Plasma Alzheimer's disease biomarker variability: Amyloid‐independent and amyloid‐dependent factors

Abstract

INTRODUCTION

We aimed to investigate which factors affect plasma biomarker levels via amyloid beta (Aβ)‐independent or Aβ‐dependent effects and improve the predictive performance of these biomarkers for Aβ positivity on positron emission tomography (PET).

METHODS

A total of 2935 participants underwent blood sampling for measurements of plasma Aβ42/40 ratio, phosphorylated tau 217 (p‐tau217; ALZpath), glial fibrillary acidic protein (GFAP), and neurofilament light chain (NfL) levels using single‐molecule array and Aβ PET. Laboratory findings were collected using a routine blood test battery.

RESULTS

Aβ‐independent factors included hemoglobin and estimated glomerular filtration rate (eGFR) for p‐tau217 and hemoglobin, eGFR, and triiodothyronine (T₃) for GFAP and NfL. Aβ‐dependent factors included apolipoprotein E genotypes, body mass index status for Aβ42/40, p‐tau217, GFAP, and NfL. However, these factors exhibited negligible or modest effects on Aβ positivity on PET.

DISCUSSION

Our findings highlight the importance of accurately interpreting plasma biomarkers for predicting Aβ uptake in real‐world settings.

Highlights

• We investigated factor–Alzheimer's disease plasma biomarker associations in a large Korean cohort.

• Hemoglobin and estimated glomerular filtration rate affect the biomarkers independently of brain amyloid beta (Aβ).

• Apolipoprotein E genotypes and body mass index status affect the biomarkers dependent on brain Aβ.

• Addition of Aβ‐independent factors shows negligible effect in predicting Aβ positivity.

• Adjusting for Aβ‐dependent factors shows a modest effect in predicting Aβ positivity.

1. BACKGROUND

Previous studies have shown that the plasma Alzheimer's disease (AD) biomarkers can accurately detect the amyloid beta (Aβ) burden in the brain. ¹ , ² , ³ , ⁴ Nonetheless, elucidating the factors that affect the levels of plasma AD biomarkers is crucial for their future interpretation in the clinical setting.

It is reasonable to classify certain comorbidities and factors affecting plasma biomarker levels into two different types based on their pathomechanisms: Aβ‐independent and Aβ‐dependent variabilities. Some of the comorbidities and factors may influence plasma biomarker levels through physiological processes that are independent of the Aβ burden in the brain (Aβ‐independent variability). For example, chronic kidney disease (CKD) and body mass index (BMI) may potentially influence plasma phosphorylated tau (p‐tau) or neurofilament light chain (NfL) by altering protein clearance rates and body blood volume, respectively. ⁵ , ⁶ , ⁷ , ⁸ , ⁹ In contrast, diminished renal function may affect brain atrophy ¹⁰ and BMI may be related to Aβ tracer uptake on positron emission tomography (PET; below Aβ uptake), ¹¹ indicating that these comorbidities and factors may also be related to the risk factors for AD pathologies (Aβ‐dependent variability). Furthermore, the predictive performances of plasma biomarkers for the Aβ burden in the brain may vary depending on the permutations and combinations of comorbidities and factors related to Aβ‐independent and Aβ‐dependent variabilities.

In the present study, we aimed to investigate which comorbidities and factors affect plasma biomarker levels, including the Aβ42/40 ratio; p‐tau181, ‐217, and ‐231; glial fibrillary acidic protein (GFAP); and NfL, with respect to Aβ uptake on PET. First, we evaluated whether certain comorbidities and factors affect plasma biomarker levels, irrespective of Aβ uptake. Second, we determined whether each of these factors influenced plasma biomarkers through both mechanisms (Aβ‐independent and Aβ‐dependent variabilities), and if so, to what extent each mechanism was influential through mediation analysis. Finally, we determined whether considering these comorbidities and factors would increase the predictive performance of plasma biomarkers for Aβ uptake on PET.

2. METHODS

2.1. Study population

Participants were recruited from the Korea Registries to Overcome and Accelerate Dementia research project, ¹² encompassing 25 hospitals, including Samsung Medical Center, which served as the core center between 2016 and 2023. This study enrolled 2935 individuals diagnosed with one of the following conditions: cognitively unimpaired, n = 681 (23.2%); mild cognitive impairment (MCI), n = 1451 (49.4%); dementia of the AD type (DAT), n = 613 (20.9%); or subcortical vascular cognitive impairment, n = 190 (6.5%). The detailed criteria for these diagnoses and general exclusion criteria are described in the Supplementary Methods in the Supporting Information. ¹³ , ¹⁴ , ¹⁵ , ¹⁶ , ¹⁷

2.2. Comorbidities and factors

All participants underwent APOE genotyping. APOE ε2 carriers were defined as individuals with the APOE ε2/ε2 or ε2/ε3 genotypes, and APOE ε4 carriers were defined as individuals with the APOE ε2/ε4, ε3/ε4, or ε4/ε4 genotypes. The BMI was collected from medical records. In this study, BMI status was stratified as follows: ¹⁸ , ¹⁹  < 18.5 kg/m², underweight; 18.5 to 24.9 kg/m², normal weight; and > 25 kg/m², obese. The status (presence/absence) of comorbidities, including hypertension, diabetes mellitus (DM), hyperlipidemia, ischemic stroke, coronary artery disease (CAD), and CKD, was obtained from the medical records and/or reliable informants.

