RBFOX1 association with age at onset of Alzheimer's disease

Abstract

INTRODUCTION

Genetic contributors to early onset Alzheimer's disease (AD) beyond APP and PSEN1/2 remain unknown. Identifying novel loci may reveal disease mechanisms and therapeutic targets. We investigated genetic variants influencing age at onset in early onset families from the Long‐Life Family Study (LLFS).

METHODS

Six families with at least two early onset cases (onset ≤ 65) were identified among 3476 LLFS participants. Genome‐wide linkage analysis of age at onset was followed by single nucleotide polymorphism association. Validation analyses were performed in nine independent cohorts, alongside blood and brain transcriptomic analyses.

RESULTS

Three significant linkage regions were identified, including RBFOX1 (logarithm of the odds = 4.41). RBFOX1 variants were associated with age at onset and cognitive phenotypes. A consistent association of RBFOX1 was observed across validation cohorts. Blood transcriptomics revealed RBFOX1 overexpression was associated with an earlier onset.

DISCUSSION

RBFOX1 may influence AD age of onset, nominating the gene as a potential therapeutic target for delaying or preventing dementia.

Highlights

• Genetic contributors to early‐onset Alzheimer's disease (AD) beyond APP, PSEN1, and PSEN2 remain unknown.

• We identified six families with at least two family members affected by AD at age ≤ 65 from the Long‐Life Family Study (LLFS), a family cohort characterized by exceptional longevity and reduced dementia risk.

• Genome‐wide linkage and single nucleotide polymorphism‐based association analyses identified variants within RBFOX1 associated with disease's age at onset.

• Validation analyses confirmed RBFOX1 genetic associations in nine independent early onset AD and late onset AD cohorts.

• Transcriptomic analysis revealed overexpression of RBFOX1 associated with earlier age at disease onset.

• Our results suggest RBFOX1 as a potential therapeutic target for delaying the onset of dementia.

1. BACKGROUND

Alzheimer's disease (AD) is a progressive neurodegenerative disease characterized by the gradual deterioration of cognitive capacities. Currently ranked as the seventh leading cause of death in the United States, it is the most common cause of dementia among older adults; ¹ that is, 60% to 70% of dementia cases. With a prevalence of > 50 million individuals worldwide, and a tripled projection by 2050, AD dementia represents a significant global burden. ²

Age, the strongest risk factor for developing AD, ³ , ⁴ , ⁵ is a primary way by which the disease is classified. ⁶ The traditional classification involves two categories: early onset AD (EOAD) and late onset AD (LOAD). EOAD, defined as < 65 years old, has a strong genetic basis (90%–100%); however, 90% of EOAD remains genetically unexplained. ⁷ Between 35% and 60% of EOAD cases have at least one affected first‐degree relative with an autosomal dominant inheritance. ⁸ , ⁹ , ¹⁰ Genetic analysis of exceptionally large pedigrees identified high‐penetrant mutations in the three EOAD genes, coding for amyloid precursor protein (APP) and presenilins (PSEN1 and PSEN2). However, only 10% to 15% of EOAD cases can be explained by known mutations in these genes. ¹¹ Our previously reported analyses of Caribbean Hispanic families ¹² , ¹³ , ¹⁴ with at least one EOAD case (onset ≤ 65 years) identified a novel founder PSEN1 mutation, Gly206Ala. Although the EOAD families were unrelated, genetic variation flanking the PSEN1 gene indicated a common Puerto Rican founder. The same mutation was reported in additional individuals from Puerto Rico, but not in other ethnic groups. ¹⁴ , ¹⁵ , ¹⁶

In the current study, we investigated six families with at least two family members affected by AD at age ≤ 65 from the Long‐Life Family Study (LLFS), a cohort characterized by exceptional longevity and reduced dementia risk compared to the general population. ¹⁷ , ¹⁸ Families in LLFS were selected using the Family Longevity Selection Score (FLoSS), ¹⁹ which reflects birth cohort survival probabilities of probands and their siblings. Eligible families (FLoSS ≥ 7) included at least one living sibling and one offspring. Compared to general population cohorts like the Framingham Heart Study, LLFS families showed strong familial clustering of exceptional longevity, higher survival across generations, and evidence of heritable longevity traits. ¹⁹ , ²⁰ These EOAD families exhibited familial aggregation, with at least two family members affected at age ≤ 65, suggesting causal Mendelian risk variants.

Our aims were to: (1) identify genetic variants affecting the age at onset in EOAD families, (2) examine the association of these variants with cognitive function and gene expression, and (3) replicate the genetic associations in independent EOAD and LOAD cohorts. The inclusion of both EOAD and LOAD cohorts enhances our statistical power to detect genetic variants contributing to AD susceptibility across the age spectrum.

