Development of a Novel Insulin Sensor for Clinical Decision-Making

Abstract

Background:

Clinical decision support systems that incorporate information from frequent insulin measurements to enhance individualized diabetes management remain an unmet goal. The development of a disposable insulin strip for fast decentralized point-of-care detection replacing the current centralized lab-based methods used in clinical practice would be highly desirable to improve the establishment of individual insulin absorption patterns and algorithm modeling processes.

Methods:

We carried out the development and optimization of a novel decentralized disposable insulin electrochemical sensor focusing on obtaining high analytical and operational performance toward achieving a true point-of-care insulin testing device for clinical on-site application.

Results:

Our novel insulin immunosensor demonstrated an attractive performance and efficient user-friendly operation by providing high sensitivity capability to detect endogenous and analog insulin with a limit of detection of 30.2 pM (4.3 µiU/mL), rapid time-to-result, stability toward remote site application, and scalable low-cost fabrication with an estimated cost-of-goods for disposable consumables of below $5, capable of near real-time insulin detection in a microliter (≤10 µL) sample droplet of undiluted serum within 30 minutes.

Conclusions:

The results obtained in the optimization and characterization of our novel insulin sensor illustrate its suitability for its potential application in remote clinical environments for frequent insulin monitoring. Future work will test the insulin sensor in a clinical research setting to assess its efficacy in individuals with type 1 diabetes.

Introduction

Diabetes is a complex chronic medical condition that requires the undivided attention of both clinician and patient. Today, individualization of diabetes treatment for each person is recognized as a key therapeutic goal, and requires innovation by the research community. ¹ Clinical decision support systems (CDSS) based on machine learning and algorithm modeling for enhanced patient-tailored diabetes management approaches have been brought to the forefront of diabetes care and will become essential to guide personalized glycemic control platforms such as automated insulin delivery (AID) systems.^{2 -5} People with type 1 diabetes (T1D) require lifelong replacement of insulin to maintain glucose in a healthy range, to prevent diabetes-related complications, and to maintain overall health. Due to the narrow therapeutic window for insulin replacement therapy, individuals with T1D benefit from frequent monitoring of blood glucose levels to minimize both hyperglycemia and hypoglycemia.

Many other metabolites and hormones, beyond glucose, can be disrupted by T1D or its treatment. Multibiomarker-based monitoring of events and metabolic changes induced by events of daily living have the potential to prevent potential hypo/hyperglycemic events, and thus, perspective has boosted the development of CDSS.^1,5,6 For example, insulin dosing for meals needs to be accurately estimated while considering multiple factors, such as the insulin-to-carbohydrate ratio, meal size and macronutrient composition, insulin delivery history, blood glucose measurements, and correction factor. While these factors and their integration are challenging to discern, other individual-specific factors, such as stress, illness condition, and physical activity, can also impact patient’s insulin sensitivity and thus modulate insulin bolus needs. ⁶ For years, blood glucose measurement-based CDSS have been successfully implemented in clinical practice, thanks to major advances in continuous glucose monitoring and decentralized self-testing technology. In addition, bolus calculators rely also on the estimated active insulin on board (IOB), which is determined through the approximation of insulin decay curves. This approach provides only a rudimentary approximation of IOB, as it does not rely on actual insulin levels which may reflect interindividual variation in insulin clearance. Currently available methods for measurement of insulin rely on centralized lab-based enzyme-linked immunosorbent assay (ELISA) methodologies whose sample-to-result time can take days to weeks; this delay hinders the analysis of personal insulin variation profiles. Thus, new approaches based on mobile, low-cost, easy-to-use, rapid, and reliable technology allowing rapid data acquisition for frequent insulin measurements are urgently required. ⁴ To meet this need, a disposable insulin strip for fast decentralized point-of-care (POC) detection compatible with a handheld compact meter would be valuable.

