Bioanalytical sensors using the heat-transfer method HTM and related techniques

Abstract

This review provides an overview on bio- and chemosensors based on a thermal transducer platform that monitors the thermal interface resistance R _th between a solid chip and the supernatant liquid. The R _th parameter responds in a surprisingly strong way to molecular-scale changes at the solid–liquid interface, which can be measured thermometrically, using for instance thermocouples in combination with a controllable heat source. In 2012, the effect was first observed during on-chip denaturation experiments on complementary and mismatched DNA duplexes that differ in their melting temperature. Since then, the concept is addressed as heat-transfer method, in short HTM, and numerous applications of the basic sensing principle were identified. Functionalizing the chip with bioreceptors such as molecularly imprinted polymers makes it possible to detect neurotransmitters, inflammation markers, viruses, and environmental pollutants. In combination with aptamer-type receptors, it is also possible to detect proteins at low concentrations. Changing the receptors to surface-imprinted polymers has opened up new possibilities for quantitative bacterial detection and identification in complex matrices. In receptor-free variants, HTM was successfully used to characterize lipid vesicles and eukaryotic cells (yeast strains, cancer cell lines), the latter showing spontaneous detachment under influence of the temperature gradient inherent to HTM. We will also address modifications to the original HTM technique such as M-HTM, inverted HTM, thermal wave transport analysis TWTA, and the hot-wire principle. The article concludes with an assessment of the possibilities and current limitations of the method, together with a technological forecast.

1. Introduction

Biosensors are a class of analytical devices for detecting a wide variety of “targets” such as small molecules, DNA fragments, proteins, enzymes, viruses, bacteria and entire cells. Their major field of application is in medical diagnostics while there is also considerable potential for food-safety analysis and environmental monitoring. While the vast majority of biosensors is still in the research & development state, there are prominent examples that are broadly known: the glucose sensors for diabetes patients, the pregnancy test, and the antigen test for the Covid-19 corona virus [1–3]; other applications are upcoming. All tests have in common that they can be handled by non-trained persons, it is facile to interpret the result and, most importantly, analysis takes only a few minutes in most cases. According to the definition by the International Union of Pure and Applied Chemistry, a biosensor is “A device that uses specific biochemical reactions mediated by isolated enzymes, immunosystems, tissues, organelles or whole cells to detect chemical compounds usually by electrical, thermal or optical signals” [4]. This means that the receptors are derived from nature, which holds also for antibodies and single-stranded DNA with known base sequences, as well as aptamers [5, 6]. Aptamers are specially selected oligo-nucleotides that undergo self-folding in a way that they can recognize and bind, in most of their applications, protein molecules from solution or proteins located on cell membranes. To comply with the broad range of analytical targets, there are also biomimetic receptors, sometimes referred to as “artificial antibodies”, which can be roughly divided into molecularly imprinted polymers MIPs [7, 8], and surface-imprinted polymers SIPs [9, 10]. These polymer-based materials can operate under harsh conditions, show little or no degradation over time, and their selectivity (expressed in terms of the affinity constant) comes gradually close to the performance of natural antibodies.

A key aspect of biosensors, including the biomimetic ones, is whether they can provide quantitative information on the respective target concentration, or simply result in a yes-or-no answer. The pregnancy- and Covid-antigen tests are of the second type; however, target concentrations need to exceed a threshold value to trigger a positive sensor response. In the majority of other applications, including the detection of glucose in blood, it is clear that accurate, quantitative results are required. To achieve this, the receptors must be coupled to a transducer that translates the biomolecular recognition into an absolute concentration value. Generally, literature mentions four overarching transducer categories, being i) optical, ii) mass sensitive, iii) electrochemical, and iv) thermal [11, 12]. Each category has several subtypes so that certain sources refer to even seven different transducer families, some of them being combinations of the four fundamental ones. In the present overview article, we will mainly focus on thermal transducers, and specifically on novel embodiments that exploit thermal interface resistance as measurand. Thermal transducers in bioanalysis have a long tradition in the form of calorimeters that were used, for example, to study the stability and melting temperature of DNA duplexes [13], the enzymatic decomposition of glucose, urea, penicillin and ascorbic acid [14, 15], metabolic heat emerging from biofilms [16], and glycolytic oscillations in yeast cultures [17]. Calorimeters are instruments traditionally used for fundamental research, instead of for point-of-care applications. Yet the results of these early studies currently play a key role in gene analysis and the design of DNA microarrays [18, 19]. It is worth mentioning that calorimeters probe reaction enthalpies and the signals are therefore transient in nature, changes in temperature or heating power are only observed at the moment that a (bio-) chemical reaction takes place.