RESEARCH IN CONTEXT

1. Systematic review: Only 10% to 15% of early onset Alzheimer's disease (AD) cases can be explained by known mutations in APP, PSEN1, and PSEN2 genes. We investigated six families with at least two family members affected at age ≤ 65 from the Long‐Life Family Study, a cohort characterized by exceptional longevity and reduced dementia risk. We identified a significant linkage and association between RBFOX1 locus and age at onset in long‐lived families, supported by consistent associations across multiple independent early onset AD (EAOD) and late onset AD cohorts. RBFOX1 variants were further associated with cognitive phenotypes, and transcriptomic analyses revealed gene expression downregulation in blood samples.

2. Interpretation: The RBFOX1 gene encodes a neuron‐specific splicing factor regulating splicing networks implicated in neurodevelopmental and neuropsychological disorders. Our findings support RBFOX1 as a contributor to EOAD pathogenesis and highlight its potential as a therapeutic target for delaying or preventing dementia onset.

3. Future directions: Further research should include comprehensive multi‐omics approaches that may enhance our understanding of the underlying causes of heterogeneity in the age at onset of AD.

2. METHODS

2.1. Study design, participants, and phenotype definitions

This study analyzed six families from the LLFS with at least two cases of EOAD (onset ≤ 65 years), drawn from a cohort of 3476 US participants. Validation analyses were conducted across independent cohorts, encompassing EOAD and LOAD cohorts (Estudio Familiar de Influencia Genetica en Alzheimer [EFIGA], Early‐Onset Alzheimer's Disease Whole‐Genome Sequencing Project [EOAD ADSP], National Alzheimer's Coordinating Center [NACC], Wisconsin Registry for Alzheimer's Prevention [WRAP], Alzheimer's Disease Neuroimaging Initiative [ADNI], The National Institute on Aging Alzheimer's Disease Family Based Study [NIA‐AD FBS], Washington Heights/Inwood Columbia Aging Project [WHICAP], and Religious Orders Study and Memory and Aging Project [ROSMAP]). Across all study cohorts, age at onset for AD cases and age at last clinical assessment for cognitively intact participants served as primary outcomes. The only exception was for ROSMAP, wherein brain samples were used, and age at death was modeled as the outcome.

The LLFS is a longitudinal, family‐based study designed to assess genetic and environmental risk factors associated with exceptional longevity. ²⁰ As previously described, AD status was determined primarily via dementia review committee as well as by a previously validated diagnostic algorithm. ¹⁸ Participants included in the current analyses included those diagnosed with probable AD with and without stroke. Detailed description of whole genome sequencing and processing was previously published. ¹⁸ We excluded Danish participants who lacked assessments of everyday function required for the clinical consensus dementia diagnosis. ¹⁷ A total of 3476 US participants were included.

As previously described, ²¹ the EFIGA study was designed to identify genetic variants increasing LOAD risk in Caribbean Hispanic individuals. EFIGA consists of LOAD families of ≥ 2 affected individuals recruited in New York and the Dominican Republic. The AD diagnosis is based on the National Institute of Neurological and Communicative Disorders and Stroke and the Alzheimer's Disease and Related Disorders Association (NINCDS‐ADRDA) criteria. ²² Non‐demented subjects were defined as individuals with no dementia, and subjects with mild cognitive impairment (MCI) were excluded from analyses. Details of genome‐wide association study (GWAS) data, quality control, and imputation procedures can be found elsewhere. ²³ , ²⁴ We used two independent EFIGA samples, the first one consisted of 505 individuals from 84 Puerto Rican families (32% are EOAD cases carriers of the PSEN1 Gly206Ala founder mutation). ²⁵ The second EFIGA sample (EFIGA‐DR) corresponds to 2451 members from 623 predominantly Dominican Republic families.

The NACC serves as a central repository for data collected from the Alzheimer's Disease Research Centers (ADRCs) in the United States. ²⁶ Clinical AD cases were defined as cases whose cognitive status was labeled as dementia with a primary etiologic diagnosis of AD dementia in the NACC Uniform Data Set. The analysis sample consisted of 1383 NACC participants.

WRAP is an ongoing longitudinal observational cohort study of non‐demented individuals aged 40 to 65 at baseline. Since 2001, WRAP has enrolled > 1700 individuals, 73% of whom had a parental history of probable AD dementia. ²⁷ AD cases status was determined with autopsy confirmed or probable AD as defined by the NINCDS‐ADRDA criteria. ²² A total of 625 WRAP participants were included.

ADNI is a public–private partnership to develop a multisite, longitudinal, prospective study of normal cognitive aging, MCI, and early AD to facilitate the scientific evaluation of neuroimaging and other biomarkers for the disease's onset and progression. ²⁸ Detailed information can be found at www.adni‐info.org. The subjects classified as AD had to meet the NINCDS‐ADRDA criteria. ²² A subsample of 266 ADNI participants were included.

The EOAD ADSP is a collaborative initiative to generate whole‐genome sequence data with extensive harmonized clinical, neuropathological, and biomarker data of EOAD cases of diverse ancestries. As previously described, ²⁹ , ³⁰ the project used genetic data from > 70,000 (non‐Hispanic White, African American, and East Asian) individuals enrolled in the Alzheimer Disease Genetics Consortium (ADGC) and the Charles F. and Joanne Knight ADRC (Knight‐ADRC) for gene identification in EOAD. ³¹ The present analysis includes data from 19,668 non‐Hispanic White individuals.

