The UNC13A cryptic exon associates with cognitive impairment in Alzheimer’s disease Authors: Madison M. Reeves, Sara Aviles Carpintero, Matteo Zanovello, Karen Jansen-West, Mei Yue, Anna Calliari, Yuping Song, Judith Dunmore, Candela Maroto Cidfuentes, Sophie Ball, Jennifer M. Kachergus, Heather E. Killeen, Bailey Rawlinson, Erica Engelberg-Cook, Michael DeTure, Gregory S. Day, Neill R. Graff-Radford, Bradley F. Boeve, David S. Knopman, Ronald C. Petersen, Melissa Murray, Aubrey E. Thompson, Dennis W. Dickson, Pietro Fratta, Leonard Petrucelli, Keith A. Josephs, Mercedes Prudencio Published: 2026 ABSTRACT BackgroundTAR DNA-binding protein of 43 kDa (TDP-43) is often found in the brains of patients with Alzheimer’s disease (AD), where it co-occurs with amyloid β plaques and tau neurofibrillary tangles, and associates with accelerated cognitive decline and brain atrophy. TDP-43’s function of repressing the inclusion of cryptic exons (CEs) during RNA splicing is compromised in AD. A single-nucleotide polymorphism (SNP) located within the CE in the UNC13A gene [rs12973192 (C > G)] is associated with higher disease risk and reduced survival in amyotrophic lateral sclerosis (ALS) and frontotemporal dementia (FTD) by weakening TDP-43 binding promoting CE inclusion.ObjectiveTo investigate the influence of the rs12973192 UNC13A CE SNP and UNC13A cryptic splicing on TDP-43 pathology, survival and cognitive impairment in AD.MethodsWe evaluated the UNC13A CE SNP in a cohort of 1,672 AD, including 643 AD brains with available cognitive measurements and 73 AD cases for which we measured cryptic RNA levels in the amygdala. We also evaluated a cohort of 466,517 from the UK Biobank to determine associations between the UNC13A CE SNP and dementia diagnosis.ResultsIn AD, the UNC13A CE SNP associated significantly with cognitive decline, but not with TDP-43 pathology or with survival. UNC13A cryptic RNA levels in the amygdala were a better predictor of cognitive decline than the UNC13A CE SNP itself, while STMN2–another well-known CE target–exhibited no such association.ConclusionsThese findings point to UNC13A cryptic splicing as a specific driver of cognitive decline in AD, outperforming both genetic risk and other cryptic targets.Supplementary InformationThe online version contains supplementary material available at 10.1186/s13195-026-02157-7. Background Single nucleotide polymorphisms (SNPs) in UNC13A, a gene with important roles in synaptic function, increase the risk of developing amyotrophic lateral sclerosis (ALS) and frontotemporal dementia (FTD) [1, 2]. Both neurodegenerative diseases share a common pathological mechanism: the nuclear depletion, aggregation and posttranslational modifications (i.e., phosphorylation and ubiquitination) of transactive response DNA-binding protein of 43 kDa (TDP-43) in neurons and glia [3–5]. TDP-43 is a heterogeneous nuclear ribonucleoprotein that primarily resides in the nucleus where it regulates multiple functions related to RNA metabolism, including splicing, stability, trafficking and microRNA synthesis [6]. A crucial role of TDP-43 is to repress the inclusion of cryptic exons (CE), intronic sequences that are normally excluded from mature messenger RNA [7–15]. Thousands of splicing targets are regulated by TDP-43, including STMN2, a key regulator of axonal regeneration [16, 17], and UNC13A, which plays a crucial role in the vesicle priming prior synaptic vesicle fusion [18–20]. The nuclear clearance of TDP-43 triggers inclusion of CEs in these transcripts, ultimately leading to the loss of functional protein expression [11, 14]. The presence of UNC13A and STMN2 CEs has been reported not only in FTD and ALS [11, 14, 16, 17, 21–23], but also in brain regions affected by TDP-43 pathology in patients with Alzheimer’s disease (AD) [24] and limbic-predominant age-related TDP-43 encephalopathy (LATE) [25]. Notably, the relevance of the UNC13A CE is further reinforced by genetics where the rs12608932 (A > C) SNP in UNC13A is strongly linked with an increased risk for FTD/ALS [1, 26, 27]. This SNP is in linkage disequilibrium with the nearby rs12973192 (C > G, R2: 0.49, D’: 0.99; Χ 2: 2477.02, p < 0.0001) [28], situated within the CE itself. The presence of the risk allele in rs12973192 (G), reduces TDP-43 binding and potentiates CE inclusion in conditions of TDP-43 dysfunction [11, 14, 29]. Moreover, increased copies of the risk allele on either rs12608932 or rs12973192 are associated with shorter survival after onset in FTD/ALS [11, 26, 30–32] and with cognitive deficits [32–34]. However, little is known about the role of UNC13A SNPs in AD. We analyzed the frequency of the rs12973192 (C > G) UNC13A CE SNP in a large cohort of 1,672 AD cases that are well characterized by the presence (AD/LATE) or absence (AD-no-TDP) of TDP-43 pathology to elucidate the relationship between UNC13A CE (SNP and cryptic RNA) and TDP-43 pathology, survival, and cognitive impairment in AD. Materials and methods Participants Postmortem AD cases from both males and female were obtained from the Mayo Clinic Brain Bank, the Mayo Clinic Alzheimer’s Disease Research Center and the Mayo Clinic Study of Aging. Brain