Conversely, sensors measuring the thermal interface resistance, R _th, probe signals that are persistent in time as observed in on-chip thermal denaturation experiments on double-stranded DNA by van Grinsven et al. [20]: the double-stranded (ds) state has a low R _th, the single-stranded (ss) state has a high one. Hence, thermal resistance will not change unless a transition between both states is induced on purpose. Within this review, we will show how the “heat transfer method HTM” has evolved over the last 10 years with a broad range of bioanalytical applications and a variety of new instrumental designs. Throughout, all devices and applications discussed within this article focus on the thermal interface resistance, unless stated explicitly otherwise. The main idea is to make HTM ready from proofs-of-concept in the laboratory to on-site applications in medical diagnostics, drug testing, food-safety analysis, and environmental monitoring. We will critically benchmark the performance of HTM with respect to other label-free biosensing methods and discuss current limitations and future perspectives. Several applications, such as the detection of bacteria with thermometers, will look surprisingly easy; we will also discuss innovations that are not yet fully understood, but most promising from both a fundamental and an applied perspective. This article is arranged along the lines of selected examples in a close to chronicle order from the beginnings in 2012 to date with a technical forecast on what may be expected in the years to come.

2. Instrumental aspects of HTM and DNA characterization

The fact that the binding of biomolecules changes the thermal interface resistance between a solid and a liquid was not predicted by theory, but came in as an unexpected observation. van Grinsven et al. studied the denaturation of double-stranded (ds) DNA on synthetic diamond layers when exposing them to a NaOH solution [21]. Following the impedance signal during NaOH exposure and rinsing steps, they found that molecular brushes of complementary DNA duplexes required a longer time to denature than duplexes with weaker bonds due to base-pair mismatches. The chip was temperature-stabilized from its backside to 19.3 °C by using a copper block to which a power resistor was attached, see the basic geometry in Figure 1. The time constants for DNA denaturation correlated with their theoretical melting temperatures. Then, an attempt was made to denature thermally by ramping up the chip temperature using only PBS buffer without chemicals. Thereby, the chip temperature T ₁ was measured with a miniature thermocouple (type K, 0.5 mm diameter), a second thermocouple measured the temperature T ₂ of the liquid inside the sensor compartment [20]. By heating a chip functionalized with ds-DNA from T ₁ = 35 to 90 °C at a constant rate, T ₂ increased linearly as expected but with a tiny deviation from linearity at around T ₁ ≈ 60 °C, see Figure 2a. Using fluorescence microscopy, it was confirmed that the small anomaly of T ₂ was related to DNA denaturation, but this signal appeared still unobtrusive. The picture changed completely when plotting the data in the sense of thermal resistance R _th that is defined by the following equation used to describe heat removal from packaged electronic components [22].

$$R_{\text{th}}=\frac{T_1-T_2}{P}$$ (1)

Figure 1:

Overview on heat-transfer devices for bioanalytical applications. (a) Scheme of the HTM concept with the chip heated from the backside to a temperature T ₁ and measuring the temperature T ₂ of the liquid. The heat-transfer resistance R _th responds sensitively to the properties of the solid–liquid interface. The examples illustrate DNA analysis, small-molecule detection with MIPs, cell detection with SIPs, and lipid vesicles. (b) Photograph of a heat-transfer device. (c) Cross section of the design with the sensing compartment, thermocouples, and the in- and outlet for samples. Thermal-insulation elements reduce parasitic heat loss. Panel (a) is reprinted with permission from ref. [23], copyright 2014 American Chemical Society. Panel (b) is reprinted with permission from ref. [24], copyright 2017 John Wiley & Sons. Panel (c) is adapted from ref. [25] under Creative Commons CC BY license.

