Wearable electrochemical sensors for real-time monitoring in diabetes mellitus and associated complications

Enzyme

Enzymes are crucial in constructing wearable electrochemical biosensors. Through simple enzymatic reactions targeting biomarkers such as glucose, lactate, and uric acid, biosensors offer numerous advantages, manifesting in cost-effectiveness, rapid result acquisition, and necessitating minimal sample volume for measurement.

The fundamental concept of an enzyme-based biosensing electrode involves immobilizing enzyme molecules in close proximity to an electrode surface. Among the enzyme families associated with biomarkers for DM and its related complications, glucose oxidase (GO_X) is commonly employed for glucose monitoring in DM management^[101]. Similarly, lactate oxidase (LO_X) and lactate dehydrogenase (LDH) are used in developing lactate biosensors^[102], while uricase is employed for uric detection^[103]. According to the difference of electron transfer mechanisms, various architectural designs of enzyme-based electrochemical biosensors measuring free electrons emitted from enzymatic reactions are illustrated in Figure 2A-C.

Figure 2. Response mechanisms of enzymatic electrochemical sensors and enzyme immobilization methods. (A) Schematic illustration of the first-generation enzymatic sensor utilizes the concept of electrocatalytic detection; (B) Schematic representation of the second-generation enzymatic sensor incorporates redox mediators; (C) The third-generation enzymatic sensor; (D) Schematic illustration for physical enzyme immobilization of soaking method, drying method, entrapment method, and ionic binding, respectively; (E) Diagram illustrating enzyme immobilization of chemical methods of covalent bonding and crosslinking, respectively.

First-generation electrochemical biosensors utilize enzymes that catalyze reactions involving either the consumption of electroactive reactants (e.g., O₂) or the production of electroactive species (e.g., H₂O₂) [Figure 2A]. The depletion of the target substrate (e.g., glucose, lactate, etc.) or the increase in the product (electrochemically active H₂O₂) is then monitored to determine the concentration of the target analyte. The current, which is directly proportional to the concentration of biomarkers, arises from the oxidation of H₂O₂ at the WE. It is important to note that oxygen serves as the physiological electron acceptor in this oxidase-based concept. First-generation biosensors have demonstrated remarkable sensitivity and are distinguished by exceedingly short response times, typically on the order of one second ^[104]. Nevertheless, the stoichiometric constraints of oxygen and the inherent fluctuations in its levels within biofluids may introduce inaccuracies in this initial conceptualization^[105]. Additionally, the exigency of a high potential for directly detecting targeted analytes impedes selectivity toward the desired analyte by inducing the unintended oxidation of untargeted analytes^[106].

Second-generation enzyme-based sensors incorporate redox mediators that directly interact with enzymes, addressing the challenge of high potential requirements. These mediators, such as redox dye compounds or transition metal-based compounds, facilitate electron transfer from the active site of the enzyme to the electrode surface at a lower potential than needed for H₂O₂ oxidation^[107,108]. Among the most prevalent and widely recognized mediators in electrochemical biosensing are ferricyanide and ferrocene (Fc). Additionally, other notable mediators include methylene blue, phenazines, methyl violet, alizarin yellow, Prussian blue, thionin, azure A and C, toluidine blue, and inorganic redox ions, all of which find extensive utilization in this field^[109]. Additional enhancements are achieved by substituting oxygen with an electron acceptor capable of facilitating electron transfer from the redox center of the enzyme to the electrode. As depicted in Figure 2B, the MED_RED undergoes oxidation at the electrode surface, thereby generating a current signal that correlates with the concentration of detected biomarkers. Here, MED_OX and MED_RED represent the oxidized and reduced forms of the mediator, respectively^[85]. This integration improves electron transfer acceleration, mitigates the oxygen effect, and enhances stability, sensitivity, and selectivity of biosensors compared to first-generation oxidase-based models^[110]. Careful mediator selection, considering attributes such as solubility and biocompatibility, is crucial for in-vivo applications.

Third-generation enzyme-based sensors utilize either redox or engineered enzymes with modified structures to enable direct electron transfer between the redox center of the enzyme and the electrode surface^[111] [Figure 2C]. The absence of mediators in these biosensors is a notable advantage, operating within a potential window closer to the redox potential of the enzyme, resulting in reduced susceptibility to interfering reactions and enhanced selectivity^[112]. The simplicity of the reaction system, with no additional reagent, contributes to the appeal of third-generation biosensors. Customization of biosensor properties is achievable through protein modification using advanced genetic or chemical engineering techniques and interfacial technologies^[113]. However, challenges such as high costs and technical intricacies, including protein denaturation and renaturation in genetic engineering processes, are significant drawbacks^[114]. Moreover, third-generation biosensors are still being developed and are not commonly used for analysis.

Most commercial sensors and research efforts predominantly use enzyme-based techniques from the first- and second- generations. The issue of needing an oxygen supply has been addressed using semipermeable membranes, which are both easily and cost-effectively produced. There is, however, still a strong demand for advanced sensor designs that can detect biomarkers such as glucose and lactate without requiring oxygen, offering selectivity and energy efficiency^[101,115].

Immobilization of enzyme toward the fabrication of wearable biosensors

Enzyme immobilization refers to the process of constraining the mobility of an enzyme either entirely or significantly within a given space. The utilization of immobilized enzymes presents several advantages, including facile separation from the target analyte upon completion of the reaction, the ability to catalyze reactions iteratively, and the potential for multiple reuses^[116]. However, biocatalysts immobilized on electrodes encounter challenges such as limited substrate accessibility and constrained mass transfer. The overall properties and efficacy of an enzyme are contingent upon factors such as the enzyme type, the immobilization matrix employed, and the methodology applied for enzyme immobilization^[117]. Consequently, ensuring effective immobilization of the enzyme molecule onto the electrode is crucial for facilitating electron transport on the electrode surface. Table 3 summarizes the various immobilization methods for enzymes, highlighting their respective advantages and limitations in enhancing electrode performance. Methods for immobilizing enzymes can be either physical [Figure 2D] or chemical [Figure 2E].

