Exploratory Analysis of APOE ε4-Stratified Associations Between Plasma Alzheimer's Biomarkers and Retinal Imaging Metrics in Cognitively Normal Adults

Abstract

Purpose

Retinal imaging offers a noninvasive window into microvascular and neurodegenerative changes, yet its relationship to circulating plasma biomarkers of Alzheimer's disease (AD) remains incompletely understood. We evaluated associations between plasma AD biomarkers and retinal imaging metrics in cognitively normal adults and whether these differ by apolipoprotein E (APOE) ε4 status.

Methods

This cross-sectional analysis included cognitively normal participants prospectively enrolled from the Duke/UNC Alzheimer's Disease Research Center. Eighty-three eyes from 44 participants (20 APOE ε4 carriers, 24 noncarriers) underwent optical coherence tomography (OCT) and OCT angiography (OCTA) within 1 year of plasma sampling. Plasma AD biomarkers were assessed for associations with structural OCT and microvascular OCTA metrics. Linear mixed-effects models adjusted for age, sex, treated hypertension, and years of education were used, with interaction terms for APOE ε4 status followed by stratified analyses.

Results

Nominally significant APOE ε4–dependent interactions were observed between Aβ42/40 and neurofilament light chain (NfL) with ganglion cell–inner plexiform layer (GCIPL) thickness (uncorrected P = 0.026 and P = 0.028). For OCTA, multiple nominally significant APOE ε4–dependent interactions were identified between plasma biomarkers (NfL, pTau217/Aβ42, and pTau217) and perfusion and vessel density across macular regions (uncorrected P = 0.009–0.048). Stratified analyses suggested stronger or more positive relationships in noncarriers, with attenuated or inverse relationships in carriers. None of the associations remained significant after false discovery rate (FDR) correction.

Conclusions

In this exploratory analysis, the APOE ε4 genotype may modify associations between plasma AD biomarkers and retinal structural and microvascular metrics in cognitively normal adults.

Translational Relevance

These exploratory findings support further investigation of whether APOE ε4 genotype-status should be considered in future studies integrating retinal imaging and plasma biomarkers in preclinical Alzheimer's disease.

Introduction

Alzheimer's disease (AD) is a leading cause of cognitive decline worldwide, affecting approximately 57 million people globally in 2021, with prevalence expected to triple by 2050 as populations age.¹ Pathologic changes begin decades before the onset of clinical symptoms, underscoring the need for biomarkers that can detect early disease processes.² Apolipoprotein E (APOE) ε4 is the strongest genetic risk factor for sporadic AD and is associated with increased risk of progression from preclinical AD to symptomatic disease, identifying a particularly vulnerable population.³

The retina, as an extension of the central nervous system, offers a noninvasive approach to studying neurogenerative and microvascular changes associated with AD. Prior studies from our group and others have demonstrated retinal thinning and microvascular alterations associated with symptomatic AD and APOE ε4 positivity.^4–7 However, retinal imaging alone does not directly capture the underlying molecular pathology of AD, and findings have been variable, particularly in preclinical populations.⁸

In parallel, blood-based biomarkers, including plasma phosphorylated tau (pTau), amyloid-beta (Aβ) ratios, neurofilament light chain (NfL), and glial fibrillary acidic protein (GFAP), have emerged as accessible indicators of AD pathology. These biomarkers show strong concordance with established cerebrospinal fluid and positron emission tomography measures of AD pathology and have been recently incorporated into clinical guidelines for AD diagnosis.⁹ These advances raise the possibility that combining plasma biomarkers with retinal imaging could provide complementary insights into early AD-related neurodegenerative and microvascular changes.

Despite these advances, the relationship between circulating AD biomarkers and retinal imaging metrics remains incompletely defined, particularly in cognitively normal individuals. Moreover, given that APOE ε4 influences both AD pathology and early vascular function, it is unclear whether APOE carrier status modifies these plasma-retinal associations. To date, few studies have examined these relationships in cognitively normal, genotype-stratified populations. Accordingly, this study evaluated the relationship between plasma AD biomarkers and retinal structural and microvascular imaging metrics in cognitively normal older adults and tested whether these associations differ by APOE ε4 carrier status.

Methods

Study Design and Participants

This cross-sectional study was approved by the Duke Health Institutional Review Board (Pro00111831) and adhered to the tenets of the Declaration of Helsinki. Written informed consent was obtained from all study participants or a legally authorized representative prior to enrollment.