Electrochemical sensors combining advanced detection and fabrication methods have resulted in potential POC solutions, with distinct advantages such as miniaturization, inexpensive consumables, portability, and nonexpert workflows. Recent efforts related to decentralized diabetes monitoring have led to the development of two electrochemical dual-analyte biosensor platforms for simultaneous detection of insulin along with glucose ⁷ and cortisol ⁸ in undiluted body fluids. Although there are also reports describing electrochemical in vitro insulin chip approaches,^{9 -18} relevant operational features for clinical on-site application, such as stability outside controlled lab settings or sensitivity for detection of insulin analogs, were not explored in these cases. Furthermore, complexity of technology, sensor fabrication, and assay analysis represent significant obstacles to their future practical implementation as low cost, scalable POC insulin tests.

In this article, we report on the development, optimization, and attractive performance of a novel disposable insulin electrochemical sensor that can provide near real-time insulin detection in a microliter sample of undiluted serum using a microchip-based immunoassay within 30 minutes (Figure 1). The microchip insulin-sensing platform was designed to achieve not only efficient user-friendly operation but also high sensitivity to reliably detect insulin and insulin analogs in human serum, along with rapid time-to-result, stability, and scalable low-cost fabrication to achieve a potential POC sensor prototype ready for routine use in serum insulin determinations in individuals with T1D.

Figure 1.

Schematic of the novel insulin sensor chip showing the sensor fabrication and electrochemical assay workflow and operation. (a) Images of the chips array and the single sensor comprising two gold-sputtered electrodes, one employed as working electrode (WE) and one as joint reference/counter (RE/CE) electrode. (b) Schematic of the insulin immunosensor and amperometric transduction principle: HRP-tagged sandwich immunoassay and the electrochemical reaction of the H₂O₂/HRP/TMB redox probe that generates a reduction current. (c) Schematic of the insulin analysis protocol, with direct one-step sandwich-type immunoassay for insulin antibodies/antigen binding in a serum sample droplet and representative chronoamperometric measurement for timely insulin levels tracking.

Materials and Methods

Equipment and Instruments

All electrochemical measurements were performed at room temperature using homemade metal-sputtered two-electrode chips. ⁸ Chronoamperometric measurements were realized using a “EmStat3 Blue” potentiometer (Palmsens, The Netherlands) controlled by “PSTrace” software version 5.8., and performed in 20 µL of commercial 3,3′,5,5′-tetramethylbenzidine (TMB) solution, applying −0.1 V for 150 seconds. A Barnstead Thermolyne (Type 16700 Mixer) vortex, and pH/ISE meter (Orion Star A214 model) were used. Reagents used in the assay are listed in the Supplemental Material.

Insulin Sensor Fabrication and Modification

The sensor chips were batch-fabricated as electrode arrays using a previously developed proprietary photolithography-free masking method,^7,8 with each sensor consisting of two rectangular 21-mm² area electrodes on a plastic substrate to generate the working electrode (WE) and a joint reference/counter electrode (RE/CE) (Figure 1a). For each single chip, the WE was modified by immobilizing commercial-grade insulin-specific capture antibodies (CAb) onto the self-assembled monolayer-functionalized gold surface and performing an electrode surface deactivation reaction to prevent nonspecific binding. More detailed fabrication and electrode modification procedures can be found in the Supplemental Information (Figure S1).

Sensor and Assay Protocol Optimization

The optimization of the insulin assay was performed using recombinant human insulin (Sigma-Aldrich) as the representative insulin peptide standard. Testing of the immunoassay protocol was performed in undiluted serum samples to achieve the best sensitivity in this body fluid while conducting short sample analysis times. All parameters impacting the preparation and bioassay protocols of the insulin sensor were systematically evaluated in both buffer and serum solutions using chronoamperometry and are described in detail in the Supplemental Information (Table S1).

Final Insulin Immunosensor Analysis Protocol

The immunoassay protocol optimized for serum insulin detection is illustrated in Figure 1. First, a microliter droplet (2.5 or 10 µL) of serum sample nonspiked or spiked with recombinant human insulin standard and supplemented with horseradish peroxidase (HRP)-labeled anti-insulin detector antibody (HRP-DAb) solution in a final concentration of 1 µg/mL was incubated for 30 minutes onto the CAb-modified WE, to sandwich the insulin peptide between the CAb and the HRP-DAb. Then, a 0.05% sodium dodecyl sulfate (SDS) washing step (repeated three times), and another PB washing step (repeated three times) was used to stop the immunoreaction and remove unbound HRP-DAb from the electrode surface. Upon drying with compressed air, the sensor was connected to an “EmStat3 Blue” potentiometer, attached to a laptop running “PSTrace” software (Palmsens, The Netherlands). A 20-µL drop of TMB/H₂O₂ was then placed over both WE and RE/CE electrodes, and a cathodic current of −0.1 V was applied for 2.5 minutes to record the amperometric signal. The current registered from the potentiometer corresponded to the reduction of the oxidized TMB, generated by the coupled reduction of H₂O₂ catalyzed by the HRP tag on the DAb, which was directly proportional to the amount of the HRP-tagged DAb bound to the sensor, corresponding to the insulin level in the sample.

