Antibody-detecting resistive biosensor for protein biomarkers

Abstract

Proteins serve as important biomarkers for clinical diagnostics, and a simple detection method for point-of-care applications is highly desirable. We developed an antibody-detecting resistive biosensor that utilizes a conductive network of gold (Au) nanoparticles for the detection of protein biomarkers. The proposed biosensor comprises two electrodes bridged by a functional film comprising Au nanoparticles, a polymer matrix, and target-specific antibodies, and the nanoparticle concentration is selected within an experimentally identified percolation transition region to achieve high sensitivity to small changes in resistance. In experiments, the electrical response of the proposed biosensor showed reliable concentration-dependent behavior, consistent with charge-mediated modulation of electron transport within the conductive network. When applied to detecting urinary growth differentiation factor-15 (GDF-15) as the target protein, the proposed biosensor exhibited a detection range of 23.4–1500 pg/mL and a fast response time of approximately 30 s, which would allow the tracking of dynamic changes in protein concentration. These results demonstrate that the proposed biosensor provides a simple and effective method for detecting various biomarkers in real time. The sensing concept may be extendable to other antigen–antibody systems, providing a potential general platform for biomolecular sensing.

Introduction

Proteins play various important roles in living organisms and are central to antigen–antibody reactions and immune functions. Multiple diseases have been observed to be associated with abnormal protein concentrations^1–3. Consequently, the concentrations of protein biomarkers are a useful clinical diagnostic indicator. Antibody-based methods are frequently used to identify target-specific protein biomarkers and are widely employed for various techniques including Western blotting⁴, immunofluorescence⁵, immunoturbidimetry⁶, and enzyme-linked immunosorbent assay (ELISA)^7–9. ELISA is the most popular immunoassay used in clinical practice because of its high sensitivity and selectivity¹⁰, and various readout methods can be employed including colorimetric⁷, chemiluminescent⁸, and fluorescent⁹ approaches. However, ELISA is tedious and time-consuming, which limits its convenience. Another conventional approach to protein detection utilizes mass spectrometry to directly identify and quantify protein analytes based upon their mass-to-charge ratio¹¹. This approach exhibits high quantitative accuracy, specificity, and universal applicability across all organisms without requiring specific binding reagents¹². However, both immunoassays and mass spectrometry often require costly instrumentation and complex procedures that limit their applicability. Therefore, there is a strong demand for a simple, sensitive, and rapid method for protein quantification in point-of-care applications.

Biosensors are simple structures that integrate a biorecognition element with a transducer¹³. The design of the biorecognition element is flexible and can be an antibody, enzyme, nucleic acid, aptamer, or molecularly imprinted polymer¹⁴, which allows the detection of target analytes with high efficiency and specificity. The design of the transducer is also flexible and can take different forms, such as optical, mechanical, and electrochemical, each with its own advantages. Optical biosensors are characterized by high precision¹⁵ while mechanical biosensors are well-suited for rapid detection¹⁶, and both optical and mechanical biosensors permit label-free analysis. However, these biosensors often require complex design and fabrication processes that limit their practical applicability. Electrochemical biosensors are one of the most cost-effective options^17,18 and can be subdivided according to their measurement parameters into categories including amperometric biosensors¹⁹, potentiometric biosensors²⁰, and conductometric biosensors^21,22. Chemiresistive biosensors are another class of electrical biosensor in which biomolecular interactions are transduced as changes in resistance. Depending on the material system, the response can arise from modulation of surface charge, carrier density, or electron-transport pathways. In metal nanoparticle–organic composite films, charge transport proceeds by tunneling or hopping across the junctions between neighboring particles, so that the junction conductance depends exponentially on both the interparticle separation and the height of the potential barrier; this mechanism has been established as the basis of chemiresistive transduction since the introduction of the colloidal metal–insulator–metal ensemble sensor^23,24. Because these perturbations can be small, one strategy for enhancing transduction sensitivity is to confine electron transport to nanoscale channels or junctions, where minor changes in the local electrical environment can produce large resistance changes^25–27. Nanogap devices functionalized with recognition elements have been reported to transduce antigen–antibody binding into a concentration-dependent change in current²⁶. Such tunneling-based transduction is also not restricted to the gas phase: gold nanoparticle films have been operated as chemiresistors directly in high-conductivity aqueous electrolytes, by ensuring that the resistance of the nanoparticle film rather than the surrounding electrolyte dominates the measured impedance^28,29. However, precise nanogap electrodes require sophisticated nanofabrication techniques, which increases manufacturing complexity and limits practical implementation.

