Clinical Evaluation of a Novel Insulin Immunosensor

Abstract

Background:

The estimation of available active insulin remains a limitation of automated insulin delivery systems. Currently, insulin pumps calculate active insulin using mathematical decay curves, while quantitative measurements of insulin would explicitly provide person-specific PK insulin dynamics to assess remaining active insulin more accurately, permitting more effective glucose control.

Methods:

We performed the first clinical evaluation of an insulin immunosensor chip, providing near real-time measurements of insulin levels. In this study, we sought to determine the accuracy of the novel insulin sensor and assess its therapeutic risk and benefit by presenting a new tool developed to indicate the potential therapeutic consequences arising from inaccurate insulin measurements.

Results:

Nine adult participants with type-1 diabetes completed the study. The change from baseline in immunosensor-measured insulin levels was compared with values obtained by standard enzyme-linked immunosorbant assay (ELISA) after preprandial injection of insulin. The point-of-care quantification of insulin levels revealed similar temporal trends as those from the laboratory insulin ELISA. The results showed that 70% of the paired immunosensor-reference values were concordant, which suggests that the patient could take action safely based on insulin concentration obtained by the novel sensor.

Conclusions:

This proposed technology and preliminary feasibility evaluation show encouraging results for near real-time evaluation of insulin levels, with the potential to improve diabetes management. Real-time measurements of insulin provide person-specific insulin dynamics that could be used to make more informed decisions regarding insulin dosing, thus helping to prevent hypoglycemia and improve diabetes outcomes.

Keywords

insulin measurement point-of-care immunosensor type-1 diabetes

Introduction

Individuals with type-1 diabetes (T1D) require lifelong replacement of insulin to maintain healthy metabolism, prevent ketosis, and maintain glucose in a desired range to minimize long-term disease complications. Advances in technology, development of fast-acting insulin analogues, and closed-loop control algorithms have enabled automated insulin delivery (AID) for the management of T1D.¹ Even with these inputs, there is still a significant amount of decision-making needed by the user. As a result, adults with T1D generally only achieve, on average, ~70% continuous glucose monitoring (CGM) time in range 70 to 180 mg/dL using one of the most advanced AID systems,^2,3 while children and adolescents achieve a slightly lower percentage time in range using such advancements.^4,5

Frequent, real-time insulin monitoring using a primary point-of-care (POC) insulin sensor would represent an advance over estimations of residual circulating insulin levels that are currently used to inform insulin dosing. Bolus calculators rely on the current glucose concentration, target glucose concentration, amount of carbohydrate to be consumed, insulin-to-carbohydrate ratio, sensitivity factors, as well as the estimated residual insulin remaining active in the circulation, so-called insulin on board (IOB) or active insulin.⁵

Unfortunately, estimates for insulin action curves and IOB are challenging due to interindividual differences and often the lack of specificity to insulin type. Moreover, insulin action varies within each individual as a function of many factors, for example, time of day, stress, exercise, illness, presence of insulin antibodies, and so on. Thus, there is no “generic” insulin action curve for all. Quantifying insulin levels to determine available circulating insulin in individual patients rather than relying on predicted patterns would allow real-time computation of IOB to permit further refinement of insulin dosing, especially under different physiological conditions, such as changing insulin sensitivity as often occurs with exercise. Insulin concentration measurements from the insulin immunosensor can enable the near real-time measurement of the circulating insulin concentration, which cannot otherwise be measured without laboratory-based methods⁶ and, as such, can better inform insulin dosing algorithms with more robust estimation of active insulin.⁷

Given the critical need to enhance glucose control for patients and to improve performance of AID systems, we performed the first clinical evaluation of a novel insulin sensor that can provide near real-time measurement of insulin levels using a microchip-based immunoassay of serum insulin within 30 minutes. The applicability of the insulin immunosensor as a POC device was evaluated by conducting validation tests in adults with T1D at the clinical research unit at the Sansum Diabetes Research Institute, Santa Barbara, CA. Direct on-the-spot quantification of insulin levels was carried out in untreated serum. In addition, sensor results were compared with insulin assays from similarly collected venous samples measured by commercial assays.

Methods

The near real-time insulin sensor chip was developed in controlled laboratory settings at the University of California San Diego. The study protocol was reviewed and approved by the Institutional Review Board at the Joslin Diabetes Center for this multicenter clinical study. Due to the COVID-19 pandemic, only a single site (Sansum Diabetes Research Institute, Santa Barbara, CA) was able to carry out the clinical studies.

Data Collection

Participants were ≥18 years of age with T1D for at least one year and had used an insulin pump for at least three months. Participants were not permitted to use any form of long-acting insulin for at least two weeks before the sensor evaluation sessions and wore a Dexcom G6 CGM (Dexcom, Inc., San Diego, CA) the night before and throughout the study session. Before eating breakfast, participants were given an injection of their usual mealtime insulin (lispro or aspart) per their carbohydrate ratio by the study investigator while their insulin pump continued to deliver their normal basal rate. No bolus insulin was given within 6 hours of the AM dose.