Figure 2:

Experimental data and schematic concept of the heat-transfer effect in case of thermally-induced DNA denaturation. (a) Temperature profile of an on-chip thermal denaturation experiment on ds-DNA. The chip temperature is repetitively ramped from T ₁ = 30 °C to 90 °C and back, the temperature T ₂ of the liquid follows the triangular profile. The small anomaly of T ₂ during the first heating (see circle) corresponds to thermal denaturation [20]. (b) R _th signature of thermal DNA denaturation on a synthetic sapphire chip, showing an R _th increase upon the transition from ds- to ss-DNA. The melting temperature is highest for complementary duplexes (blue) and reduced in presence of base-pair mismatches (green, yellow) [26]. (c) Difference in chip-surface coverage between ds- and ss-DNA, explaining the thermal insulation effect after denaturation [20]. The panels (a) and (c) are reprinted with permission from ref. [20]. Copyright 2012 American Chemical Society. Panel (b) is reprinted with permission from ref. [27]. Copyright 2016 Elsevier.

In this application, T ₁ is the temperature of the copper block, T ₂ is the slightly lower temperature in the liquid, and P is the power provided by the PID controlled power resistor [23]. When exact dimensions are known, Eq. (1) can be used to determine the thermal conductivity, κ of a material. Many HTM applications described within this article use a typical chip size of 10 × 10 mm² and an O-ring with ca. 7 mm inner diameter defines the active area in a sample compartment that is 4–5 mm high on the inside. The tip of the thermocouple for T ₂ is located in the centre of the compartment, at a height of ≈ 2.0 mm above the chip surface. In case of DNA, Figure 2b shows a steplike increase from low R _th in the double-stranded state towards high R _th values for single-stranded probe DNA with its random-coil structure after denaturation (see Figure 2c), which increases the surface coverage. The inflection point of the R _th(T ₁) curves corresponds to the DNA-melting temperature that decreases when single-nucleotide polymorphisms are present in the target DNA. The amplitude of the R _th change has also been used to determine the hybridization efficiency between probe- and target DNA [26].

We note that the design of Figure 1a is prone to heat loss and only ±10 % of the applied heating power P passes the actual interface between the chip and the liquid, the rest is dissipated to the environment. This also holds true for recent designs such as the compartment depicted in Figure 1b [24]. Therefore, R _th is not an absolute value but a nominal indicator that still responds sensitively enough to changes at the chip surface-liquid interface. One can write the heater power as P = P _int + P _diss with P _int the power that passes the chip surface-liquid interface and P _diss the dissipated heat flowing to other directions. Hence, the R _th values calculated with Eq. (1) underestimate the true thermal interface resistance by assuming that the complete power P passes the interface of interest. Nevertheless, it is possible to calibrate HTM devices by quantifying the dissipated power with an empty (air-filled) sensor compartment. This results in the real power flux through the interface and Stilman et al. determined the thermal conductivity of water as a reference liquid with a value close to literature data by this way [24]. Since calibration is tedious, several authors publish HTM data in the form of plotting T ₂ or the relative effect size (see Section 3.2) as a function of time, or as a function of target concentration: T ₂ can be measured accurately with an uncertainty ≤0.1 °C when using thermocouples, hereby avoiding the uncertainty regarding the exact P value. Sections 4 and 8 discuss more refined techniques for temperature measurements.