Analytical and Operational Characteristics

We evaluated the performance of the insulin sensor by measuring ultralow pM insulin increments first in buffer, and then in serum samples, to assess whether the sensitivity provided by the immunosensor would be suitable for monitoring of insulin levels in individuals with T1D. Increasing concentrations ranged from 0 to 400 pM in increments of 50 pM were detected in 0.1 M phosphate buffer solution (PBT) supplemented with 0.02% Tween-20 buffer solutions, and 250-pM increments were measured in the concentration range between 0 and 1000 pM, along with 100-pM increments from 0 to 400 pM in serum samples. These calibration curves were constructed by spiking recombinant human insulin in increasing concentrations into buffer or into commercial human serum and were employed to estimate the limit of detection (LOD) values provided by the insulin sensor in each matrix.

The reproducibility of the optimized insulin immunoassay, reflecting reliability of the immunosensor’s fabrication and analysis protocols that will ensure reproducible insulin measurements, was assessed by comparing the chronoamperometric measurements obtained from six optimized insulin immunosensors, batch-fabricated and tested using 200-pM insulin standards prepared in buffer (n = six replicates) and serum samples spiked with 100-pM insulin standard (n = six replicates).

The storage stability of the chips modified with CAb illustrates the large-scale fabrication capability of the technology aiming at a ready-to-use disposable insulin strip. This property was evaluated by batch-fabricating sensors and attaching the CAb to the WE surface, then storing them at 4°C and frequently testing the sensors over 30 days. Assays were conducted in triplicate with 200-pM insulin spiked into PBT buffer, and amperometric measurements were recorded and averaged. Furthermore, the stability of the CAb-modified immunostrips at room temperature, illustrating their possible use hours after the termination of refrigerated storage and thus their ability to adapt to the unpredictable workflow of a clinical site, was evaluated by comparing the calibration slopes obtained in the measurements in triplicate of 0-, 200-, and 400-pM insulin standards prepared in PBT buffer, realized the day of fabrication and after 24 hours. In addition, to determine the stability of the CAb-modified immunostrips for shipping to remote clinical sites and field testing, the amperometric responses and signal-to-blank (S/B) ratios obtained in the measurements in triplicate of 0- and 200-pM insulin standards prepared in PBT buffer were assessed on both the day of fabrication and after 7 days of shipping transit and receiving back to our laboratory in UC San Diego.

The feasibility of using lower sample volume on a reduced WE surface area for serum incubations, which would add convenience to the test for the patient and practitioners by reducing the required sample volume collected for analysis, was assessed by comparing the S/B ratios obtained in the measurement in triplicate using three sensor chips of 0- and 400-pM human insulin standard spiked into serum samples for each volume (10, 5, and 2.5 µL) or area (21, 15.75, and 10.5 mm²) tested.

A key element for feasibility of insulin monitoring in individuals with T1D would be optimal sensitivity for multiple insulin analogs used in clinical diabetes treatment. This characteristic was evaluated in buffer and serum samples by comparing chronoamperometric measurements obtained in the analysis of samples spiked with 0- and 200-pM levels of recombinant human insulin and insulin aspart (Novolog), along with the measurement of the additive signal after spiking with both insulins. Assays were conducted in triplicate and amperometric readouts were recorded and averaged.

Statistical Analysis

Chip assays for optimization and characterization were evaluated in triplicate; error bars were estimated as standard deviation of the three replicates. The LOD values were calculated using the 3 Sb/m criterion, where Sb is the standard deviation for 10 blank signal measurements and m is the slope value of the calibration plot. Linear regression analyses and concentration curves were performed using Excel (Microsoft) and Origin (OriginLab). In addition, analysis of variance (ANOVA) and t test analyses were also performed for comparison of signals for human insulin and insulin aspart amperometric and insulin sensor stability at room temperature evaluation, respectively.