Participants were enrolled at the Duke/UNC Alzheimer's Disease Research Center (ADRC) between January 1, 2020, and January 1, 2026, and subsequently referred to the Eye Multimodal Imaging in Neurodegenerative Disease (iMIND) lab. Before referral to the iMIND lab, participants attended an ADRC study visit that included formal neurocognitive testing using the Uniform Data Set Version 3 or 4 (UDS3/UDS4) to assess executive function, language, and memory, in addition to the collection of a plasma sample. During the subsequent iMIND lab visits, participants underwent Mini-Mental State Examination (MMSE), visual acuity testing, and retinal imaging at the study visit. Inclusion criteria were cognitively normal status at the time of plasma collection, availability of high-quality retinal imaging, and known APOE genotype. Exclusion criteria included cognitive impairment, diabetes, glaucoma, or other vitreoretinal pathology. Participants with best-corrected visual acuity < 20/40 on the day of image acquisition or refractive error ± 6 diopters (D) or greater were also excluded.

Plasma Biomarker Sampling

Plasma samples were collected via venipuncture at the ADRC study visit and processed according to a protocol provided by the National Centralized Repository for Alzheimer's Disease and Related Dementias (NCRAD) at Indiana University. Plasma and buffy coat were aliquoted and sent to NCRAD per protocol. NCRAD performed APOE testing and biomarker testing including plasma pTau217, Aβ40, Aβ42, NfL, and GFAP. The tests were performed on the LUMIPULSE G1200 (Fujirebio, Tokyo, Japan) and Simoa HD-X Analyzer (Quanterix, Billerica, MA) platforms, and results were transmitted to the Duke/UNC ADRC.

Retinal Imaging Acquisition

Optical coherence tomography (OCT) and OCT angiography (OCTA) imaging was performed within 1 year of plasma biomarker sampling. OCT/OCTA imaging was done using the CIRRUS 5000 HD-OCT with AngioPlex (version 11.0.0.29946; Carl Zeiss Meditec, Dublin, CA). For OCT metrics, a 512 × 128 macular cube scan was used to measure central subfield thickness (CST) and ganglion cell–inner plexiform layer (GCIPL) thickness. The CST was defined as the thickness between the inner limiting membrane and the retinal pigment epithelium at the fovea. The GCIPL thickness was quantified over a 14.13-mm² elliptical annulus area centered on the fovea. A 200 × 200 optic disc cube scan was used to measure average retinal nerve fiber layer (RNFL) thickness, calculated over a 3.46-mm diameter circle centered on the optic disc. All measurements were automatically calculated via the AngioPlex software.

OCTA images were acquired using the same platform; 3 × 3-mm and 6 × 6-mm circular images of the superficial capillary plexus centered on the fovea were obtained. Perfusion density (PD) and vessel density (VD) were quantified using the AngioPlex software with the Early Treatment Diabetic Retinopathy Study (ETDRS) grid overlay for 3 × 3-mm and 6 × 6-mm circular and ringed regions of interest (ROIs). PD was defined as the total area of perfused vasculature per unit area in a ROI. VD was defined as the total length of perfused vasculature per unit area in a ROI. All OCT/OCTA images were carefully reviewed for image quality by trained graders. Images with poor signal strength (<7/10), poor centration, shadow or motion artifacts, or segmentation errors were excluded.

Statistical Analysis

Demographic characteristics were compared between APOE ε4 carriers and noncarriers using independent samples t-tests for continuous variables and χ² tests or Fisher's exact tests for categorical variables, as appropriate. Associations between plasma biomarkers and retinal imaging parameters were evaluated using linear mixed-effects models with a random intercept for participant to account for inter-eye correlation, fit by maximum likelihood with Satterthwaite-approximated degrees of freedom. Models were adjusted for age, sex, treated hypertension, and years of education. Interactions by APOE ε4 status were tested for plasma biomarker–retinal metric pairs. Each model included a two-way interaction term between the plasma biomarker and APOE ε4 status; a separate model was fit for each plasma biomarker–retinal metric pair. Models were checked for convergence and singular fit. Stratified analyses were performed by APOE ε4 status, with interpretation limited to those with significant interactions. Given the large number of interaction terms tested (n = 65), P values were corrected for multiple comparisons using the Benjamini–Hochberg false discovery rate (FDR) procedure; statistical significance was defined as q < 0.05. Both uncorrected P values and FDR-corrected q values are reported. All statistical analyses were performed using R 4.5.0 (R Foundation for Statistical Computing, Vienna, Austria).