To determine accuracy of the novel insulin sensor, we collected venous samples for subsequent insulin assay at Quest Diagnostics/Nichols Institute, San Juan Capistrano, CA, using the Beckman Access immunoassay.

Likewise, other analytes affecting and/or reflecting insulin action (serum cortisol and beta-hydroxybutyrate [BHB], and plasma lactate) were assayed in venous samples at Quest Diagnostics/Nichols Institute.

To test the feasibility of capturing frequent capillary measurements of insulin, capillary blood samples were obtained from the fingertip using the Ram Scientific Safe-T-Fill serum gel capillary collection tubes (RAM Scientific, Inc., Nashville, TN). Blood sampling for sensor and capillary blood insulin assays was obtained at baseline and 60, 120, and 240 minutes after breakfast, and additional venous samples were obtained at 15, 30, 60, 90, 120, 180, 240, and 300 minutes. Fasting C-peptide and insulin antibodies were measured at baseline.

The on-the-spot quantification of insulin levels was carried out in untreated serum samples obtained from venous blood using the standard addition-based calibration method.^8,9 See Supplementary Material for details on standard additions methodology.

The error in the insulin concentration quantification $S_{X E}$ in the calibration protocol was evaluated, and the concordance of the immunosensor and enzyme-linked immunosorbant assay (ELISA) insulin levels was assessed.

Given the presence of insulin antibodies in some participants and differences in baseline insulin levels across the nine participants, evaluation of the change in insulin levels following the breakfast bolus was the focus of the analysis of the insulin levels and ascertainment of the sensor’s performance. Concordance of immunosensor and ELISA insulin levels was assessed in the evaluation of changes from baseline.

Insulin Measurement Error Grid

As the IOB estimation can affect the insulin bolus calculation, errors in insulin measurements and estimation of IOB could result in inaccurate insulin bolus dosing and risk for either hyperglycemia or hypoglycemia following overestimation or underestimation of circulating insulin, respectively. An insulin measurement error (IME) therapeutic risk analysis is proposed to quantify the clinical accuracy of the novel sensor as compared with results obtained using standard commercial insulin assays. Inspired by the methodology developed by Clarke et al,¹⁰ the IME therapeutic risk assessment grid is a novel analysis tool focused on the risk that can be derived from the clinical outcome that might occur if the patient’s insulin self-management was based on the insulin concentration obtained by the novel sensor versus the insulin concentration measured by a laboratory assay. We thus sought to understand how an absolute error in quantification of the insulin concentration using the novel sensor could impact suggested insulin delivery and to define the risk resulting from inaccurate insulin data that would guide one of several possible actions, such as recommending treatment of hypoglycemia, pump suspension, decreased insulin dose, increased insulin dose, or no action, that is, only maintaining administration of the basal insulin dose.

Therapeutic risk arises when a correction drives the resulting glucose values outside the target range or worsens an existing hypoglycemia or hyperglycemia condition.

We redefined the five risk zones of the Clark-Error analysis, adapting to the consequences derived by the sensor-based measurements. Risk zone A represents the sensor concentrations that deviate from the reference values by no more than 20%, while zone B includes those insulin measurements that are outside of zone A’s percent error but would lead to the same adjustment to the therapy.

In Clarke et al error grid,¹⁰ a measurement error of ±20% for glucose was considered a tolerable level of error, especially for regulatory purposes. In our case, a 20% error may not have the same meaning due to the impact of interindividual insulin sensitivities. A gain factor was introduced to convert the ±20% error for glucose to the corresponding percentage error for insulin measurements. Our gain factor aims to include the effect of insulin sensitivity, normalizing the relative insulin error using the individual’s correction factor (CF).

Zone C represents the overcorrection condition, in which data would result in an aggressive decrease or increase in the insulin dose instead of a mild correction. Zone D includes those insulin sensor measurements which could lead to a potentially dangerous situation, due to suggestion of no action when one was needed. Zone E includes samples with a ±70% measurement error, leading to a correction that causes hypoglycemia from hyperglycemia.

As the insulin pumps use the IOB to adjust the insulin dose, we inferred the error on the IOB estimate from the deviation of the insulin concentration from the reference value. It is reasonable to assume that if the insulin percent error is negative, the sensor measurement suggests that the percent insulin still active in the plasma will be lower than the IOB value derived using the reference insulin concentration. Correspondingly, if a positive insulin percent error occurs, the IOB estimate using the sensor will be higher than the corresponding value obtained by the reference. Therefore, both positive and negative percent IMEs between the sensor measurements and reference values were divided into three levels based on clinical considerations: (i) within 20%, (ii) between 20% and 70%, and (iii) greater than 70%.

As the risk associated with the insulin error may vary with the glucose level, three possible scenarios were considered: hypoglycemia (glucose <70 mg/dL), euglycemia (70-180 mg/dL), and hyperglycemia (>180 mg/dL). As the basal infusion rates would remain unmodified in the euglycemic range, the IOB estimate would not impact the therapy. However, in a hypoglycemic state, a low IOB value may suggest a slight reduction of the insulin dose while a high IOB estimate would lead to suspension of insulin delivery and suggested treatment of hypoglycemia. Conversely, in a hyperglycemic state, a low IOB estimate may lead to an increase of the insulin dose, while high IOB estimate could lead to unchanged insulin dosing. The insulin measurement error grid (IMEG) is detailed in Table 1, where the error levels are associated with the defined glucose levels, as well as with the risk derived by the adjustment of the insulin therapy.