As illustrated in Figure 2D, the physical adsorption method is widely employed to develop enzyme-based biosensors due to its simplicity. Various studies have reported using this method, which involves straightforwardly applying the enzyme solution onto the electrode surface through soaking or drop-casting. The process typically includes an incubation period of overnight or 24 h to facilitate the occurrence of physical adsorption^[122-125]. In this method, non-covalent linkages such as van der Waals forces, hydrophobic interactions, and hydrogen bonding allow the enzyme to adsorb onto the electrode surface without requiring pre-activation of the surface^[118].

Despite maintaining the natural conformation of enzymes, this method faces notable drawbacks, including enzyme leakage and desorption triggered by changes in temperature, pH, and the nature of the analyte solution^[126].

An additional example of a physical method is entrapment [Figure 2D]. In this method, enzymes are physically confined within a porous polymer matrix by linking side chains of the enzyme surface (amino acids) to the polymer surface. Importantly, the enzymes are not directly affixed to the electrode surface, allowing only the traverse of the targeted analyte and products^[106]. The entrapment process involves mixing the enzyme into a monomer solution, followed by the polymerization of the monomer solution through a chemical reaction or by altering experimental conditions. Various procedures are employed in an entrapment method, depending on the entrapment type, such as electropolymerization, photopolymerization^[127,128], the sol-gel process^[129,130], and microencapsulation^[131,132].

This method provides several advantages, including simplicity of process, preservation of intrinsic enzyme characteristics, enhanced stability, reduced enzyme leaching and denaturation, absence or required chemical modification, minimal enzyme demand, various matrix choices (e.g., chitosan, alginate, poly-acrylamide, etc.), and the ability to optimize the microenvironment for the enzyme by modifying the encapsulating materials to achieve the optimal pH and/or polarity^{[118,131,133]}. However, there are some drawbacks, such as enzyme leakage due to the pore size of the support matrix, the occurrence of mass transfer resistance due to increased gel matrix thickness, limited diffusion of the targeted analyte due to size restrictions, and a relatively low enzyme loading capacity^{[120,134-136]}.

In addition to physical immobilization methods, chemical immobilization techniques are also explored to enhance the characterization of immobilized enzymes. Immobilizing enzymes through chemical methods involves irreversible processes where covalent or ionic bonds are established between the enzyme and the support (or surface of the electrode). One of the prevalent methods is chemical covalent bonding [Figure 2E], wherein stable functional groups on enzyme molecules interact with a support matrix^[137]. The functional group on the enzyme should be non-essential for enzymatic activity, typically involving binding through side chains of the ε-amino, thiol, and carboxylic groups^[137,138]. The covalent bonding process typically involves activating the support using glutaraldehyde or carbodiimide as linker molecules, followed by enzyme covalent coupling to the activated sites. Linker molecules serve as multifunctional reagents that act as bridges between the support and enzyme through covalent bonding.

This chemical bonding method has advantages, such as minimal anticipation of conformational changes when linking non-functional amino acids to the support and enhanced tolerance of immobilized enzymes to severe physical and chemical conditions (e.g., temperature, denaturants, and organic solvents). On the other hand, there are notable concerns regarding the harsh conditions during immobilization and the possibility of similar acids at the active site coinciding with the support linkage site, potentially leading to drastic changes in conformation and diminished catalytic properties of the enzyme^[138].

Crosslinking stands out as another chemical method for enzyme immobilization, presenting an irreversible strategy involving establishing intermolecular crosslinks among enzymes [Figure 2E]. This process entails the creation of a robust enzyme network by forming numerous covalent bonds. Employing bi-or multifunctional reagents, such as glutaraldehyde^[139,140], glyoxal^[141,142], and others^[143,144], achieves the desired crosslinking effect. Using cross-linkers in covalently linking enzymes to electrodes ensures heightened durability and stability, surpassing the efficacy of van der Waals or hydrophobic interactions and preventing enzyme leaching. Its widespread adoption in industrial applications is attributed to the simplicity and cost-effectiveness of the method. However, a lack of meticulous regulation in the procedure may lead to substantial enzyme loss. Furthermore, using multifunctional reagents in crosslinking introduces the risk of denaturing or altering the enzyme structure, potentially resulting in a loss of enzymatic activity. Lastly, this method contends with diffusional delays in mass transport within the system, contributing to slow reaction rate and extended equilibrium times^[145,146].

Ionic binding, a less commonly employed strategy for enzyme immobilization in wearable electronics, capitalizes on ionic interactions between the charged surface of the support matrix and amino acids carrying opposite charges on the enzyme surface [Figure 2E]. The quantity of enzyme bound to the support matrix is positively correlated with the surface charge density of the materials. The support matrix materials, including polysaccharide derivatives (e.g., diethylaminoethyl cellulose^[147], carboxymethyl cellulose^[148], and chitosan^[149]), synthetic polymers (e.g., polystyrene derivatives^[150] and polyethylene vinyl alcohol^[151]), and inorganic substances such as silica gel^[152,153], are utilized in this method. In certain instances, physical adsorption may also occur. The key differentiation between ionic binding and physical adsorption pertains to the strength of their interactions. Ionic binding demonstrates a distinctly stronger interaction, albeit not as sturdy as that observed in covalent binding. Ionic binding offers the advantage of minimizing the conformational alterations in the enzyme structure, thereby ensuring the preservation of proper enzymatic activity. However, factors such as pH, temperature, and enzyme concentration can influence ionic interactions. Consequently, maintaining optimal ionic strength becomes imperative to prevent leaching of the immobilized enzyme in suboptimal conditions^[154,155].