Results

Study Cohort and Demographics

A total of 498 Duke/UNC ADRC participants were screened. Of those, 121 had retinal imaging and 107 of them had known APOE genotype. After restricting the cohort to cognitively normal individuals who met all eligibility criteria and had high-quality retinal imaging within 1 year of plasma biomarker sampling, 44 participants (83 eyes) were included in the final analysis (Fig. 1). Five eyes among these participants were excluded based on the image-quality criteria described above (signal strength < 7/10, poor centration, or segmentation error).

Figure 1.

Flow diagram of participant selection for analysis. AD, Alzheimer’s Disease; ADRC, Alzheimer’s Disease Research Center; APOE, apolipoprotein E; MCI, mild cognitive impairment.

The final cohort included 20 APOE ε4 carriers and 24 noncarriers, with a mean age of 64.0 ± 7.9 years; 68.2% of the cohort was female. Mean MMSE scores confirmed cognitively normal status in both carriers and noncarriers and did not differ significantly between groups. The mean interval between plasma biomarker sampling and retinal imaging acquisition was 164 ± 107 days and was similar between groups. Full demographic data are shown in Table 1.

Table 1.

Demographic Characteristics of Participants by APOE ε4 Carrier Status

Variable	Overall (n = 44)	Noncarriers (n = 24)	Carriers (n = 20)	P a
Age (y), mean ± SD	64.0 ± 7.9	65.7 ± 7.5	61.9 ± 8.1	0.12
Female sex, n (%)	30 (68.2)	16 (66.7)	14 (70)	1.0
Race, n (%)				0.18
White	40 (90.9)	20 (83.3)	20 (100)
African American	3 (6.8)	3 (12.5)	0 (0)
Hispanic/Latino	1 (2.3)	1 (4.2)	0 (0)
Treated hypertension, n (%)	6 (14.3)	5 (21.7)	1 (5.3)	0.28
MMSE, mean ± SD	28.93 (2.19)	29.42 (1.18)	28.28 (2.99)	0.10
Education (y), mean ± SD	17.32 ± 2.52	17.67 ± 3.16	16.90 ± 1.41	0.32
APOE genotype, n (%)				—
ε2/3	6 (13.6)	6 (25)	—
ε3/3	18 (40.9)	18 (75)	—
ε2/4	2 (4.5)	—	2 (10)
ε3/4	15 (34.1)	—	15 (75)
ε4/4	3 (6.8)	—	3 (15)
Days between plasma sampling and retinal imaging, mean ± SD	164 ± 107	158 ± 111	171 ± 104	0.70

MMSE, Mini-Mental State Examination; standard deviation.

^a P values were calculated using independent samples t-tests for continuous variables and χ² or Fisher's exact tests for categorical variables comparing carriers to noncarriers.

OCT/OCTA Parameters and Plasma Biomarkers

Mean OCT/OCTA parameters and plasma biomarker values by APOE ε4 status are summarized descriptively in Table 2. APOE ε4 carriers tended to have higher plasma pTau217 and pTau217/Aβ42 ratios along with lower Aβ42/40 ratios compared to noncarriers, consistent with expected AD-related plasma biomarker profiles. OCT metrics were broadly similar between groups; OCTA metrics showed a modest trend toward higher PD and VD in carriers across macular regions.

Table 2.