NIA‐AD FBS is the largest worldwide collection of longitudinally assessed AD families. ³² AD cases were defined using NINCDS‐ADRDA criteria. ²² Non‐demented subjects were defined as any cognitively normal with no evidence of AD, and MCI subjects were excluded from analyses. Genome‐wide microarray data ³³ were imputed using the Haplotype Reference Consortium panel through the Michigan Imputation Server. ³⁴ The analysis sample consisted of 3773 LOAD participants.

Participants in WHICAP were drawn from a community‐based multiethnic cohort of Medicare beneficiaries in New York. ³⁵ Based on information obtained at the baseline and follow‐up visits, dementia diagnosis was made by consensus during diagnostic conferences. The dementia diagnosis was based on the Diagnostic and Statistical Manual of Mental Disorders ³⁶ as well NINCDS‐ADRDA criteria. ²² Generation of the GWAS data was previously described. ³⁷ A total of 2356 Caribbean‐Hispanic participants were included.

The ROSMAP study combines two publicly available datasets, the ROS and the MAP. Participants from the cohorts were cognitively normal upon enrollment, agreed to annual blood tests and cognitive evaluations, and to organ donation after death. Post mortem evaluations were performed to assess AD pathology (Consortium to Establish a Registry for Alzheimer's Disease and Braak staging). ³⁸ Detailed methodology of genome wide genotype data generation can be found elsewhere. ³⁹ A total of 864 brain samples were included.

2.2. Statistical analyses

Among the quality control filters applied to whole genome sequencing data were: minor allele frequency threshold of 5%, minimum minor allele count of 10, and linkage disequilibrium (LD) pruning with PLINK using a sliding window of 50 single nucleotide polymorphisms (SNPs), a step size of five variants, and an LD threshold of r ² ≥ 0.8. The final number of variants considered for analyses purposes was 2,896,971 SNPs.

As previously described, ⁴⁰ centiMorgan (cM) positions for the variants were obtaining using the deCODE sex‐specific SNP meiotic map. ⁴¹ Variants present on the map were assigned directly to their reported cM positions. For variants absent from the map, we estimated positions by linear interpolation based on their base pair (bp) coordinates relative to nearby mapped SNPs. We excluded a subset of SNPs showing discrepancies between bp order in GRCh38 and cM order in the deCODE map (reflecting differences in genome builds). To retain only informative variants for linkage analysis, within‐family variation was quantified for each SNP, and those with an average value > 0.1 were selected.

Multipoint linkage analysis was performed using SOLAR software. ⁴² To maximize statistical power, all available members (EOAD and LOAD cases, and non‐demented subjects) from the six selected EOAD families were included in the analysis. The analysis outcome (i.e., age) corresponds to the age at onset for AD cases and age at last evaluation for the non‐demented family members. Age was inverse transformed to fit normality.

Covariates included sex, field center, the three first principal components, genetic relationship matrix, and AD diagnosis. Among these, only AD diagnosis met the significance threshold and was subsequently retained in the final multipoint linkage model. A multipoint logarithm of the odds (LOD) score ≥ 3 was considered statistically significant. ⁴³ cM genetic distances were converted to physical positions (bps) using the Kosambi mapping function. ⁴⁴

Linkage regions yielding significant LOD scores were further evaluated using association analysis via GMMAT software. ⁴⁵ Linear regression models included age at onset as an outcome and were adjusted for AD diagnosis, sex, education, recruitment site, three first principal components, and kinship matrix. Cognitive outcomes included a previously described composite endophenotype, ⁴⁶ , ⁴⁷ and an episodic memory endophenotype. The composite endophenotype was derived from factor analysis of 28 traits across five health domains, including cognitive performance. ⁴⁷ The episodic memory domain score was calculated by standardizing raw test scores against the average of demographically adjusted scores in the non‐demented offspring of the LLFS probands. ⁴⁸

Validation analyses were carried out for RNA binding fox‐1 homolog 1 (RBFOX1) variants in nine independent EOAD (EFIGA Gly206Ala, ADSP, NACC, WRAP, and ADNI) and LOAD (EFIGA‐DR, NIA‐AD FBS, WHICAP, and ROSMAP) cohorts. All linear regression models included sex, AD diagnosis, genetic relationship matrix, and the three principal components as covariates. In the EFIGA familial cohort, the PSEN1 p. Gly206Ala mutation status was additionally modeled as a covariate. When testing cognitive outcomes, linear models were additionally adjusted for age at onset, and education. Pairwise measures of LD between top associated SNPs across datasets were computed using 1000 Genomes data within the LDlink web‐based tool (https://ldlink.nih.gov/?tab=home).