Antibodies

Immunological sensors, leveraging the intrinsic specificity of antibody-antigen interactions, are instrumental in detecting biomolecules ranging from nanomolar to femtomolar concentrations^[156-159]. These sensors excel in detecting large biomolecules, such as proteins and cells, by harnessing the changes in surface properties that occur upon antigen-antibody binding. Such interactions are pivotal for signal transduction, primarily through electrochemical mechanisms within the domain of electrochemical immunosensors. It is important to note that while antibodies are highly effective in detecting numerous biomolecules such as some small molecules, the specific detection of certain analytes such as glucose, lactate, potassium, sodium, and uric acid often relies on alternative sensing elements and mechanisms that may not involve traditional antibody-antigen interactions. A diverse array of methodologies is employed in this field, with the competitive and sandwich formats being particularly prominent^[160]. These methodologies adeptly translate biochemical interactions into quantifiable electrical signals, thereby capitalizing on the specificity and sensitivity inherent in antigen-antibody interactions for precise biomarker detection.

The competitive format of electrochemical immunosensors [Figure 3A (i)], primarily involves label-tagged targets that compete with analytes for binding sites on specific antibodies. In this setup, the analyte and a labeled analog are introduced to the system. The more analytes in the sample, the less labeled analog binds to the antibody. The detection and quantification are negatively correlated with the analyte concentration, as the signal decreases when the analyte concentration increases. This format is particularly useful for small molecule detection where the epitope is limited, allowing for precise quantification even at low analyte concentrations.

Figure 3. Response mechanisms of non-enzymatic electrochemical sensors. (A) Scheme for aptamer-based electrochemical sensors of competitive format and sandwich format; (B) An electrochemical aptamer-based sensor blinds and releases a target molecule; (C) Scheme for the synthesis and re-binding of molecular imprinted polymers; (D) Representation of an electrochemical sensor for ion measurement having a polymeric ion-selective membrane, showing the movement of ion and charge separation that results in the surface potential which is measured.

On the other hand, the sandwich format [Figure 3A (ii)], is characterized by its utilization of two antibodies: a capture antibody and a labeled detector antibody. This method is tailored for larger molecules that possess multiple epitopes. Initially, the target analyte is captured by the immobilized antibody on the sensor surface. Subsequently, a second labeled antibody binds to a different epitope on the analyte. The sandwich format’s signal amplification, resulting from multiple bindings, enhances the sensitivity and specificity of the sensor. It is particularly effective for complex samples, providing robust detection even in the presence of interfering substances.

However, these high-precision sensors necessitate meticulous bioreagent incubation and washing steps before detection, underpinning the need for innovative solutions to streamline and adapt these processes for wearable applications^[161,162].

Aptamers

Diverging from traditional antibody-based sensors, electrochemical aptamer-based (E-AB) sensors represent a significant advancement and a paradigm shift in biosensing technology. Aptamers, single-stranded DNA or RNA molecules, are discovered through the systematic evolution of ligands by exponential enrichment (SELEX) process, which involves selecting specific ligands. They are known for their ability to fold into unique spatial configurations, granting them the specificity to bind to a wide array of target molecules such as proteins, small molecules, and even cells. This binding capability is especially valuable for targets that are challenging for antibodies to recognize due to their intricate structures or other biochemical complexities^[163]. This high affinity and specificity make them particularly suitable for use in sensors. A common mechanism associated with E-AB sensors is illustrated in Figure 3B. Initially, aptamers are immobilized on the electrode surface, maintaining their ability to undergo conformational changes upon binding to the target analyte. Upon introduction of the target molecule, the aptamer binds to it, undergoing a conformational change that alters the spatial orientation or the distance between the electrode surface and a reporter molecule, typically a redox-active compound, which is either part of the aptamer structure or closely associated with it. This change in spatial configuration or distance affects the electron transfer rate between the redox reporter and the electrode, a phenomenon meticulously captured by the sensor. Specifically, the target binding event leads to either an increase or decrease in the electron transfer rate, depending on the nature of the conformational change. This, in turn, causes a change in the electrochemical signal, typically measured as a change in current, potential, or impedance. The electrochemical signal is then recorded and quantified, correlating directly with the concentration of the target analyte in the sample. The specificity of the aptamer for its target ensures that the sensor response is highly selective, while the versatility of the aptamer structure allows for the detection of a wide range of targets, from metal ions^[164,165] to protein molecules^[166,167]. Despite the impressive capabilities of E-AB sensors, challenges such as the necessity for specific aptamer sequences for each target and potential for non-specific binding do exist^[168]. However, ongoing advancements in aptamer selection^[169-171], strategies for preventing non-specific binding^[172], and immobilization strategies^[173,174] are progressively refining these sensors, enhancing their specificity, sensitivity, and practical applicability.

MIPs

MIPs are engineered as synthetic receptors, offering a high specificity for target molecules in electrochemical biosensing applications. The synthesis of MIPs entails arranging monomers around a preselected template molecule. Following polymerization and the subsequent removal of the template, the process yields tailored cavities that correspond in shape and functional group orientation to the target molecule. These cavities serve as binding sites for the analyte, causing a detectable electrochemical signal shift that is directly proportional to the analyte concentration in the sample^[175].

Figure 3C delineates the MIP synthesis stages. The process begins with the strategic assembly of functional monomers around a template molecule. After polymerization, with a cross-linker to reinforce the structure, the template is extracted, leaving behind cavities that are primed for the selective re-association with the target analyte^[176].