In our previous study³⁰, we demonstrated a chemiresistive biosensor based on a conductive network of platinum nanoparticles in which enzyme-mediated biochemical reactions were transduced into electrical signals. The conductive network was used to detect creatinine via the enzymatic generation of hydrogen peroxide, which led to measurable changes in electrical resistance. In the present study, we extend this sensing mechanism from enzyme-based detection to the detection of protein biomarkers based on antigen–antibody interactions. The antibody-detecting resistive biosensor employs a metallic nanoparticle network operated within the experimentally identified percolation transition region to achieve high sensitivity to small changes in resistance. Instead of relying on a single fabricated nanogap for electron transport, the proposed biosensor forms numerous nanoscale electron-transport junctions spontaneously within a metallic nanoparticle conductive network. Owing to its simple structure and electrical readout, the proposed biosensor offers a promising platform for the portable and convenient detection of protein biomarkers.

Results

Working principle

Figure 1a illustrates the antibody-detecting resistive biosensor, which comprises two interdigitated electrodes bridged with a functional layer (i.e., bioresistor). The bioresistor comprises 5-nm-diameter gold (Au) nanoparticles, target-specific antibodies, and a polyvinyl alcohol (PVA)-based polymer matrix. As shown in Fig. 1b, the proposed sensing model relates antigen–antibody binding events to changes in electrical resistance within the nanoparticle conductive network. When the nanoparticle concentration is selected within the experimentally identified percolation transition region (see Supplementary Note 1), the conductive network is expected to be highly sensitive to small variations in interparticle spacing and local charge distribution³¹. Under such conditions, molecular interactions associated with antigen–antibody binding can modulate electron transport pathways such as hopping or tunneling conduction between nanoparticles, as shown in Fig. 1c. Variations in the protein concentration affect the extent of antigen–antibody binding within the bioresistor, which in turn influences the effective interparticle separation and local electrostatic environment and thus affects electron transport processes.

Fig. 1. Working principle of the antibody-detecting resistive biosensor.

a Schematic diagram. The proposed biosensor comprises interdigitated electrodes coated with a biosensitive film containing the target-specific antibody and Au nanoparticles. b Sensing mechanism. The Au nanoparticle volume fraction was selected within the experimentally identified percolation transition region to form a conductive network that is sensitive to small perturbations in the local charge environment. c Antibody detection. Antigen–antibody binding events may induce local charge variations that can modify the potential barrier between Au nanoparticles and affect electron transport processes

Experimental validation

A series of experiments was conducted to analyze the impedance characteristics and concentration-dependent electrical response of the proposed biosensor. Urinary growth differentiation factor-15 (GDF-15) was selected as the target protein. GDF-15 is a divergent member of the transforming growth factor beta (TGF-β) superfamily³², and it has been demonstrated to exert multiple effects under physiological and pathological conditions³³, which suggests that it may be a valuable biomarker for healthcare applications. GDF-15 is conventionally detected by immunoassays^33,34, which typically require complex procedures and instrumentation.

Figure 2a shows the dimensions of the proposed biosensor. Figure 2b and c, respectively, present photographs of the electrodes before and after being coated with the bioresistor. Figure 2d shows the experimental setup. The electrode pads of the biosensor were connected to an AC impedance analyzer to evaluate the impedance characteristics or a Schmitt trigger oscillator to evaluate the electrical response. The target protein was transferred onto the bioresistor by using a pipette.

Fig. 2. Experimental setup.

a Dimensions of the proposed biosensor. The electrodes were deposited on a glass substrate. The sensing area corresponded to the region with an interdigitated pattern and measured 3.5 mm × 3.5 mm. The sensing area was covered by the bioresistor. b Photograph of the sensing area before being coated with the bioresistor. c Photograph of the sensing area after being coated with the bioresistor. d Experimental setup for evaluating the impedance characteristics (using an impedance analyzer) or electrical response (using a Schmitt trigger oscillator) of the proposed biosensor. A solution containing the target protein was dropped onto the sensing area by using a pipette