Mean Retinal Imaging and Plasma Biomarker Values by APOE ε4 Carrier Status

	Overall (n = 83 Eyes)	Noncarriers (n = 46 Eyes)	Carriers (n = 37 Eyes)
OCT metrics, mean ± SD
Macular CST (µm)	266.8 ± 26.7	266.7 ± 27.9	266.9 ± 25.6
GCIPL thickness (µm)	78.5 ± 8.4	76.8 ± 10.0	80.6 ± 5.4
RNFL thickness (µm)	90.8 ± 10.2	90.3 ± 12.2	91.3 ± 7.4
OCTA metrics, mean ± SD
3-mm circle PD	0.34 ± 0.04	0.33 ± 0.04	0.34 ± 0.04
3-mm ring PD	0.35 ± 0.04	0.35 ± 0.04	0.36 ± 0.04
3-mm circle VD (mm−1)	18.4 ± 2.3	17.8 ± 2.3	19.0 ± 2.3
3-mm ring VD (mm−1)	19.4 ± 2.3	18.8 ± 2.3	20.0 ± 2.3
6-mm circle PD	0.40 ± 0.05	0.39 ± 0.06	0.41 ± 0.04
6-mm outer ring PD	0.41 ± 0.05	0.40 ± 0.06	0.42 ± 0.05
6-mm inner ring PD	0.39 ± 0.06	0.37 ± 0.07	0.40 ± 0.05
6-mm circle VD (mm−1)	16.4 ± 2.0	16.0 ± 2.2	16.8 ± 1.7
6-mm outer ring VD (mm−1)	16.7 ± 2.0	16.4 ± 2.2	17.1 ± 1.8
6-mm inner ring VD (mm−1)	16.2 ± 2.5	15.6 ± 2.8	16.9 ± 1.8
	Overall (n = 44 Participants)	Noncarriers (n = 24 Participants)	Carriers (n = 20 Participants)
Plasma biomarkers
pTau217 (pg/mL), median (IQR)	0.309 (0.242–0.547)	0.269 (0.217–0.365)	0.345 (0.270–0.707)
Aβ42/40 ratio, median (IQR)	0.083 (0.073–0.091)	0.086 (0.078–0.092)	0.074 (0.070–0.084)
pTau217/Aβ42 ratio, median (IQR)	0.014 (0.010–0.025)	0.012 (0.008–0.016)	0.019 (0.013–0.036)
log_NfL (pg/mL), mean ± SD	2.27 ± 0.49	2.32 ± 0.48	2.20 ± 0.52
log_GFAP (pg/mL), mean ± SD	4.31 ± 0.64	4.29 ± 0.66	4.32 ± 0.63

Aβ42/40, amyloid beta 42/40 ratio; CST, central subfield thickness; GCIPL, ganglion cell-inner plexiform layer; GFAP, glial fibrillary acidic protein; NfL, neurofilament light chain; PD , perfusion density; pTau217, phosphorylated tau 217; RNFL, retinal nerve fiber layer; VD, vessel density.

Plasma NfL and GFAP values are log transformed.

Values are presented descriptively; no formal between-group comparisons were performed.

Plasma Biomarker–Structural OCT Associations

Nominally significant interactions by APOE ε4 status were observed between plasma Aβ42/40 and GCIPL thickness (uncorrected P = 0.026) and between plasma NfL and GCIPL thickness (uncorrected P = 0.028), although neither survived FDR correction (q = 0.21 for both) (Table 3).

Table 3.

Associations Between Plasma Biomarkers and OCT Structural Metrics With Interaction by APOE ε4 Carrier Status

Plasma Biomarker	OCT Metric	Noncarrier β (95% CI)	P	Carrier β (95% CI)	P	Interaction Pa	FDR qb
pTau217	Macular CST	38.5 (–38.4, 115.3)	0.31	11.3 (–20.6, 43.2)	0.47	0.71	0.78
Aβ42/40	Macular CST	−421.4 (–1672.5, 829.7)	0.49	−771.5 (–1542.0, –1.0)	0.050	0.50	0.62
pTau217/Aβ42	Macular CST	528.0 (–1116.9, 2172.9)	0.51	270.2 (–413.0, 953.4)	0.42	0.99	0.99
log_NfL	Macular CST	13.3 (–13.7, 40.2)	0.32	−5.7 (–29.9, 18.6)	0.63	0.86	0.90
log_GFAP	Macular CST	14.0 (–4.9, 32.8)	0.14	8.0 (–11.6, 27.5)	0.41	0.62	0.71
pTau217	GCIPL thickness	3.3 (–19.3, 26.0)	0.77	4.5 (–0.8, 9.8)	0.090	0.93	0.95
Aβ42/40	GCIPL thickness	−349.8 (–684.1, –15.5)	0.041	−104.1 (–245.5, 37.3)	0.14	0.026	0.21
pTau217/Aβ42	GCIPL thickness	98.9 (–379.3, 577.2)	0.67	124.5 (18.0, 231.0)	0.024	0.73	0.78
log_NfL	GCIPL thickness	6.3 (–1.1, 13.6)	0.092	−2.1 (–6.2, 2.1)	0.31	0.028	0.21
log_GFAP	GCIPL thickness	4.5 (–0.8, 9.9)	0.094	3.8 (0.9, 6.8)	0.014	0.48	0.62
pTau217	RNFL thickness	17.3 (–14.4, 49.0)	0.31	3.4 (–5.9, 12.7)	0.46	0.28	0.44
Aβ42/40	RNFL thickness	−76.6 (–639.4, 486.2)	0.78	−86.9 (–330.5, 156.7)	0.47	0.69	0.77
pTau217/Aβ42	RNFL thickness	383.4 (–303.1, 1070.0)	0.26	115.8 (–78.0, 309.7)	0.23	0.28	0.44
log_NfL	RNFL thickness	2.4 (–8.1, 12.8)	0.64	−1.2 (–8.2, 5.8)	0.73	0.59	0.71
log_GFAP	RNFL thickness	5.5 (–1.9, 12.8)	0.14	4.6 (–0.7, 9.9)	0.086	0.91	0.94