In the context of electrochemical biosensing, MIPs offer several advantages over biological receptors such as antibodies and aptamers. Unlike these biological counterparts, MIPs are synthesized to be inherently more stable across a broad spectrum of environmental conditions, including extreme pH, temperatures, and solvent compositions^[177]. Their robust nature ensures longevity and reusability, which are significant benefits for practical applications.

However, MIPs also exhibit certain limitations. Their inherent insulating nature can inhibit sensitivity due to poor mass transfer, or challenges such as incomplete removal or saturation of the MIP matrix when exposed to a sample containing the target^[178]. These issues not only affect the sensitivity of the sensor but also its reusability, often necessitating design modifications to enhance performance.

Antibodies, Aptamers, and MIPs as bioaffinity elements, along with their target specificity, stability, and cost-effectiveness in various sensing applications, are comparatively analyzed in Table 4. This detailed comparison aims to equip researchers with a clearer insight into selecting the most suitable bioaffinity element for their specific sensor applications.

Ion-selective membrane

Ion-selective electrodes (ISEs) constitute a well-established potentiometric sensor technology extensively applied across environmental, industrial, and clinical domains to analyze crucial electrolytes^[181-183]. The burgeoning field of wearable sensors is currently centered on using ISEs for real-time, non-invasive monitoring of ions in biological fluids, showing recent advancements geared toward evaluating personal physiological states. Potentiometric sensors, specifically designed for ion sensing, have garnered attention owing to their compact size, exceptional selectivity, prompt response, low energy consumption, user-friendly operation, and economic viability^[184,185].

These potentiometric sensors, comprising a WE and RE, are presumed to have their potentiometric response modeled under open-circuit state, specifically under the condition of zero current. In ISEs, the analytical insight is derived by translating an ion-exchange event into a voltage signal^[100]. The electrode design focuses on perturbations in the local equilibrium at the interface between an ISM and the sample solution. Variations in the activity of the primary ion induce changes in membrane potential and that provided by the RE^[185,186] [Figure 3D].

The potential response of ISEs can be explained by the phase-boundary potential mode, which is based on total equilibrium assumptions^[187,188]. It relies on charge separation at the solution-membrane interface, influenced by primary ion activity through surface chemisorption and equilibrium partitioning of primary ions. The energy required for charge separation stems from the chemical driving force, determined by Gibbs free energy. Through the perselectivity and ion selectivity of ISMs, primary ions can move across the membrane phase unhindered. Upon reaching a state where chemical driving forces counterbalance the opposing electrical Coulombinc driving forces, electrochemical equilibrium is attained, forming an electrical double layer and establishing a potential difference. The basic functional relationship between the potential and activity follows the Nernst equation, denoted as^[188-190]

where z_i is charge on the ion, Q is activity in the solution, E₀ is the standard potential, R is the universal ideal gas constant (8.134 J·K^-1·mol^-1), T is temperature in Kelvins, and F is the Faraday constant (96,485.332 C·mol^-1).

ISMs encompass several crucial constituents, including a matrix/supporting material, plasticizer, anion/cation excluder, and the ionophore. The matrix/supporting material, often high molecular weight polyvinyl chloride (PVC), is chosen for its low toxicity, chemical inertness, and strength. The plasticizer is pivotal in facilitating the integration of the ionophore into the polymeric matrix, thereby exerting significant influence over the ion adsorption/absorption dynamics and the resultant redox characteristics. These characteristics may manifest either in a Nernstian response, characterized by equilibrium partitioning of analyte ions between the sample and the membrane at their interface, or in a non-Nernstian manner, typified by a nonequilibrium steady state condition^[191-193]. The dielectric constant of the plasticizer is crucial, and the redox behavior of ISMs is highly sensitive to plasticizer chemistry and concentrations. Anion/cation excluders are employed to minimize competitive coordination between the ionophore and the counter ion of the analyte. The sensing component, the ionophore, is responsible for ion immobilization, and the redox/double layer capacitance at the ISM-electrode substrate interface, which is determined by the inherent characteristics of WE composed of functional materials acting as ion-to-electron transducers, is expected to demonstrate a stable potentiometric response to changes in ion activity^[194,195].

Electrochemical and physiological biomarkers

As the pioneering study for multiplexed analysis of electrochemical and physiological biomarkers, the hybrid wearable patch was developed to simultaneously measure biochemical and electrophysiological parameters [Figure 8A]. Patch-type electrodes for the lactate biosensor and electrocardiogram were fabricated by leveraging a screen-printing method on the polyester sheet. Human experiments collecting the hybrid signals during 15-30 min cycling activity demonstrated concurrent lactate and heart rate measurement, extracted from electrocardiogram data, with negligible cross-talk^[246]. In addition to patch-type wearable sensors, an integrated wearable health management system with a glucose sensing strip and smart band was reported in 2018 [Figure 8B]. A sweat-based glucose sensing strip was attached on the forehead, and a smart band with an electronic circuit was worn on the wrist for measuring physiological signals, such as blood oxygen saturation level, heart rate, and physical activity. The sensing strip, which collected the sweat, was inserted into the smart band for electrochemical glucose measurements. Daily activity studies with human subjects showed pre- and post-exercise monitoring of glucose and physical activities, which is important to prevent hypoglycemic shock during a high-intensity workout^[247]. Additionally, an ear-worn device was presented as one of multiplexed wearable platforms [Figure 8C]. Sweat-based lactate concentration and pH levels were measured using amperiometric and potentiometric sensors around both ears. At the same time, electrocardiogram (ECG) was recorded around the ears with a SNR level of 18 dB. Acquired signals were transmitted wirelessly to the mobile phone via Bluetooth Low Energy (BLE) communication^[248]. These wearable sensing platforms have limitations in reducing their sizes due to batteries for power supply. As shown in Figure 8D, Yu et al., from the group led by Gao, reported the fully perspiration-powered integrated electronic skin (PPES) operated by highly efficient lactate biofuel cells with a power density of 3.5 mW/cm². During a 60-hour operation, the PPES monitored multiplexed signals including metabolic analytes and the skin temperature. Moreover, with a CNTs-PDMS elastomer-based strain sensor, on-body evaluation of the PPES was performed by measuring muscle contraction and validated the promising use of human-machine interaction for the prosthetic control^[249]. In addition to electrical sensing methods for measuring physical signals, acoustic sensing with ultrasonic transducers was used to acquire more complex hemodynamic parameters. As shown in Figure 8E, a wearable device could continuously and simultaneously monitor multiple biomarkers via enzymatic chemical sensors and physical parameters, such as blood pressure and heart rate, via ultrasonic transducers. Glucose in ISF and lactate, caffeine and alcohol in sweat were measured during the different daily activities. This multiple analysis demonstrated that these skin-worn devices could ensure the understanding of various activities, such as food or caffeine intake, digestion and exercise, and the forecasting of physiological changes^[250].