tissues (cerebellum or cortex) were obtained from 1,672 individuals with neuropathological diagnosis of AD for genomic DNA extraction, based on availability. For a subset of individuals in our AD cohort (N = 643), we had information on cognitive and motor function, including the Clinical Dementia Rating scale Sum-of-Boxes (CDR-SB) [35], the Mini-Mental State Examination (MMSE) [36, 37]. The CDR-SB assesses the severity of functional impairment (range: 0–18), with higher scores indicating greater impairment. The MMSE is a 30-point bedside screening tool of global cognition, where lower scores (typically < 24) suggest the presence of cognitive impairment. Out of the 643 cases with cognitive scores available, we were able to obtain amygdala from 73 cases to measure cryptic RNA levels as described below. Data from the UK Biobank included 466,517 participants with available rs12973192 UNC13A genotype. ICD-10 codes (UK Biobank Field ID 41202), self-reported diagnoses (UK Biobank Field ID 20002), and causes of death (UK Biobank Field ID 40001) were retrieved and dementia was coded based on the descriptor “dementia” or “Alzheimer” (F00-F03 or G30). Neuropathological characterization of the postmortem cohort All participants had died and underwent standard neuropathological examination performed by a board-certified neuropathologist (DWD). Each case had been assigned a Braak neurofibrillary tangle (NFT) stage [38] using anti-tau antibodies (clone AT8, 1:1,000 dilution; Endogen, Woburn, MA), and a Thal phase [39] using thioflavin-S microcopy, modified Bielschowsky silver impregnation or antibodies to beta-amyloid (clone 6 F/3D, 1:10 dilution; Novocastra Vector Labs, Burlingame, CA). All patients were given a Braak NFT stage based on published criteria as follows: Braak stage 0 = no NFTs found throughout the entire brain; Braak stage I = NFTs are present and confined to the transentorhinal cortex; Braak stage II = NFTs extends into the entorhinal cortex, Braak stage III = NFTs extends beyond entorhinal into fusiform cortex; Braak stage IV = NFTs extend beyond fusiform cortex into inferior and medical temporal cortex, Braak stage V = NFTs extend beyond lateral temporal cortex into occipital and frontal association cortices and Braak stage VI = NFTs are present in peristriate and striate areas of the occipital cortex. All patients were given a Thal β-amyloid phase based on published criteria as follows: Thal phase 0 = no β-amyloid deposition is present throughout the entire brain; Thal phase 1 = β-amyloid positive senile plaques are present and confined to cortical regions; Thal phase 2 = β-amyloid senile plaques are identified in neocortex as well as in the cornu ammonis region 1 or subiculum of the hippocampus; Thal phase 3 = β-amyloid senile plaques are found in the basal ganglia; Thal phase 4 = β-amyloid senile plaques are found in the cornu ammonis region 4 or midbrain and Thal phase 5 = β-amyloid deposition extends into the cerebellar molecular layer. To determine LATE neuropathological changes, amygdala, hippocampus, and frontal cortex sections were immunostained for TDP-43 (using an antibody that detects TDP-43 phosphorylated at serines 409/410, 1:500, CAC-TIP-PTD-M01, Cosmo Bio USA). Cases without TDP-43 immunoreactivity were considered LATE stage 0 (AD-no-TDP), those with amygdala only inclusions (stage 1), those with TDP-43 immunoreactivity extending into the hippocampus (stage 2), and those with TDP-43 immunoreactivity in the frontal cortex as (stage 3) were considered AD/LATE cases. TDP-43 subtyping was based on whether there were typical TDP-43-immunoreactive inclusions (type α) or inclusions colocalizing with NFTs (type β) [40]. DNA extraction and genotyping Genomic DNA was extracted from ~ 15 mg of cerebellar tissue (or cortex if cerebellum was unavailable) from AD brains using Wizard Genomic DNA Purification kit, following manufacturer instructions (Promega, Cat. No. A1620). Concentrations were assessed with a Nanodrop. Approximately 40 ng of genomic DNA was used in a TaqMan SNP genotyping assays (Thermo Fisher Scientific, Cat. No. 4351376) for UNC13A (rs12973192 ID: C_11514504_10) and APOE (rs429358 ID: C_3084793_20 and rs7412 ID: C_904973_10), using TaqMan™ Genotyping Master Mix (Cat. No. 43-713-53, Thermo Fisher Scientific) and run in a QuantStudio 7 Flex Real-Time PCR system (Applied Biosystems), according to the manufacturer’s instructions. The APOE diplotype (rs429358, rs7412) was established by evaluating an Infinium Global Screening Array v3.0 (Illumina). Paired raw probe intensity files (.idat files) were obtained for each sample (N = 1,672). Conversion of .idat files to genotype call files (GTCs), subsequently to variant call files (VCFs), was performed using Illumina’s DRAGEN Array Software v1.2. Sample VCFs were merged and processed using bcftools [41]. Additional sample- and variant-level QC was performed using PLINK2 [42] (and PLINK1 for identity-by-descent analyses). Samples were excluded for missingness > 5%, excess heterozygosity (± 3 SD), relatedness (pi-hat > 0.98), sex discordance, or non-ancestry outlier behavior in principal component analyses. Variants were removed for call rate < 98%, Hardy–Weinberg equilibrium (HWE) < 1 × 10⁻⁶, or MAF < 1%. The cleaned dataset was phased using Eagle (2.4.1) [43] using the 1000 Genomes Phase 3 version 5 reference panel (GRCh37) [44] and the HapMap-reformatted genetic map (hg19) provided by Eagle. Genotype imputation was performed using Minimac (4.1.6) [45] using the same 1000 Genomes Phase 3 version 5 reference panel for the full chromosome 19 (GRCh37) [44]. Imputation accuracy for rs7412 (imputation R² = 0.994) was evaluated in the QC-filtered variant calls across samples by comparing panel-based genotyping and imputations. Only one discrepancy was observed between panel-based and imputed calls, which corresponded to a no-call in the panel-based assay, which was subsequently imputed. Imputation was also performed for rs429358, resulting in a similarly high imputation quality of R² = 0.985. The six APOE variants were assigned as follows: E2/E2 = rs7412 (T/T) & rs429358 (T/T), E2/E3 = rs7412 (C/T) & rs429358 (T/T), E2/E4 = rs7412 (C/T) & rs429358 (T/C), E3/E3 = rs7412 (C/C) & rs429358 (T/T), E3/E4 = rs7412 (C/C) & rs429358 (T/C), E4/E4 = rs7412 (C/C) & rs429358 (C/C). RNA extraction and NanoString analyses on human postmortem tissues The RNAeasy Plus Mini Kit (Qiagen) was used to extract RNA, per manufacturer’s instructions and as previously described [12, 46]. Briefly, only extractions with high RNA integrity number (RIN), assessed using an Agilent 2100 bioanalyzer (Agilent Technologies), were kept for downstream analyses. RNA concentrations were determined using Nanodrop technologies (Thermo Fisher). RNA extractions were repeated up to three times if RIN values were < 6.5 or concentrations were too low. The samples with highest RIN and concentrations were selected for downstream analysis. UNC13A and STMN2 cryptic RNA levels were measured using 500 ng of total RNA from 73 AD/LATE amygdalae, in the NanoString Elements platform. Transcript level abundance was determined using nSolver 4.0 software (NanoString Technologies). The target sequences for the NanoString probes used in this study are included in Supplementary Table 1. Statistical analysis Statistical analyses were performed and visualized using R [47] or GraphPad Prism v.10.2.0. For all groups of similar statistical tests, a Bonferroni-corrected significance threshold (α/n) was applied to determine the threshold of statistical significance, which are indicated in the corresponding tables. Detailed statistical analyses are described in supplementary materials and methods. We evaluated the relationship between the UNC13A CE SNP and the odds of exhibiting TDP-43 pathology in AD/LATE compared to AD-no-TDP using additive [CC = 0 (reference), CG = 1, GG = 2] genetic and allelic [C = 0 (reference), G = 1] binary logistic regression models. We also evaluated the odds of exhibiting TDP-43 type β pathology, compared separately to type α in AD/LATE, as well as to AD-no-TDP cases. Unadjusted and adjusted models were included for all analyses, with covariate adjustments selected based on biological relevance. No multicollinearity (variance inflation factors) was detected in any models. Kaplan-Meier survival curves were generated in GraphPad Prism v.10.2.0 to visualize unadjusted differences in survival times between groups based on the UNC13A CE SNP genotype. Cox proportional hazards regression models were used to assess the link between the rs12973192 UNC13A CE SNP and survival using additive [CC = 0 (reference), CG = 1, GG = 2], dominant [CC = 0 (reference), CG + GG=1] genetic models. Models were adjusted by age at symptom onset and sex for assessment in AD/LATE and AD-no-TDP groups, or by age at symptom onset, sex, and TDP-43 status for the full cohort. Assessment of the proportional hazards assumption was conducted for all covariates across all models. The association between the UNC13A CE SNP and dementia diagnosis in the UK Biobank cohort was performed using generalized linear regression analysis, where sex, age at participation, the first 10 genetic principal component, smoking status (never, past, or current; UK Biobank Field ID 20116), years of education (UK Biobank Field ID 6138), and body mass index (BMI, UK Biobank Field ID 21001) were used as covariates. The associations between the rs12973192 UNC13A CE SNP and the cognitive and functional baseline and follow-up assessments of MMSE and CDR-SB were determined using additive [CC = 0 (reference), CG = 1, GG = 2] genetic and allelic [C = 0 (reference), G = 1] linear regression models. To improve the normality of model residuals while retaining interpretability, we applied variable-specific transformations. MMSE scores were reflected and log-transformed using log[(maximum MMSE + 1)–MMSE] to achieve greatest normality, and CDR-SB scores were log-transformed using log(CDR-SB + 1). To preserve the directionality of MMSE modeling results on the original scoring scale, the signs of the estimated MMSE outcomes were reversed post hoc. Both unadjusted and models adjusted for time from cognitive test administration to death, sex and TDP-43 status were constructed. Linear regression models were tested for normality of residuals, heteroscedasticity (Breusch–Pagan