RESEARCH IN CONTEXT

1. Systematic review: We searched PubMed for articles in English on “Alzheimer's disease (AD) plasma biomarker” and “comorbidity.” While several studies have investigated the effects of specific comorbidities and factors on AD plasma biomarkers, they are predominantly focused on non‐Hispanic White populations. Furthermore, it is not fully investigated if these effects relate to the amyloid beta (Aβ) burden in the brain or not.

2. Interpretation: Our study identified which comorbidities and factors affect the plasma biomarker levels and whether these effects are independent of or mediated by Aβ burden. Additionally, we found that adjusting for factors related to Aβ‐dependent variability as covariates in Aβ positivity prediction models resulted in modest performance improvement, while factors related to Aβ‐independent variability had insignificant impact.

3. Future directions: More studies are needed to further explore additional factors or approaches to enhance the predictive accuracy of plasma biomarkers in assessing Aβ burden.

The laboratory findings within 7 days from the date of plasma sampling were collected from the medical records of participants who underwent the routine blood test battery including the complete blood count, blood biochemistry, inflammatory markers, and thyroid function test. The specific items were hemoglobin, platelet, plasma glucose, low‐density lipoprotein cholesterol (LDL‐C), high‐density lipoprotein cholesterol (HDL‐C), aspartate transaminase (AST), alanine transaminase (ALT), total cholesterol, estimated glomerular filtration rate (eGFR), high‐sensitivity C‐reactive protein (hs‐CRP), erythrocyte sedimentation rate (ESR), triiodothyronine (T₃), free thyroxine, and thyroid stimulating hormone levels. The findings were acquired from three different laboratories, namely, Samsung Medical Center's clinical data warehouse, Seoul clinical laboratories, and EONE laboratory, which all meet the College of American Pathologists standards for accreditation. Therefore, the laboratory findings were z transformed within each laboratory.

2.3. Plasma collection and plasma biomarker assays

We obtained 8 mL of blood from each participant, placed the sample into a 0.5 M ethylenediaminetetraacetic acid–containing tube, and mixed it for 5 minutes. Plasma was extracted from the blood sample after centrifugation (1300 × g, 4°C) for 10 minutes and dispensed into 5 or 10 vials at a volume of 0.3 mL each. All plasma samples were stored frozen at −75°C until analysis. This process complied with the manual for human resource collection and registration of the National Biobank of the Republic of Korea.

The frozen plasma samples were shipped to the Department of Psychiatry and Neurochemistry, University of Gothenburg, for analysis. These samples were thawed on wet ice and centrifuged at 500 × g for 5 minutes at 4°C. Plasma Aβ40, Aβ42, GFAP, and NfL were quantified using the commercial Neurology 4‐Plex E kit (Quanterix). Plasma p‐tau181 and p‐tau231 were analyzed using in‐house single‐molecule array (Simoa) assays developed at the University of Gothenburg, ²⁰ , ²¹ while p‐tau217 was analyzed using the commercial ALZpath p‐tau217 assay (ALZpath Inc.).

2.4. Brain magnetic resonance imaging acquisition

All participants underwent brain magnetic resonance imaging (MRI) at the respective centers using a standardized common imaging protocol for 3‐dimensional (3D) T1 turbo field echo images and fluid‐attenuated inversion recovery using a 3.0‐T MRI scanner. T1‐weighted images were acquired using a standard isotropic voxel size of 1 mm³ on all MRI scanners. All images were centralized at the Samsung Medical Center. The median interval between Aβ PET imaging and plasma collection was 4 days (interquartile range, 0–69 days).

2.5. Aβ PET imaging acquisition and analysis

All participants underwent Aβ PET imaging with ¹⁸F‐florbetaben or ¹⁸F‐flutemetamol, according to the manufacturer's imaging guidelines. We then quantified Aβ uptake using the global MRI‐ or CT‐based regional direct comparison Centiloid (rdcCL) method. ²² Aβ positivity (+) on PET was defined using a global MRI‐based rdcCL threshold of 25.5. According to our previous study, the CT‐based rdcCL threshold of 25.1 was applied for the 28 cases in which 3D T1 MR images were lacking. ²² All imaging analyses were conducted at the Alzheimer's Disease Convergence Research Center at Samsung Medical Center. The detailed protocol for PET imaging, quantification, and obtaining Aβ (+) cutoff points is described in the Supplementary Methods.

2.6. Statistical analyses

The demographic and clinical characteristics were presented as means ± standard deviations for continuous variables and as numbers (percentages) for categorical variables.