All linear mixed-effects models were adjusted for age, sex, treated hypertension, and years of education.

Noncarrier and carrier β coefficients and P values were derived from stratified analyses performed separately within each subgroup.

N = 44 participants (79 eyes) for models with macular CST or GCIPL thickness; n = 40 participants (73 eyes) for RNFL thickness.

All primary interaction models converged without warnings and showed no evidence of singular fit.

^aInteraction P values represent the significance of the plasma biomarker × APOE ε4 interaction term.

^bFDR q values reflect Benjamini–Hochberg correction of interaction P values across all plasma biomarker × retinal metric interaction tests.

In stratified analyses, higher Aβ42/40 was significantly associated with lower GCIPL thickness in noncarriers (β = –349.8; 95% confidence interval [CI], –684.1 to –15.5; P = 0.041), whereas this association was attenuated and not significant in carriers (β = –104.1; 95% CI, –245.5 to 37.3; P = 0.14). For NfL, associations with GCIPL differed in direction by genotype, with a positive trend in noncarriers (β = 6.3; 95% CI, –1.1 to 13.6; P = 0.092) and an inverse trend in carriers (β = –2.1; 95% CI, –6.2 to 2.1; P = 0.31).

A sensitivity analysis restricting the time frame of retinal imaging to within 6 months of plasma biomarker sampling showed a similar pattern (Supplemental Table S1). No nominally significant interactions were observed between plasma biomarkers and other OCT structural metrics (Table 3).

Plasma Biomarker–Microvascular OCTA Associations

Nominally significant interactions by APOE ε4 status were observed between plasma biomarkers and OCTA microvascular metrics, particularly for NfL, pTau217/Aβ42 ratio, and pTau217 across 3-mm and 6-mm macular regions (uncorrected P = 0.009–0.048) (Fig. 2), although none survived FDR correction (all q > 0.05) (Table 4).

Figure 2.

Heatmap of associations between plasma biomarkers and OCTA metrics with interaction by APOE ε4 carrier status. Heatmap displays uncorrected interaction P values from linear mixed-effects models evaluating associations between plasma biomarkers and OCTA parameters. Interaction effects by APOE ε4 carrier status are shown. The color scale represents statistical significance, with darker shades indicating lower uncorrected P values. None of the interactions shown retained significance after Benjamini–Hochberg FDR correction for multiple comparisons. Aβ42/40, amyloid beta 42/40 ratio; GFAP, glial fibrillary acidic protein; NfL, neurofilament light chain; PD, perfusion density; pTau217, phosphorylated tau 217; VD, vessel density.

Table 4.

Associations Between Plasma Biomarkers and OCTA Microvascular Metrics With Nominally Significant Interaction by APOE ε4 Carrier Status

