provided (admittedly limited) support for the hypothesis that ADNC and cancer are indeed reciprocally associated (Figure 3). Among the three specific alleles that were queried for associations between ADNC and cancer (2 BIN1 SNPs, and APOE alleles), one of the BIN1 SNPs (rs6733839) had a dementia risk allele with marginal trend (p < 0.06) in meta-analyzed data for negative association with cancer risk (Figure 7). There are also a number of prior published studies that have indeed indicated a potential role for BIN1 in neoplasia transformation or progression,90–92 possibly via inflammatory mechanisms, and a cancer-protective role for the dementia-risk associated allele(s) of BIN1 remains an area for future studies. Our study has some limitations. Firstly, cancer history was ascertained using the NACC UDS version 3.0 Form D2, in which cancer is defined as “cancer present in the last 12 months (excluding non-melanoma skin cancer), primary or metastatic.” Because this variable was introduced in March 2015, individuals who were treated for and survived cancer prior to March 2014 may have been coded as cancer-free despite having a true prior cancer history.30 Moreover, the 12-month time restriction was specified in the coding guidebook but not explicitly included on the data collection form, so there is potential for heterogeneity in data collection across contributing ADRCs, where some may have considered cancers from earlier in life while others did not. Therefore, we consider this a measure of later life cancer history that likely has higher sensitivity than specificity. However, misclassification probability is almost certainly nondifferential with respect to genetic profiles, since genetic profiles would have been unknown to both participants and clinicians at the time of data collection. As such, we anticipate that our observed results for associations with later life cancer may be biased toward the null. Our study design also lacked direct histopathological correlations between tumor tissue subtypes and dementia-related neuropathology. Moreover, cancer type–specific analyses were under-powered statistically, highlighting the need for larger studies to clarify potential cancer subtype-specific effects. In principle, the molecular “fingerprinting” of cancer subtypes should go beyond histological classification. Likewise, on the other side of the associative comparison, fully quantitative neuropathological measures (eg, TDP-43–immunopositive area or cell counts), along with more detailed qualitative assessment of TDP-43 pathology, will be also required in future investigations. Our analyses were also limited to participants with European ancestry. This study design feature is commonly employed to minimize genotype variation due to ethnoracial genetic architectural differences. However, this consideration is doubly relevant because we previously described that the rs5848 risk allele has far higher population frequency in persons of African ancestry than in those with European ancestry.23 Theoretically this may indicate a genetic factor associated with lower cancer risk and higher dementia risk in person of African ancestry. Again, we emphasize that larger and more diverse cohorts are required to study these research questions in a more generalizable manner. We also acknowledge the possibility of survival bias and recruitment biases. These are known features of the dementia-associated study cohorts that were used in the present study.32 As such, these observations are preliminary, and future analyses incorporating formal survival modelling and longitudinal follow-up will be necessary to determine whether age-related, socioeconomic factors, and other cohort-specific selection effects contribute to the phenomena that were described in the current study. Despite these limitations, the proportion of participants with a documented later life cancer history in the present dataset (∼28%) is within a plausible range when compared to population-based estimates of lifetime cancer risk (∼39%) and cancer mortality (∼17%).93 Considered together, these observations suggest modest under-recruitment, and/or under-reporting, of individuals with later life cancer rather than severe misclassification. Future studies using datasets with more comprehensive lifetime cancer ascertainment may help further clarify these associations. In summary, our study provides evidence that the PGRN-lowering rs5848 T allele is associated with reduced cancer burden while simultaneously conferring increased susceptibility to LATE-NC. The robustness of the rs5848 associations across multiple cohorts, together with the absence of comparable signals for TMEM106B, APOE ε4 or BIN1 variants in relation to cancer, underscores at least partial specificity of this genetic association. Although survival bias cannot be fully excluded, the opposing associations suggest that rs5848 may differentially influence two common aging-related phenotypes, each with large public health impact. Future studies across larger, diverse populations, with more comprehensive data gathered including detailed cancer phenotyping, will be essential to clarify the GRN-mediated disease risk trade-off and to help inform how GRN/PGRN-related pathways could be targeted for therapeutic development. Supplementary Material