The main analyses followed several steps. First, linear regression analyses that were to investigate the effects of comorbidities and factors on the plasma biomarkers after adjusting for age and sex (model 1) and further for age, sex, and Aβ uptake on PET (global rdcCL) as covariates (model 2). Factors related to Aβ‐independent variability would be significant in both model 1 and model 2, while factors related to Aβ‐dependent variability would show significance in model 1 but lose significance in model 2. Second, to select predictors for subsequent mediation analyses, linear regression analyses of comorbidities and factors on Aβ uptake on PET after controlling for age and sex were conducted (model 3). We selected predictors that showed significant association in both model 1 and model 3. Third, we performed mediation analyses to determine whether Aβ uptake on PET (mediator) mediated the association between the comorbidities and factors (predictors) and plasma biomarkers (outcomes), controlling for age and sex. Bootstrapping was used to verify the significance of indirect effects. Finally, to investigate whether the factors related to Aβ‐independent and Aβ‐dependent variability affect performance of plasma biomarkers for predicting Aβ (+) on PET, we performed logistic regressions, followed by receiver operating characteristic (ROC) analysis, after controlling for age and sex (model A), and further considering the factors related to Aβ‐independent variability (model B) and Aβ‐dependent variability to the covariates of model A (model C). The areas under the curves (AUCs) of multiple models were compared using the DeLong method. ²³

For the main analyses, the plasma biomarker values were log‐transformed and values beyond three standard deviations were removed (39 values for the Aβ42/40 ratio, 4 for p‐tau217, 14 for GFAP, and 32 for NfL were removed). For laboratory findings, we did not specifically remove outliers because all values were within the medically possible ranges. All reported P values were two‐tailed, and statistical significance was determined using a threshold of P < 0.05 after Bonferroni correction for multiple comparisons. All analyses were performed using R version 4.2.3 (The R Foundation for Statistical Computing) and Mplus version 8.1 (Muthén & Muthén).

3. RESULTS

3.1. Baseline characteristics of the study population

The detailed demographic and clinical characteristics of the 2935 participants are presented in Table 1. The participants’ mean age was 71.6 ± 8.7 years, 63.6% were women, and 38.4% were APOE ε4 carriers. The associations of age and sex with the plasma biomarker levels are described in Figures S1 and S2 in the Supporting Information.

TABLE 1.

Baseline characteristics of the study population.

	Total participants (N = 2935)	CU (N = 681)	MCI (N = 1451)	DAT (N = 613)	SVCI (N = 190)
Demographics
Age, years	71.6 ± 8.7	70.0 ± 8.2	72.2 ± 8.3	70.5 ± 9.7	76.1 ± 8.1
Sex, female	1867 (63.6%)	434 (63.7%)	900 (62.0%)	398 (64.9%)	135 (71.1%)
Education, years	10.7 ± 4.9	11.5 ± 4.7	10.5 ± 4.8	10.6 ± 4.8	8.9 ± 5.5
APOE genotype a (N = 2926)
ε2 carrier/ε3ε3/ε4 carrier	222 (7.6%)/1579 (54.0%)/1125 (38.4%)	71 (10.4%)/435 (64.2%)/172 (25.4%)	105 (7.3%)/747 (51.6%)/595 (41.1%)	31 (5.0%)/269 (44.0%)/311 (50.9%)	15 (7.9%)/128 (67.4%)/47 (24.7%)
BMI status b (N = 2914)
Underweight/normal weight/obesity	110 (3.7%)/1919 (65.4%)/885 (30.2%)	11 (1.6%)/422 (62.8%)/239 (35.6%)	59 (4.1%)/974 (67.3%)/415 (28.7%)	36 (5.9%)/415 (68.4%)/156 (25.7%)	4 (2.1%)/108 (57.8%)/75 (40.1%)
Comorbidities
Hypertension	1466 (49.9%)	304 (44.6%)	757 (52.2%)	263 (42.9%)	142 (74.7%)
DM	715 (24.4%)	144 (21.1%)	379 (26.1%)	128 (20.9%)	64 (33.7%)
Hyperlipidemia	1374 (46.8%)	358(52.6%)	690 (47.6%)	243 (39.6%)	83 (43.7%)
Stroke	302 (10.3%)	61 (9.0%)	123 (8.5%)	48 (7.8%)	70 (36.8%)
Coronary artery disease	334 (11.4%)	94 (13.8%)	176 (12.1%)	35 (5.7%)	29 (15.3%)
CKD	76 (2.6%)	18 (2.6%)	39 (2.7%)	10 (1.6%)	9 (4.7%)
Laboratory findings
Hemoglobin, g/dL (N = 1646)	13.3 ± 1.4	13.5 ± 1.3	13.3 ± 1.4	13.2 ± 1.3	13.0 ± 1.4
Platelet, 1000/µL (N = 1646)	228.0 ± 61.0	224.5 ± 62.5	229.4 ± 60.5	229.3 ± 61.6	224.8 ± 58.1
Glucose, plasma, mg/dL (N = 1665)	107.7 ± 31.0	103.1 ± 21.9	107.5 ± 27.4	110.3 ± 44.2	116.7 ± 33.1
LDL‐C, mg/dL (N = 1570)	104.1 ± 33.2	103.3 ± 31.2	102.6 ± 33.2	109.9 ± 34.6	101.4 ± 34.9
HDL‐C, mg/dL (N = 1572)	57.0 ± 14.3	57.4 ± 13.9	56.8 ± 14.0	57.6 ± 15.4	54.8 ± 14.9
Total cholesterol, mg/dL (N = 1667)	176.7 ± 39.9	176.5 ± 39.6	176.7 ± 39.9	180.6 ± 39.9	164.0 ± 39.0
AST, U/L (N = 1663)	26.6 ± 11.4	27.3 ± 13.6	26.7 ± 10.5	25.8 ± 11.4	25.0 ± 9.4
ALT, U/L (N = 1662)	22.3 ± 17.2	23.7 ± 25.4	22.0 ± 12.7	21.7 ± 15.9	21.8 ± 14.6
eGFR, mL/min (N = 1660)	80.8 ± 18.3	83.2 ± 16.4	79.8 ± 18.3	81.9 ± 19.2	76.3 ± 21.2
hs‐CRP, mg/L (N = 1586)	2.2 ± 10.5	1.6 ± 6.7	2.0 ± 10.6	2.5 ± 12.2	5.8 ± 14.7
ESR, mm/hr (N = 247)	17.4 ± 16.5	17.3 ± 18.5	15.7 ± 14.4	18.0 ± 17.2	23.8 ± 18.0
T3, ng/dL (N = 1358)	103.7 ± 20.5	106.3 ± 19.0	103.1 ± 20.7	101.9 ± 21.6	103.4 ± 21.0
Free T4, ng/dL (N = 1286)	1.2 ± 0.5	1.2 ± 0.7	1.2 ± 0.4	1.2 ± 0.2	1.2 ± 0.2
TSH, µIU/mL (N = 1530)	2.6 ± 5.2	2.5 ± 2.1	2.7 ± 5.7	2.3 ± 1.9	3.7 ± 13.5