Impedance characteristics

Figure 3a shows Nyquist plots of the proposed biosensor in the high-frequency domain at protein concentrations of 23.4–1500 pg/mL and pH 7. The Nyquist plots reveal a characteristic semicircle followed by a linear impedance component, and the overall impedance dimensions decreased as the protein concentration increased. This behavior can be phenomenologically represented as an equivalent circuit by the R-CPE model³⁵, which comprises a resistor (R) in parallel with a constant-phase element (CPE) as illustrated in the inset. Because of the distortion of the semicircle at higher protein concentrations, a geometric method³⁶ was adopted as an alternative to equivalent-circuit fitting to estimate the resistance, which involved extrapolating the high-frequency semicircular component and the lower-frequency linear impedance component within the measured range. Their intersection with the real axis was used to determine the apparent resistance based on the conventional interpretation of Nyquist plots. Figure 3b shows that the effective resistance ($R_{eff}$) of the proposed biosensor clearly decreased as the protein concentration increased, which is consistent with enhanced electrical transport associated with target-protein binding. The impedance data alone, however, do not identify a unique microscopic transport mechanism.

Fig. 3. Impedance characteristics of the proposed biosensor.

a Nyquist plots obtained at different concentrations of the target protein. The inset shows the R-CPE model. b Dependence of the effective resistance $R_{eff}$ on the concentration of the target protein. Each data point represents the mean of measurements obtained by five separate biosensors

Concentration-dependent electrical response

The electrical response of the biosensor to variations in the protein concentration was evaluated by using a Schmitt trigger oscillator. A higher oscillation frequency is associated with lower resistance. The sensing area was cleaned with deionized (DI) water and dried with a blower after which prepared protein solutions with concentrations of 23.4–1500 pg/mL were deposited onto the sensing area and allowed to remain for 100 s. Measurements were repeated five times for each protein concentration. Because the bioresistor was formed by drop-casting a fixed volume of solution onto the sensing area, variations in film morphology and thickness were expected, which may result in device-to-device variation in the biosensor performance. Therefore, the initial frequency response in DI water was used as an internal reference for each device, thereby reducing the influence of device-to-device variation and improving the consistency of the measured sensing response. For practical implementation and scale-up, further reduction in baseline variation should be achievable by replacing manual drop-casting with standardized bioresistor deposition methods such as automated dispensing or printing.

Figure 4a shows the oscillation frequency response of the proposed biosensor according to the protein concentration at pH 7. The frequency response was consistent following alignment in DI water, decreased immediately after surface cleaning, and then rapidly rose when the protein solution was applied, showing a rapid electrical response upon application of the target-protein solution. Together with the antibody-free control described below, this behavior is consistent with antibody-dependent protein recognition affecting the bioresistor response. The inset of Fig. 4a illustrates a magnified view of the frequency response for the final 10 s. Higher protein concentrations were clearly associated with a higher frequency.

Fig. 4. Concentration-dependent electrical response of the proposed biosensor according to the protein concentration.

a Oscillation frequency response to the protein concentration over time. The inset shows the frequency response for the last 10 s of measurements. b Frequency response to the protein concentration of biosensors prepared with and without antibodies at pH 7. c Frequency response to the protein concentration of biosensors prepared with and without antibodies at pH 8

To confirm that the frequency response of the proposed biosensor can be attributed to protein binding, the same experimental procedure was conducted for biosensors prepared without antibodies. Measurements were repeated three times for each protein concentration. Figure 4b compares the frequency responses to the protein concentration with and without antibodies at pH 7. Biosensors containing antibodies demonstrated an increase in frequency with the protein concentration. In contrast, biosensors without antibodies showed no notable dependence on the protein concentration, which indicates that the frequency response is influenced by protein binding. The increase in frequency with the protein concentration indicates enhanced conductivity, which is consistent with the AC impedance measurements showing a reduction in resistance at higher protein concentrations.

To evaluate the effects of the solution pH, Fig. 4c shows the frequency responses to the protein concentration of biosensors with and without antibodies at pH 8. No notable differences were observed. Independent trends in response to the protein concentration were exhibited by both biosensors with and without antibodies. However, the baseline frequency at pH 8 was marginally higher than at pH 7. This is likely because a higher solution pH increases the ionic concentration, which in turn increases the background conductivity.