It is possible to perform HTM measurements in two different orientations of the measuring compartment: in refs. [20, 26, 27], the functionalized side of the chip pointed downwards with the heater element on top. This way, one obtains a well-defined temperature gradient without convective movement; this orientation can also be used to detect molecules and particles that stay in suspension. Cells and bacteria have a tendency to sediment and therefore the functionalized side of the chip should point upwards, with the heating element placed underneath. This second configuration will cause convective movement of the fluid, which stimulates bio-recognitions at the expense of a less clearly defined temperature gradient. However, by adjusting the aspect ratio of the measuring compartment, it is possible to avoid convection [24]. In both configurations, there is a temperature gradient at the interface: to maintain it, HTM measurements are performed under no-flow conditions with stagnant liquid except for minor convective flow.

3. Detection of small molecules

3.1. Standard HTM method

Molecularly imprinted polymers (MIPs) are a versatile, comparably new type of biomimetic receptors. Their combination with HTM was demonstrated for the first time by Peeters et al. [28]. Kept at dry conditions, their shelf life is several years and they also retain their recognition functionality at non-physiological pH, temperature, and salinity conditions, which is usually not the case for antibodies. The synthesis protocol for e.g. nicotine MIPs can be found in ref. [28]. In brief, methacrylic acid (MAA) is polymerized in the presence of template molecules, a cross linker, an initiator, and a solvent, creating a porous bulk material with embedded templates. Then, the bulk material is ground to micron-sized particles, followed by removing the templates by Soxhlet extraction. This leaves behind molecular cavities that can rebind target molecules through steric arrangement of hydrogen bonds (in case of, e.g. nicotine) or other non-covalent interactions. For HTM, the MIP particles were fixated on aluminium chips by using a thin PPV (polyphenylene vinylene) polymer layer, PVC (polyvinyl chloride) is also a possibility, resulting in the schematic configuration shown in Figure 3a. The sensing effect is empiric and can be tentatively explained by the different heat-transfer properties of the extracted particles, in which the cavities are filled with water, and less efficient transport when the pores are filled with target molecules. Besides of nicotine, MIPs were also made for the inflammation marker histamine and the neurotransmitter serotonin. Comparing HTM side-by-side with impedance analysis, the detection limit for nicotine was 100 nM with both techniques. Serotine had an LoD of 20 nM with HTM and 5 nM with impedance, for histamine 30 nM (HTM) and 15 nM (impedance), see Figure 3b. The decisive step towards reaching detection limits below 100 nM was the stringent optimization of the PID parameters used to stabilize the temperature of the heating element as described in ref. [29]. Hence the detection limits are widely comparable. Furthermore, it was possible to detect spiked nicotine in untreated saliva samples with 0.25 mM as the lowest measured concentration. Follow-up work on an alternative polymerization scheme, multiplexing towards a sensor array, and single-shot analysis of serotonin in full blood are published in refs. [30–32]. The same concept was furthermore refined towards the detection of melamine residues in milk and glucose in human urine samples [33, 34].

Figure 3:

Heat-flow blocking by the recognition of target molecules in case of MIP-type receptors. (a) Schematic illustration of the heat flow through an extracted, water filled MIP microparticle and after recognition of target molecules. This “pore-blocking effect” gives a qualitative understanding of the observed decrease of the temperature T ₂, which depends on the concentration [28]. (b) Dose-response curve for histamine detection using histamine-imprinted MIPs. The sensor does not respond to the chemically related histidine molecules, non-imprinted particles (NIP) do not show a binding effect for histamine targets but can be used to correct the sensor output for non-specific adsorption in complex samples. The figure is reprinted with permission from ref. [28]. Copyright 2013 Springer Nature.