In the LLFS cohort, RNA extraction from blood samples was processed at the McDonnell Genome Institute at Washington University. RNA sequencing design, quality control, oversight, pipelines, and filters were done at the Division of Computational & Data Sciences at Washington University, using published detailed protocols. ⁴⁹ LLFS transcriptomic analyses were performed on two separate strata: EOAD families and LOAD families, that is, without early onset cases (familial or sporadic). Within each stratum, only affected individuals were included in the analysis. Generalized estimating equation linear models were used to assess the association between the categorized RNA expression levels and age at onset, adjusting for sex and study site. In the ROSMAP cohort, post mortem dorsolateral prefrontal cortex (DLPFC) samples were used for RNA extraction and sequencing. Specific protocols can be found in the AD Knowledge portal (https://adknowledgeportal.synapse.org/). Linear regression models were used to assess the association between the categorized RNA expression levels and age at death, adjusting for sex and AD diagnosis. In both cohorts (LLFS and ROSMAP), RNA expression levels were dichotomized into “high” and “low” categories based on whether they fell above or below the mean expression level (using the average and standard deviation of the RNA expression distribution).

3. RESULTS

Table 1 summarizes the demographic characteristics of the study cohorts. Across all cohorts, the majority of the participants were women. Mean age ranged from 60 years old in EFIGA families with the PSEN1 Gly206Ala mutation to 89 years old in the ROSMAP brain samples. The highest proportion of EOAD cases was observed in the PSEN1 EFIGA families (32%), while the rest of the cohorts showed very similar EOAD prevalence (≈ 16%). As expected for a familial cohort, the NIA‐AD FBS cohort had the largest number of LOAD (45%) individuals, followed by ROSMAP (38%).

TABLE 1.

Characteristics of the study cohorts.

Datasets	AD	Study	N	%	EOADcases			LOADcases
				women	N	%	Avg ± SD	N	%	Avg ± SD
Discovery	EOAD	LLFS	77	55	14	14	58 ± 5	20	19	86 ± 9
	LOAD	LLFS	2867	56				606	25	89 ± 8
Validation	EOAD	EFIGA	505	56	148	29	55 ± 7	57	11	75 ± 6
		ADSP	19,668	56	3182	52	60 ± 5	6282	52	64 ± 6
		NACC+WRAP+ADNI	2272	56	354	16	59 ± 5
	LOAD	EFIGA‐DR	2451	63	354	14	60 ± 5	487	20	77 ± 7
		NIA‐AD FBS	3773	62	238	6	62 ± 3	2082	55	72 ± 11
		WHICAP	3783	68				995	26	85 ± 7
		ROSMAP	864	66				595	69	90 ± 6

Notes: The AD column indicates the most predominant form of AD represented in the study; N and % correspond to number and percentage of cases (EOAD, LOAD); Avg and SD represent the average and standard deviation of the age at onset.

Abbreviations: AD, Alzheimer's disease; ADNI, Alzheimer's Disease Neuroimaging Initiative; ADSP, Alzheimer's Disease Sequencing Project; EFIGA‐DR, Estudio Familiar de Influencia Genetica en Alzheimer Dominican Republic; EOAD, early onset Alzheimer's disease; LLFS, Long‐Life Family Study; LOAD, late onset Alzheimer's disease; NACC, National Alzheimer's Coordinating Center; NIA‐AD FBS, The National Institute on Aging Alzheimer's Disease Family Based Study; ROSMAP, Religious Orders Study and Memory and Aging Project; SD, standard deviation; WHICAP, Washington Heights/Inwood Columbia Aging Project; WRAP, Wisconsin Registry for Alzheimer's Prevention.

Genome‐wide multipoint linkage analysis in LLFS EOAD families identified three regions with significant LOD scores on chromosomes 4, 7, and 16 (Figure 1; Figure S1 in supporting information). The strongest linkage was observed on chromosome 16, with the SNHG9‐RNF151 intergenic region yielding LOD of 5.04. A 7cM‐13cM linkage region comprised multiple significant LOD scores (strongest LOD = 4.41) corresponding to the RBFOX1 gene. On chromosome 4, the highest score (LOD = 3.89) was observed within an intergenic region encompassing an uncharacterized locus and pseudogene (LOC105374559‐RN7SL101P). On chromosome 7, a LOD score of 3.04 was detected at the sidekick cell adhesion molecule 1 (SDK1) gene. Full LOD score values under the significant peaks are provided in Table S1 in supporting information.

FIGURE 1.

Genome‐wide multipoint linkage analysis of age at onset in early onset families from the Long‐Life Family Study cohort. The Y axis represents LOD score values, with the red horizontal line indicating the genome‐wide significance threshold (LOD ≥ 3). The X axis indicates the physical positions (in base pairs) of the single nucleotide polymorphisms included in the analysis. Vertical lines demarcate chromosomal boundaries. bp, base pair; Chr, chromosome; LOD, logarithm of the odds

Table 2 presents the results of the SNP‐based association analysis for genes located under the significant linkage peaks identified in EOAD LLFS families. Further supporting RBFOX1, the intronic variant rs76959607 exhibited the strongest genetic effect on age at onset across all identified linkage peaks, (β = 40.07, standard error [SE] = 9.84, P = 4.7 × 10⁻⁵). Among LLFS members, carriers of the G allele in this variant exhibited an average delay of 5 years in age at onset (71 ± 2 vs. 66 ± 2, Figure 2A). Moreover, the association between age at onset and RBFOX1 was further supported by additional variants within the locus, which also showed significant associations (Table 3).