test), multicollinearity (variance inflation factors), and influential points (Cook’s distance). The associations between the UNC13A and STMN2 cryptic RNA levels and the cognitive and functional assessments of MMSE and CDR-SB were investigated using Spearman’s partial correlations and linear regression modeling. Separate analyses for baseline and final cognitive assessment scores, as well as for cognitive rate of change over time (calculated as [(Final Score – Baseline Score) / Time between Baseline and Final testing)] were performed. Spearman’s partial correlation analyses were performed using the ppcor package in R [48], including time from testing to death and sex. 95% confidence intervals (CI) for nonparametric correlations were estimated by percentile bootstrapping (1,000 resamples) using the boot package in R [49, 50]. For linear regression models examining the relationship between UNC13A RNA levels and baseline and final cognitive assessment scores, UNC13A RNA levels were untransformed, and MMSE and CDR-SB baseline and follow-up scores were log-transformed to achieve greatest normality. Adjusted models included sex and age at death as covariates. For linear regression models examining cognitive assessment rate of change, a rank-ordered quantile normalization was applied to both RNA levels and cognitive outcomes to improve normality of model residuals and to enable the parametric assessment of linear associations between the rank-based, normalized data. Adjusted models included sex as a covariate. Linear regression models were tested for normality of residuals, heteroscedasticity (Breusch–Pagan test), multicollinearity (variance inflation factors), and influential points (Cook’s distance). Of note, in cognitive and functional analysis the term “baseline” refers to the first available cognitive testing timepoint within our dataset. Further, since cognitive scores are highly dependent on the disease stage at the time of testing, we adjusted our models for ‘time from testing to death’, since death represents a definitive, biological uniform end-stage for all patients, making ‘time to death’ a highly robust proxy for disease progression. In contrast, age at disease onset is subjective, prone to recall bias, and missing for a larger number of cases in our cohort. Due to missing data, adjusting age at disease onset reduces statistical power, though the overall trends remain consistent. These results are included in the Supplementary Tables for completeness. Results The rs12973192 UNC13A CE SNP does not predict TDP-43 pathology or survival in AD To investigate whether the rs12973192 UNC13A CE SNP is associated with increased risk of developing TDP-43 pathology in AD, we genotyped a large cohort of 1,672 postmortem cases with a neuropathological diagnosis of AD (median Braak: 6, median Thal: 5, Table 1). The cohort included patients that were histologically confirmed to lack TDP-43 pathology in limbic regions including the amygdala and hippocampus (AD-no-TDP, N = 928) or have TDP-43 pathology (AD/LATE, N = 744). In AD, 44.4% (N = 742) were homozygous for the major allele (C/C), 44.1% were heterozygous (N = 738, C/G) and 11.5% (N = 192) were homozygous for the minor allele (G/G) (Fig. 1A). The minor allele frequency (MAF) in our AD cohort (N = 1,672) was 33.6%, similar to what has been observed in a general European population [MAF: 34.2%, N = 1,006 (1000genomes [51]), ], and in a cohort from the UK Biobank (MAF: 34.3%, N = 466,517; Supplementary Table 2) [52, 53]. Further, MAFs were remarkably similar between AD-no-TDP (MAF: 33.6%, N = 928) and AD/LATE cases (MAF: 33.5%, N = 744; Supplementary Table 3). We also observed similar proportions of individuals that were homozygous and heterozygous for the rs12973192 UNC13A SNP between AD-no-TDP (44.2% C/C, 44.4% C/G, 11.4% G/G) and AD/LATE (44.6% C/C, 43.8% C/G, 11.6% G/G) (Fig. 1A). Logistic regression analysis demonstrated no significant associations between the UNC13A CE SNP genotype and the odds of exhibiting TDP-43 pathology in AD, even when adjusting for potential confounding variables such as age of death, sex and APOE ε4 status (Fig. 1A). Table 1Characteristics of the AD cohortVariableAll AD(N = 1,672)AD-no-TDP(N = 928)AD/LATE(N = 744)Sex (female)906 (54.2%)476 (51.3%)430 (57.8%)Age at onset (years)72 (34, 101)70 (34, 98)74 (34, 101)Age at death (years)83 (38, 110)80 (38, 110)86 (54, 104)Survival after onset (years)9 (0.1, 36)9 (1, 26)10 (0.1, 36)Race (White)1,643 (98.4%)908 (98.2%)735 (98.8%)Ethnicity (non-Hispanic or Latino)1,627 (97.3%)902 (97.2%)725 (97.4%)TDP-43 pathology744 (44.5%)0 (0%)744 (100%) Type α327 (44.0%)-327 (44.0%) Type β287 (38.6%)-287 (38.6%)Braak stage6 (2, 6)6 (2, 6)6 (3, 6)Thal phase5 (2, 5)5 (2, 5)5 (2, 5)Lewy body pathology699 (63.1%)377 (66.6%)322 (59.5%)Vascular dementia709 (43.1%)370 (40.3%)339 (46.6%)The sample median (minimum, maximum) is given for continuous variables, categorical variables were summarized with number and percentage of patients (calculated from total of patients with information in each category). Age at death was missing for 3 individuals, and age of disease onset and survival time after onset were not