Note: Unless otherwise noted, values are expressed as N (%) or mean ± standard deviation.

Abbreviations: ALT, alanine transaminase; AST, aspartate transaminase; Aβ, amyloid beta; BMI, body mass index; CKD, chronic kidney disease; CU, cognitively unimpaired; DAT, dementia of the Alzheimer's disease type; DM, diabetes mellitus; eGFR, estimated glomerular filtration rate; ESR, erythrocyte sedimentation rate; HDL‐C, high‐density lipoprotein cholesterol; hs‐CRP, high‐sensitivity C‐reactive protein; LDL‐C, low‐density lipoprotein cholesterol; MCI, mild cognitive impairment; SVCI, subcortical vascular cognitive impairment; T₃, triiodothyronine; T₄, thyroxine; TSH, thyroid stimulating hormone.

^a APOE genotype ε2 carriers were defined as individuals with the APOE ε2/ε2 or ε2/ε3 genotypes. APOE genotype ε4 carriers were defined as individuals with the APOE ε2/ε4, ε3/ε4, or ε4/ε4 genotypes.

^b Underweight was defined as BMI < 18.5 kg/m², normal weight as 18.5 kg/m² ≤ BMI < 25 kg/m², and obesity as BMI ≥ 25 kg/m².

3.2. Effects of comorbidities and factors on plasma biomarkers

Figure 1 and Table S1 show the effects of comorbidities and factors on the plasma biomarkers (Aβ42/40 ratio, p‐tau217, GFAP, and NfL). A lower Aβ42/40 ratio was associated with APOE ε4 carrier status and LDL‐C, whereas a higher Aβ42/40 ratio was associated with obesity (p < 0.05, model 1). The relationships of APOE ε4 carrier status with the Aβ42/40 ratio remained significant after controlling for Aβ uptake on PET (model 2).

FIGURE 1.

Forest plots for comorbidities and factors as predictors. This figure shows the result of linear regression for each plasma biomarker and Aβ uptake on PET using comorbidities and biological factors as predictors. APOE ε2 carriers were defined as individuals with the APOE ε2/ε2 or ε2/ε3 genotypes. APOE ε4 carriers were defined as individuals with the APOE ε2/ε4, ε3/ε4, or ε4/ε4 genotypes. Underweight was defined as BMI < 18.5 kg/m², normal weight as 18.5 kg/m² ≤ BMI < 25 kg/m², and obesity as BMI ≥ 25 kg/m². The laboratory findings that were measured by three different laboratories were z transformed within each laboratory, and the plasma biomarker levels were log‐transformed. *p < 0.05 after Bonferroni correction. Aβ, amyloid beta; ALT, alanine transaminase; AST, aspartate transaminase; BMI, body mass index; CAD, coronary artery disease; CKD, chronic kidney disease; DM, diabetes mellitus; eGFR, estimated glomerular filtration rate; ESR, erythrocyte sedimentation rate; GFAP, glial fibrillary acidic protein; Hb, hemoglobin; HDL‐C, high‐density lipoprotein cholesterol; HL, hyperlipidemia; hs‐CRP, high‐sensitivity C‐reactive protein; HTN, hypertension; LDL‐C, low‐density lipoprotein cholesterol; NfL, neurofilament light chain; O, obesity; PET, positron emission tomography; p‐tau217, phosphorylated tau 217; rdcCL, regional direct comparison Centiloid; T₃, triiodothyronine; T₄, thyroxine; TC, total cholesterol; TSH, thyroid stimulating hormone; UW, underweight

Higher p‐tau217 levels were associated with APOE ε4 carrier status, being underweight, and the presence of CKD, whereas lower p‐tau217 levels were associated with APOE ε2 carrier status, obesity, the presence of hyperlipidemia, CAD, elevated hemoglobin, ALT, and eGFR levels (p < 0.05, model 1). The relationships of APOE ε4 carrier status, being underweight, obesity, the presence of CKD, level of hemoglobin, and eGFR retained significance after controlling for Aβ uptake on PET (model 2).