TABLE 2.

LLFS EOAD families SNP‐based association with age at onset analysis for genes within significant linkage peaks (LOD ≥ 3.0).

Gene	Chr	SNP	bp	A1	AF	N	β	SE	P
PCDH7	4	rs185949611	30,884,625	A	0.007	74	45.91	15.81	0.004
DTHD1	4	rs28714091	36,291,797	G	0.007	74	21.35	6.67	0.001
RELL1	4	rs3849019	37,653,768	T	0.367	75	−10.5	3.5	0.003
SDK1	7	rs140320232	3,362,401	C	0.064	70	45.76	15.81	0.004
SNRNP25	16	rs9936292	54,754	C	0.253	73	−2.73	3.26	0.402
IFT140	16	rs2294622	1,546,258	T	0.053	75	−14.88	5.44	0.006
OR2C1	16	rs11644228	3,338,386	T	0.225	71	−8.98	3.95	0.023
CREBBP	16	rs2054833048	3,851,341	T	0.007	68	33.59	14.1	0.017
ADCY9	16	rs7190663	3,961,222	T	0.149	74	12.18	4.2	0.004
SRL	16	rs11076825	4,203,991	A	0.273	75	−9.53	3.05	0.002
ANKS3	16	rs141421236	4,730,200	C	0.014	74	22.36	10.22	0.029
EEF2KMT	16	rs34385676	5,084,499	T	0.053	75	−6.68	5.6	0.233
RBFOX1	16	rs76959607	5,443,268	G	0.054	74	40.07	9.84	4.7 × 10−5
USP7	16	rs139628004	8,943,537	C	0.007	75	45.72	15.81	0.004

Notes: bp = base pair location based on GRCh38/hg38 reference genome; A1 = associated minor allele; AF = associated allele frequency; N = sample size; β = regression model beta coefficient.

Abbreviations: EOAD, early onset Alzheimer's disease; LLFS, Long‐Life Family Study; LOD, logarithm of the odds; SE, standard error; SNP, single nucleotide polymorphism.

FIGURE 2.

Boxplot of RBFOX1 top SNP associated with age at onset and cognitive phenotypes in the Long‐Life Family Study early onset Alzheimer's disease families: (A) association with age at onset; (B) association with cognitive endophenotype; (C) association with episodic memory. The Y axis corresponds to the age at onset; the X axis represents the SNPs’ genotypes. RBFOX1, RNA binding fox‐1 homolog 1; SNP, single nucleotide polymorphism

TABLE 3.

RBFOX1 top SNP associations with age at onset in LLFS EOAD families (P ≤ 5.0 × 10⁻⁴).

SNP	bp	A1	AF	N	β	SE	P
rs76959607	5,443,268	G	0.054	74	40.07	9.84	4.7 × 10−5
rs150211829	5,654,832	G	0.013	75	43.67	12.19	3.4 × 10−4
rs146102655	5,660,475	G	0.014	73	43.88	12.16	3.1 × 10−4
rs145791081	5,715,120	G	0.014	73	43.58	12.14	3.3 × 10−4
rs117369744	5,729,000	T	0.014	74	43.45	12.29	4.1 × 10−4
rs9934817	6,086,218	G	0.034	74	35.7	10.19	4.6 × 10−4
rs35886152	6,506,129	A	0.014	74	43.8	12.48	4.5 × 10−4
rs35090327	6,528,045	T	0.014	74	43.55	12.43	4.6 × 10−4

Notes: bp = base pair location based on GRCh38/hg38 reference genome; A1 = associated minor allele; AF = associated allele frequency; N = sample size; β = regression model beta coefficient.

Abbreviations: EOAD, early onset Alzheimer's disease; LLFS, Long‐Life Family Study; LOD, logarithm of the odds; SE, standard error; SNP, single nucleotide polymorphism.

To further assess the functional relevance of RBFOX1, we examined its association with cognitive performance endophenotype, and an episodic memory domain score within EOAD LLFS families. As shown in Table 4, an RBFOX1 variant appeared nominally associated with cognitive function (rs4787053, β = 1.66, SE = 0.48, P = 5.1 × 10⁻⁴). LLFS members carrying one or two G alleles (Figure 2B) demonstrated better cognitive performance (mean = 1.36 ± 0.17) compared to non‐carriers, (mean = −0.21 ± 0.05). A broader RBFOX1 effect was observed on episodic memory, with multiple SNPs showing significant associations. The strongest association was at rs17135053 (β = −2.63, SE = 0.71, P = 1.9 × 10⁻⁴). Individuals homozygous for the G allele performed significantly better (mean = −0.01 ± 0.05) compared to non‐carriers (mean = −2.48 ± 0.29, Figure 2C).

