Portable platform for measuring amyloid beta 42/40 ratio via photooxidation-induced fluorescence amplification

Abstract

Alzheimer's disease (AD) is a serious health condition that exacerbates with age. Among various AD biomarkers, the measurement of the Aβ 42/40 ratio (i.e., ratio of Aβ 42 and Aβ 40 concentrations) has garnered prominence for the early identification of AD patients and the development of disease-modifying treatments. Approaches such as positron emission tomography examinations and biomarker measurements in cerebrospinal fluid and blood plasma face constraints related to expense and procedural complexity. To address these issues, we developed and evaluated the efficacy of a portable platform for detecting amyloid beta (Aβ) 40 and 42 in plasma, utilizing photooxidation-induced fluorescence amplification (PIFA). We conducted a comparative analysis of Aβ 42/40 measurements between the PIFA and single-molecule immunoassay (SiMoA) platforms. By measuring 38 cases of subjective cognitive decline (SCD), 24 cases of amnestic mild cognitive impairment (aMCI), and 46 cases of AD dementia samples, we observed a significant difference in Aβ 42/40 ratios between the SCD and aMCI groups. The PIFA platform demonstrated an area under the curve compared to that of the SiMoA platform, which is currently the most precise method for Aβ 42/40 ratio measurement. Consequently, the PIFA platform presents a viable cost-effective tool for detecting the Aβ 42/40 ratio.

---

Introduction

Alzheimer's disease (AD) is the leading cause of dementia globally, affecting approximately 6.7 million Americans aged 65 and older as of 2023.¹ AD remains a significant public health challenge, highlighting the urgent need for reliable, noninvasive diagnostic instruments such as blood-based biomarkers to facilitate early detection and intervention. Following the approval of two amyloid-clearing medications by the Food and Drug Administration, and with 96 disease-modifying treatments (DMTs) under development as of 2024, the significance of assessing biomarkers to identify eligible patients for DMT and to monitor their status is increasing.² Amyloid beta (Aβ) plaques and neurofibrillary tangles in the brain are characteristics of AD.^3,4 Although amyloid positron emission tomography (PET) scans, are established amyloid biomarkers, they are costly and time-consuming.⁵ A reduced Aβ 42/40 ratio in bodily fluids is advantageous for identifying amnestic moderate cognitive impairment (aMCI) or early-stage AD.^6–8 A reduced Aβ 42/40 ratio in cerebrospinal fluid (CSF) correlates with amyloid PET positivity but necessitates an invasive procedure.^9–12 Consequently, a less intrusive blood biomarker for screening during the preclinical phases of AD is essential.

The evaluation of the Aβ 42/40 ratio in blood plasma is considered an alternative to CSF measurement, with a decreased Aβ 42/40 ratio associated with aMCI.¹³ Further studies demonstrate that a decline in Aβ 42/40 levels in blood plasma is associated with positive amyloid PET scans.^14,15 Techniques for quantifying the Aβ 42/40 ratio in blood plasma include immunoassays such as enzyme-linked immunosorbent assay, single-molecule immunoassay (SiMoA), and immunoprecipitation-coupled mass spectrometry.¹⁶ Although these techniques are more practical than amyloid PET or CSF tests, they are limited to laboratory settings due to the necessity for highly calibrated and costly equipment.

To address current challenges in early AD identification and monitoring, we developed an innovative immunoassay platform incorporating a cartridge and handler based on photooxidation-induced fluorescence amplification (PIFA)¹⁷ for quantifying Aβ 40 and Aβ 42 in plasma. This platform, optimized for field applications, features a cost-effective disposable cartridge and a fully automated portable handler. We analyzed 108 samples using the PIFA immunoassay platform and compared the results with those from a SiMoA platform. The results from both platforms were consistent with patients' medical data, particularly in evaluating the Aβ 42/40 ratio in correlation with amyloid PET positivity and clinical diagnoses.

Materials and methods

Development of a cartridge and a handler for PIFA immunoassay platform

The designed handler can simultaneously manage two cartridges, one designated for Aβ 40 and the other for Aβ 42 (Fig. 1A). The handler incorporates a detecting module featuring a light-emitting diode (LED) and photodiode, a tip handling module, a handle, a central processing unit (CPU), and fans. The dimensions of the handler are 240 × 367 × 270 mm³, with a weight of 5.6 kg (Fig. 1B). The cartridge was produced via polystyrene injection molding, and includes one tip container and eight reagent wells. The tip container comprises a polypropylene tip, which was pretreated with oxygen plasma (under 500 sccm of O₂ flow for 320 s at 0.7 kW RF and 0.14 Torr) to enhance binding with the biotin-BSA.

Fig. 1 Overview of the cartridge and the handler for Aβ 40 and Aβ 42 measurement in human plasma. (A) Schematic of the developed handler for two cartridges simultaneously for detecting Aβ 40 and Aβ 42 separately. The first well of each cartridge contains the detection antibody, and the sample is initially loaded into well 1. The antigens in the plasma sample bind with the detection antibodies before being analyzed. (B) Internal structure of the handler shows the detecting module (including LED and photodiode), tip handling module, cartridge bed, CPU, and fan. (C) Upper view of the cartridge and layout of the eight wells in the developed cartridge.