available for 317 cases. Non-White individuals included 3 American Indian/Alaskan Native, 6 Asian, and 15 African Americans. A total of 5 individuals had undisclosed or unknown racial status, and 3 cases had unknown ethnic status. AD/LATE cases included 130 individuals that were positive for TDP-43 pathology, but TDP-43 subtyping classification was different from α or β subtypes or not assessed. The presence of Lewy body pathology was not available for 565 cases, and vascular disease for 26 cases was unknown Table 1Characteristics of the AD cohortVariableAll AD(N = 1,672)AD-no-TDP(N = 928)AD/LATE(N = 744)Sex (female)906 (54.2%)476 (51.3%)430 (57.8%)Age at onset (years)72 (34, 101)70 (34, 98)74 (34, 101)Age at death (years)83 (38, 110)80 (38, 110)86 (54, 104)Survival after onset (years)9 (0.1, 36)9 (1, 26)10 (0.1, 36)Race (White)1,643 (98.4%)908 (98.2%)735 (98.8%)Ethnicity (non-Hispanic or Latino)1,627 (97.3%)902 (97.2%)725 (97.4%)TDP-43 pathology744 (44.5%)0 (0%)744 (100%) Type α327 (44.0%)-327 (44.0%) Type β287 (38.6%)-287 (38.6%)Braak stage6 (2, 6)6 (2, 6)6 (3, 6)Thal phase5 (2, 5)5 (2, 5)5 (2, 5)Lewy body pathology699 (63.1%)377 (66.6%)322 (59.5%)Vascular dementia709 (43.1%)370 (40.3%)339 (46.6%)The sample median (minimum, maximum) is given for continuous variables, categorical variables were summarized with number and percentage of patients (calculated from total of patients with information in each category). Age at death was missing for 3 individuals, and age of disease onset and survival time after onset were not available for 317 cases. Non-White individuals included 3 American Indian/Alaskan Native, 6 Asian, and 15 African Americans. A total of 5 individuals had undisclosed or unknown racial status, and 3 cases had unknown ethnic status. AD/LATE cases included 130 individuals that were positive for TDP-43 pathology, but TDP-43 subtyping classification was different from α or β subtypes or not assessed. The presence of Lewy body pathology was not available for 565 cases, and vascular disease for 26 cases was unknown Characteristics of the AD cohort The sample median (minimum, maximum) is given for continuous variables, categorical variables were summarized with number and percentage of patients (calculated from total of patients with information in each category). Age at death was missing for 3 individuals, and age of disease onset and survival time after onset were not available for 317 cases. Non-White individuals included 3 American Indian/Alaskan Native, 6 Asian, and 15 African Americans. A total of 5 individuals had undisclosed or unknown racial status, and 3 cases had unknown ethnic status. AD/LATE cases included 130 individuals that were positive for TDP-43 pathology, but TDP-43 subtyping classification was different from α or β subtypes or not assessed. The presence of Lewy body pathology was not available for 565 cases, and vascular disease for 26 cases was unknown Fig. 1The rs12973192UNC13ACE SNP does not predict TDP-43 pathology or shorter survival in AD. A Genotype frequencies of the rs12973192 UNC13A CE SNP in AD patients stratified by TDP-43 pathology status. Odds ratio (OR) and 95% CI were calculated using binary logistic regression analysis using the additive genetic [CC = 0 (reference), CG = 1, GG = 2] and allelic (C = 0 (reference), G = 1) models and adjusting for age of death, sex and APOE ε4. ORs can be interpreted as the odds of a patient exhibiting TDP-43 pathology vs. no TDP-43 pathology. After applying Bonferroni correction, adjusted p-values < 0.025 are considered statistically significant. B Survival curves of all AD cases in this study stratified by rs12973192 UNC13A CE SNP. Hazard ratios (HR) and 95% CI were calculated using the additive model [CC = 0 (reference), CG = 1, GG = 2], adjusting for age at symptom onset and sex for AD-no-TDP and AD/LATE groups, and age at symptom onset, sex and TDP-43 status for all AD cases. After applying Bonferroni correction, adjusted p-values < 0.0167 are considered statistically significant The rs12973192UNC13ACE SNP does not predict TDP-43 pathology or shorter survival in AD. A Genotype frequencies of the rs12973192 UNC13A CE SNP in AD patients stratified by TDP-43 pathology status. Odds ratio (OR) and 95% CI were calculated using binary logistic regression analysis using the additive genetic [CC = 0 (reference), CG = 1, GG = 2] and allelic (C = 0 (reference), G = 1) models and adjusting for age of death, sex and APOE ε4. ORs can be interpreted as the odds of a patient exhibiting TDP-43 pathology vs. no TDP-43 pathology. After applying Bonferroni correction, adjusted p-values < 0.025 are considered statistically significant. B Survival curves of all AD cases in this study stratified by rs12973192 UNC13A CE SNP. Hazard ratios (HR) and 95% CI were calculated using the additive model [CC = 0 (reference), CG = 1, GG = 2], adjusting for age at symptom onset and sex for AD-no-TDP and AD/LATE groups, and age at symptom onset, sex and TDP-43 status for all AD cases. After applying Bonferroni correction, adjusted p-values < 0.0167 are considered statistically significant TDP-43 pathology in AD may present in the form of TDP-43 subtypes α and β [40]. Therefore, we analyzed the UNC13A CE SNP distribution between TDP-43 types α and β in AD/LATE. Interestingly, we observed an increase in the proportion of cases harboring the minor (G) allele in AD/LATE type β cases (MAF: 36.2%, N = 287) compared to AD/LATE type α (MAF: 30.7%, N = 327; Supplementary Table 4). Similarly, the frequency of the G/G (12.9%) and C/G (46.7%) genotypes was slightly higher, and C/C (40.4%) lower in type β cases compared to type α (9.5% G/G, 42.5% C/G, 48% C/C), although these differences did not reach statistical significance (Figure S1). Moreover, the UNC13A CE SNP did not increase the odds of just having TDP-43 type β compared to AD-no-TDP (Supplementary Fig. 1). The UNC13A CE SNP minor allele is associated with shorter survival after disease onset in FTD/ALS [2, 11, 26, 30, 31, 54]. Thus, we wanted to investigate whether the UNC13A CE SNP impacts survival in our AD cohort. The median survival time after symptom onset in our AD cohort was 9 years, with the AD/LATE group showing longer survival times (median: 10 years) compared to AD-no-TDP cases (median: 9 years, Table 1). Interestingly, we observed no associations between the UNC13A CE SNP and survival across all AD cases, including when adjusting for sex and age of symptom onset or sex, age of onset and TDP-43 status, or when evaluating AD/LATE and AD-no-TDP separately (Fig. 1B). The UNC13A CE SNP associates with cognitive dysfunction Previous studies have linked other UNC13A SNPs to higher prevalence of FTD in ALS, as well as with worse cognitive symptoms [32–34]. Indeed, analysis of the UK Biobank dataset shows that homozygosity for the UNC13A CE SNP minor allele is significantly associated with a diagnosis of dementia. While the genotype distributions were highly similar between individuals with dementia (43% C/C, 43% C/G, 14% G/G; N = 1,505) and those without (43% C/C, 45% C/G, 12% G/G; N = 465,012), a recessive model confirmed a significant enrichment of the minor allele homozygotes in the dementia cohort (OR: 1.1732; 95% CI: 1.0074–1.3591; P value: 0.0364; Supplementary Table 2). Further, we evaluated whether the UNC13A CE SNP was associated with greater cognitive impairment in our neuropathologically confirmed AD postmortem cohort. We analyzed baseline and last available clinical assessments before death, available for 643 cases (Supplementary Table 5) of our AD cohort using two clinical measures: MMSE and CDR-SB. Performance on these measures of cognitive function worsened (lower MMSE and higher CDR-SB scores) over time (Supplementary Table 5). The UNC13A CE SNP minor allele was associated with lower MMSE scores at baseline (N = 636) in linear regression additive genetic (Fig. 2A, Supplementary Table 6) and allelic (Supplementary Table 7) models, even after adjusting for sex, time from test to death, and TDP-43 status (Fig. 2A, Supplementary Tables 6 and 7). In contrast, this association was absent when evaluating the last follow-up MMSE values (N = 585) prior death (Fig. 2A, Supplementary Tables 6 and 7). Similarly, the UNC13A CE SNP minor allele was associated with higher CDR-SB scores at baseline (N = 637) in the unadjusted and adjusted for sex, time from testing to death and TDP-43 status in the genetic (Fig. 2B, Supplementary Table 6) and allelic (Supplementary Table 7) models. No associations were found between the UNC13A CE SNP and follow-up CDR-SB scores (N = 594, Fig. 2B, Supplementary Tables 6 and 7). Fig. 2Increasing copies of the risk UNC13A CE allele is associated with worse cognitive function in AD. Baseline (blue) and follow-up (black) scores for MMSE (A) and CDR-SB (B) stratified by presenting 0, 1 or 2 copies of the minor UNC13A CE SNP (G) allele (middle graphs). Linear regression models were used to determine the association between baseline (blue) and follow-up (black) scores and the UNC13A CE SNP using the additive [CC = 0 (reference), CG = 1, GG = 2] genetic model, unadjusted (squares) adjusted (circles) for time from testing to death, sex and TDP-43 status (right graphs). After applying Bonferroni correction, adjusted p-values < 0.0125 are considered statistically significant Increasing copies of the risk UNC13A CE allele is associated with worse cognitive function in AD. Baseline (blue) and follow-up (black) scores for MMSE (A) and CDR-SB (B) stratified by presenting 0, 1 or 2 copies of the minor UNC13A CE SNP (G) allele (middle graphs). Linear regression models were used to determine the association between baseline (blue) and follow-up (black) scores and the UNC13A CE SNP using the additive [CC = 0 (reference), CG = 1, GG = 2] genetic model, unadjusted (squares) adjusted (circles) for time from testing to death, sex and TDP-43 status (right graphs). After applying Bonferroni correction, adjusted p-values < 0.0125 are considered statistically significant The levels of UNC13A cryptic RNA associate with cognitive decline in AD The presence of the UNC13A CE SNP enhances the accumulation of the UNC13A cryptic RNA under conditions of TDP-43 dysfunction in FTD/ALS [11, 14, 29], and UNC13A cryptic RNA levels are associated with insoluble phosphorylated TDP-43 protein burden in the brains of FTD/ALS