Higher GFAP levels were associated with APOE ε4 carrier status, being underweight, increased HDL‐C, and total cholesterol, whereas lower GFAP levels were associated with obesity, the presence of DM, hyperlipidemia, CAD, increased level of hemoglobin, plasma glucose, ALT, eGFR, and T₃ (p < 0.05, model 1). The relationships of BMI status, level of hemoglobin, plasma glucose, eGFR, and T₃ with GFAP remained significant after controlling for Aβ uptake on PET (model 2).

A higher NfL level was associated with being underweight, and the presence of DM and CKD, whereas a lower NfL level was associated with obesity, the presence of hyperlipidemia, hemoglobin, eGFR, and T₃ (p < 0.05, model 1). The relationships between NfL and all comorbidities, except being underweight and the presence of hyperlipidemia, remained significant after controlling for Aβ uptake (model 2).

In addition to the analyses for the above‐mentioned plasma biomarkers, the results of linear regression analyses for p‐tau181 and p‐tau231 are presented in Figure S3 and Table S2 in the Supporting Information.

For sensitivity analyses, we performed linear regression analyses after controlling for additional covariates or subgroups. Even after additional control for APOE genotypes and the presence of CKD, the statistical significance of factors with Aβ‐independent variability remained largely unchanged (Figure S4 in the Supporting Information). Among factors with Aβ‐dependent variability, the statistical significance of the effects of obesity on the Aβ42/40 ratio disappeared (Figure S4). We also performed linear regression analyses after additionally controlling for laboratory sites. We found that the statistical significance of the predictors remained largely unchanged (Figure S5 in the Supporting Information). To determine whether the association between comorbidities and plasma biomarker levels in DAT could be derived from a simple proxy, we performed linear regression analyses in DAT participants. The statistical significance of the factors remained largely unchanged (Figure S6 in the Supporting Information).

3.3. Relationships among the factors, Aβ uptake, and plasma biomarkers

Predictors that showed significant association after Bonferroni correction in both model 1 and model 3 were selected for mediation analyses. The findings of model 3 are provided in Table S3 in the Supporting Information.

Figure 2 shows the results of mediation analyses examining the relationships among various factors (predictors), Aβ uptake (mediators), and plasma biomarkers (outcome). Aβ uptake completely meditated the relationship between Aβ42/40 ratio and obesity; the respective relationship between p‐tau217 and APOE ε2 carrier status and the presence of CAD; and the respective relationship between GFAP and APOE ε4 carrier status, the presence of DM, and CAD. Aβ uptake partially meditated the relationship between the Aβ42/40 ratio and APOE ε4 carrier status; the respective relationships of p‐tau217 with APOE ε4 carrier status, being underweight, and obesity; the respective relationships of GFAP with being underweight and obesity; and the respective relationships of NfL with APOE ε4 carrier status, being underweight, obesity, and the presence of DM. The mediation percentages for the models with partial mediation are provided in Table S4 in the Supporting Information.

FIGURE 2.

Effect of comorbidities and factors on plasma biomarkers with respect to the Aβ burden. This figure shows the results of mediation analyses for investigating whether each biological factor affected plasma biomarkers in relation to Aβ burden in the brain. The dashed lines indicate associations that were statistically insignificant. The β (standard error) value for each association is written on the line. APOE ε2 carriers were defined as individuals with the APOE ε2/ε2 or ε2/ε3 genotypes. APOE ε4 carriers were defined as individuals with the APOE ε2/ε4, ε3/ε4, or ε4/ε4 genotypes. Underweight was defined as BMI < 18.5 kg/m², normal weight as 18.5 kg/m² ≤ BMI < 25 kg/m², and obesity as BMI ≥ 25 kg/m². *p < 0.05; **p < 0.01; ***p < 0.001. Aβ, amyloid beta; ALT, alanine transaminase; APET rdcCL, amyloid positron emission tomography regional direct comparison Centiloid; BMI, body mass index; CAD, coronary artery disease; DM, diabetes mellitus; ε2, APOE ε2 carrier; ε3, APOE ε3/ε3 homozygote; ε4, APOE ε4 carrier; GFAP, glial fibrillary acidic protein; NfL, neurofilament light chain; NW, normal weight; O, obesity; p‐tau217, phosphorylated tau 217; TC, total cholesterol; UW, underweight

3.4. Predictive performance of combined plasma biomarkers and the factors for Aβ (+) on PET

Table 2 and Figure 3 depict the comparison of the predictive performance of plasma biomarkers for Aβ (+) on PET. The AUC of model B that was adjusted with the factors related to Aβ‐independent variability was improved for p‐tau217 and GFAP compared to model A (AUC 0.950–0.959 for p‐tau217, AUC 0.837–0.848 for GFAP, and AUC 0.657–0.669 for NfL). By contrast, adding the factors related to Aβ‐dependent variability to the covariates significantly increased the predictive performances of all plasma biomarkers (AUC 0.836–0.870 for the Aβ42/40 ratio, AUC 0.950–0.958 for p‐tau217, AUC 0.837–0.885 for GFAP, AUC 0.657–0.796 for NfL).

TABLE 2.

Comparison of the performance of prediction models with plasma biomarkers for Aβ positivity.