Plasma Biomarker	OCTA Metric	Noncarrier β (95% CI)	P	Carrier β (95% CI)	P	Interaction Pa	FDR qb
pTau217	6-mm inner ring PD	0.18 (−0.01, 0.36)	0.056	−0.01 (−0.06, 0.04)	0.78	0.043	0.21
pTau217/Aβ42	6-mm circle PD	3.85 (0.73, 6.96)	0.018	0.22 (−1.05, 1.49)	0.72	0.037	0.21
pTau217/Aβ42	6-mm inner ring PD	4.43 (0.53, 8.32)	0.028	−0.05 (−1.11, 1.02)	0.93	0.020	0.21
pTau217/Aβ42	6-mm inner ring VD	170.7 (15.3, 326.1)	0.033	7.36 (−35.4, 50.1)	0.73	0.029	0.21
log_NfL	3-mm circle VD	1.08 (−0.53, 2.69)	0.18	−0.74 (−2.97, 1.49)	0.49	0.046	0.21
log_NfL	3-mm ring VD	0.96 (−0.65, 2.57)	0.23	−0.61 (−2.88, 1.67)	0.58	0.048	0.21
log_NfL	6-mm circle PD	0.03 (−0.02, 0.07)	0.28	−0.02 (−0.06, 0.02)	0.34	0.042	0.21
log_NfL	6-mm inner ring PD	0.05 (−0.01, 0.10)	0.096	−0.02 (−0.06, 0.01)	0.21	0.009	0.21
log_NfL	6-mm circle VD	0.99 (−0.88, 2.85)	0.28	−0.73 (−2.46, 1.00)	0.37	0.044	0.21
log_NfL	6-mm inner ring VD	1.75 (−0.41, 3.91)	0.11	−0.73 (−2.17, 0.71)	0.31	0.013	0.21

All linear mixed-effects models were adjusted for age, sex, treated hypertension, and years of education.

Noncarrier and carrier β coefficients and P values were derived from stratified analyses performed separately within each subgroup.

N = 41 participants (64 eyes) for models with 3-mm OCTA metrics; n = 38 participants (63 eyes) for 6-mm OCTA metrics.

All primary interaction models converged without warnings and showed no evidence of singular fit.

^aInteraction P values represent the significance of the plasma biomarker × APOE ε4 interaction term.

^bFDR q values reflect Benjamini–Hochberg correction of interaction P values across all plasma biomarker × retinal metric interaction tests.

In stratified analyses, a consistent pattern of differing directional associations was observed between groups. Among noncarriers, higher plasma biomarker levels were generally associated with increased perfusion and vessel density. For example, a higher pTau217/Aβ42 ratio was associated with higher 6-mm inner ring perfusion density (β = 4.43; 95% CI, 0.53–8.32; P = 0.028). In contrast, among APOE ε4 carriers, these associations were attenuated or reversed (e.g., pTau217/Aβ42 with 6-mm inner ring PD: β = –0.05; 95% CI, –1.11 to 1.02; P = 0.93), with similar trends observed across multiple OCTA parameters (Table 4). Because image quality and scan availability varied by modality and metric, the number of eyes included in individual OCT and OCTA models differed slightly by outcome; model-specific participant and eye counts are provided in the footnotes for Tables 3 and 4. Complete plasma biomarker–OCTA metric association data are provided in Supplemental Table S2.

Discussion

In this exploratory study of cognitively normal adults, we observed nominal associations between plasma AD biomarkers and OCT/OCTA retinal imaging metrics that differed by APOE ε4 carrier status, although none survived correction for multiple comparisons. To our knowledge, this is among the first studies to evaluate these relationships in a fully cognitively normal, genotype-stratified cohort.

For structural OCT metrics, nominally significant interactions were observed between Aβ42/40 and NfL with GCIPL thickness, although neither survived FDR correction. In stratified analyses, lower Aβ42/40, reflecting higher amyloid burden, was associated with thicker GCIPL in noncarriers, with an attenuated association in the same direction in carriers. Although this pattern may appear paradoxical given the well-established relationship between amyloid pathology and retinal ganglion cell loss in symptomatic AD,^10,11 a potential explanation consistent with emerging evidence is a biphasic retinal response to early amyloid accumulation, in which early amyloid deposition triggers transient thickening rather than thinning. Increased inner macular thickness has been reported in Aβ positron emission tomography (PET)–positive individuals with subjective cognitive decline.¹² Also, dynamic structural changes in the GCIPL and RNFL, including areas of thickening adjacent to areas of thinning, have been reported in mild cognitive impairment (MCI) and AD patients.¹³ Evidence for this model specific to APOE ε4 is mixed, with thicker macular volume reported in asymptomatic, high AD-risk carriers at midlife,¹⁴ whereas others have reported macular thinning in similar at-risk populations.¹⁵ At the cellular level, retinal ganglion cell loss in MCI and AD appears to be accompanied by soma hypertrophy in surviving cells, one possible cellular mechanism underlying compensatory swelling.¹⁶ The attenuation of this association in carriers may reflect additional APOE-mediated mechanisms beyond amyloid, such as impaired lipid metabolism, neuroinflammation, and/or direct vascular effects.^17,18 NfL showed a divergent directional trend by genotype, with a positive trend in noncarriers and an inverse trend in carriers. If replicated, this pattern would be consistent with findings from Ng et al.,¹⁹ who demonstrated that higher plasma NfL was associated with accelerated cognitive decline specifically in APOE ε4 carriers, and Benedet et al.,²⁰ who found that gray matter atrophy associated with plasma NfL in cognitively normal individuals was specific to APOE ε4 carriers. If replicated, these findings would suggest that retinal associations with NfL may differ by APOE genotype. One possible explanation for this genotype-dependent pattern is reduced capacity among APOE ε4 carriers to repair or compensate for neuronal damage; APOE ε4 has been shown to impair lipid recycling and myelin maintenance and is associated with greater neurodegeneration even in cognitively unimpaired adults.^17,21 These nominal findings are consistent with, but do not establish, a biphasic model of retinal changes in preclinical AD in which structural responses may differ by disease stage and genotype. Given the cross-sectional design of the study and the lack of significance after correction for multiple comparisons, this model remains speculative and requires longitudinal confirmation.