Results

Sensor and Assay Protocol Optimization

Operation parameters involved in the sensor preparation and electrochemical immunoassay performance were studied using recombinant human insulin as the representative insulin peptide standard. These were evaluated and optimized in PBT buffer solutions by comparing the S/B ratios obtained from the amperometric measurements corresponding to positive control signal (S), in the presence of recombinant human insulin standard, and negative control signal (B), in the absence of standard spike, selecting as optimal value for each parameter the conditions that provided the largest S/B ratio. To optimize the analytical performance in the insulin detection directly in undiluted serum samples, reevaluation of the assay parameters greatly influencing the analysis time and sensitivity was carried out, such as the number of assay steps, the sample incubation time, and the concentration of the HRP-DAb. Supplemental Figure S2 displays the experimental parameters studied in PBT buffer, whereas those evaluated in serum samples are shown in Figure 2 and Figure S3; in all cases, the amperometric responses and S/B ratios are recorded for each condition, and Table S1 summarizes the final parameter values selected in both matrices. The initial immunosensor optimizations in buffer solutions suggested that a convenient one-step assay involving simultaneous insulin peptide binding to CAb and HRP-Ab performed in a single incubation was perfectly feasible and even more favorable than the two-step sandwich immunoreaction. Such operation greatly simplifies the analysis protocol by reducing the washing steps required between incubations and hence reducing the sample analysis time. This one-step assay condition was confirmed for the serum assay, thus providing a more user-friendly workflow for clinical practice. The lower amperometric responses recorded for both nonspiked and insulin standard-spiked serum samples compared with buffer solutions indicated the distinct electrochemical sensor performance obtained in this body fluid, which can be ascribed to the electrode surface biofouling commonly observed in biological matrices. The reoptimization of the insulin immunoassay in serum samples allowed reduction of the sample incubation time to 30 minutes, bringing the assay closer to near real-time monitoring, while keeping the highest analytical performance toward ultrasensitive insulin detection.

Figure 2.

Results of optimization experiments for critical insulin immunosensor preparation and assay parameters in serum. Chip assays were evaluated in triplicate; error bars were estimated as standard deviation of the three replicates. (a) Immunoreagents involved in the developed one-step insulin assay. Chronoamperometric signals obtained from evaluated parameters of a novel insulin assay using human serum (San Diego Blood Bank) as a negative control or blank (labeled “B” and shown with gray bars) and human serum spiked with 1 nM of recombinant human insulin standard (Sigma-Aldrich) as a positive control or signal (labeled “S” and shown with blue bars). Graphs show change in signal-to-blank (S/B) ratio (red lines) with different assay parameters: (b) Number of incubation steps; (c) sample incubation time; (d) HRP-DAb concentration.

Analytical and Operational Characteristics

We generated calibration curves with the optimized insulin sensor in PBT buffer (Figure 3a) and serum (Figure 2b and c) to establish the analytical characteristics of the developed methodology in both matrices; resulting figures of merit are summarized in Table 1. The estimated LOD derived from the calibration curves obtained in the detection of insulin concentrations in undiluted serum with the optimized immunosensor was 30.2 pM (4.3 µiU/mL), indicating the suitability of the developed insulin immunosensor approach for insulin detection in serum samples from T1D individuals. ¹⁸

Figure 3.

Electrochemical performance and characterization of the developed immunosensor using insulin standards spiked into buffer and serum. (A) Schematic of the homemade sputtered sensor chip for insulin determination along with the immunoreaction and electrochemical transduction processes involved. Chronoamperograms and calibration plots obtained for increasing insulin concentrations in buffer (B) from 0 (a) to 400 pM (i) with 50-pM increments, and in serum (C) from 0 (a) to 1000 pM (e) with 250-pM increments, and (D) from 0 (a) to 400 pM (e) with 100-pM increments.

Table 1.