Reagents of the Aβ 40 cartridge were prepared as follows: well 1 contains 20 μL of a solution with 8 μg mL⁻¹ horseradish peroxidase (HRP)-conjugated anti-beta-amyloid 1–16 antibody (BioLegend, catalog number 803012) and 50 μg mL⁻¹ human anti-mouse antibody blocker (Fitzgerald, catalog number 85R-1003) diluted in antibody stabilizer phosphate-buffered saline (PBS; Candor, catalog number 131500). Well 2 contains 100 μL of 0.05 mg mL⁻¹ biotin-labeled bovine albumin (Sigma, catalog number A8549-10MG) diluted in distilled water. Well 3 contains 100 μL of 0.05 mg mL⁻¹ recombinant Streptavidin-plus® (Agilent, catalog number SA26) diluted in distilled water. Well 4 contains 100 μL of 2 μg mL⁻¹ polyclonal antibody to Aβ peptide 1–40 (Cloud-Clone Corporation, catalog number PAA864Ra08) biotinylated using 0.005M EZ-Link sulfo-NHS-LC-LC-biotin (Thermo Fisher Scientific, catalog number 21338) and Slide-A-Lyzer™ Dialysis Cassettes, 20K MWCO (Thermo Fisher Scientific, catalog number 66005), diluted in antibody stabilizer PBS (Candor, catalog number 131500). Wells 5 and 6 contain 100 μL of 1X PBS at pH 7.4 (HanLAB, catalog number HPBS-1010-74) mixed with 0.1% Tween 20. Wells 7 and 8 contain 100 μL of QuantaRed™ Enhanced Chemifluorescent HRP substrate (Thermo Fisher Scientific, catalog number 15159).

Reagents of the Aβ 42 cartridge: well 1 contains a 20 μL mixture comprising 6 μg mL⁻¹ HRP-conjugated anti-beta-amyloid 1–16 antibody (BioLegend, catalog number 803012), and 20 μg mL⁻¹ human anti-mouse antibody blocker (Fitzgerald, catalog number 85R-1003) diluted in antibody stabilizer PBS. Well 4 contains 100 μL of 2 μg mL⁻¹ biotinylated anti-beta-amyloid x-42 antibody (BioLegend, catalog number 812103) diluted in antibody stabilizer PBS. The reagents in the remaining wells are identical to those in the wells of the Aβ 40 cartridge (Fig. 1C).

Test procedures

All reagents, except for the sample and QuantaRed™ solution, were loaded into the designated wells, and the cartridge was sealed with a film. The sealed cartridges were stored at 4 °C until use. Prior to measurement, the sealing film was removed from both the Aβ 40 and Aβ 42 cartridges. 60 μL of sample was added to well 1, followed by adding of 100 μL of QuantaRed™ solution in wells 7 and 8. The cartridges were mounted on the handler's beds, which were inward at the same time, allowing the tip to move to follow the programmed sequence. After 45 minutes, the signals were amplified and quantified by the detection module (Fig. S1 and S2†).

Signal detection and calculation

The sensor in the handler measures light-amplified signals produced during the chemical reaction in which 10-acetyl-3,7-dihydroxyphenoxazine (ADHP) is converted to resorufin. The measured signal correlates with the time-dependent generation of resorufin. These data are fitted to a four-parameter logistic curve (eqn (1)), illustrating the initial concentration of resorufin (RSF₀) and ADHP (AR₀) in wells 7 and 8 (the latter serving as the reference well) immediately following the reaction with the tip and reagent in well 7. The parameter k is the reaction constant, representing the slope of in the linear region of the resorufin vs. time curve.To normalize the resorufin value obtained (RSF₀), we divide RSF₀ by the sum of RSF₀ and AR₀, as shown in eqn (2).Next, we subtract the normalized values from wells 7 and 8, and define the resulting difference as the amplification index (AI). The AI thus quantifies the additional resorufin generated in the test well (well 7) relative to the reference well (well 8), reflecting how much HRP-conjugated antibody, bound to the target antigens, drives resorufin production. Consequently, AI values provide an indirect measure of the concentration of Aβ40 or Aβ42 in the sample. We conducted a quality-control test of the PIFA methodology by spiking the cartridge's test well with known concentrations of HRP (Fig. S3†).

Validation and calibration of measured signal

Standards of beta-amyloid 1–40 HFIP (r-peptide, A-1153) and beta-amyloid 1–42 HFIP (r-peptide, A-1163) were prepared and spiked into SeraCon™ I negative diluent (Seracare, 1800-0009). The Aβ 40 standard was diluted to 500, 250, 100, 50, 25, and 12.5 pg mL⁻¹, and the Aβ 42 standard was diluted to 70, 35, 17.5, 10.5, 5.3, and 2.6 pg mL⁻¹. All standards were serially diluted 1 : 1, and the concentrations were verified using an N3P Kit (Quanterix, catalog number 101995) on an HD-X analyzer (Quanterix). Each standard was measured three times and quantified as AI. Finally, calibration curve was drawn fitted to the 4-parameter-logistic (4PL) model, concentration – AI graph.

The Aβ 40 and Aβ 42 levels in collected plasma samples (N = 108) were also measured using the N3P Kit on the HD-X analyzer.

We performed a quality-control test for PIFA methodology by spiking the test well with HRP standards at defined concentrations, confirming reliable quantification (Fig. S3†). To ensure comprehensive quality control, including sandwich-complex formation and immobilization, we assessed control samples. For Aβ 40, negative, 50 pg mL⁻¹, 125 pg mL⁻¹, 250 pg mL⁻¹, and 500 pg mL⁻¹ of Aβ 40 controls were measured twice to verify proper reagent performance. For Aβ 42, negative, 5.3 pg mL⁻¹, 10.5 pg mL⁻¹, 17.5 pg mL⁻¹, and 35 pg mL⁻¹ of Aβ 42 controls were likewise measured twice (Fig. S4†). We additionally tested the assay's repeatability and reproducibility by measuring Aβ 40 and Aβ 42 on different date. Specifically, four calibration standards for Aβ 40 were each measured three times, and the same procedure was applied for four calibration standards of Aβ 42 (Fig. S5†).