TABLE 4.

RBFOX1 top SNPs associated with cognitive outcomes in the LLFS EOAD families (P ≤ 10 × 10⁻⁴).

Outcome	SNP	bp	A1	AF	N	β	SE	P
CE	rs4787053	7,646,603	G	0.268	56	1.66	0.48	5.1 × 10−4
EM	rs79571980	5,491,087	G	0.008	63	−2.68	0.81	9.1 × 10−4
EM	rs75849590	5,567,446	A	0.008	63	−2.7	0.81	8.2 × 10−4
EM	rs74332379	5,580,326	C	0.008	64	−2.68	0.81	8.2 × 10−4
EM	rs111949269	5,620,985	T	0.008	62	−2.69	0.81	8.6 × 10−4
EM	rs7205638	5,635,814	A	0.008	64	−2.68	0.81	8.7 × 10−4
EM	rs72765144	5,684,684	T	0.008	62	−2.66	0.81	9.5 × 10−4
EM	rs11644707	5,704,757	T	0.008	64	−2.68	0.81	8.7 × 10−4
EM	rs145982118	5,752,029	T	0.008	61	−2.69	0.81	8.6 × 10−4
EM	rs17187180	5,757,894	A	0.008	62	−2.7	0.81	8.3 × 10−4
EM	rs137879397	5,768,739	G	0.008	63	−2.67	0.81	9.1 × 10−4
EM	rs117727430	5,769,887	C	0.008	63	−2.69	0.81	8.2 × 10−4
EM	rs140076130	5,789,711	A	0.008	62	−2.68	0.81	8.7 × 10−4
EM	rs149659393	5,953,964	T	0.008	61	−2.67	0.81	9.6 × 10−4
EM	rs190415463	6,002,070	T	0.008	62	−2.7	0.81	8.4 × 10−4
EM	rs111803384	6,159,366	T	0.008	62	−2.68	0.81	9.2 × 10−4
EM	rs118098404	7,304,092	C	0.016	63	2.14	0.63	6.2 × 10−4
EM	rs17143994	7,537,453	T	0.008	61	−2.71	0.81	7.8 × 10−4
EM	rs111655413	7,587,929	A	0.016	63	2.13	0.62	6.6 × 10−4
EM	rs77019581	7,611,129	C	0.017	60	2.13	0.63	6.7 × 10−4
EM	rs112638224	7,618,051	G	0.016	64	2.14	0.62	6.1 × 10−4
EM	rs77365627	7,654,850	G	0.033	60	−2.62	0.71	2.1 × 10−4
EM	rs17135053	7,661,894	T	0.031	64	−2.63	0.71	1.9 × 10−4
EM	rs4787062	7,687,003	A	0.008	64	−2.68	0.81	8.7 × 10−4

Notes: bp = base pair location based on GRCh38/hg38 reference genome; A1 = associated minor allele; AF = associated allele frequency; N = sample size; β = regression model beta coefficient.

Abbreviations: CE, cognitive endophenotype; EM, episodic memory; EOAD, early onset Alzheimer's disease; LLFS, Long‐Life Family Study; SE, standard error; SNP, single nucleotide polymorphism.

Validation analyses demonstrated strong evidence of an RBFOX1 association with age at onset, applicable to both EOAD and LOAD cohorts (Table 5). Table 5 also shows measures of LD between top SNPs within each cohort and the strongest SNP reported in LLFS, rs76959607 (P = 4.7 × 10⁻⁵). Given the low allele frequencies of some of the RBFOX1 variants, LD was assessed using D′ rather than r ², due to D′’s reduced sensitivity to allele frequency. ⁵⁰ , ⁵¹ The strongest association was seen in NIA‐AD FBS (rs9936825, β = 7.29, SE = 1.57, P = 3.4 × 10⁻⁶). Significance levels were similar across other cohorts, except for meta‐analysis of NACC, WRAP, and ADNI (P = 0.001). The most significant SNP‐based association results (P ≤ 10⁻⁴) within cohort are displayed in Tables S2–S8 in supporting information.

TABLE 5.

RBFOX1 top SNPs associated with age at onset across additional independent cohorts.

Cohort	SNP	bp	A1	AF	N	β	SE	P	D'
EFIGA Gly206Ala	rs140898228	6,151,541	T	0.008	505	−14.81	3.89	1.4 × 10−4	1.00
EFIGA‐DR	rs4786798	5,983,841	G	0.340	2451	1.28	0.29	1.1 × 10−5	0.29
ADSP‐EOAD	rs1493886	6,914,338	T	0.211	19,668	−0.13	0.03	1.8 × 10−5	0.30
NACC+WRAP+ADNI	rs113423126	6,574,716	T	0.087	2274	1.09	0.34	0.001	0.77
NIA‐AD FBS	rs9936825	7,682,716	A	0.005	3773	7.29	1.57	3.4 × 10−6	1.00
WHICAP	rs10163423	6,844,095	T	0.160	3783	1.18	0.29	5.2 × 10−5	0.58
ROSMAP	rs17760843	7,671,967	G	0.006	864	7.99	2.02	7.5 × 10−5	1.00

Notes: bp = base pair location based on GRCh38/hg38 reference genome; A1 = associated minor allele; AF = associated allele frequency; N = sample size; β = regression model beta coefficient; D'^h measures linkage disequilibrium between the corresponding variant and the top associated SNP in LLFS cohort (rs76959607, chr16:5443268).