Figure 8. Multiplexed biosensors coupling electrochemical and physiological biomarkers for monitoring diabetes mellitus. (A) A chemical-electrophysiological hybrid bio-sensing system with screen-printed lactate sensor and ECG electrodes. Reproduced with permission^[246]. Copyright 2016, The Authors; (B) A wearable smart band for detecting sweat-based glucose and vital signs. Reproduced with permission^[247]. Copyright 2016, Springer Nature Limited; (C) A wireless ear-worn device capable of pH, lactate and ECG sensing during physical exercise. Reproduced with permission^[248]. Copyright 2018, WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim; (D) Scheme of biofuel-powered soft biosensor with multiplexed metabolic sensing system. Reproduced with permission^[249]. Copyright 2020, The American Association for the Advancement of Science; (E) An epidermal patch capable of monitoring sweat-based lactate, caffeine, alcohol and ISF-based glucose and ultrasound-based hemodynamic information. Reproduced with permission^[250]. Copyright 2021, The Authors, under exclusive license to Springer Nature Limited, part of Springer Nature. WE: Working electrode; RE: reference electrode; CE: counter electrode; ISF: interstitial fluid; ECG: electrocardiogram.

Multi-electrochemical biomarkers

For more accurate analysis using a wearable non-invasive biosensor, Gao et al., from the group led by Javey, showed the fully integrated biosensor array for in situ perspiration measurement, which is integrated with functional components such as conditioning, processing and wireless transmission in the wristband platform and measures sweat metabolites of glucose and lactate and electrolytes of potassium and sodium [Figure 9A]. On-body perspiration analysis during cycling activity demonstrated real-time and selective measurement of multi-electrolytes and metabolites of a human subject wearing a smart headband and a smart wristband. In addition to flexible band-type configurations, various efforts have been made to develop a form of wearable biosensor that can be easily integrated with existing accessories^[34]. As shown in Figure 9B, Sempionatto et al., from the research group led by Wang, presented a fully integrated eyeglasses platform for real-time monitoring of glucose, lactate and potassium in sweat. The screen-printed electrochemical sensors were placed on the nose pads of eyeglasses, and a wireless electronic system was attached on the arms of the eyeglasses frame^[251]. Efforts toward in-vivo sensing of multiplexed electrochemical parameters involved developing needle-based microscale biosensor arrays [Figure 9C]. Active layers were electrodeposited on a polymeric polyimide substrate and integrated into the curved surface of the medical needle. By measuring real-time glucose, lactate concentration, electrical conductivity and pH, the needle-based biosensor array discriminated cancer from the normal tissues^[252]. For improving the measurement accuracy of multiplexed glucose and lactate sensing, Yokus et al., from the group led by Daniele, designed the wristwatch-shaped analog front-end of the electrochemical cells capable of measuring twelve WE [Figure 9D]. Each WE collected chronoamperometric responses, and an accurate measurement was obtained as an average of all collected data. This measurement strategy showed sensitivities of 26.31 μA·cm^-2·mM^-1 for glucose, 1.49 μA·cm^-2·mM^-1 for lactate, 54 mV·pH^-1 for pH and 0.002 °C^-1 for temperature^[253].

Figure 9. Multiplexed biosensors coupling multi-chemical biomarkers for monitoring diabetes mellitus. (A) Fully integrated sensor arrays for multiplexed monitoring of glucose, lactates, sodium, potassium and skin temperature in sweat. Reproduced with permission^[34]. Copyright 2016, Springer Nature Limited; (B) Eyeglasses-based sensor platform for monitoring glucose, lactate and potassium ions. Reproduced with permission^[251]. Copyright 2017, Royal Society of Chemistry; (C) A flexible biosensor integrated with curved needle for the measurement of electrical conductivity, pH, glucose and lactate. Reproduced with permission^[252]. Copyright 2020, American Chemical Society; (D) A band-type wearable platform with flexible sensor array for simultaneous monitoring of glucose, lactate, pH, and temperature. Reproduced with permission^[253]. Copyright 2020, Elsevier B.V.; (E) Schematic representation of microneedle platform for real-time continuous glucose/ketone monitoring in ISF. Reproduced with permission^[257]. Copyright 2020, American Chemical Society; (F) Touch-based β-HB/glucose sweat sensing device for personalized health monitoring. Reproduced with permission^[258]. Copyright 2022, American Chemical Society; (G) A microfluidic chip integrating dual assays for simultaneous measurements of glucose and insulin. Reproduced with permission^[259]. Copyright 2019 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim; (H) Portable aptamer-based biosensor chip for non-invasive simultaneous monitoring of glucose and insulin. Reproduced with permission^[260]. Copyright 2022, Elsevier B.V. All rights reserved; (I) Epidermal patch-type biosensor platform capable of simultaneous alcohol sensing in sweat and glucose sensing in ISF. Reproduced with permission^[262]. Copyright 2018 The Authors. Published by WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim; (J) Wearable microneedle-based biosensor arrays for real-time sensing of alcohol and glucose in ISF. Reproduced with permission^[263]. Copyright 2022, The Authors, under exclusive license to Springer Nature Limited; (K) Flexible biosensor for label-free detection of cortisol and glucose using non-faradaic electron-ionic charge transfer. Reproduced with permission^[265]. Copyright 2016, Elsevier B.V. All rights reserved; (L) Wearable cortisol and glucose biosensor using monoclonal antibody-oriented approach. Reproduced with permission^[266]. Copyright 2023, American Chemical Society; (M) A wearable electrochemical biosensor for monitoring of multiple metabolites and nutrients. Reproduced with permission^[267]. Copyright 2022, The Author(s), under exclusive license to Springer Nature Limited. RE: Reference electrode; CE: counter electrode; PVA: poly(vinyl alcohol); ISF: interstitial fluid; β-HB: β-hydroxybutyrate.