Cohorts

This study was performed following the guidance of the institutional review board of Myongji Hospital (IRB no. 2021-04-016-032, 2022-12-023-009). Written informed consent was obtained from all of 108 participants or their guardians between June 2023 and July 2024. The participants reporting memory decline were prospectively enrolled for Brain Bank (IRB no. 2021-04-016-032) and the Study on the effects of thyroid hormones on cognitive decline and correlation with Alzheimer's disease biomarkers (IRB no. 2022-12-023-009) at the memory clinic of Myongji Hospital.

Patients underwent comprehensive evaluations, including interviews, neurological assessments, and neuropsychological tests such as the Mini-Mental State Examination (MMSE), Clinical Dementia Rating (CDR), brain magnetic resonance imaging, and amyloid PET scans. In clinical diagnosis, subjective cognitive decline (SCD) encompasses patients exhibiting normal results in neurocognitive assessments. Patients met Petersen's clinical modified criteria for mild cognitive impairment (MCI), which included:¹⁸ (1) subjective memory complaints from the patient or caregiver, (2) normal activities of daily living (ADL) as evaluated through a clinician interview, (3) objective memory decline below the 16th percentile or the normative standard set by neuropsychological tests, and (4) absence of dementia. The AD dementia (ADD) cohort comprised individuals who met the NIA-AA diagnostic guidelines for Alzheimer's disease,¹⁹ thus satisfying the clinical criteria for probable ADD. Cognitive function was assessed using the clinical deterioration scale the Korean version of the MMSE (ESI† 2).

Sample collection

Blood samples were collected using 6 mL K2 EDTA vacutainer tubes (BD, catalog number: 367863) with 21G needles, ensuring that each tube was filled to the specified capacity to maintain the anticoagulant concentration. The blood was gently mixed by inverting the tube 8–10 times and processed within 2 hours to prevent degradation. Plasma was obtained by centrifugation at 1500 g for 15 min at 4 °C. Post-separation, plasma was carefully aspirated to avoid contamination with blood cells and pooled if multiple tubes were used. Subsequent 1 mL aliquots were transferred to 1.5 mL Eppendorf tubes labeled with sample details and stored at −80 °C.

Statistical analysis

For statistical analysis, we employed Prism 10 (version 10.3.1). Group differences were assessed using Welch's test or Dunnett's T3 multiple comparisons test, while categorical variables were analyzed with the Fisher exact test. Adjusted p-values below 0.05 were considered statistically significant. The sensitivity and specificity of the Aβ 42/40 ratio were evaluated via receiver operating characteristic (ROC) curve analysis. Adjusted ROC models incorporating sex, age, years of education, apolipoprotein E4 (APOE4) status, and the Aβ 42/40 ratio were obtained using logistic regression for ROC curve analysis.

Cohort descriptions

Cohort samples were collected as follows: SCD (N = 38), aMCI (N = 24), and ADD (N = 46) (Table 1). The percentage of APOE 4 carriers in the aMCI and ADD groups exhibits significant differences when compared to the SCD group: aMCI (p = 0.032) and ADD (p = 0.019). The average MMSE scores significantly differed between the SCD and ADD groups (p < 0.0001) and the SCD and aMCI (p = 0.009) groups. The median (inter quartile range, IQR) CDR sum of boxes scores were significantly higher in the ADD group compared to the SCD group (p < 0.0001). Additionally, the prevalence of amyloid PET positive in the SCD and ADD cohorts was 20 and 64.29%, respectively, indicating a statistically significant difference between the two groups (p = 0.014).

Table 1 Sample demographics. SCD, aMCI, and ADD groups are considered. Regarding amyloid PET, 20, 2, and 14 samples from the SCD, aMCI, and ADD groups have the corresponding exams, respectively. All the variables are tested under Dunnett's T3 multiple comparisons test, except for categorical variables. Sex, APOE4 status, and amyloid PET status are tested using the Fisher exact test. (****, p < 0.0001; ***, p < 0.001; **, p < 0.01; *, p < 0.05)

		SCD^a (n = 38)	aMCI (n = 24)	ADD (n = 46)
a Not available data: 1 in sex and age; 7 in education and MMSE; 2 in APOE 4; 8 in CDR sum of boxes.
Female	(%)	64.86	79.17	67.39
	n/total (number)	24/37	19/24	31/46
Age in years	Mean	70.2	71.9	76.5
	SD	7.7	7.6	8.9
Years of education	Mean	9.4	8.1	8.7
	SD	4.1	6.2	5.6
APOE 4 carrier	(%)	13.9	41.7*	37.0*
	n/total (number)	5/36	10/24	17/46
MMSE score	Mean	26.5	23.0**	16.4****
	SD	3.3	3.5	7.7
CDR sum of boxes score	Median	1	2	5.5****
	IQR	0.5–1.5	1.6–3.0	2.9–10.0
Amyloid PET positivity	(%)	20	50	64.29*
	n/total (number)	4/20	1/2	9/14

---

Results

Operation principle of PIFA immunoassay platform

The developed PIFA immunoassay platform is composed of two main components: a cartridge and a PIFA handler. The PIFA handler, in turn, consists of key modules such as the tip handler module and the detection module. The operation process is also divided into two major steps: the ELISA reaction and the PIFA reaction.