Real-time response

To evaluate the real-time response of the proposed biosensor to the protein concentration, the same biosensor was used to measure a series of protein solutions. Each concentration was measured for 100 s. The surface of the biosensor was cleaned before each measurement. Prior to the initial protein solution, a baseline measurement was conducted in DI water to provide a reference. Figure 5a shows the oscillation frequency response as a function of the protein concentration at pH 7. The frequency clearly increased with the protein concentration. Figure 5b shows the frequency response at pH 8, which indicates that increasing the solution pH reduced the dependence of the frequency on the protein concentration. To ensure a reliable comparison between the frequency response and protein concentration under these conditions, the mean frequency response at each concentration was calculated by averaging the frequency values over the final 10 s of measurement. Figure 5c shows that consistent trends were observed between static and real-time measurements, which indicates that the proposed biosensor can respond to dynamic changes in the protein concentration.

Fig. 5. Real-time electrical response of the proposed biosensor according to the protein concentration.

a Oscillation frequency response at pH 7. b Frequency response at pH 8. c Averaged frequency response at each concentration calculated from the final 10 s of measurement

Discussion

The proposed biosensor exhibited consistent responses across different measurement methods, with an increase in oscillation frequency and a decrease in effective resistance as the concentration of the target protein increased at pH 7. These observations are consistent with antigen–antibody binding perturbing the conductive network, but they do not by themselves directly resolve the microscopic transport mechanism. In contrast, the response showed little dependence on the protein concentration at pH 8, which is likely due to the reduced charge of the target protein near its isoelectric point (pI) and the increased background conductivity from the buffer solution. GDF-15 was estimated to have a theoretical pI of approximately 8.17 using the ExPASy ProtParam tool³⁷, which suggests that it carries a slightly positive charge at pH 7. Such charge effects may influence local electron transport within the conductive network and contribute to the observed resistance change.

The observed decrease in effective resistance despite the presence of relatively large protein molecules can be interpreted within the proposed model as arising from the combined effects of percolation sensitivity and charge-mediated modulation of electron transport (see Suppl. Note 2). Within the percolation transition region, electrical conduction can be dominated by a limited number of critical pathways³⁸. Small perturbations such as local charge effects induced by protein binding can greatly alter these pathways. In addition, the presence of charged biomolecules may reduce the effective tunneling barrier or modify the local potential landscape³⁹ to facilitate electron transport across the conductive network⁴⁰. Within this model, these effects could outweigh the expected increase in resistance from steric hindrance. The electrical conductance $G$ can be approximated as a function of the protein concentration $C$ (see Suppl. Note 2):

$$G\left(C\right)\approx G_{base}+Sln\left(1+KC\right),$$ 1

where $G_{base}$ is the baseline conductance, $S$ is the logarithmic sensitivity coefficient, and $K$ is the effective association constant. Within the proposed model, increasing antigen–antibody binding can increase conductance when charge-mediated barrier modulation outweighs steric hindrance. In contrast, the protein is near its pI and has a reduced net charge at pH 8, which diminishes its influence on the biosensor response. Furthermore, the higher ionic strength at pH 8 can increase the background response and electrostatic screening, reducing the relative contribution of binding-induced perturbations. The small differences observed between biosensors with and without antibodies at pH 8 may arise from device-to-device variation and matrix-dependent contributions in addition to any biomolecular effects. These observations are therefore consistent with stronger charge-mediated sensitivity at pH 7 and greater masking by background and matrix effects at pH 8, rather than demonstrating a complete change in the underlying transport mechanism.

Additional experiments were performed to directly examine the role of Au nanoparticle network connectivity in the sensing response. Biosensors with Au nanoparticle volume fractions spanning 0.02–0.10% showed a non-monotonic dependence of the GDF-15 response on nanoparticle loading: the concentration-dependent response was largest at 0.07%, within the experimentally identified percolation transition region, and was markedly suppressed at 0.10% (see Suppl. Note 6 and Fig. S5 in the Supplementary Information). This result provides experimental support for percolation-enhanced electrical transduction, while not by itself identifying tunneling or hopping as the unique microscopic transport mechanism.