Table 1.

Levels of the Percent Insulin Measurement Errors Coupled With the Therapy Adjustment Based on the Novel Sensor Measurement Are Listed With the Resulted Clinical Risk, as Well as the Assigned Zone.

	Normalized percent insulin measurement error $(e_{%})$	Adjustment to the therapy based on the reference value	Adjustment to the therapy based on the sensor value	Associated clinical risk	Zone
Hypoglycemia	$e_{%} > 70 %$	Suspend the pump Hypo-treatment	May suspend the pump	Risk severe hypoglycemia due to a failure in the treatment	Zone D
	$20 % < e_{%} \leq 70 %$	Suspend the pump Suggest hypo-treatment	May suspend the pump May suggest hypo-treatment	Risk severe hypoglycemia due to a failure in the treatment	Zone D
	$- 20 % \leq e_{%} \leq 20 %$	Same adjustment to the therapy		None	Zone A
	$- 70 % < e_{%} < - 20 %$	May suspend the pump May suggest hypo-treatment	Suspend the pump Suggest hypo-treatment	Overcorrection	Zone C
	$e_{%} < - 70 %$	Suspend the pump	Suspend the pump Suggest hypo-treatment	Overcorrection	Zone C
Hyperglycemia	$e_{%} > 70 %$	No action	Increase the insulin	Trigger the opposite of the correct adjustment	Zone E
	$20 % < e_{%} \leq 70 %$	May increase the insulin	Increase the insulin	Overcorrection	Zone C
	$- 20 % \leq e_{%} \leq 20 %$	Same adjustment to the therapy		None	Zone A
	$- 70 % < e_{%} < - 20 %$	Increase the insulin	May increase the insulin	Risk severe hyperglycemia due to failure in the treatment	Zone D
	$e_{%} < - 70 %$	Increase the insulin	No action	Trigger the opposite of the correct adjustment	Zone E
In euglycemic range	$e_{%} > 20 %$	No action		No applicable adjustment	Zone B
	$- 20 % \leq e_{%} \leq 20 %$			None	Zone A
	$e_{%} < - 20 %$			No applicable adjustment	Zone B

Abbreviation: IOB, insulin on board.

The cases have been repeated for each glucose state. It is reasonable to assume that if the insulin percent error is negative, the sensor measurement suggests that the percent insulin still active in the plasma is lower than the IOB value derived using the reference insulin concentration, and vice versa, if a positive insulin percent error occurs. If in a hypoglycemia state, a low IOB value may suggest a slight reduction of the insulin dose, while a high IOB estimate would lead to suspension of insulin delivery and suggested treatment of hypoglycemia. Conversely, in hyperglycemia, a low IOB estimate may lead to an increase of the insulin dose, while high IOB estimate could lead to unchanged insulin dosing.

Results

Nine adult participants with T1D completed the study. Demographics of the study participants are shown in Table 2. Seven individuals were using insulin lispro and two were using insulin aspart in their pumps. Samples used for the sensor glucose determinations were drawn at baseline before breakfast, then at 60, 120, and 240 minutes after breakfast. Elevated insulin antibodies were found in seven of nine participants, while C-peptide levels were low or undetectable in all participants, consistent with the diagnosis of T1D as reported in Table 3.

Table 2.

Subject Demographics and Baseline Characteristics (mean ± SD) for N = 9 T1D Adults.

Age (years)	41.2 ± 15.4
Sex
Male	7
Female	2
Weight (kg)	80.9 ± 23.1
Body mass index (kg/m²)	26.7 ± 3.7
HbA1c (%)	6.7 ± 0.8
Duration of diabetes (years)	24.1 ± 13.1
Total daily insulin (units/day)	49.2 ± 18.0

Table 3.

Insulin Concentrations Quantified in Serum and Capillary Blood Samples Collected From Nine Subjects With T1D.