Into well 1 of both Aβ 40 and Aβ 42 cartridge, 60 μL of plasma sample was introduced. The antigen in the sample interacted with the preloaded detection antibody in well 1 of each cartridge. Subsequently, the cartridges were placed into the handler. The handling module seized the tip from the tip container and elevated it vertically, while the cartridge bed shifted laterally to facilitate the access of tips to wells. The tip coating and enzyme reaction process are depicted in Fig. 2A. The tip was immersed sequentially in well 2 and 3 to form a chemical bridge composed of biotin conjugated bovine serum albumin and streptavidin (S.A). In well 4, the capture antibody was bound. The tip was then cleansed in wells 5 and 6, followed by the binding of the detection antibody paired with the antigen in well 1. To mitigate nonspecific binding, the tip was rinsed in wells 5 and 6 again. It subsequently reacted with the substrate in well 7, yielding RSF₀. Wells 7 and 8 were irradiated by LEDs, and fluorescence intensity was monitored in real-time using photodiodes. The greater the quantity of RSF₀ produced in well 7, the more rapid the fluorescence amplification rate. The fluorescent amplification rate of the test well, compared to the reference well, increases proportionally with the quantity of RSF₀ produced in the test well. This results in a disparity in the fluorescence amplification graphs between the test and reference wells (Fig. 2B).

Fig. 2 Process of tip coating and quantification of Aβ 40 and Aβ 42. (A) Sequence of tip coating for measurement. (B) Schematic of the signal measurement process. The reactions in wells 7 (red line: signal of high concentration for test, blue line: signal of low concentration for test) and 8 (black line: signal for reference) are amplified by LEDs. In the test well, RSF existed due to the previous reaction between ADHP and HRP in well 7 in Fig. 2A. The greater the amount of RSF₀ in well 7 the more rapidly the RSF concentration increases (high concentration), and vice versa (low concentration). The fluorescence amplification rate difference between the test and the reference wells is converted to antigen concentration for quantification.

When the antigen concentration in the sample is low, a corresponding decrease is noted in the amount of detection antibody immobilized on the tip. Consequently, this results in a reduced binding of HRP to the tip, leading to a lower generation of RSF₀. Therefore, the fluorescence amplification rates between the test and reference wells exhibits no significant difference. Conversely, a high antigen concentration leads to an increased generation of RSF₀, causing a notably higher fluorescence amplification rate in the test well compared to the reference well. The difference in these rates manifests as distinct variations in the graphs. By analyzing these variations, the concentration of RSF₀ can be determined inversely as the AI value. A low AI value indicates a low antigen concentration, while a high AI value corresponds to a high antigen concentration (Fig. 2B). Each standard concentration was measured three times using the PIFA immunoassay platform. The Amplification index (AI) values for Aβ 40 and Aβ 42 are depicted in Fig. 3A and B, with error bars representing the coefficient of variation (CV%). The data were fitted to a sigmoidal, 4-parameter-logistic (4PL) model, yielding R² values exceeding 0.99 for both. The dynamic ranges, where the CV% remained below approximately 10% of Aβ 40 and Aβ 42 were 25–500 and 10.5–70 pg mL⁻¹, respectively.

Fig. 3 Amplification index versus as a function of concentration of Aβ 40 and Aβ 42 concentrations. Seven standard concentrations were tested, with red bars indicating the coefficient of variation (CV). For clarity, the four lowest concentration standards are magnified in the insets. (A) Aβ 40, (B) Aβ 42.

Aβ 42/40 measurement in patient cohort

Analysis of the Aβ 42/40 ratio in N = 108 samples revealed a statistically significant distinction between the SCD and aMCI groups on both the PIFA and SiMoA immunoassay platforms (Fig. 4A and B; p = 0.0314 for PIFA and p = 0.0257 for SiMoA). Conversely, no significant differences were detected when comparing the SCD and ADD groups or the aMCI and ADD groups (p = 0.1612 and 0.5272, respectively, for PIFA; p = 0.2409 and 0.6680, respectively, for SiMoA). Additionally, the Aβ 42/40 did not exhibit significant variations between amyloid PET-negative and PET-positive cohorts on either platform (Fig. 4C and D; p = 0.4832 for PIFA and p = 0.2325 for SiMoA).

Fig. 4 Aβ 42/40 ratios by disease status and amyloid PET. Aβ 42/40 ratios for (A) PIFA and (B) SiMoA in SCD, aMCI, and ADD groups. Ratios in amyloid PET-negative (PET-) and PET-positive (PET+) groups using (C) PIFA and (D) SiMoA immunoassay platforms. (*, p < 0.05; ns, no significant difference).

ROC curves and adjusted models on PIFA and SiMoA platforms

Fig. 5 illustrates the analysis of ROC curves for the Aβ 42/40 ratio on both the PIFA and SiMoA immunoassay platforms. In Fig. 5A, comparing the SCD and aMCI groups using the PIFA platform, the AUC (95% CI) was 0.713 (0.583–0.843). For the same comparison using the SiMoA platform, the AUC (95% CI) was 0.6535 (0.512–0.795). Fig. 5B shows the comparison between SCD and ADD, where the AUC was lower than that of the SCD versus aMCI comparison, with values of 0.6001 (0.477–0.723) for PIFA and 0.6733 (0.556–0.791) for SiMoA. Fig. 5C and D present models where the Aβ 42/40 ratio is adjusted by sample-specific factors (sex, age, years of education, and APOE4 status). These adjusted models demonstrate improved performance over the unadjusted Aβ 42/40 ratio. For the SCD versus aMCI groups (Fig. 5C), the AUC (95% CI) increased to 0.7759 (0.636–0.916) for PIFA and 0.7500 (0.597–0.903) for SiMoA. For the SCD versus ADD groups (Fig. 5D), the AUC (95% CI) was 0.7675 (0.653–0.882) for PIFA and 0.7781 (0.656–0.900) for SiMoA.