and AD [11, 24]. We hypothesized that UNC13A cryptic RNA levels in the AD brain, resulting from TDP-43 loss of function and which can be enhanced by the presence of the risk UNC13A CE SNP, may better predict the degree of cognitive decline. Thus, we measured the levels of UNC13A cryptic RNA in the amygdala available from 73 cases for which baseline or both baseline and follow-up clinical measurements of MMSE (Fig. 3A) and CDR-SB (Fig. 3B) were available. Overtime, MMSE scores were lower (Fig. 3A), and CDR-SB scores were higher (Fig. 3B), demonstrating worse cognition closer to death. Higher UNC13A cryptic RNA levels in the amygdala strongly associated with lower cognitive performance at baseline and follow-up based on both MMSE (Fig. 3C, Supplementary Tables 8 and 9) and CDR-SB (Fig. 3D, Supplementary Tables 8 and 9). We also calculated rate of cognitive decline for the cases with both baseline and follow-up measurements for MMSE (N = 71) and CDR-SB (N = 72) and amygdala UNC13A cryptic RNA levels. We found that the rate of decline significantly associated with higher levels of UNC13A cryptic RNA for both MMSE (Fig. 3E, Supplementary Table 10) and CDR-SB (Fig. 3F, Supplementary Table 10). Of interest, these associations were specific for the UNC13A cryptic RNA, as the levels of STMN2 cryptic RNA did not associate with either MMSE (Spearman’s r: 0.0097; 95% CI: -0.2334 to 0.2758; P value: 0.9367) or CDR-SB (Spearman’s r: 0.0784; 95% CI: -0.1564 to 0.3153; P value: 0.5159) rate of change (Fig. 3G-H, Supplementary Table 11). Fig. 3Higher UNC13A cryptic RNA levels associate with cognitive decline in AD. A, B Temporal trajectories, represented by each individual lines, of cognitive scores for each AD/LATE case for which we had baseline and follow-up measurements of MMSE (A, N = 71) and CDR-SB (B, N = 72) and matching amygdala tissue. Higher MMSE and lower CDR-SB scores closer to death indicate poorer cognitive performance. C, D Spearman’s partial correlation analysis between UNC13A cryptic RNA levels in the AD/LATE amygdala and corresponding baseline (blue) and follow-up (black) measurements for MMSE (C, circles) or CDR-SB (D, squares) and controlling for time from testing to death and sex. E-H Spearman’s partial correlation analysis between UNC13A (E, F) or STMN2 (G, H) cryptic RNA levels measured in the AD/LATE amygdala and corresponding rate of cognitive decline for MMSE (E, G) or CDR-SB (F, H) and controlling for sex Higher UNC13A cryptic RNA levels associate with cognitive decline in AD. A, B Temporal trajectories, represented by each individual lines, of cognitive scores for each AD/LATE case for which we had baseline and follow-up measurements of MMSE (A, N = 71) and CDR-SB (B, N = 72) and matching amygdala tissue. Higher MMSE and lower CDR-SB scores closer to death indicate poorer cognitive performance. C, D Spearman’s partial correlation analysis between UNC13A cryptic RNA levels in the AD/LATE amygdala and corresponding baseline (blue) and follow-up (black) measurements for MMSE (C, circles) or CDR-SB (D, squares) and controlling for time from testing to death and sex. E-H Spearman’s partial correlation analysis between UNC13A (E, F) or STMN2 (G, H) cryptic RNA levels measured in the AD/LATE amygdala and corresponding rate of cognitive decline for MMSE (E, G) or CDR-SB (F, H) and controlling for sex Discussion We evaluated the relationship between the UNC13A CE SNP (rs12973192), TDP-43 proteinopathy, and clinical outcomes in patients with neuropathology confirmed AD. UNC13A is risk gene identified in genome-wide association studies for FTD and ALS [1, 26, 27]. In our AD cohort, the frequency of the UNC13A rs12973192 allele did not differ from other populations of similar European descent, nor has it emerged as a primary hit in AD genome-wide association studies. However, both AD and FTD/ALS share a common downstream mechanism: the loss of nuclear TDP-43 function, which leads to the aberrant incorporation of a CE into the UNC13A transcript and a subsequent reduction in functional protein levels [11, 14, 24]. Since the rs12973192 variant enhances UNC13A CE inclusion under conditions of TDP-43 dysfunction [11, 14, 29], we hypothesized that this SNP acts as a genetic modifier of TDP-43 pathology and clinical progression in AD. The reported prevalence of TDP-43 pathology in AD cases is highly variable, ranging from as low as 22–25% [55, 56] to nearly 70% [57, 58]. This variability could be explained, in part, by non-standardized regional assessment of TDP-43 pathology, as accurate staging requires specific brain regions, such as the amygdala, that are not always available for analysis. In our study cohort, TDP-43 pathology was present in 44.5% of AD cases. While we found no overall association between the UNC13A CE SNP and the broad incidence of TDP-43 pathology, the risk allele was slightly more prevalent in AD/LATE cases with TDP-43 type β than type α pathology. TDP-43 type β is characterized by the co-occurrence or co-aggregation of TDP-43 pathology with neurofibrillary tangles [40]. Intriguingly, a recent study observed a modest positive correlation between UNC13A cryptic RNA levels and phosphorylated