DKA is considered as a severe complication of both type 1 and type 2 DM and still has high rates of mortality and morbidity^[254]. β-Hydroxybutyrate (β-HB) is well known as a key factor for the diagnosis of DKA and a dominant biomarker of ketone formation. Because of this, many studies have been introduced to fulfill the demands for self-testing of ketone^[255,256]. Despite major improvements in DM management, a real-time continuous ketone bodies monitoring (CKM) has not been as developed as CGM. Recently, Teymourian et al., from the group led by Wang, presented a microneedle-based device that enables the amperometric monitoring of HB and glucose, respectively [Figure 9E]. This CKM microneedle using HB dehydrogenase (HBD) enzymatic reaction solved the major challenges associated with the reliable confinement of the enzyme/cofactor couple (HBD/NAD⁺) and the fouling-free detection of nicotinamide adenine dinucleotide (NADH). The potential feasibility of CKM microneedles was demonstrated by showing an analytic performance, with good stability and high selectivity and sensitivity with a low detection limit of 50 μM in artificial ISF^[257]. Furthermore, Moon et al. from the same group reported a touch-based sweat HB detection method, which relies on collecting fingertip sweat and transporting the sweat to a biocatalytic layer through a porous poly(vinyl alcohol) (PVA) hydrogel [Figure 9F]. The response on sweat HB was enhanced by mediating the oxidation reaction of NADH. A dual disposable biosensing device, integrating with ketone and glucose sensors, was demonstrated within healthy human subjects by simultaneously measuring the change of sweat HB and glucose levels during the intake of ketone supplements and glucose drinks^[258].

Insulin is another significant biomarker for diagnosing DM, but the detection of insulin using a non-invasive wearable platform has some limitations due to extremely low concentrations in biofluids and interference from constituents in blood. Moreover, for detecting glucose and insulin simultaneously to improve the glycemic control and personalized health monitoring, a major challenge relies on integrating enzymatic and immunosensing electrochemical detection methods into a single system. Vargas et al., from the group led by Wang, addressed the different assay requirements by fabricating dual diabetes biomarker chip [Figure 9G]. Glucose detection was performed amperometrically through the biocatalytic oxidation of glucose, and insulin detection was conducted through the sandwich immunoreaction assay using horseradish peroxidase (HRP) as an enzymatic label and 3,3’,5,5’-tetramethylbenzidine (TMB)/H₂O₂ as the mediator/detection system. The glucose and insulin chip showed a considerable promise by accurately measuring two biomarkers with large concentration differences in blood and saliva^[259]. Recently, Liu et al. developed an electrochemical aptasensor on carbon electrode, which enables real-time detection of glucose and insulin in saliva using two thiolated aptamers targeting glucose and insulin, respectively [Figure 9H]. The real-time detection was realized with a portable and wireless interface and demonstrated the sensing performance of the linearity in the range of 0.1-50 mM for glucose [limit of detection (LOD), 0.08 mM] and 0.05-15 nM for insulin (LOD, 0.85 nM)^[260].

Cortisol is a glucocorticoid hormone that is significant for regulating glucose metabolism. In order to keep the glucose amounts in plasma, it increases the expression of glucogenesis-related enzymes. Moreover, according to recent studies, insufficient pancreatic insulin secretion may result from the disturbance in the circadian rhythm caused by cortisol^[261]. Therefore, the simultaneous monitoring of glucose and cortisol can play an important role in better forecasting the onset of diabetes and/or stress-related symptoms. Munje et al., from the group led by Prasad, developed a wearable electrochemical biosensor for the multi-biomarker detection of glucose and cortisol in sweat [Figure 9I]. Ultrasensitive detection of low-volume glucose and cortisol was realized with stacked metal/metal-oxide thin films within porous polyamide substrates. Monoclonal GO_X and cortisol antibodies were used to bind to GO_X enzyme and cortisol antigen molecules, respectively. Combinatorial detection of glucose and cortisol was demonstrated through electrochemical impedance spectroscopy with LOD of 0.1mg/dL in human sweat^[262]. Very recently, Tian et al., from the group led by Zhang, engineered a fully integrated system for cortisol biosensing with an antibody-oriented immunosensor in human sweat [Figure 9J]. This system was demonstrated with several volunteers that it has a good linear range from 1 pg/mL to 1 μg/mL (LOD, 0.26 pg/mL). Furthermore, the immunosensor showed relatively long-term stability with only 4.1% decay after nine days of storage^[263].