Abbreviations: ADSP EOAD, Alzheimer's Disease Sequencing Project Early Onset Alzheimer's disease; EFIGA, Estudio Familiar de Influencia Genetica en Alzheimer; EFIGA‐DR, Estudio Familiar de Influencia Genetica en Alzheimer Dominican Republic; EOAD, early onset Alzheimer's disease; LLFS, Long‐Life Family Study; NACC, National Alzheimer's Coordinating Center; NIA‐AD FBS, The National Institute on Aging Alzheimer's Disease Family Based Study; ROSMAP, Religious Orders Study and Memory and Aging Project; SE, standard error; SNP, single nucleotide polymorphism; WHICAP, Washington Heights/Inwood Columbia Aging Project; WRAP, Wisconsin Registry for Alzheimer's Prevention.

The blood transcriptomic of EOAD LLFS families (Table 6A) revealed that overexpression of RBFOX1 was associated with a 2 year earlier age at disease onset (Table 6B, 72 ± 1 vs. 76 ± 1) in the EOAD cases strata (β = 4.39, SE = 0.09, P < 0.001). However, when gene expression was examined in LOAD families, no significant association was observed in any of the two considered strata (P > 0.050). In brain samples from ROSMAP (Table 6C), the expression pattern of RBFOX1 showed an inverse trend, for which later age at death (Table 6D, 90 ± 0) was nominally associated with its downregulation (β = −0.95, SE = 0.49, P = 0.050).

TABLE 6.

Association of RBFOX1 RNA expression levels with age at onset. Table 6a Comparison of blood gene expression in EOAD and LOAD samples from LLFS cohort.

RNA seq	EOAD cases (n = 18)			LOAD cases (n = 196)
	β	SE	P	β	SE	P
RBFOX1	4.39	0.09	<0.001	0.5	1.11	0.651

Note: β = regression model beta coefficient.

Abbreviations: EOAD, early onset Alzheimer's disease; LLFS, Long‐Life Family Study; LOAD, late onset Alzheimer's disease; SE, standard error; SNP, single nucleotide polymorphism.

TABLE 6b.

Differences in average age at onset by gene expression strata in blood samples from Long‐Life Family Study.

Blood transcriptomics	Age at onset
High RBFOX1 expression	72 ± 1
Low RBFOX1 expression	76 ± 1

Note: Age is expressed as average ± standard deviation.

TABLE 6c.

Gene expression in brain samples from Religious Orders Study and Memory and Aging Project cohort.

RNA seq	β	SE	P
RBFOX1	−0.95	0.49	0.050

Note: β = regression model beta coefficient.

Abbreviation: SE, standard error.

TABLE 6d.

Differences in average age at onset by gene expression strata in brain samples from Religious Orders Study and Memory and Aging Project.

Brain transcriptomics	Age at onset
High RBFOX1 expression	89 ± 0
Low RBFOX1 expression	90 ± 0

Note: Age is expressed as average ± standard deviation.

4. DISCUSSION

Our study identified a significant linkage between RBFOX1 locus and EOAD in long‐lived families, supported by consistent associations across multiple independent EAOD and LOAD cohorts. Variants within RBFOX1 appeared associated with cognitive function with a more pronounced effect on episodic memory. RBFOX1 overexpression in blood was associated with earlier disease, while brain downregulation showed a trend toward later disease onset. These findings support RBFOX1 as a contributor to EOAD pathogenesis and highlight its potential in future therapeutic strategies focusing on gene downregulation for delaying dementia onset.

The RBFOX1 gene encodes a neuron‐specific splicing factor regulating splicing networks implicated in neurodevelopmental and neuropsychological disorders. ⁵² , ⁵³ Using genome‐wide RNA sequencing in differentiating neural stem cells, Fogel et al. ⁵⁴ observed that RBFOX1 regulates a wide range of alternative splicing events implicated in neuronal development and maturation, including transcription factors, other splicing factors, and synaptic proteins. Their work highlighted RBFOX1 as a key factor in the development of human neurons. The gene's pleiotropic effect has been attributed to its control of splicing patterns and subsequent alterations in gene expression. ⁵² Previous studies have demonstrated that downregulation of RBFOX1 leads to the destabilization of mRNA encoding proteins involved in synaptic transmission, potentially contributing to the synaptic dysfunction observed in AD. ⁵⁵ Moreover, its role in alternative splicing of the amyloid precursor protein has linked RBFOX1 to AD. ⁵⁶ Further supporting its potential implication in AD pathology, RBFOX1 has been localized in dystrophic neurites and tau tangles. ⁵⁶ It has also been suggested to have an impact on dementia risk through vascular‐related pathways. ⁵⁷