The total electrical response can include electron transport through the conductive network together with ionic and dielectric contributions from the surrounding hydrated medium. Under the present impedance and oscillator-based measurement conditions, slow ionic diffusion is expected to contribute less strongly to the measured response, particularly in the high-frequency impedance measurements, making modulation of transport within the nanoparticle network a plausible major contribution⁴¹. At pH 8, the increased ionic strength and electrostatic screening can increase the background response and reduce the relative contribution of binding-induced perturbations, thereby masking the concentration dependence. Thus, the loss of concentration dependence at pH 8 does not necessarily indicate the absence of antigen–antibody binding. For practical applications, variations in pH and ionic conductivity could be mitigated using a differential configuration in which a target-specific sensing region is paired with a reference region carrying a nonbinding control antibody, or otherwise lacking target-specific recognition, with both regions exposed to the same sample. Differential readout would help suppress common-mode changes caused by the sample matrix. The AC impedance measurements likewise showed a decrease in the effective resistance with increasing protein concentration at pH 7, which is consistent with enhanced electrical transport within the conductive network.

Since the two electrodes are bridged by the bioresistor and the exposed metal outside the sensing region is insulated from the solution, the measured response is less likely to be dominated by bulk electrolyte conduction. In addition, all measurements at a given pH were performed using the same supporting electrolyte while only the GDF-15 concentration was varied. Because the GDF-15 concentration is much lower than the concentration of ionic species in the supporting electrolyte, direct changes in bulk electrolyte conductivity caused by the added GDF-15 are expected to be small. Nevertheless, ionic transport within the hydrated polymer, electrical double-layer effects, dielectric polarization, and polymer swelling may contribute to the measured response by altering the local environment of the conductive network. The present experiments do not completely separate these contributions. Taken together, the high-frequency impedance response, pH dependence, and target/control measurements are consistent with charge-mediated modulation of transport within the Au nanoparticle network, while secondary ionic, dielectric, and polymer-related contributions cannot be completely excluded.

Additional dry-state I–V measurements showed that measurable electrical conduction persisted after removal of the liquid electrolyte (see Suppl. Note 7 and Fig. S6 in the Supplementary Information). This observation indicates that the device response cannot be attributed solely to bulk ionic conduction through the surrounding solution and supports an intrinsic conduction pathway within the Au nanoparticle/polymer bioresistor. However, the dry-state I–V curves were nonlinear and hysteretic; therefore, these measurements do not uniquely establish direct tunneling, and charge trapping, dielectric polarization, or residual mobile species within the dried film may also contribute.

The transport picture adopted here is also consistent with the behavior previously reported for related systems. Interparticle tunneling has long been established as the dominant conduction mechanism in metal nanoparticle–organic composite films used as chemiresistors^23,24, and such films have been operated directly in high-conductivity aqueous electrolytes, indicating that transport within a nanoparticle network can dominate the measured response even in ionic media^28,29. The direction of the observed response is likewise not anomalous: analyte-induced decreases in resistance have been reported for gold nanocluster chemiresistors, in which the sign of the response depends on the nanoscale dimensions of the film⁴². In addition, nanogap devices functionalized with antibodies exhibit an increase in current upon antigen–antibody binding that scales with the target concentration²⁶, and solid-state protein junctions have been shown to support efficient electron transport by tunneling^43–45. A bound protein, therefore, need not act solely as an insulating spacer that increases the effective interparticle separation. Within this context, the decrease in effective resistance observed here with increasing GDF-15 concentration is consistent with, rather than contrary to, the reported behavior of tunneling-mediated nanoscale junctions that incorporate biomolecular recognition. We note that agreement with these previous reports supports the plausibility of the proposed model, but does not by itself constitute direct microscopic verification, which will require further mechanistic study.