	Model A		Model B (with factors of Aβ‐independent variability)			Model C (with factors of Aβ‐dependent variability)
	Covariates	AUC	Covariates	AUC	p	Covariates	AUC	p
Aβ42/40 ratio	Age + sex	0.836	Not applicable	–	–	Age + sex + APOE ε4 carrier + obesity	0.870	<0.001
p‐tau217	Age + sex	0.950	Age + sex + CKD + Hb + eGFR	0.959	<0.001	Age + sex + APOE genotype + BMI status + CAD	0.958	<0.001
GFAP	Age + sex	0.837	Age + sex + Hb + glucose + eGFR + T3	0.848	0.004	Age + sex + APOE ε4 carrier + BMI status  + DM + CAD	0.885	<0.001
NfL	Age + sex	0.657	Age + sex + CKD + Hb + eGFR + T3	0.669	0.082	Age + sex + APOE ε4 carrier + BMI status + DM	0.796	<0.001

Note: APOE ε4 carriers were defined as individuals with the APOE ε2/ε4, ε3/ε4, or ε4/ε4 genotypes.

Abbreviations: Aβ, amyloid beta; APOE, apolipoprotein E; BMI, body mass index; CAD, coronary artery disease; CKD, chronic kidney disease; DM, Diabetes mellitus; eGFR, estimated glomerular filtration rate; GFAP, glial fibrillary acidic protein; Hb, hemoglobin; HDL‐C, high‐density lipoprotein cholesterol; hs‐CRP, high sensitivity C‐reactive protein; NfL, neurofilament light chain; p‐tau217, phosphorylated tau 217.

FIGURE 3.

Improvement in performance of amyloid positivity prediction models by adding factors as covariates. The AUCs represent the performance of each plasma biomarker predicting Aβ (+) on PET. Model A included each plasma biomarker, age, and sex. Model B added factors related to Aβ‐independent variability to the covariates of model A (the presence of CKD, hemoglobin, and eGFR for p‐tau217; hemoglobin, plasma glucose, eGFR, and T₃ for GFAP; the presence of CKD, hemoglobin, eGFR, and T₃ for NfL). Model C added factors related to Aβ‐dependent variability to the covariates of model A (APOE ε4 carrier status and obesity for Aβ42/40 ratio; APOE genotypes, BMI status, and the presence of CAD for p‐tau217; APOE ε4 carrier status, BMI status, the presence of DM and CAD for GFAP; APOE ε4 carrier status, BMI status, and the presence of DM for NfL). Aβ, amyloid beta; AUC, area under the curve; BMI, body mass index; CAD, coronary artery disease; CKD, chronic kidney disease; DM, diabetes mellitus; eGFR, estimated glomerular filtration rate; GFAP, glial fibrillary acidic protein; HDL‐C, high‐density lipoprotein cholesterol; LDL‐C, low‐density lipoprotein cholesterol; NfL, neurofilament light chain; PET, positron emission tomography; p‐tau217, phosphorylated tau 217; T₃, triiodothyronine

4. DISCUSSION

In the present study, we investigated whether certain comorbidities and factors affected plasma biomarker levels (Aβ42/40 ratio, p‐tau217, GFAP, and NfL) with respect to Aβ uptake on PET in a large Korean cohort. We found that specific comorbidities and factors affected plasma biomarker levels either independent of the Aβ burden in the brain (Aβ‐independent variability) or by affecting the Aβ burden (Aβ‐dependent variability). However, incorporating these comorbidities and factors as covariates in the models with plasma biomarkers to predict Aβ uptake in the brain resulted in a negligible to modest improvement in their performance. Therefore, our findings suggested that while certain biological factors and comorbidities influence plasma biomarker levels related to Aβ uptake, their inclusion as covariates offers only a negligible to modest enhancement in predictive performance.

Our first major finding was that several comorbidities and factors affect plasma biomarkers, independent of the Aβ burden in the brain. The presence of CKD, hemoglobin, and eGFR affected p‐tau217; hemoglobin, plasma glucose, eGFR, and T₃ affected GFAP; and the presence of CKD, hemoglobin, eGFR, and T₃ affected NfL. Our findings on CKD and eGFR are consistent with those of previous studies; that is, the presence of CKD or lower eGFR levels are associated with higher plasma biomarker levels, which may be related to their reduced clearance. ⁶ , ²⁴ , ²⁵ Our novel finding was that hemoglobin and T₃ affected plasma biomarker levels independent of the Aβ burden in the brain. In particular, the effects of hemoglobin on plasma biomarker levels remained significant after adding the presence of CKD to the covariates (Table S5 in supporting information). For hemoglobin, there is one study that showed a negative relationship between hemoglobin and plasma biomarkers (NfL and p‐tau181) within cognitively impaired patients with heart failure. ²⁶ Considering that T₃ levels are related to systemic inflammation, ²⁷ , ²⁸ these factors may affect plasma biomarker levels through inflammatory changes. However, the mechanisms by which these factors are related to Aβ‐independent variability remain unclear, necessitating further investigation. For Aβ42/40 ratio, it appears that there are no factors related to Aβ‐independent variability because it was used in ratio form, which can cancel out individual variabilities.

