Clinically accurate diagnosis of Alzheimer's disease via multiplexed sensing of core biomarkers in human plasma

Abstract

Alzheimer's disease (AD) is the most prevalent neurodegenerative disorder, affecting one in ten people aged over 65 years. Despite the severity of the disease, early diagnosis of AD is still challenging due to the low accuracy or high cost of neuropsychological tests and neuroimaging. Here we report clinically accurate and ultrasensitive detection of multiple AD core biomarkers (t-tau, p-tau₁₈₁, Aβ₄₂, and Aβ₄₀) in human plasma using densely aligned carbon nanotubes (CNTs). The closely packed and unidirectionally aligned CNT sensor array exhibits high precision, sensitivity, and accuracy, evidenced by a low coefficient of variation (<6%), a femtomolar-level limit of detection, and a high degree of recovery (>93.0%). By measuring the levels of t-tau/Aβ₄₂, p-tau₁₈₁/Aβ₄₂, and Aβ₄₂/Aβ₄₀ in clinical blood samples, the sensor array successfully discriminates the clinically diagnosed AD patients from healthy controls with an average sensitivity of 90.0%, a selectivity of 90.0%, and an average accuracy of 88.6%.

Fig. 1. Schematic illustration of a densely aligned CNT sensor array for AD biomarkers.

The unidirectional aligned and highly dense CNTs facilitate femtomolar sensitivity and high sensor-to-sensor reliability. The densely aligned CNT sensor array exhibited diagnostic accuracy of ~88.6% in discriminating AD patients from healthy individuals.

Fig. 2. Characterization of densely aligned CNT sensor array.

a SEM and b AFM images of densely aligned CNT film. The scale bars in SEM and AFM images indicate 250 nm and 500 nm, respectively. c Polarized Raman spectra of the densely aligned CNT film recorded at various angles between 633 nm incident laser and an alignment direction of CNT film. The inset shows the angular dependence of the Raman intensity at 1595 cm⁻¹. The blue solid line fits cos²α function. d Sensing performance of densely aligned CNT device in comparison with random-network CNT devices that have similar CNT densities. The densities of densely aligned and random-network CNT samples were 345 and 339 CNTs μm⁻¹, respectively. For each data point, a different set of devices were used. Data reproducibility was confirmed by two additional experiments. All reported values represent the mean ± SD. e The values of coefficient of variation (CV) of resistance change in the densely aligned and random-network CNT device array. The CV of densely aligned device array was approximately six times lower than that in random-network CNT device array. Source data are provided as a Source Data file.

Fig. 3. Densely aligned CNT sensor arrays’ sensitivities toward AD core biomarkers.

Changes in resistance of the densely aligned CNT sensor array with the increasing concentration of (a) Aβ₄₂, (b) Aβ₄₀, (c) t-tau, and (d) p-tau. Each data point was attained using a different set of devices. The human IgG was used as a negative control. e Selectivity of the densely aligned CNT sensor array toward various individual AD biomarkers and their mixtures in human plasma. The concentrations of Aβ₄₂ and Aβ₄₀ were 22.2 and 23.1 fM, respectively. In the case of t-tau and p-tau, the concentrations were 21.8 fM and 360 fM, respectively. The measurement was performed in triplicate and all reported values represent the mean ± SD. Source data are provided as a Source Data file.

Fig. 4. Clinical relevance of the densely aligned CNT sensor array.

Waterfall plots and box plots showing the estimated levels of (a, b) Aβ₄₂/Aβ₄₀, (c, d) t-tau/Aβ₄₂, and (e, f) p-tau/Aβ₄₂ in the plasma of AD patients (n = 20) and healthy controls (n = 20). Statistical analysis was carried out by means of one-way analysis of variance (ANOVA). ****p < 0.000001. In the boxes, the 25th, 50th (median), and 75th percentiles of the data are indicated. The whiskers represent mean ± 1.5 SD. g Receiver operating characteristic curves of the sensor array when composite biomarkers (Aβ₄₂/Aβ₄₀, t-tau/Aβ₄₂, and p-tau/Aβ₄₂) and single biomarker (Aβ₄₂, t-tau, p-tau) were used as a predictor, respectively. Source data are provided as a Source Data file.