Despite its significant neurological relevance, genomic studies of RBFOX1 remain limited. Previous studies have linked rare private deletions overlapping the RBFOX1 locus to EOAD. ⁵⁸ Genetic variation in RBFOX1 has been associated with increased brain amyloid burden in both preclinical and early stages of AD, and elevated AD risk among Black American populations. ⁵⁹ , ⁶⁰ , ⁶¹ , ⁶² Brain expression studies from a Mount Sinai cohort associated RBFOX1 with amyloid and tau pathology, ⁶³ while transcriptomics of post mortem brain tissue ⁶⁴ highlighted its role among splicing regulators implicated in AD susceptibility. Mouse models further support RBFOX1 as part of a dysregulated splicing network in AD brains. ⁶⁵

To our knowledge, this is the first human genetic study to report an association between RBFOX1 variants and age at onset of AD. Interestingly, in C. elegans models, ⁶⁶ , ⁶⁷ , ⁶⁸ the RBFOX1 homolog, host cell factor 1 (HCF‐1), has been shown to interact with SIR‐2.1 (the ortholog of mammalian SIRT1), forming a regulatory loop that modulates aging processes in both worms and mammals. These findings suggest that RBFOX1 may represent a novel factor influencing mammalian aging, and age‐related pathologies.

Publicly available RNA sequencing data from the Accelerating Medicines Partnership Alzheimer's Disease (AMP‐AD, https://agora.adknowledgeportal.org) consortium revealed that RBFOX1 is differentially expressed between AD cases and cognitively healthy controls across multiple brain regions. Moreover, when comparing differential expression of RBFOX1 to that of EOAD genes and apolipoprotein E (APOE), RBFOX1 demonstrates a balanced contribution from both genetic and multi‐omics data, highlighting its potential biological relevance in AD pathogenesis (Figure S2 in supporting information).

We observed contrasting patterns of RBFOX1 gene expression when comparing EOAD families with a LOAD cohort (ROSMAP). These RBFOX1 expression patterns might suggest distinct regulatory mechanisms. The gene overexpression observed in earlier onset EOAD families might indicate neurodevelopmental vulnerability. Conversely, in ROSMAP, an observed trend of RBFOX1 downregulation might reflect age‐related neurodegenerative processes characteristic of LOAD. It is also important to consider that these differences may be influenced by the sample source, that is, plasma in EOAD families versus post mortem brain tissue in ROSMAP. Gene expression in peripheral tissues does not always mirror gene expression in brain tissue, due to factors such as cell‐type specificity and post‐transcriptional regulation, among others.

Our study has some limitations. First, given the limited sample size of biomarker data in EOAD families, the identification of cases relied entirely on clinical diagnostic criteria. Future work must prioritize the integration of diagnostic biomarkers. Second, methodological differences across genotyping platforms used in the validation cohorts precluded the possibility of conducting a formal meta‐analysis. Nonetheless, we observed that RBFOX1‐associated variants across cohorts were in moderate to strong LD, reinforcing our results. Third, we did not observe a significant association between APOE and age at onset in the EOAD families. The lack of association of the ε4 allele with AD in LLFS is likely due to its significantly lower frequency compared to the non‐Hispanic White general population. ⁶⁹ This lower ε4 allele frequency could be explained by the detrimental effect that it has on longevity, as previously described in other centenarian studies. ⁷⁰ , ⁷¹ , ⁷² Fourth, comorbid conditions that may influence age at onset of AD were not accounted for, potentially confounding the observed associations. Fifth, we cannot rule out the contribution of environmental factors or gene–environment interactions to the variability in age at onset. Finally, further functional analyses are warranted to elucidate the mechanistic role of RBFOX1 in modulating disease onset.

In conclusion, our findings support a novel association between RBFOX1 and AD age at onset. Further research using comprehensive multi‐omics approaches may enhance our understanding of the underlying causes of heterogeneity in age at onset of AD. Identifying genes that influence age of disease onset, such as RBFOX1, could offer new therapeutic options for delaying or preventing onset of AD.

CONFLICT OF INTEREST STATEMENT

T.J. Hohman serves as scientific advisory board member for Vivid Genomics, as a consultant for Circular Genomics, as deputy editor for Alzheimer's & Dementia: TRCI, and as senior associate editor for Alzheimer's & Dementia. C. Cruchaga is a member of the scientific advisory board of Circular Genomics and owns stocks and is on the scientific advisory board of ADmit and Alamar; consults for Sanofi, NovoNordisk, and Owkin; and has received research support from GSK, Danaher, and EISAI. The remaining authors declare no competing financial interests.

CONSENT STATEMENT

Individual studies were approved by the institutional review boards at the respective universities and adhered to the Declaration of Helsinki. Informed written consent was obtained from all participants.

Supporting information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information