Out of our expectations, we did not find positive relationships between vascular risk factors, such as DM and CAD, and Aβ uptakes in the brain. Although these vascular risk factors are known to contribute to the development of dementia, numerous studies—including cross‐sectional analyses, biomarker assessments, and autopsy data—have reported an inverse correlation between vascular factors and Aβ burden. This inverse relationship may be related to the fact that these vascular risk factors could contribute to dementia through various mechanisms. Furthermore, rather than merely the presence of vascular risk factors, various parameters associated with them—especially their variability—showed distinct associations with AD and vascular markers. A previous study from our group suggested that glucose variability exhibits more significant associations with AD and vascular markers than the mere presence of hypertension and DM. ²⁹

Our second major finding was that the effects of APOE genotypes and BMI status on plasma biomarker levels decreased or disappeared after controlling for Aβ uptake. In fact, our mediation analyses showed that APOE genotypes and BMI status affected plasma biomarker levels via the mediation of Aβ uptake. Specifically, Aβ uptake partially mediated the relationships between APOE ε4 carriers and all types of plasma biomarker levels, except the GFAP level, and the relationships between BMI status and all types of plasma biomarker levels, except the Aβ42/40 ratio. Thus, our findings suggested that not only Aβ‐independent variability but also Aβ‐dependent variability in the APOE ε4 carrier status and BMI affect plasma biomarkers. Indeed, the APOE ε4 allele is a well‐known risk factor for increased brain Aβ burden, which also influences the transport of Aβ across the blood–brain barrier by altering barrier integrity. ³⁰ , ³¹ Regarding BMI status, being underweight is associated with a higher brain Aβ burden, whereas obesity is associated with a lower Aβ burden. ¹¹ , ³² , ³³ , ³⁴ , ³⁵ BMI status is also known to affect plasma biomarker levels through changes in body blood volume. ⁸ , ³⁶ , ³⁷

Our final major finding was that some factors related to Aβ‐dependent variability improved the predictive performance of plasma biomarkers for Aβ (+) on PET. This aligns with our previous studies, which showed that adding APOE as covariates improved the predictive performances of Aβ42/40 ratio. ⁹ , ³¹ , ³⁸ However, for p‐tau217, the most prominent AD plasma biomarker, the improvement was relatively minor because it already showed superior predictive performance compared to other plasma biomarkers. These findings are consistent with previous literature, which showed that adding comorbidities and genotypes as covariates has only modest effects on the predictive performance of Aβ uptake on PET. ⁸ , ³⁹ The findings indicate that p‐tau217 is relatively robust for predicting Aβ (+) on PET across various medical conditions compared to other plasma biomarkers. Conversely, incorporating factors with Aβ‐independent variability as covariates had minimal impact on the Aβ (+) predicting performance of plasma biomarkers, except for NfL. This indicates that, aside from NfL, plasma biomarkers effectively reflect Aβ tracer uptake, rendering the influence of confounders negligible, particularly those related solely to Aβ‐independent variability. Nonetheless, the comorbidities and factors related to both Aβ‐dependent and ‐independent variabilities might still be considered in clinical settings when applying plasma biomarkers in quantitative analyses and longitudinal follow‐up for patients with significant anemia, kidney or liver disease, or active inflammatory conditions.

The strengths of the present study include its prospective design; standardized Aβ PET, MRI, and plasma biomarker acquisition protocols; and standardized genotype‐phenotyping of participants in a large Asian cohort. However, this study also has some limitations. First, there was no pathologic confirmation of the diagnosis. Therefore, the presence of other pathologies such as argyrophilic grain disease or hippocampal sclerosis could not be excluded. Second, rdcCL is different from Klunk's original Centiloids, although the basic concept of rdcCL is almost the same as Klunk's original Centiloids. ²² , ⁴⁰ We calculated the rdcCL based on ≈ 100 participants, consisting of young age controls, old age controls, MCI, and dementia participants, who were imaged simultaneously with ¹⁸F‐florbetaben and ¹⁸F‐flutemetamol. We found a very high correlation between Klunk's original Centiloids and the global values of rdcCL. Third, we obtained the laboratory findings from three different laboratories. However, this issue might have been mitigated by use of z transformed values within each laboratory. Finally, this study only captured the cross‐sectional relationship between the given medical conditions and plasma biomarkers. This may obscure the long‐term modulating effect of certain medical conditions on brain Aβ pathology or on other pathways that affect plasma biomarker levels. Nevertheless, our findings are noteworthy because they support the expansion of the use of plasma biomarkers and highlight their potential as valuable tools for early detection of AD, especially with respect to various comorbidities and biological factors.

In conclusion, our findings indicate that while specific comorbidities and factors do impact plasma biomarker levels associated with Aβ uptake, incorporating these variables as covariates in predictive models yields only a negligible to modest improvement in performance. This underscores the need for further research to explore additional factors or alternative approaches that may enhance the predictive accuracy of plasma biomarkers in assessing Aβ uptake in the brain.

AUTHOR CONTRIBUTIONS

Drs Eun Hye Lee, Sun Hoon Kang, Hyemin Jang, and Sang Won Seo had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Concept and design: Eun Hye Lee, Sun Hoon Kang, Hyemin Jang, and Sang Won Seo. Acquisition of data: Eun Hye Lee, Sun Hoon Kang, Daeun Shin, Jihwan Yun, Kyoung Cheon, Sang Won Seo, on behalf of the K‐ROAD study group. Statistical analysis: Eun Hye Lee, Young Ju Kim, and Heejin Yoo. Interpretation of data: Eun Hye Lee, Sun Hoon Kang, Daeun Shin, Kim, Hyemin Jang, and Sang Won Seo. Drafting of the manuscript: Eun Hye Lee. Critical revision of the manuscript for important intellectual content: Eun Hye Lee, Sun Hoon Kang, Henrik Zetterberg, Kaj Blennow, Fernando Gonzalez‐Ortiz, Nicholas J. Ashton, Kim, Kim, Duk L. Na, Hyemin Jang, and Sang Won Seo. Supervision: Hyemin Jang and Sang Won Seo.