Fig. 5 ROC curves for PIFA and SiMoA immunoassay platforms. (A and B) Unadjusted Aβ 42/40 ROC curves comparing (A) SCD vs. aMCI and (B) SCD vs. ADD. (C and D) Adjusted ROC curves (sex, age, years of education, and APOE4 status) comparing (C) SCD vs. aMCI and (D) SCD vs. ADD.

Fig. 6 shows the predicted probabilities from the adjusted multiple regression models comparing aMCI and ADD with SCD. A logistic regression model incorporating sex, age, years of education, APOE4 status, and the Aβ 42/40 ratio was employed. The model estimates the probability of disease, ranging from 0 (normal) to 1 (Alzheimer's disease). Four logistic regression models were assessed, and the Hosmer–Lemeshow test (p > 0.05) confirmed good fit for the SCD vs. aMCI and SCD vs. ADD models on the PIFA platform, as well as the SCD vs. aMCI model on the SiMoA platform. In contrast, the SCD vs. ADD model on the SiMoA platform had a poor Hosmer–Lemeshow result (p = 0.0085), indicating potential misfit. Specifically, for the PIFA platform, p = 0.0007 for SCD vs. aMCI and p = 0.0001 for SCD vs. ADD (Fig. 6A and B). For the SiMoA platform, p = 0.0018 for SCD vs. aMCI (Fig. 6C) and p = 0.0002 for SCD vs. ADD (Fig. 6D).

Fig. 6 Predicted probability (0–1) from multiple logistic regression of Aβ 42/40, adjusted for sex, age, years of education, and APOE4 status. (A and B) Comparison of SCD vs. aMCI and SCD vs. ADD on the PIFA immunoassay platform. (C and D) Comparison of SCD vs. aMCI and SCD vs. ADD on the SiMoA immunoassay platform. (***, p < 0.001; **, p < 0.01).

Discussion

In this study, we quantified blood plasma Aβ 40 and Aβ 42 levels in 108 participants experiencing cognitive decline, stratifying them into three groups: ADD, aMCI, and SCD. Our primary aim was to evaluate the discriminative power of the Aβ 42/40 ratio among these groups using both the PIFA and SiMoA immunoassay platforms. Consistent with prior reports, Aβ 40 and Aβ 42 concentrations in our samples were within the operational range of the PIFA platform, maintaining a coefficient of variation (CV) below 10%.^15,20,21 This is notable as bioanalytical guidelines often recommend for a CV under 20% to ensure acceptable assay performance.^22–24 Comparative cross-section analysis of the PIFA and SiMoA platforms demonstrated a strong correlation in measuring Aβ 40, Aβ 42, and the Aβ 42/40 ratio (Fig. S6†). This aligns with earlier research indicating that these assays generate comparable results.²⁵ Both platforms exhibited significant discriminative capacity in differentiating the SCD group from the aMCI group, corroborating previous studies that identify the Aβ 42/40 ratio as an early-stage biomarker for AD.^26–28 As a result, the AUC values of SCD vs. aMCI were higher than SCD vs. ADD on both platforms measuring Aβ 42/40. Additionally, we analyzed various clinical information (Fig. S7 and S8†). Our results reaffirm earlier reports^15,20,29,30 that incorporating demographic and genetic information (sex, age, years of education, and APOE4 status) enhanced the area under the curve (AUC) for the Aβ 42/40 ratio models on both platforms. Specifically, the AUC of the PIFA platform rose from 0.713 to 0.776 and the AUC of the SiMoA platform increased from 0.654 to 0.750 post-adjustment, highlighting the multifactorial nature of AD and the benefit of integrating multiple risk factors.^20,27

Despite promising outcomes, several limitations merit attention. First, the absolute values of Aβ measurements vary depending on the specific immunoassay technique employed.²⁸ This underscores the need for a standardized protocol to validate the PIFA platform for routine use in AD diagnostics. Second, although both platforms effectively differentiated aMCI from SCD, our study lacked sufficient amyloid PET data to evaluate the ability of each assay to identify amyloid PET–positive cases. Only 33% of the samples included PET information, resulting in insufficient statistical power to confirm differences in PET status. The power analysis was conducted by Prism ver.10.4.1 (GraphPad Software), and it recommended that the sample size be 96 per group for power values = 0.803 on SiMoA platform. On PIFA platform, however, for getting power values = 0.801, N = 291 were required per group. As a result, more PET samples were needed for proving valid relationship with PIFA platform and PET measurement. Future studies should incorporate extensive cohorts and standardized immunoassay protocols to further elucidate the clinical relevance and reproducibility of PIFA measurements concerning amyloid PET findings. Notably, aMCI may precede clinical AD by several years, represents a critical window for potential intervention.^1,27,31 SiMoA is often used as a benchmark assay for measuring the Aβ 42/40 ratio in human blood plasma,^32,33 but its high cost can limit broad adoption.^34,35 By contrast, the PIFA immunoassay platform offers a more affordable and compact alternative, suggesting its potential as a next-generation tool for large-scale or point-of-care screening. If further validated, it could expand access to early biomarker detection, facilitating timely clinical decision-making and potentially improving patient outcomes.