Analytical Characteristics of the Optimized Insulin Sensor for the Detection of Different Concentrations of Insulin Standards Prepared in 0.1 M phosphate buffer solution supplemented with 0.02% Tween-20 (PBT) Buffer and Human Blood Serum.

Variable	Buffer	Serum
R2	0.9997	0.9996
Slope (nA/pM)	2.97 ± 0.02	0.118 ± 0.001
Intercept (nA)	270 ± 4	15.2 ± 0.3
LOD (pM)	11.2	30.2
LOD (µiU/mL)	1.6	4.3

Abbreviations: LOD, limit of detection.

Next, aspects related to the potential use of the developed insulin sensor as reliable decentralized POC device were assessed (Figure 4a). The reproducibility of the developed insulin assay was evaluated by estimation of the relative standard deviation (RSD) of the amperometric measurements realized with six different chips in the detection of insulin standards spiked in buffer and serum samples; these yielded RSD values of 4.7% and 7.8%, respectively (Figure 4b and c). The sensors used for storage stability assessment in buffer solutions provided mean amperometric signals ranging from 727.4 to 911.8 nA with no systematic decrease observed over the 30-day period studied (Figure 4d). Moreover, the sensor stability at room temperature over time suggested acceptable stability with similar calibration curve slope values recorded in buffer of (2.97 ± 0.05) nA/pM the first day and (3.07 ± 0.09) nA/pM at 24 hours after batch fabrication (Figure 4e). Such stability quality was further explored after seven days of shipping transit by ordinary mailing service, obtaining a slight variation (less than 5%) of the resulting signal-to-noise (S/B) ratio values (Figure 4f).

Figure 4.

Reproducibility and stability of the insulin immunosensor. (a) Image of a disposable insulin strip for decentralized POC testing; reproducibility of the insulin immunoassay for the detection of (b) 200-pM insulin standards spiked in buffer and (c) 100-pM insulin standards spiked in serum, using a six-electrode batch for each matrix; (d) amperometric responses obtained in the measurement of 200-pM insulin standards spiked in buffer on each day of test over a 30-day period after fabrication and storage at 4°C; (e) amperometric responses and resulted calibration slope values (red lines) obtained in the measurement of 0- (black bars), 200- (blue bars), and 400-pM (green bars) insulin standards spiked in buffer on each day of test over a 24-hour period after fabrication and storage at room temperature; (f) amperometric responses and resulted signal-to-blank (S/B) values (red lines) obtained in the measurement of 0- (blank, black bars) and 200-pM (signal, blue bars) insulin standards spiked in buffer on each day of test over a seven-day period after fabrication and ordinary mailing shipping transit. Error bars in (d-f) were estimated as the standard deviation of three replicates.

The reduction of the serum sample volume and electrode surface area for immunoreaction incubation led to a systematic drop of the sensor sensitivity evaluated through the S/B ratios recorded for each condition, ranging from 2.5, obtained for conventional optimized 10-µL volume deposited on 21-mm² WE area, to 1.7 registered for 2.5-µL sample droplet placed on a chip with 10.5-mm² WE area (see Figure S4). Such S/B ratio decrease is explained by a direct signal dependence on electrode surface area and analyte amount incubated. However, an acceptable S/B ratio of 2.4 was obtained when a low volume of 2.5 µL was placed on an adjusted electrode area of 15.75 mm², allowing serum assays with satisfactory reproducibility.

Insulin aspart analog measurements with optimized insulin sensors in assays using 200-pM insulin standard-spiked buffer and serum samples yielded mean amperometric responses of 851.3 and 36.2 nA, respectively, which are remarkably similar to the results of 875.4 and 35.1 nA obtained for recombinant human insulin detection (ANOVA test: buffer assay—P > .05; serum assay— P > .05). Furthermore, we successfully recorded the additive signals corresponding to the presence in the sample of insulin and the analog in same concentration (Figure 5).

Figure 5.

Immunosensor response toward insulin analog using insulin standards spiked into buffer and serum. Chip assays were evaluated in triplicate; error bars were estimated as standard deviation of the three replicates. (a) Schematic of the insulin peptide along with fractions of the amino acids sequences corresponding to human insulin and insulin aspart. Chronoamperograms and signal results obtained in the measurement of (b) buffer solutions and (c) serum samples nonspiked (black lines and bars), and spiked with 200 pM of human insulin “HI” (Sigma-Aldrich) (red lines and bars), 200 pM of insulin aspart analog “NI” (Novolog) (blue lines and bars), and 200 pM of both human insulin and insulin aspart (yellow lines and bars).