The most numerous nominal associations were observed in OCTA microvascular metrics, where multiple plasma biomarkers, including NfL, pTau217/Aβ42 ratio, and pTau217, showed APOE ε4–dependent interactions across several 3-mm and 6-mm macular regions, although none survived FDR correction. Stratified analyses suggested a consistent pattern of differing directional associations, with noncarriers showing positive relationships between AD-related plasma biomarkers and retinal PD/VD, whereas carriers demonstrated inverse or null associations. These exploratory observations warrant evaluation in larger, independent cohorts to determine whether the APOE genotype is associated with differing retinal microvascular associations with AD-related plasma biomarkers. One possible, although speculative, explanation for the positive associations in noncarriers is early compensatory neuroinflammatory or metabolic upregulation in response to subclinical or age-related AD pathology that increases retinal perfusion. Cognitively normal Aβ-positive individuals have been shown to exhibit higher cerebral blood flow than Aβ negative controls,²² and an “inverted-U pattern” of cerebral blood flow has been described across AD stages, with early hyperperfusion attributed to neurovascular dysregulation requiring higher perfusion to maintain cognitive function.²³ Whether a comparable pattern exists in the retina specifically among APOE ε4 carriers is unclear, with reports of both compensatory retinal blood flow enhancement and reduced retinal capillary density in cognitively normal carriers, suggesting that carriers’ microvascular response to early pathology may be variable or stage dependent rather than uniformly compensatory or impaired.^24,25 If our pattern in noncarriers reflects such a compensatory response in the retina, the negative or null associations in carriers may reflect early vascular damage that blunts this response. APOE ε4 has been linked to blood–brain barrier (BBB) breakdown through activation of the cyclophilin A–matrix metalloproteinase-9 (MMP-9) pathway and pericyte injury in cognitively unimpaired carriers, before amyloid and tau AD biomarkers emerge.²⁶ Such early BBB disruption has been observed to precede synaptic deficits and cognitive decline.²⁷ Such vascular dysfunction could plausibly blunt a compensatory hyperperfusion response in carriers, which may contribute to the attenuated or inverse associations observed in our cohort. Together with our structural OCT observations, these exploratory findings raise the possibility that APOE ε4 status may be associated with differing retinal structural and microvascular patterns to AD pathology through shared mechanisms involving disruption of cerebral signaling, neuroinflammation, and direct vascular effects. Longitudinal data are needed to test this hypothesis directly.

Prior studies have demonstrated retinal structural thinning and microvascular alterations in symptomatic AD patients and in APOE ε4 carriers.^4,5,7,28 However, only a select few have looked at the association between circulating plasma biomarkers and retinal imaging. Two recent studies have examined this plasma–retina relationship in symptomatic AD patients. In a cross-sectional study of 107 AD patients stratified by APOE ε4 genotype, Liu et al.²⁹ found that APOE ε4 homozygotes exhibited the most severe reductions in retinal microvascular density, with vascular dysfunction correlating with plasma biomarker abnormalities (lower Aβ42, higher pTau217/Aβ42) and worse cognitive scores. Similarly, Kwapong et al.³⁰ found that reduced microvascular density and less branching complexity correlated with plasma Aβ42, Aβ42/40, pTau181, and pTau217 in a cross-sectional study of 72 AD patients. Two other recent studies have evaluated this plasma–retina link in cognitively normal populations. In a pilot study of 11 cognitively healthy individuals at risk for AD (defined as plasma Aβ42/40 < 0.10), Abboud et al.³¹ found associations between retinal layer thickness and plasma NfL, GFAP, pTau181, and pTau217, with photoreceptor inner segment thinning showing the largest effect size with NfL. In the Atlas of Retinal Imaging in Alzheimer's Study (ARIAS) cohort of 82 cognitively unimpaired participants, Ravichandran et al.³² found a significant correlation between putative retinal gliosis and plasma pTau217. Our findings extend this literature by stratifying by genotype in cognitively normal individuals, suggesting that plasma-retina associations may not be uniform across all cognitively unimpaired patients but instead may be influenced by APOE genotype in this population. To our knowledge, this is among the first studies to explore the APOE genotype as a potential modifier of plasma biomarker–retinal imaging relationships, highlighting the importance of incorporating genetic context into biomarker interpretation.