Subject	Sample	$S_{X E}$ (µiU/mL)	Venous blood samples		Capillary blood samples (µiU/mL)	$C - P e p t i d e$ $(n g / m L)$	Insulin anti-bodies µiU/mL $i U / m L$	SMBG (mg/dL)
Subject	Sample	$S_{X E}$ (µiU/mL)	Novel sensor (µiU/mL)	ELISA (µiU/mL)	Capillary blood samples (µiU/mL)	$C - P e p t i d e$ $(n g / m L)$	Insulin anti-bodies µiU/mL $i U / m L$	SMBG (mg/dL)
HS1-01	Baseline	N/A	-	14.2	16	<0.10	37.2	123
	1 hour^a	N/A	29.34	57.8	49.7
	2 hours	N/A	41.11	36.6	38.8
	4 hours	N/A	22.01	25.6	24.2
HS1-02	Baseline	N/A	16.04	18.6	49.4	<0.10	<0.04	145
	1 hour	0.34	36.06	52.7	40.4
	2 hours	0.15	24.13	31.3	43.1
	4 hours	N/A	-	16.3	25.4
HS1-03	Baseline	N/A	12.80	10.8	23.9	<0.10	8.4	167
	1 hour	0.43	50.43	86.8	67.6
	2 hours	1.14	25.21	51.7	69.8
	4 hours	0.4	19.94	14.4	20.6
HS1-04	Baseline	N/A	18.76	5.8	269.4	<0.10	3.2	190
	1 hour	0.96	44.89	23.0	-
	2 hours	0.42	28.20	15.2	210.1
	4 hours	0.26	25.74	9.1	60.3
HS1-05	Baseline	N/A	27.25	374.0	122.8	<0.10	43.5	92
	1 hour	0.91	47.64	121.7	178.4
	2 hours	0	33.69	103.9	261.7
	4 hours	0.68	29.70	78.8	110.2
HS1-06	Baseline	N/A	26.27	14.9	56.7	<0.10	10.8	150
	1 hour	0.06	55.93	26.3	34.9
	2 hours	0.08	48.82	27.7	23.8
	4 hours	0.09	31.46	25.4	31.8
HS1-07^b	Baseline	N/A	39.22	42.2	-	<0.10	50	156
	1 hour	0.25	89.03	71.6	118
	2 hours	0.07	55.70	66.5	169.3
	4 hours	0.23	34.53	47.8	176.1
HS1-08	Baseline	N/A	57.97	5.6	10.23	0.24	<0.04	205
	1 hour	0.16	80.71	27.1	27.27
	2 hours	0.16	59.57	24.3	29.26
	4 hours	0.22	59.60	15.9	14.81
HS1-09	Baseline	N/A	104.28	6.1	11.03	0.31	9	252
	1 hour	0.92	167.68	37.2	38.43
	2 hours	0.87	134.38	25.5	93.2
	4 hours	1.09	128.69	11.2	16.27

Insulin concentration values estimated using the immunosensor from the corresponding standard additions calibration curve produced for each sample. Errors in the calibration $(S_{\frac{y}{x}})$ and in the analyte concentration quantification $(S_{X E})$ are reported. Capillary, insulin antibodies, and C-peptide concentrations, as well as self-monitoring blood glucose (SMBG) values are reported. Participants were using insulin Lispro, but HS1-06 and HS1-09 were using insulin Aspart.

Insulin concentrations quantified in serum sample extracted after 15 minute from the baseline.

Subject tested performing 2.5 µL sample incubations on 15.75-mm² working electrode (WE) immunostrips.

Additional analytes measured during this study included cortisol, BHB, and lactate. As there was no significant stress or exercise component in this clinical study, we saw no significant changes in these analytes. Cortisol levels were generally highest at baseline and decreased throughout the day. Lactate and BHB levels were generally unchanged throughout the course of the study with no significant variation.

We assessed insulin concentrations using the optimized novel chip assay in 34 serum samples from nine individuals with T1D. Our immunosensor provided a limit of detection of 30.2 pM (4.3 µiU/mL) insulin in undiluted serum, indicating the suitability of the sensor methodology for detecting clinically relevant concentrations of serum insulin in patients with T1D. For all samples, amperometric responses were recorded and chronoamperograms generated for the different samples, along with the temporal variation profiles traced from the averaged amperometric signals and calibration plots corresponding to each sample (see Figure 1). We registered similar temporal trends for all the participants, characterized by a rise in serum insulin levels after insulin injection and then a decrease over time. Similarly, for each person, the standard additions calibration plots showed very similar slope values, having an average variation of 0.09 ± 0.05 µiU/mL, suggesting satisfactory reproducibility of the decentralized serum immunoassays. Note that for participant HS1-01 and HS1-02, some determinations are missing (basal level for HS1-01, and after 4 hours for HS1-02) due to issues related to immunostrip transportation conditions. Insulin concentrations quantified by extrapolation from the derived calibration equations, and by subsequent insulin concentration unit conversion, ranged from 12.8 µiU/mL to 167.6 µiU/mL (Table 3).

Figure 1.

Amperometric results obtained with the immunosensor in the decentralized monitoring of insulin levels variation over time using serum samples from nine subjects with T1D. Chronoamperograms (a) and mean amperometric signals (b) corresponding to serum samples without standard spike, and standard-addition calibration plots considering spiking with insulin standard solutions of 200 and 400 pM concentrations (c) recorded for serum samples collected before (0 h) and at certain times (1, 2, 4 h) after insulin injection from nine different subjects. All subjects evaluated using the optimized serum assay, except for subject HS1-07 for whom incubations of 2.5 µL sample on 15.75 mm² working electrode (WE) immunostrips were performed.

In general, as shown in Figure 2, similar variation trends were obtained in the insulin concentrations between both immunosensor and laboratory methodologies for most of the samples (Table 3), with an error Gaussian distributed with zero mean and standard deviation equal to 16 µiU/mL. However, differences between sensor and ELISA results are distinct between individuals. For example, HS1-05 had substantially greater levels assessed by the ELISA method vs. immunosensor; HS1-06 showed a more consistent variation over time; and HS1-08 and HS1-09 had much higher insulin concentrations estimated by the immunosensor relative to ELISA. These high concentration values derived from the distinctly high amperometric responses registered for these participants with the sensor. However, as shown in Figure 2, when delta insulin concentration change from baseline is displayed, similar variation trends were obtained with both methods for these two individuals, as comparable delta concentration variations were found.