Discussion

Commercial assays of insulin now exist but can take hours to perform and conventionally are limited to centralized laboratories providing delayed analytical results (the sample-to-answer time can take weeks), and therefore cannot offer timely measurements which could be used to assess the dynamically changing concentrations of insulin during insulin therapy.

We measured satisfactorily low pM insulin concentrations in undiluted human serum samples using our developed amperometric immunosensor relying on one-step bioassays completed within 30 minutes. This provided reproducible and well-defined amperometric responses that yielded an LOD of 30.2-pM (4.3 µiU/mL) insulin, indicating the suitability of the sensor methodology for near real-time detection of clinically relevant concentrations of serum insulin for individuals with T1D. In previous studies, the concentration of insulin has been quantified in the ultralow concentration range.^{9 -18} Although these insulin sensors presented lower LODs, the complexity of these bioassay approaches and their sensor preparation are significant obstacles to their future potential as low-cost, scalable POC insulin tests. Moreover, the evaluation of the analytical and operational characteristics of insulin strips using these approaches was primarily focused on application in controlled lab settings. To our knowledge, this is the first report of a decentralized electrochemical insulin sensor prototype whose operational characteristics have been widely explored toward real application for on-the-spot detection of clinically relevant concentrations of serum insulin at the primary POC. The demonstrated attractive analytical performance and adequate immunostrip stability for large-scale fabrication and remote site application suggest its considerable potential as a true POC in vitro insulin test. Furthermore, the possibility of reducing the serum sample analysis volume required, without compromising the ultrasensitive detection capability, adds convenience for clinical application, versatility toward feasible miniaturization, and potential for integration into microfluidic systems, further simplifying the analysis workflow. It is worth to highlight that although the insulin immunosensor was designed to detect recombinant human insulin in serum, the developed assay showed the same sensitivity toward insulin aspart, as representative synthetic insulin analog. This analytical feature would represent an advantage as usually different ELISA assay kits are required for each insulin analog, and detection of native and analog insulin would be required for a clinically meaningful insulin assay.

Future work will focus on critical validation of the developed sensor in clinical trials as a proof-of-concept prototype for on-the-spot insulin detection in serum samples from individuals with T1D conducted in a remote clinical location. Our approach to the development of effective near real-time monitoring and modeling will continue to emphasize sensor accuracy and sensitivity insulin analogs during in vivo analysis and will include comparison of sensor results with those obtained with a centralized lab-based ELISA methodology.

Conclusions

The results obtained in the optimization and characterization of our novel insulin sensor illustrate its suitability for its potential application in decentralized clinical settings for frequent insulin monitoring. The developed insulin immunosensor provides ultrasensitive detection to determine clinically relevant serum insulin levels; the low-cost bioassay, featured by an estimated cost of consumables of less than $5, using 30-minute incubations with ultralow undiluted serum sample volumes (≤10 µL) can be easily used in remote locations in combination with a low-cost compact potentiostat for readout of measurements. Preliminary data indicate that the current immunosensor provides adequate sensitivity and specificity for detecting serum insulin in samples from individuals with T1D, and attractive decentralized in vitro analytical performance. Thus, the insulin sensor may be a potentially mobile alternative to the gold standard reference ELISA methodology and could transform the prospect for POC diabetes monitoring.

Supplemental Material

sj-docx-1-dst-10.1177_19322968211071132 – Supplemental material for Development of a Novel Insulin Sensor for Clinical Decision-Making

Supplemental material, sj-docx-1-dst-10.1177_19322968211071132 for Development of a Novel Insulin Sensor for Clinical Decision-Making by Eva Vargas, Eleonora M. Aiello, Jordan E. Pinsker, Hazhir Teymourian, Farshad Tehrani, Mei Mei Church, Lori M. Laffel, Francis J. Doyle, Mary-Elizabeth Patti, Eyal Dassau and Joseph Wang in Journal of Diabetes Science and Technology