Conclusion

We developed and evaluated a cost-effective, user-friendly PIFA immunoassay platform capable of measuring the early AD biomarker Aβ 42/40 in diverse samples. Its performance was comparable to the resource-intensive SiMoA platform. The reliability and affordability of the PIFA platform for quantifying Aβ 42 and Aβ 40, makes it a promising option for early AD detection and risk assessment in clinical settings. However, standardization efforts and extensive longitudinal studies will be crucial to establish its clinical utility and explore its potential for preclinical stages of AD detection.

Data availability

The data supporting this article have been included as part of the ESI.†

Ethics approval

All participants' samples were collected under Institutional Review Board of Myongji Hospital (Approval IRB no. 2021-04-016-032, 2022-12-023-009). This study was approved by the Institutional Review Board of Myongji Hospital (Approval IRB no. 2024-09-004). Written informed consent was obtained from all of 108 participants or their guardians between June 2023 and July 2024.

Author contributions

S. K: conceptualization, methodology, investigation, data curation, formal analysis, visualization, writing – original draft, writing – review and editing. H. B: formal analysis, investigation, methodology, visualization, writing – original draft, writing – review and editing. D. K: formal analysis, methodology. Y. L: resources. I. C: methodology. D. L: formal analysis. D. C: formal analysis. H. C. K: resources. S. Y. S: resources. Y. J: formal analysis, methodology, writing – review and editing. S. C: conceptualization, supervision, project administration, writing – review and editing, funding acquisition. Y. H. J: conceptualization, supervision, writing – review and editing, resources, funding acquisition.

Conflicts of interest

There are no conflicts to declare.

Acknowledgements

This work was supported by Ministry of Science and ICT (grant numbers RS-2024-00450828 and RS-2024-00440577), Ministry of Trade, Industry and Energy (grant number: 20023308), Ministry of SMEs and Startups (grant number RS-2024-00439056), Ministry of Health and Welfare (grant number B0080430001724), Korea Disease Control and Prevention Agency and KNIH (grant number 2023-ER1003-01) and the faculty grant of Myongji Hospital (grant numbers 2205-09-01 and 2202-09-02). We also gratefully acknowledge BK Electronics for their assistance in the fabrication of the device used in this study.