CONFLICT OF INTEREST STATEMENT

Zetterberg is a Wallenberg Scholar and a Distinguished Professor at the Swedish Research Council supported by grants from the Swedish Research Council (#2023‐00356; #2022‐01018 and #2019‐02397), the European Union's Horizon Europe research and innovation programme under grant agreement No 101053962, Swedish State Support for Clinical Research (#ALFGBG‐71320), the Alzheimer Drug Discovery Foundation (ADDF), USA (#201809‐2016862), the AD Strategic Fund and the Alzheimer's Association (#ADSF‐21‐831376‐C, #ADSF‐21‐831381‐C, #ADSF‐21‐831377‐C, and #ADSF‐24‐1284328‐C), the Bluefield Project, Cure Alzheimer's Fund, the Olav Thon Foundation, the Erling‐Persson Family Foundation, Stiftelsen för Gamla Tjänarinnor, Hjärnfonden, Sweden (#FO2022‐0270), the European Union's Horizon 2020 research and innovation programme under the Marie Skłodowska‐Curie grant agreement No 860197 (MIRIADE), the European Union Joint Programme–Neurodegenerative Disease Research (JPND2021‐00694), the National Institute for Health and Care Research University College London Hospitals Biomedical Research Centre, and the UK Dementia Research Institute at UCL (UKDRI‐1003); has served on scientific advisory boards and/or as a consultant for Abbvie, Acumen, Alector, Alzinova, ALZPath, Amylyx, Annexon, Apellis, Artery Therapeutics, AZTherapies, Cognito Therapeutics, CogRx, Denali, Eisai, Merry Life, Nervgen, Novo Nordisk, Optoceutics, Passage Bio, Pinteon Therapeutics; has received payments or honoraria for lectures, presentations, speakers bureaus, manuscript writing, or educational events from Prothena, Red Abbey Labs, reMYND, Roche, Samumed, Siemens Healthineers, Triplet Therapeutics, Wave, Cellectricon, Fujirebio, Lilly, Novo Nordisk, and Roche; has served on scientific advisory boards and/or as a consultant for Abbvie, Acumen, Alector, Alzinova, ALZPath, Amylyx, Annexon, Apellis, Artery Therapeutics, AZTherapies, Cognito Therapeutics, CogRx, Denali, Eisai, Merry Life, Nervgen, Novo Nordisk, Optoceutics, Passage Bio, Pinteon Therapeutics, Prothena, Red Abbey Labs, reMYND, Roche, Samumed, Siemens Healthineers, Triplet Therapeutics, and Wave, with payments for these roles; is a co‐founder of Brain Biomarker Solutions in Gothenburg AB (BBS), which is part of the GU Ventures Incubator Program, with payments; is chair of the Alzheimer's Association Global Biomarker Standardization Consortium. Blennow has served as a consultant and on advisory boards for Abbvie, AC Immune, ALZPath, AriBio, BioArctic, Biogen, Eisai, Lilly, Moleac Pte. Ltd, Neurimmune, Novartis, Ono Pharma, Prothena, Roche Diagnostics, and Siemens Healthineers; has served on data monitoring committees for Julius Clinical and Novartis; has given lectures, produced educational materials, and participated in educational programs for AC Immune, Biogen, Celdara Medical, Eisai, and Roche Diagnostics; and is a co‐founder of Brain Biomarker Solutions in Gothenburg AB (BBS), which is a part of the GU Ventures Incubator Program, outside the work presented in this paper. Ashton has received consulting fees from Quanterix, and has also received payments for lectures, presentations, speakers bureaus, manuscript writing, or educational events from Alamar Biosciences, Biogen, Eli‐Lilly, and Quanterix. Ashton is listed as an inventor on a patent application (Application No.: PCT/US2024/037834, WSGR Docket No. 58484‐709.601) related to methods for remote blood collection, extraction, and analysis of neuro biomarkers; serves on the advisory board for Biogen, TargetALS, and TauRx; and receives payments for this role. Na and Seo are co‐founders of BeauBrain Healthcare, Inc. Other authors have no conflicts of interest to disclose. Author disclosures are available in the supporting information.

CONSENT STATEMENT

This study was approved by the institutional review board of Samsung Medical Center (approval no. 2021‐02‐135). All participants provided informed consent to participate in the study and data were collected in accordance with the Declaration of Helsinki.

Supporting information

Supporting Information

Supporting Information

Supporting Information

Lee EH, Kang SH, Shin D, et al. Plasma Alzheimer's disease biomarker variability: Amyloid‐independent and amyloid‐dependent factors. Alzheimer's Dement. 2025;21:e14368. 10.1002/alz.14368

DATA AVAILABILITY STATEMENT

Anonymized data for our analyses presented in the present report are available from the corresponding authors upon request.