Figure 2.

Delta change from baseline of insulin concentrations obtained by the decentralized immunosensor and a centralized ELISA method in serum samples collected from nine individuals with T1D. Comparison of the results obtained on-the-spot with the developed insulin sensor (blue circles) and in a centralized external laboratory with an ELISA kit (orange circles) in the serum insulin quantization in the blood samples collected from the participants at each time point when insulin measurements were available, that is, at 1, 2, and 4 hours after the mealtime measurement. Error bar of immunosensor measurements is also included at each delta. All subjects evaluated using the optimized serum assay, except for subject HS1-07 for whom incubations of 2.5 µL sample on 15.7mm² working electrode immunostrips were performed. Participants were using insulin Lispro, but HS1-06 and HS1-09 were using insulin Aspart.

Evaluation of the Novel Sensor Using the IMEG

Based on the absolute percent error normalized by the individual insulin sensitivity, 18 serum samples were classified in zone A, 7 in zone B, 2 in zone C, 2 in zone D and 5 in zone E. As more than 70% of the data appear in zones A and B, the novel sensor results may provide clinical utility in most situations; however, specific cases need further discussion.

In hypoglycemic conditions, it is evident how the error appears to be a direct indicator of the clinical risk. Figure 3 shows that two samples were in zone D: the error size indicates that using the sensor concentration, the IOB estimate is more than 70% less than the IOB value obtained by the reference insulin level. In this case, the sensor measurement may lead to only a small decrease of the insulin bolus, while the reference concentration would suggest suspending the pump to avoid severe hypoglycemia.

Figure 3.

Insulin measurement error (IME) therapeutic risk assessment grid on the serum samples collected from nine individuals with T1D. The IMEs (e_%) are reported in relation to the glucose levels. The error values were computed by normalizing the relative IMEs using the individual CF. Zone A (light green) includes samples with IME less than 20%. Zone B (dark green) includes samples having an IME larger than 20% provided that the error does not compromise the adjustment to the therapy. Zone C (yellow) represents the overcorrection condition associated with an IME larger than 20%. Zone D (orange) represents the suggestion of no action when one was needed having an IME larger than 20 %. Zone E represents a correction that causes hypoglycemia from hyperglycemia and vice versa associated with an IME larger than 70%. The dashed horizontal lines separate the three glycemic levels: hypoglycemia (glucose <70 mg/dL), euglycemia (70-180 mg/dL), and hyperglycemia (>180 mg/dL). The error thresholds for a 20% and 70% measurement error are displayed by the dashed vertical lines. A zoomed detail of the right-side part has been included in the central area.

In the euglycemic range, the error size is irrelevant because the at-target glucose concentration is the dominant factor for the dose decision. Both the sensor and the reference measurements would not suggest adjusting the basal therapy: the samples with a percent error less than 20% fall in zone A, the rest in zone B, as shown in Figure 3.

The hyperglycemia state represents the most challenging condition to assess the clinical risk. Results from five of twelve serum samples fell into zone E as shown in Figure 3. The corresponding percent errors are greater than 70% and occurred mostly in a patient having high insulin sensitivity. In this case, measurement errors can lead to a wrong decision and could create hypoglycemia in the patient.

Discussion

We detected serum insulin in low pM concentrations in sera collected from individuals with T1D using our novel immunosensor within 30 minutes. In previous studies, the concentration of insulin has been quantified in the ultra-low concentration range.^11-22 However, only a few of these publications report application in real clinical samples,^12,15,19,22 and only a few have been validated by comparison with a reference analytical method.^15,22 Although the previously published insulin sensor data presented lower LODs, the complexity of those immunoassay approaches and their sensor preparation remain significant obstacles to their future potential as low cost, scalable POC insulin tests. Moreover, their use was limited to controlled laboratory settings. To our knowledge, this is the first report of a decentralized microchip-based immunoassay prototype for on-the-spot, POC detection of clinically relevant concentrations of insulin directly from undiluted serum clinical samples.

The addition of analyte measurements beyond glucose holds the potential to greatly improve AID technology. Currently, insulin pumps calculate active insulin using either a curvilinear or linear decay curve, with different pumps using different methodologies.⁶ It has been reported that the absorption of insulin analogs can vary 10% to 30% within an individual and 20% to 50% between individuals,²³ leading to the potential to increase risk of hypo- or hyperglycemia if the estimate of insulin action time is not accurate. Bolus calculators also do not consider bolus size, exercise, or heat.²⁴ These factors can all affect duration of insulin action. As a result, insulin pump users often rely on personal experience to adjust the IOB or active insulin time setting, without a clear understanding of what the setting really does in each pump or exactly how it affects insulin delivery.²⁵

We have previously reported on dynamically adjusting active insulin based on time of day or other factors,¹⁰ but this approach can be vastly improved by actual measurement of insulin levels conveying a true representation of its dynamic concentrations. An insulin state-based observer has been developed for modeling patient-specific insulin PK behavior. Starting from the subcutaneous absorption insulin PK model proposed by Schiavon et al.²⁶, we expanded the model to include insulin in a remote compartment, which represents the delayed appearance of insulin from the plasma compartment and serves as the measured state.⁶

Once the outcome of the next measurement is observed, the estimated plasma insulin concentration is updated and can be used to derive the percent insulin remaining active in the circulation providing a real-time quantification of the IOB.