From a translational perspective, these exploratory findings highlight the need to evaluate whether APOE genotype influences retinal imaging and plasma biomarker relationships in larger cohorts before genotype-informed interpretation can be considered clinically useful. An umbrella review of retinal biomarkers found that the most widely studied retinal parameters (peripapillary RNFL thickness, foveal avascular zone) achieved areas under the curve (AUCs) in the range of 0.70 to 0.73 for distinguishing AD from controls, described as “poor to moderate diagnostic accuracy.”³³ Plasma biomarkers alone also show good but not excellent accuracy, with plasma pTau217 alone achieving AUCs around 0.88 to 0.95, depending on the cohort, and combined plasma biomarker models (pTau217, cognitive tests, APOE) reaching AUCs of 0.90 to 0.91 for predicting AD progression.^34,35 Ravichandran et al.³² described an exciting finding where a multimodal model combining retinal and plasma biomarkers achieved an AUC of 0.97 for distinguishing preclinical AD from controls compared to unimodal models of retinal or plasma biomarkers, highlighting the potential of combining these two emerging diagnostic modalities. The differing nominal patterns observed between carriers and noncarriers in our study raise the possibility that a single, “one-size-fits-all” approach to retinal imaging and plasma biomarker interpretation may not capture genotype-related heterogeneity, although this remains speculative. Whether accounting for APOE genotype improves interpretation of combined retinal imaging and plasma biomarker data is an important question for future adequately powered studies to address.

Strengths of this study include prospective enrollment of a well-characterized cohort of cognitively normal patients who had comprehensive plasma biomarker sampling and multimodal retinal imaging with both OCT and OCTA. For retinal imaging metrics, we conducted rigorous quality control to minimize confounding or noise from poor quality images. The use of linear mixed-effects models allowed for appropriate accounting of inter-eye correlation. Conducting overall interaction testing followed by stratified analyses enabled identification of genotype-specific associations that may not have been apparent in pooled analyses or separate regression models by subgroup.

Limitations of this study include a relatively small sample size, which likely limited statistical power and may explain why none of the nominally significant associations survived correction for multiple comparisons. Given that no interaction remained significant post-FDR correction, findings should be considered hypothesis generating. The cross-sectional design of the study does not allow for determination of temporal relationships or causality; the compensatory mechanisms proposed above, although consistent with prior literature, are speculative interpretations of cross-sectional associations and require longitudinal confirmation. Other potential confounders associated with retinal vasculature and APOE ε4 biology, including smoking pack-years and lipid levels, were not available with sufficient granularity; residual confounding contributing to the observed associations cannot be excluded. Additionally, although retinal imaging was performed within 1 year of plasma sampling and the plasma-to-imaging interval was not significantly different between groups, variability in the timing between measurements may have nonetheless introduced some noise. Finally, this was a single-center study in a predominantly white, female population, and our findings should be validated in larger, more diverse cohorts.

Future studies that examine the longitudinal trajectories of plasma–retinal relationships could determine whether these APOE-dependent patterns predict progression along the AD continuum or subsequent cognitive decline. Integration with established biomarkers such as amyloid and tau PET imaging may further clarify the mechanisms underlying these associations.

Conclusions

The APOE ε4 genotype may be associated with differing plasma biomarker–retinal imaging relationships in cognitively normal adults. These exploratory findings suggest that retinal manifestations of systemic AD pathology may differ by genetic background and support further investigation of genotype-informed, multimodal approaches to early AD detection and risk stratification in larger, longitudinal cohorts.