tau levels in the hippocampus [25]. It is possible that local tau/TDP-43 interactions in TDP-43 type β co-pathology drive distinct molecular phenotypes that set them apart from AD/LATE type α or AD-no-TDP cases. Although the odds of predicting TDP-43 type β pathology via the UNC13A CE SNP did not reach statistical significance here, our cohort may be underpowered to detect this nuance. This limitation underscores the need for larger, multi-center studies to routinely incorporate TDP-43 subtyping into the pathological assessment of AD. Previous studies have linked UNC13A SNPs to shortened survival in ALS and FTD [11, 26, 30–32]. In contrast, we found no associations between the UNC13A CE SNP and survival in AD, even when stratified by the presence of TDP-43 pathology. Notably, our AD/LATE cases exhibited longer survival times than AD-no-TDP cases. While consistent with previous literature [59, 60], this observation remains highly counterintuitive because TDP-43 is strongly associated with more severe cognitive deficits and pronounced brain atrophy than AD hallmark pathology (i.e. amyloid β plaques and tau neurofibrillary tangles) alone [40, 61–64]. Importantly, the UNC13A CE SNP was significantly associated with poorer cognitive performance in our AD cohort, suggesting that this genotype impacts clinical presentation. This effect was evident in baseline cognitive and functional measurements (i.e., MMSE and CDR-SB) but dissipated later in the disease course (follow-up measurement). Thus, these findings suggest that the UNC13A CE SNP, which exacerbates UNC13A CE inclusion following TDP-43 dysfunction, primarily affects the onset and expression of early cognitive symptoms in AD. This role in shaping early cognitive impairment aligns with previous observations that UNC13A variants associate with impaired cognitive performance, frontotemporal cortical atrophy, and an increased risk of FTD in patients with ALS [32–34]. Because UNC13A CE inclusion ultimately causes UNC13A protein deficiency [11, 14], the accumulation of UNC13A cryptic RNAs in AD/LATE brains [24] is predicted to deplete normal, functional UNC13A. As a presynaptic protein essential for vesicle maturation and fusion, UNC13A plays a central role in neurotransmitter release [18–20]. Genetic deletion of all Unc13 proteins in mice leads to complete paralysis at birth and death [65]. Mice with deletion of the Unc13a isoform form normal hippocampal synapses, but these are unable to complete the vesicle maturation step [19], leading to a significant decrease in spontaneous and evoked glutamatergic excitatory synaptic transmission [19, 65]. Previous studies showed that the loss of TDP-43 impairs synaptic transmission in mice [66] and in induced pluripotent stem cell-derived neurons [67], a deficit that can be reversed by correcting UNC13A missplicing [67]. Synaptic dysfunction and synaptic loss are among the earliest changes that occur in AD, occurring even before the appearance of clinical symptoms or the formation of amyloid plaques [68–70]. Notably, synapse loss remains highly correlated with cognitive decline in AD [68–71]. Further studies emphasized that presynaptic terminals represent a primary site of early pathogenesis in AD [72]. The early onset of these events aligns with our findings, helping to explain why baseline measurements of CDR-SB and MMSE show stronger correlations with the presence of the UNC13A CE SNP than follow-up assessments. Given that the risk rs12973192 UNC13A CE SNP enhances UNC13A cryptic RNA accumulation in the context of TDP-43 dysfunction [11, 14, 29], we hypothesized that UNC13A cryptic RNA levels could serve as a more effective biomarker for cognitive decline than genotype alone. Our analysis of AD amygdalae validated this, revealing a strong association between higher UNC13A cryptic RNA levels and worse cognitive scores at both baseline and follow-up measurements. This indicates that absolute quantitative expression is far more sensitive to cognitive decline than the binary presence of the UNC13A CE SNP. Notably, this clinical trajectory was highly specific to UNC13A cryptic RNA accumulation, as no such cognitive correlations were observed with STMN2 cryptic RNA, another well-established downstream target of TDP-43 missplicing. Conclusions Our study establishes the UNC13A CE SNP rs12973192 as a critical genetic modifier of AD that actively shapes early cognitive and functional phenotypes. More importantly, we demonstrate that tissue-specific accumulation of UNC13A cryptic RNA, rather than genetic background alone or parallel cryptic targets like STMN2, serves as a highly specific predictor of accelerated cognitive decline. By revealing that UNC13A missplicing directly bridges TDP-43 dysfunction to early synaptic vulnerability in AD, these findings open up exciting new avenues for precision medicine. Ultimately, this work highlights UNC13A cryptic transcripts as both sensitive, proximal biomarker for cognitive deterioration and a highly viable, targetable therapeutic avenue to preserve synaptic integrity in TDP-43-linked neurodegenerative diseases. Supplementary Information Supplementary Material 1. Supplementary Material 1.