Overall, capillary measurements matched very well with values for serum insulin levels measured by a clinical reference laboratory (Table 2). While capillary measurements were not assayed in real-time in this study, future real-time measurements of insulin in samples of capillary blood or in interstitial fluid using the novel sensor assay as POC hold great potential to derive active insulin levels. The feasibility test on the capillary measurements represents an enabling result for the insulin observer algorithm, whose formulation was expanded to integrate the capillary exchange model between plasma and remote compartments in the insulin state-based observer, leading to more accurate models for the plasma insulin concentration, which would provide more precise IOB curves.⁷

Our evaluation of a small set of samples from participants with T1D is an indication of the potential of the insulin immunosensor as a monitoring tool. As a near real-time quantitative methodology applicable in remote locations, the insulin assay can enable additional insight to glycemic regulation and reduce the uncertainty around IOB when dosing the insulin bolus. It is worth highlighting that although the insulin immunosensor was designed to detect human insulin, the developed assay showed the same sensitivity toward two distinct insulin analogs, offering future application to multiple individuals with T1D.

For further work, it would be important to include a larger set of clinical samples by recruiting a larger number of participants treated with different insulin types to analyze any potential differential detection of particular insulin analogues,²⁴ as well as to refine the reproducibility, sensitivity, and specificity of the novel insulin assay. In addition, as high anti-insulin antibodies can impact not only physiologic insulin action but also accurate measurement of free and total insulin levels, the impact of autoantibody titers on the accuracy of the sensor insulin measurement needs to be determined.²³ On the contrary, the sensing protocol can be improved by reducing the sample-to-answer time, which will bring this sensing approach even closer to real-time monitoring.^11,12 In this regard, insulin quantization with the immunosensor based on single-point calibrations^25-29 or calibration-free methodologies^30,31 should be explored in future experiments.

Conclusions

Results from this proof-of-concept study indicate our novel insulin sensor has significant potential to be developed and implemented as a POC in vitro chip for near real-time insulin monitoring. The current immunosensor provides adequate sensitivity and specificity for detecting serum insulin in samples from individuals with T1D, and an attractive on-the-spot POC performance. The reliability of the immunosensor was also shown using the proposed IMEG, which is an innovative tool to assess the therapeutic risk of insulin measurements obtained with various monitoring systems. Our effort to design this new assessment tool goes in the direction of developing a new methodological approach to evaluate the accuracy of novel insulin sensors.

To perform real-time insulin monitoring, designing and developing new insulin sensing approaches based on alternative technologies that do not require enzymatic labeling for analyte detection will likely be required. In this regard, the use of artificial recognition entities, such as aptamers^9,32-38 or molecularly imprinted polymers,^14,17,34 represents promising strategies.

Supplemental Material

sj-docx-1-dst-10.1177_19322968221074406 – Supplemental material for Clinical Evaluation of a Novel Insulin Immunosensor

Supplemental material, sj-docx-1-dst-10.1177_19322968221074406 for Clinical Evaluation of a Novel Insulin Immunosensor by Eleonora M. Aiello, Jordan E. Pinsker, Eva Vargas, Hazhir Teymourian, Farshad Tehrani, Mei Mei Church, Lori M. Laffel, Francis J. Doyle, Mary-Elizabeth Patti, Joseph Wang and Eyal Dassau in Journal of Diabetes Science and Technology

Footnotes

Acknowledgements

The authors would like to acknowledge Kelly Burke, Molly Piper, and Jimena Perez (Sansum Diabetes Research Institute) for their work supporting performance of the clinical trials, and Jessica Desmond (Joslin Diabetes Center) for capillary measurements.

Abbreviations

AID, automated insulin delivery; CGM, continuous glucose monitoring; MDI, multiple daily injections; T1D, type-1 diabetes; IOB insulin on board; CF, correction factor; PD, pharmacodynamics; PK, pharmacokinetic; LOD, limit of detection; IMEG, insulin measurement error grid.