Notes and references

1. Alzheimer’s Association, Alzheimer’s Dement., 2023, 19, 1598–1695,  DOI:10.1002/alz.13016.
2. J. Cummings, Y. Zhou, G. Lee, K. Zhong, J. Fonseca and F. Cheng, Alzheimer's Dement., 2024, 10, e12465,  DOI:10.1002/trc2.12465.
3. G. F. Chen, T. H. Xu, Y. Yan, Y. R. Zhou, Y. Jiang, K. Melcher and H. E. Xu, Acta Pharmacol. Sin., 2017, 38, 1205–1235,  DOI:10.1038/aps.2017.28.
4. B. T. Hyman, C. H. Phelps, T. G. Beach, E. H. Bigio, N. J. Cairns, M. C. Carrillo, D. W. Dickson, C. Duyckaerts, M. P. Frosch, E. Masliah, S. S. Mirra, P. T. Nelson, J. A. Schneider, D. R. Thal, B. Thies, J. Q. Trojanowski, H. V. Vinters and T. J. Montine, Alzheimer's Dementia, 2012, 8, 1–13,  DOI:10.1016/j.jalz.2011.10.007.
5. A. M. Catafau and S. Bullich, Clin. Transl. Imaging, 2015, 3, 39–55,  DOI:10.1007/s40336-014-0098-3.
6. J. D. Doecke, V. Pérez-Grijalba, N. Fandos, C. Fowler, V. L. Villemagne, C. L. Masters, P. Pesini and M. Sarasa, Neurology, 2020, 94, E1580–E1591,  DOI:10.1212/WNL.0000000000009240.
7. T. Kawarabayashi, L. H. Younkin, T. C. Saido, M. Shoji, K. H. Ashe and S. G. Younkin, J. Neurosci., 2001, 21, 372–381.
8. F. Song, A. Poljak, M. Valenzuela, R. Mayeux, G. A. Smythe and P. S. Sachdev, J. Alzheimer's Dis., 2011, 26, 365–375,  DOI:10.3233/JAD-2011-101977.
9. J. Dumurgier, S. Schraen, A. Gabelle, O. Vercruysse, S. Bombois, J. L. Laplanche, K. Peoc'h, B. Sablonnière, K. V. Kastanenka, C. Delaby, F. Pasquier, J. Touchon, J. Hugon, C. Paquet and S. Lehmann, Alzheimer's Res. Ther., 2015, 7, 30,  DOI:10.1186/s13195-015-0114-5.
10. O. Hansson, S. Lehmann, M. Otto, H. Zetterberg and P. Lewczuk, Alzheimer's Res. Ther., 2019, 11, 34,  DOI:10.1186/s13195-019-0485-0.
11. J. K. Lui, S. M. Laws, Q. X. Li, V. L. Villemagne, D. Ames, B. Brown, A. I. Bush, K. De Ruyck, J. Dromey, K. A. Ellis, N. G. Faux, J. Foster, C. Fowler, V. Gupta, P. Hudson, K. Laughton, C. L. Masters, K. Pertile, A. Rembach, M. Rimajova, M. Rodrigues, C. C. Rowe, R. Rumble, C. Szoeke, K. Taddei, T. Taddei, B. Trounson, V. Ward, R. N. Martins and AIBL Research Group, J. Alzheimer's Dis., 2010, 20, 1233–1242,  DOI:10.3233/JAD-2010-090249.
12. A. Rembach, N. G. Faux, A. D. Watt, K. K. Pertile, R. L. Rumble, B. O. Trounson, C. J. Fowler, B. R. Roberts, K. A. Perez, Q. X. Li, S. M. Laws, K. Taddei, S. Rainey-Smith, J. S. Robertson, M. Vandijck, H. Vanderstichele, K. J. Barnham, K. A. Ellis, C. Szoeke, L. MacAulay, C. C. Rowe, V. L. Villemagne, D. Ames, R. N. Martins, A. I. Bush and C. L. Masters, Alzheimer's Dement., 2014, 10, 53–61,  DOI:10.1016/j.jalz.2012.12.006.
13. A. L. Brand, P. E. Lawler, J. G. Bollinger, Y. Li, S. E. Schindler, M. Li, S. Lopez, V. Ovod, A. Nakamura, L. M. Shaw, H. Zetterberg, O. Hansson and R. J. Bateman, Alzheimer's Res. Ther., 2022, 14, 195,  DOI:10.1186/s13195-022-01117-1.
14. N. Fandos, V. Pérez-Grijalba, P. Pesini, S. Olmos, M. Bossa, V. L. Villemagne, J. Doecke, C. Fowler, C. L. Masters and M. Sarasa, Alzheimer's Dement., 2017, 8, 179–187,  DOI:10.1016/j.dadm.2017.07.004.
15. K. Yamashita, M. Miura, S. Watanabe, K. Ishiki, Y. Arimatsu, J. Kawahira, T. Kubo, K. Sasaki, T. Arai, K. Hagino, Y. Irino, K. Nagai, D. Verbel, A. Koyama, S. Dhadda, H. Niiro, S. Iwanaga, T. Sato, T. Yoshida and A. Iwata, Alzheimer's Res. Ther., 2022, 14, 86,  DOI:10.1186/s13195-022-01029-0.
16. S. Janelidze, C. E. Teunissen, H. Zetterberg, J. A. Allué, L. Sarasa, U. Eichenlaub, T. Bittner, V. Ovod, I. M. W. Verberk, K. Toba, A. Nakamura, R. J. Bateman, K. Blennow and O. Hansson, JAMA Neurol., 2021, 78, 1375–1382,  DOI:10.1001/jamaneurol.2021.3180.
17. Y. Heo, K. Shin, M. C. Park and J. Y. Kang, Sci. Rep., 2021, 11, 5831,  DOI:10.1038/s41598-021-85107-7.
18. R. C. Petersen, G. E. Smith, S. C. Waring, R. J. Ivnik, E. G. Tangalos and E. Kokmen, Arch. Neurol., 1999, 56, 303–308.
19. G. M. McKhann, D. S. Knopman, H. Chertkow, B. T. Hyman, C. R. Jack, C. H. Kawas, W. E. Klunk, W. J. Koroshetz, J. J. Manly, R. Mayeux, R. C. Mohs, J. C. Morris, M. N. Rossor, P. Scheltens, M. C. Carrillo, B. Thies, S. Weintraub and C. H. Phelps, Alzheimer's Dementia, 2011, 7, 263–269,  DOI:10.1016/j.jalz.2011.03.005.
20. I. Feinkohl, C. G. Schipke, J. Kruppa, F. Menne, G. Winterer, T. Pischon and O. Peters, J. Alzheimer's Dis., 2020, 74, 1285–1294,  DOI:10.3233/JAD-200046.