Table 1 compares the detection performance of the proposed biosensor with other methods used in the literature to detect GDF-15. Additional analytical parameters were evaluated using the independent measurements performed at pH 7. The calibration fit gave R² = 0.9647 with a slope of 0.0374 kHz/(pg/mL), which was used as the sensitivity over the tested concentration range. Because the present assay was operated in a single-use format without a regeneration step, each concentration was evaluated using independently fabricated biosensors. Inter-device reproducibility was assessed from the coefficient of variation (CV), which ranged from 1.0% to 4.2% under the tested conditions. A statistically valid blank-based limit of detection was not determined because a zero-analyte blank series was not acquired under the same assay protocol; therefore, an extrapolated LOD is not reported. The proposed biosensor exhibited a detection range comparable to existing methods while covering the physiological range for healthy adults (200–1200 pg/mL)⁴⁶. It also demonstrated a rapid response time of approximately 30 s facilitated by a simple oscillator-based readout. The sensing mechanism does not require labeling, which simplifies the detection procedure compared to conventional immunoassays. These features highlight its suitability for point-of-care diagnostics⁴⁷. To further compare the proposed biosensor with representative chemiresistive biosensors, Table 2 summarizes several reported chemiresistive biosensing mechanisms. The proposed biosensor differs from these representative approaches in that the sensing signal is associated with perturbation of electron transport within a metallic nanoparticle conductive network operated within the experimentally identified percolation transition region. The performance of the proposed biosensor was also tested in artificial urine (see Suppl. Note 4). A concentration-dependent trend was retained, but the higher baseline and larger variability relative to buffer indicated matrix effects; accordingly, these data are interpreted as a matrix-compatibility and reproducibility evaluation rather than a quantitative recovery study. Selectivity was further evaluated using human serum albumin (HSA) and immunoglobulin G (IgG) as representative non-target proteins (see Supplementary Note 5 and Fig. S4 in the Supplementary Information). HSA and IgG were each tested at four concentrations spanning 23.4–1500 pg/mL using the same real-time protocol, and for each sensor the initial DI-water response was recorded and used as that sensor’s baseline reference. In contrast to GDF-15, neither HSA nor IgG showed a comparable concentration-dependent frequency response over the tested range. These results support the contribution of target-specific antigen–antibody recognition to the sensing response under the tested conditions and indicate that selectivity is more clearly reflected in the concentration dependence than in the absolute baseline frequency. Further improvement in analytical selectivity could be pursued through optimization of antibody immobilization, suppression of nonspecific adsorption, and differential/reference sensing. The proposed platform may also be extended to other biomarkers by changing the biorecognition element.

Table 1.

Comparison of the proposed biosensor to other GDF-15 detection methods reported in the literature

Detection method	Detection range (pg/mL)	Response time (min)	Calibration R²	Inter-device CV (%)	Label required	Ref.
BLI-ELASA	10–810	30	0.996	4.4–11.3	Yes	51
FLISA	6–1600	N/A	0.9958	2.7	Yes	52
TRFIA	1000–200,000	60	0.9996	3.3–4.5	Yes	53
Electrochemical	1.5–1500	1.6	0.9898	1.9–4.2	Yes	54
Electrochemical	140–10,000	75	0.996	4.3	Yes	55
Proposed biosensor	23.4–1500	~0.5	0.9647	1.0–4.2	No	-

The bold values in Table 1 indicate the performance parameters of the proposed biosensor developed in the present study, in order to distinguish them from those of previously reported GDF-15 detection methods. The bold font does not indicate statistical significance

Table 2.

Comparison of representative chemiresistive biosensors with different transduction mechanisms for biomarker detection

Chemiresistor material	Target analyte	Bio-recognition reaction	Detection mechanism	Ref.
Calcium-doped cupric oxide	C-reactive protein	Antigen-antibody	Electrostatic surface charge modulation	56
Hydrothermally synthesized vanadium disulfide nanowires	Metalloproteinase-9	Antigen-antibody	Charge distribution modulation	57
Graphene	Influenza virus nucleoprotein	Antigen-antibody (Enzyme-labeled)	Enzymatic n-doping	58
Polystyrene sulfonate	Urea	Enzymatic	Ionic conductivity modulation	59
AuNP network	GDF-15	Antigen-antibody	Nanoparticle transport modulation (proposed)	-

Conclusion

The proposed antibody-detecting resistive biosensor integrates a simple two-electrode structure with a bioresistor comprising an Au nanoparticle conductive network, a polymer matrix, and antibodies. In experiments, the biosensor exhibited a clear concentration-dependent response at pH 7, while the response became weakly dependent on protein concentration at pH 8. The experimental observations are consistent with charge-mediated modulation of transport within the conductive network combined with sensitivity associated with operation in the percolation transition region, although ionic, dielectric, and polymer-related contributions cannot be completely excluded. The biosensor demonstrated a detection range covering physiological concentrations and a rapid response time of approximately 30 s. Additional tests using the non-target proteins HSA and IgG did not show comparable concentration-dependent responses, supporting selectivity toward GDF-15 under the tested conditions. The use of an oscillator-based readout enables a simple and compact measurement system. The present study provides a physically motivated sensing model that accounts for the observed experimental behavior and highlights the potential of nanoparticle conductive networks as a versatile platform for biomolecular sensing and simple real-time diagnostic devices. Nevertheless, further studies are required to directly verify the underlying microscopic transport mechanism.