Authors’ Contributions

J.E.P., E.M.A., E.V., H.T., E.D., M.M.C., L.M.L., M.E.P., J.W., and E.D. helped design the study protocol and ensured the regulatory approval of the study; analyzed data; and authored the manuscript. J.E.P., E.V., H.T., F.T., M.M.C., L.M.L., M.E.P., F.J.D., and E.D. performed data acquisition for the clinical study, analyzed data, and edited and revised the manuscript. E.M.A. and F.J.D. analyzed data, and edited and revised the manuscript. E.D. edited and reviewed the final manuscript and was the principal investigator of the project. E.D. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Declaration of Conflicting Interests

The author(s) declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: J.E.P. is currently an employee and shareholder of Tandem Diabetes Care, Inc. The work presented in the manuscript was performed as part of his academic appointment at Sansum Diabetes Research Institute and is independent of his employment with Tandem Diabetes Care. E.D. reports receiving grants from JDRF, NIH, and Helmsley Charitable Trust, personal fees from Roche and Eli Lilly, patents on artificial pancreas technology, and product support from Dexcom, Insulet, Tandem, and Roche. E.D. is currently an employee and shareholder of Eli Lilly and Company. The work presented in this manuscript was performed as part of his academic appointment and is independent of his employment with Eli Lilly and Company. F.J.D. reports equity, licensed IP and is a member of the Scientific Advisory Board of Mode AGC. L.M.L. reports grant support to her institution from NIH, JDRF, Helmsley Charitable Trust, Eli Lilly and Company, Insulet, Dexcom, and Boehringer Ingelheim; she receives consulting fees unrelated to the current report from Johnson & Johnson, Sanofi, NovoNordisk, Roche, Dexcom, Insulet, Boehringer Ingelheim, ConvaTec, Medtronic, Lifescan, Laxmi, and Insulogic. M.E.P. reports receiving grant support, provided to her institution, from NIH, Helmsley Charitable Trust, Chan Zuckerberg Foundation, and Dexcom, patents related to hypoglycemia and pump therapy for hypoglycemia, and advisory board fees from Fractyl (unrelated to the current report). F.T. and H.T. are currently employees of ActioX LLC. The work presented in the manuscript was performed as part of their academic appointment at UCSD. All other authors report no conflict of interest.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study was funded by a grant from the Leona M. and Harry B. Helmsley Charitable Trust (grant no. 2018PG-TID06). JEP was also supported in part by a grant from the William K. Bowes, Jr. Foundation (grant no. WKB-2020-38415). Product support was provided by Dexcom, Inc (San Diego, CA) who provided research discount pricing on continuous glucose monitoring sensors, transmitters and receivers (grant no. IIS-2019-052).

ORCID iDs

Eleonora M. Aiello

Jordan E. Pinsker

Mei Mei Church

Lori M. Laffel

Eyal Dassau

Supplemental Material

Supplemental material for this article is available online.

References

Wolkowicz

Aiello

Vargas

, et al. A review of biomarkers in the context of type 1 diabetes: biological sensing for enhanced glucose control. Bioeng Transl Med. 2021;6(2):e10201. doi:10.1002/btm2.10201.

Pinsker

Müller

Constantin

, et al. Real-world patient-reported outcomes and glycemic results with initiation of control-IQ technology. Diabetes Technol Ther. 2021;23(2):120-127. doi:10.1089/dia.2020.0388.

Brown

Kovatchev

Raghinaru

, et al. Six-month randomized, multicenter trial of closed-loop control in type 1 diabetes. N Engl J Med. 2020;381:1707-1717. doi:10.1530/ey.17.10.13.

Breton

Kanapka

Beck

, et al. A randomized trial of closed-loop control in children with type 1 diabetes. N Engl J Med. 2020;383(9):836-845. doi:10.1056/nejmoa2004736.

Isganaitis

Raghinaru

Ambler-Osborn

, et al. Closed-loop insulin therapy improves glycemic control in adolescents and young adults: outcomes from the international diabetes closed-loop trial. Diabetes Technol Ther. 2021;23(25):342-349.

Zisser

Robinson

Bevier

, et al. Bolus calculator: a review of four “smart” insulin pumps. Diabetes Technol Ther. 2008;10(6):441-444. doi:10.1089/dia.2007.0284.

Wolkowicz

Deshpande

Doyle

III Dassau

. Towards insulin monitoring: infrequent kalman filter estimates for diabetes management. IFAC-PapersOnLine. 2020;53(2):15877-15883.

Dadlani

Pinsker

Dassau

Kudva

. Advances in closed-loop insulin delivery systems in patients with type 1 diabetes. Curr Diab Rep. 2018;18:88.

Hyk

Stojek

. Quantifying uncertainty of determination by standard additions and serial dilutions methods taking into account standard uncertainties in both axes. Anal Chem. 2013;85(12):5933-5939.

10.

Clarke

Cox

Gonder-Frederick

Carter

Pohl

. Evaluating clinical accuracy of systems for self-monitoring of blood glucose. Diabetes Care. 1987;10(5):622-628.

11.

Vargas

Teymourian

Tehrani

, et al. Enzymatic/immunoassay dual-biomarker sensing chip: towards decentralized insulin/glucose detection. Angew Chem Int Ed. 2019;58:6376-6379.

12.

Ruiz-Valdepenas Montiel

Povedano

Vargas

, et al. Comparison of different strategies for the development of highly sensitive electrochemical nucleic acid biosensors using neither nanomaterials nor nucleic acid amplification. ACS Sens. 2018;3(1):211-221. doi:10.1021/acssensors.7b00869.

13.

Prasad

Madhuri

Tiwari

Sharma

. Imprinting molecular recognition sites on multiwalled carbon nanotubes surface for electrochemical detection of insulin in real samples. Electrochim Acta. 2010;55(28):9146-9156.