21. O. Hanon, J. S. Vidal, S. Lehmann, S. Bombois, B. Allinquant, J. M. Tréluyer, P. Gelé, C. Delmaire, F. Blanc, J. F. Mangin, L. Buée, J. Touchon, J. Hugon, B. Vellas, E. Galbrun, A. Benetos, G. Berrut, E. Paillaud, D. Wallon, G. Castelnovo, L. Volpe-Gillot, M. Paccalin, P. H. Robert, O. Godefroy, T. Dantoine, V. Camus, J. Belmin, P. Vandel, J. L. Novella, E. Duron, A. S. Rigaud, S. Schraen-Maschke, A. Gabelle, O. Hanon, F. Blanc, Y. Boudali, M. L. Seux, H. Lenoir, C. Bayle, S. Bombois, X. Delbeuck, F. Moulin, F. Latour, M. Plichart, S. Pichierri, G. Orvoën, F. Labourée, P. Cassagnaud, C. Paquet, F. Lala, J. Dumurgier, A. S. S. Rigaud, C. Perret-Guillaume, E. Alonso, F. du Boisgueheneuc, L. Hugonot-Diener, A. Rollin-Sillaire, O. Martinaud, C. Boully, Y. Spivac, A. Devendeville, P. H. H. Robert, L. Caillard, D. Hannequin, N. Sastre, S. Haffen, A. Kearney-Schwartz, J. L. L. Novella, V. Deramecourt, V. Chauvire, G. Abitbol, N. Schwald, C. Hommet, F. Sellal, M. A. Cariot, M. Abdellaoui, S. Benisty, S. Gherabli, P. Anthony, F. Bloch, N. Charasz, S. Chauvelier, J. Y. Gaubert, G. Sacco, O. Guerin, J. Boddaert, M. A. Mackowiak, M. T. Rabus, V. Gissot, C. Picard, C. Guillemaud, C. Gervais, J. Hugon, J. M. Michel, J. P. David, M. Paulin, P. J. Ousset, S. Pariel, A. Chawakilian, L. Joffredo, A. C. Troussiere, C. Adam, D. Dupuy, H. Briault, I. Saulnier, K. Mondon, M. A. Picat, M. Laurent, R. Daheb, S. Libercier, D. Krabchi, M. Chupin, E. Chaussade and S. Schraen-Maschke, Alzheimer's Dement., 2018, 14, 858–868,  DOI:10.1016/j.jalz.2018.01.004.
22. K. L. Cox, V. Devanarayan, A. Kriauciunas, J. Manetta, C. Montrose and S. Sittampalam, n.d, Immunoassay methods immunoassay development, optimization and validation, in Assay Guidance Manual Flow Chart 2, 2012.
23. J. W. A. Findlay and R. F. Dillard, AAPS J., 2007, 9, E260–E267.
24. US Food and Drug Administration, Bioanalytical Method Validation Guidance for Industry Biopharmaceutics Bioanalytical Method Validation Guidance for Industry Biopharmaceutics Contains Nonbinding Recommendations, 2018.
25. J. Pannee, L. M. Shaw, M. Korecka, T. Waligorska, C. E. Teunissen, E. Stoops, H. M. J. Vanderstichele, K. Mauroo, I. M. W. Verberk, A. Keshavan, P. Pesini, L. Sarasa, M. Pascual-Lucas, N. Fandos, J. A. Allué, E. Portelius, U. Andreasson, R. Yoda, A. Nakamura, N. Kaneko, S. Y. Yang, H. C. Liu, S. Palme, T. Bittner, K. G. Mawuenyega, V. Ovod, J. Bollinger, R. J. Bateman, Y. Li, J. L. Dage, E. Stomrud, O. Hansson, J. M. Schott, K. Blennow and H. Zetterberg, Alzheimer's Dement., 2021, 13, e12242,  DOI:10.1002/dad2.12242.
26. K. V. Giudici, P. De Souto Barreto, S. Guyonnet, Y. Li, R. J. Bateman and B. Vellas, JAMA Netw. Open, 2020, 3, e2028634.
27. V. Pérez-Grijalba, J. Romero, P. Pesini, L. Sarasa, I. Monleón, I. San-José, J. Arbizu, P. Martínez-Lage, J. Munuera, A. Ruiz, L. Tárraga, M. Boada and M. Sarasa, J. Prev. Alzheimer's Dis., 2019, 6, 34–41,  DOI:10.14283/jpad.2018.41.
28. J. Therriault, S. E. Schindler, G. Salvadó, T. A. Pascoal, A. L. Benedet, N. J. Ashton, T. K. Karikari, L. Apostolova, M. E. Murray, I. Verberk, J. W. Vogel, R. La Joie, S. Gauthier, C. Teunissen, G. D. Rabinovici, H. Zetterberg, R. J. Bateman, P. Scheltens, K. Blennow, R. Sperling, O. Hansson, C. R. Jack and P. Rosa-Neto, Nat. Rev. Neurol., 2024, 20, 232–244,  DOI:10.1038/s41582-024-00942-2.
29. S. Palmqvist, S. Janelidze, E. Stomrud, H. Zetterberg, J. Karl, K. Zink, T. Bittner, N. Mattsson, U. Eichenlaub, K. Blennow and O. Hansson, JAMA Neurol., 2019, 76, 1060–1069,  DOI:10.1001/jamaneurol.2019.1632.
30. T. West, K. M. Kirmess, M. R. Meyer, M. S. Holubasch, S. S. Knapik, Y. Hu, J. H. Contois, E. N. Jackson, S. E. Harpstrite, R. J. Bateman, D. M. Holtzman, P. B. Verghese, I. Fogelman, J. B. Braunstein and K. E. Yarasheski, Mol. Neurodegener., 2021, 16, 30,  DOI:10.1186/s13024-021-00451-6.
31. A. Okello, J. Koivunen, P. Edison, H. A. Archer, F. E. Turkheimer, K. Någren, R. Bullock, Z. Walker, A. Kennedy, N. C. Fox, M. N. Rossor, J. O. Rinne and D. J. Brooks, Neurology, 2009, 73, 754–760.
32. L. F. Lue, A. Guerra and D. G. Walker, Neurol. Ther., 2017, 6(1), 25–36,  DOI:10.1007/s40120-017-0074-8.
33. C. Udeh-Momoh, B. Zheng, A. Sandebring-Matton, G. Novak, M. Kivipelto, L. Jönsson and L. Middleton, J. Prev. Alzheimer's Dis., 2022, 9, 12–21,  DOI:10.14283/jpad.2021.70.
34. K. Z. Revendova, D. Zeman, R. Bunganic, K. Karasova, O. Volny, M. Bar and P. Kusnierova, Mult. Scler. Relat. Disord., 2022, 67, 104177,  DOI:10.1016/j.msard.2022.104177.
35. D. H. Wilson, D. M. Rissin, C. W. Kan, D. R. Fournier, T. Piech, T. G. Campbell, R. E. Meyer, M. W. Fishburn, C. Cabrera, P. P. Patel, E. Frew, Y. Chen, L. Chang, E. P. Ferrell, V. von Einem, W. McGuigan, M. Reinhardt, H. Sayer, C. Vielsack and D. C. Duffy, J. Lab. Autom., 2016, 21, 533–547,  DOI:10.1177/2211068215589580.

---

Footnotes

† Electronic supplementary information (ESI) available. See DOI: https://doi.org/10.1039/d5lc00235d

‡ These authors contributed to this work equally.

---