Alcohol and many other nutrients, such as amino acids and vitamins, can be additional factors for DM monitoring in terms of various food intake activities^[264]. Kim et al., from the group led by Wang, presented a wearable device capable of measuring sweat-alcohol and ISF-glucose in human subjects while intaking food and drinking alcohol [Figure 9K]. This platform allowed on-demand sampling and analysis of two different biofluids, sweat and ISF, through sweat stimulation at an anode and ISF extraction at a cathode. The dual measurements of sweat-based alcohol and ISF-based glucose demonstrated good correlations of their blood levels^[265]. With an advance of electronic systems and wearable technology, the same group developed wearable microneedle-based sensor arrays for real-time monitoring of various biomarkers, such as lactate and glucose, or alcohol and glucose in ISF biofluid [Figure 9L]. Reusability was enhanced using disposable MNA, electronics and optimized system integration. Through the comparison of gold-standard measurements in blood or breath, dual-analyte measurements using this system showed a good correlation with a commercial technology^[266]. For more sophisticated metabolic profiling and monitoring, a universal wearable biosensing strategy using molecularly imprinted polymer (MIP)-based artificial antibodies was presented by Wang et al. [Figure 9M]. Autonomous and continuous molecular analysis was realized by incorporating iontophoresis-based sweat induction and microfluidic-based sweat sampling. Several human trials demonstrated the possibility of personalized monitoring of five essential or conditionally essential amino acids for central fatigue, standard dietary intake, and COVID-19 severity^[267].

Despite these promising developments, effectively using wearable DM monitoring systems still requires more optimization and validation to thoroughly assess their reliability, versatility, and accuracy. These detailed studies will determine how well non-invasive methods can monitor biomarkers in biofluids and will compare the non-invasive data with traditional blood glucose measurements. Other limitations with wearable approaches include inconsistent collection of biofluids, contamination on the electrode surface, and the performance changes due to various physiological factors (such as temperature and pH) on the accuracy of the readings.

To enhance DM management, multiplexed analysis for the diabetes-related biomarkers, such as electrophysiological factors, insulin, ketone, alcohol, etc., can achieve a more comprehensive assessment of patient health than singular glucose measurement. Nevertheless, the implementation of multi-analyte sensing platforms entails several significant challenges, including the integration of different surface modifications and sensing modalities into a single wearable platform, the mitigation of analytical interference or cross-talk among different analytes, the necessity for receptor regeneration in bioaffinity-based assay methods, and the maximization of sensitivity for analytes in biofluids due to the lower concentration than blood. Consequently, future research should proceed in the direction of integrating glucose sensing with other novel analytes and skills for clinical use while overcoming these hurdles.

Artificial neural network

Artificial neural networks (ANN) have become a cornerstone in advancing DM monitoring, primarily due to their ability to model complex relationships within data. The structure of an ANN is inspired by the human brain, comprising layers of interconnected nodes or neurons. These include input layers, which receive data such as blood glucose levels, dietary information and physical activity, hidden layers where the computation and pattern recognition occur, and an output layer that delivers the prediction or classification. The principle behind ANNs involves the neurons in each layer processing the input data and applying weights that are adjusted during training to minimize error in the output. This process, known as backpropagation, is critical for the ANN to learn from the data. As the network processes more data, it fine-tunes its weights, enhancing its predictive accuracy. In DM monitoring, various types of ANNs, such as recurrent neural networks (RNN), long short-term memory (LSTM) and convolutional neural networks (CNN), have been employed. Among them, Gu et al. presented the SugarMate, a smartphone-based blood glucose inference system, which offers a non-invasive alternative to CGM^[272]. It integrated smartphone sensor data with food, drug, and insulin records to measure physical activity and sleep quality. Facing challenges of imbalanced and limited data, SugarMate employed Md³RNN, a deep RNN model with grouped input layers, for fine-grained blood glucose level inference. This model utilized limited personal and grouped-user data, achieving 82.14% accuracy in evaluations with 112 participants. Another recurrent convolutional neural network model was developed for precise blood glucose level forecasting in type 1 diabetes (T1D)^[274]. The model achieves a root-mean-square error (RMSE) of 9.38 ± 0.71 mg/dL over 30 min and 18.87 ± 2.25 mg/dL over 60 min for simulated cases, and 21.07 ± 2.35 mg/dL for 30 min, 33.27% ± 4.79% for 60 min in real patient cases. In terms of LSTM-based models, Beauchamp et al. introduced a novel approach for managing blood glucose levels in T1D using a dual LSTM architecture^[273]. This model inverted the traditional “what-if” scenario, instead providing recommendations for insulin dosages or carbohydrate intake to achieve target glucose levels. The technique was further enhanced by embedding the LSTM chain within a deep residual architecture, optimizing time series forecasting. This model demonstrated significant improvements for self-managing blood glucose levels in T1D over conventional methods by testing with real patient data. CNN was also used to forecast accurate glucose levels. Li et al. introduced GluNet, a personalized deep neural network framework for accurate short-term blood glucose prediction in T1D^[271]. GluNet employed deep learning techniques and historical data, including glucose measurements, meals, and insulin doses. In simulations, it achieved a RMSE of 8.88 ± 0.77 mg/dL for 30-minute predictions and 19.90 ± 3.17 mg/dL for 60-minute predictions. Clinical data tests also demonstrated its effectiveness in glucose forecasting.