14.

Luo

Davis

. The label free picomolar detection of insulin in blood serum. Biosens Bioelectron. 2013;39(1):21-25.

15.

Luo

Freeman

James

Davis

. Ultrasensitive label free electrical detection of insulin in neat blood serum. Anal Chem. 2013;83:4129-4134.

16.

Zhao

. An insulin molecularly imprinted electrochemical sensor based on epitope imprinting. Chinese J Anal Chem. 2017;45(9):1360-1366.

17.

Zhao

Zhang

Xiang

Deng

. An electrochemical dual-signaling aptasensor for the ultrasensitive detection of insulin. Anal Biochem. 2019;573:30-36.

18.

Liu

Wang

Guo

. Gold nanoparticles/Orange II functionalized graphene nanohybrid based electrochemical aptasensor for label-free determination of insulin. RSC Adv. 2016;6:30732-30738.

19.

Zhu

Kang

, et al. Dual-mode electrochemical immunoassay for insulin based on Cu₇S₄-Au as a double signal indicator. ACS Appl Mater Interfaces. 2018;10(45):38791-38798. doi:10.1021/acsami.8b14908.

20.

Abazar

Noorbakhsh

. Chitosan-carbon quantum dots as a new platform for highly sensitive insulin impedimetric aptasensor. Sens Actuators B Chem. 2020;304:127281.

21.

Sun

Zhou

, et al. Electrochemical sandwich immunoassay for insulin detection based on the use of gold nanoparticle-modified MoS₂ nanosheets and the hybridization chain reaction. Microchim Acta. 2019;186(1):6. doi:10.1007/s00604-018-3124-8.

22.

Singh

Krishnan

. Voltammetric immunosensor assembled on carbon-pyrenyl nanostructures for clinical diagnosis of type of diabetes. Anal Chem. 2015;87:2648-2654.

23.

Heinemann

Weyer

Rauhaus

Heinrichs

Heise

. Variability of the metabolic effect of soluble insulin and the rapid-acting insulin analog insulin aspart. Diabetes Care. 1998;21:1910-1914.

24.

Toffanin

Zisser

Doyle

III Dassau

. Dynamic insulin on board: incorporation of circadian insulin sensitivity variation. J Diabetes Sci Technol. 2013;7:928-940.

25.

Walsh

Roberts

Heinemann

. Confusion regarding duration of insulin action: a potential source for major insulin dose errors by bolus calculators. J Diabetes Sci Technol. 2014;8:170-178.

26.

Schiavon

Dalla Man

Cobelli

. Modeling subcutaneous absorption of fast-acting insulin. IEEE Trans Biomed Eng. 2017;65(9):2079-2086.

27.

Heurtault

Reix

Meyer

, et al. Extensive study of human insulin immunoassays: promises and pitfalls for insulin analogue detection and quantification. Clin Chem Lab Med. 2014;52(3):355-362. doi:10.1515/cclm-2013-0427.

28.

Yun

Hur

Moon

Kim

. The effects of anti-insulin antibodies and cross-reactivity with human recombinant insulin analogues in the E170 insulin immunometric assay. Korean J Lab Med. 2011;31:22-29.

29.

Peters

Maurer

. Systematic comparison of bias and precision data obtained with multiple-point and one-point calibration in six validated multianalyte assays for quantification of drugs in human plasma. Anal Chem. 2007;79:4967-4976.

30.

Khamis

Klemm

Adamko

El-Aneed

. Comparison of accuracy and precision between multipoint calibration, single point calibration, and relative quantification for targeted metabolomic analysis. Anal Bioanal Chem. 2018;410:5899-5913.

31.

Brown

Gillam

. Comparison of quantification strategies for one-point standard addition calibration: the homoscedastic case. Anal Chim Acta. 2012;716:108-111.

32.

Kościelniak

Wieczorek

Univariate analytical calibration methods and procedures. A review. Anal Chim Acta. 2016;944:14-28.

33.

Dauphin-Ducharme

Ortega

Plaxco

. Calibration-free electrochemical biosensors supporting accurate molecular measurements directly in undiluted whole blood. J Am Chem Soc. 2017;139:11207-11213.

34.

Dai

, et al. High frequency, calibration-free molecular measurements: in situ in the living body. Chemical Science. 2019;10(47):10843-10848. doi:10.1039/c9sc04434e.

35.

Zhu

Han

, et al. Insulin-binding aptamer-conjugated graphene oxide for insulin detection. Analyst. 2011;136(20):4138-4140. doi:10.1039/c1an15407a.

36.

Midinov

White

. Electrochemical aptamer-based sensor for real-time monitoring of insulin. ACS Sens. 2019;4(2):498-503.

37.

Hao

Zhu

Wang

, et al. Real-time monitoring of insulin using a graphene field-effect transistor aptameric nanosensor. ACS Appl Mater Interfaces. 2017;9(33):27504-27511. doi:10.1021/acsami.7b07684.

38.

Yin

Sempionatto

Moon

Teymourian

Wang

Tang

. Touch-based stressless cortisol sensing. Adv Mater. 2021;33:2008465.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.04 MB