Methods

Materials

All chemicals were prepared in DI-distilled water. GDF-15 powder and GDF-15 antibody were purchased from Bio-Techne (R&D Systems). Polyvinylpyrrolidone (PVP)-coated Au nanoparticle solution was purchased from Prexnet (Tanaka Kikinzoku Kogyo Co., Ltd.). PVA-based polymer (Biosurfine AWP-MRH) was purchased from Toyo Gosei company. Glutaraldehyde solution (GA), potassium dihydrogen phosphate, and disodium hydrogen phosphate were purchased from FUJIFILM Wako Pure Chemical Corporation. To evaluate the biosensor’s protein detection performance, a series of solutions with varying protein concentrations was prepared by mixing fixed amounts of GDF-15 powder with a fixed volume of phosphate buffer solution. The phosphate buffer solution was prepared by mixing solutions of 0.1 M potassium dihydrogen phosphate and 0.1 M disodium hydrogen phosphate. The proportions of the two component solutions were adjusted to achieve solution pH values of pH 7 and pH 8.

Preparation of the bioresistor

The bioresistor was prepared by mixing 10 µL of Au nanoparticle solution, 14 µL of PVA-based polymer, 10 µL of GDF-15 antibody, and 5 µL of distilled water. This ratio was selected based on electrical resistance measurements of bioresistors prepared with different Au nanoparticle volume fractions, which identified a percolation transition region (see Suppl. Note 1). The mixture was manually homogenized to prevent potential damage to the antibody. For comparison, a bioresistor lacking GDF-15 antibody was also prepared following the same process except that the GDF-15 antibody was replaced by an equivalent quantity of DI water.

Fabrication of the proposed biosensor

The proposed biosensor was fabricated by using standard photolithography techniques. Chromium–gold (Cr–Au) films with thicknesses of 20 and 200 nm, respectively, were deposited on a 300 µm-thick glass substrate via sputtering to serve as the electrodes. A positive photoresist (OFPR 34 cp) was spin-coated onto the Cr–Au films. An interdigitated pattern was achieved for the sensing area by wet etching. Electrical connections were made by using conductive paste (silver epoxy). The sensing area was formed by dropping 2.5 µL of bioresistor solution on the interdigitated pattern with a pipette. The sensing area was placed on a hotplate at 60 °C to dry the biosensor solution and form a thin film. The sensing area was then covered with a low concentration (0.0625%) of glutaraldehyde solution to stabilize the bioresistor and immobilize the biomolecules⁴⁸. To prevent undesired electrical or chemical interactions, all electrodes and metal lines outside the sensing area were covered with insulated glue. The sensing area had dimensions of 3.5 mm × 3.5 mm.

AC impedance measurement

Impedance spectroscopy was conducted using an impedance analyzer (Yokogawa Hewlett-Packard Co., 4194 A) to evaluate the response of the biosensor to the protein concentration. Measurements were conducted from 10 to 40 MHz with an AC amplitude of 10 mV and DC bias of 0 V. To minimize the influence of parasitic impedance components, including stray capacitance and inductance at high frequencies⁴⁹, a normalization procedure was applied that involved correcting negative impedance contributions and removing series parasitic components by extrapolating the impedance response toward the real axis to provide a consistent baseline for analysis.

Schmitt trigger oscillator measurement

A Schmitt trigger oscillator (12YCOS-1, Taiyo Mechanical Co., Ltd.) was used to convert changes in impedance into frequency signals⁵⁰. The oscillation frequency is governed by the effective resistance of the biosensor, which allows for indirect measurement of changes in conductivity (see Suppl. Note 3). The oscillation frequency is primarily determined by the resistor–capacitor (RC) time constant of the circuit, where the biosensor acts as a variable resistive element.

Author contributions

Y.K. and R.T. performed most experiments. Y.K. and T.O. analyzed the data, T.A. and I.V. proposed the experimental design, N.T. and I.V. analyzed the data, and Y.K. and T.O. wrote the manuscript with comments from all authors.

Data availability

All data that support the results of this study are available from the corresponding author upon request.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41378-026-01465-0.