Support vector machine

Support vector machines (SVM) are a type of supervised learning algorithm applicable to both classification and regression tasks. For classification, SVM distinguishes between two distinct classes by determining the most appropriate hyperplane that divides them, informed by the labeled training data. Depending on the task complexity and inter-feature relationships, SVM can employ either linear separation or more complex, non-linear methods. Non-linear classification challenges are managed by kernel functions that elevate the data into a higher-dimensional space, making the problem linearly separable. When it comes to regression, SVM aims to establish a function that lies within an acceptable deviation from the given targets, prioritizing the minimization of errors with a more flexible margin. This approach is similar to non-linear regression, wherein kernel functions are again leveraged to transform the data, thereby facilitating a linear approach to an intrinsically non-linear regression problem. For forecasting glycemia, CGM study from 25 DM type 1 patients was carried out in order to explore the viability of on-device computations with SVM^[268]. This study found that a smartphone could predict glucose levels over a 15-minute horizon with a RMSE of 19.90 mg/dL within 34.89 s. Moreover, for hypoglycemia prediction of DM type 2 patients, Sudharsan et al. trained a probabilistic model using self-monitored blood glucose (SMBG) values and validated their model that enables the prediction of a hypoglycemia event in the next 24 h with a sensitivity of 92 % and specificity of 70%^[269]. Bertachi et al. examined using ML to predict nocturnal hypoglycemia (NH) in T1D patients on multiple insulin doses^[280]. Ten adults were monitored over 12 weeks, with data from CGM and activity trackers. ML models, particularly SVM, effectively anticipated NH with over 70% potential avoidance, demonstrating high sensitivity and specificity. The study validates ML as a viable tool for predicting NH in T1D management.

Decision tree

A decision tree (DT) is a method used to simplify data classification through a series of binary decisions, leading to a clear partition of data. It operates by splitting data at nodes based on attribute values until the data in each partition is sufficiently similar. DTs are utilized extensively in data mining and ML for regression and classification and are favored for their interpretability, offering valuable insights into key data attributes. For discrete target values, classification trees are utilized, while regression trees handle continuous ones. DTs can be adapted to various forms, including tree ensembles, to suit specific analytical needs and outcomes. One study introduced a non-invasive sweat sensor that measures glucose levels in eccrine sweat using electrochemical impedance spectroscopy^[279]. With just 1-5 µL of sweat, the device provided real-time glucose measurements every few minutes. A ML algorithm, specifically DT regression, was used to process sensor data from three human subjects, achieving a high accuracy (R² of 0.94 and RMSE of 0.1 mg/dL).

Autoregressive integrated moving average

The autoregressive integrated moving average (ARIMA) model is a sophisticated linear method for forecasting future values in time series data, extending from past values. It integrates differencing operators with autoregressive and moving average components, making it a more generalized form of the autoregressive moving average (ARMA) model. The strength of ARIMA lies in its ability to accommodate a wide array of non-stationary series, thereby enhancing its applicability and accuracy in time series forecasting. A study monitored five DM patients for four days and forecasted glycemia by adapting ARIMA models with prediction horizons of 30 and 60 min, obtaining RMSEs of 22.9 and 42.2 mg/dL, respectively ^[281]. In addition, the ARIMA model was employed to forecast future trends in blood glucose levels, specifically targeting hypoglycemia and hyperglycemia^[282]. This predictive framework was developed using CGM data from 100 patients, encompassing both T1D and type 2 diabetes (T2D). The model effectively generated early warnings, achieving a high sensitivity of 100% while maintaining a low false rate of 9.4%.

Transfer learning

Transfer learning is a ML technique where a model developed for one task is reused as the starting point for a model on a second task. This approach leverages pre-trained models to solve similar problems, significantly addressing the issue related to the need for a large amount of training data. It is particularly effective in deep learning, where substantial datasets and extensive training are typically required. Transfer learning is instrumental in various applications, including image and speech recognition, where it utilizes knowledge gained from one domain to improve learning in another, enhancing efficiency and performance in the learning process. Regarding DM monitoring, a study developed deep-learning methods for predicting blood glucose levels in T2D patients using continuous monitoring data, aiming to improve glycemic control and reduce complications^[277]. The main challenges were small, imbalanced patient datasets and rare hypo/hyperglycemic events. By employing transfer learning and data augmentation, including techniques such as mix-up and generative models, the study achieved over 95% prediction accuracy and 90% sensitivity within a clinically relevant one-hour horizon. The approach was also effective for T1D, demonstrating its broader applicability. De Bois et al. also addressed the challenge of insufficient data in healthcare deep learning by proposing a multisource adversarial transfer learning framework^[278]. This method enhanced data transfer across multiple health sources, learning a generalized feature representation. Applied to diabetic glucose forecasting, the approach showed improved statistical and clinical accuracies, particularly effective with diverse datasets or limited data scenarios. The adversarial transfer learning framework outperformed standard methods, promoting a more universal feature representation and improving deep model training with shared healthcare data.

For the advancement of DM monitoring, signal analysis using ML or deep learning techniques has been extensively researched for purposes such as forecasting glycemia, hypoglycemia, predictive monitoring for glucose levels, and insulin recommendation. Currently, these studies mainly utilize electrochemical signals derived from a single biomarker collected via invasive or minimally invasive methods, such as blood glucose or commercial CGM devices, as inputs for model training. Although there is research on applying the decision tree (DT) model to data non-invasively collected from wearable sweat-based glucose sensors to measure actual glucose concentration from sweat, this approach still faces limitations in continuous and predictive monitoring due to insufficient samples and unreliable concentration matching. For reliable, accurate, and timely monitoring of DM using wearable electrochemical sensors, developing new artificial intelligence (AI) models is essential. These models should be capable of matching the concentrations of biomarkers from various biofluids to their actual concentrations, quantitatively comparing the concentration differences between non-diabetics and diabetics, analyzing the correlations among multi-biomarkers, and examining the trends in concentrations according to the lifestyle and dietary habits of an individual.