In a recent work, Ahmadi Tabar et al. developed a sensor with the conventional HTM geometry for detecting residues of perfluorooctanoic acid (PFOA) in spiked soil extract and river-water samples [35]. PFOA is a member of perfluoroalkyl substances (PFAS) that are by-products of the chemical industry. These compounds are found at elevated concentrations not only in soil and water, but also in vegetables, animals, and human tissue, where they are supposed to cause carcinogenic and neurotoxic effects [36]. The MIP-type receptors consisted of acrylamide microparticles fixated on aluminium chips with PVC as an adhesive. By using optical batch-rebinding experiments, the authors determined the binding capacity for target molecules as 123 μM per gram of MIP material, indicating that MIPs can bind large quantities of their molecular targets. The effect size in the sense of Eq. (2) (see Section 3.2) was ca. 0.7 % for PFOA concentrations in the order of 500 nM and the sensor showed a good selectivity with respect to other PFAS compounds. While the lowest measured concentration was 100 pM, the calculated limits of detection were 22 pM in PBS buffer, 91 pM in river water, and 154 pM in spiked soil extract, which is the relevant concentration range for environmental pollution. Reaching low detection limits is not only a matter of selective receptors with a high target affinity, but also of a low signal-to-noise ratio: Ahmadi Tabar achieved this by increasing T ₁ from 37 to 40 °C so that the noise level on T ₂ decreases in comparison to its absolute value. While HTM measurements usually need a stable ambient temperature, increasing T ₁ is efficient to keep the heat flux constant (or increase it) if measurements take place in a warmer environment than usual.

3.2. MIP-functionalized thermocouples

Next, Diliën et al. developed a thermal sensing concept that can be addressed as “inverted HTM” in the sense that the MIP microparticles are not immobilized on the chip but on the thermocouple for T ₂ instead [37]. The MIP particles are deposited on the thermocouple by roll-coating with polylactic acid as adhesive, non-imprinted particles (NIPs) served as a reference channel on a second thermocouple. Figure 4a sketches the idea: the temperature T ₁ is kept constant while the T ₂ thermocouple can be considered as a heat sink through which thermal energy dissipates to the environment. Upon binding of target molecules, the thermal insulation effect of the MIP layer increases (compare to Figure 3a). Hence, there is less heat loss along the thermocouple, which becomes apparent through an increase in T ₂ and a decrease in the heater power. The method has been used for the detection of dopamine, serotonin, and cortisol, resulting in detection limits below 10 μM in buffers, see Figure 4b. These concentrations are well within the physiologically normal range. Advantages are that small-volume samples can be analysed and only a small amount of MIP and NIP particles is needed. Due to the wire-shape geometry of the thermocouple, one may see potential for integrating the thermocouples in a catheter device for in vivo measurements; however, such far reaching application has not been demonstrated yet. When looking at Figure 4b, the effect size for a given concentration c = x is defined by Eq. (2), which does not require information on the temperature T ₁ and the heating power P, as long as both parameters are kept constant.

$$\text{Effect\hspace{0.17em}size}(\text{\%})=\frac{T_2\left(c=x\right)}{T_2\left(x=0\right)}\times 100$$ (2)

Figure 4:

Molecular detection with MIP-type receptors immobilized at the tip of a thermocouple. (a) Principle of an “inverted” HTM setup in which thermocouples are coated with MIP microparticles for dopamine recognition, respectively with non-imprinted particles (NIP) for reference [37]. After dopamine binding, the thermocouple takes up less heat from the sample, resulting in less heat loss along the thermocouple to the ambient and an increase of T ₂. The direction of heat fluxes is indicated by arrows. (b) Dose-response curve for cortisol detection with a limit-of-detection of ca. 8 μM, the NIP-coated thermocouple gives only a negligible response; the effect size is calculated with Eq. (2). Reprinted from ref. [37] with permission by the American Chemical Society under Creative Commons CC-BY-NC-ND license.

As a variation to the approach sketched in Figure 4a, Canfarotta et al. used a method that relies on functionalizing only the actual tip of the thermocouple, the sensitive region hosting the bimetal junction, with MIPs, while the rest of the thermocouple in the sample was left blank [38]. Binding of target molecules decreased the thermal flux reaching this junction so that the thermocouple recorded a decreasing value of T ₂ upon binding target analytes. This does not contradict the results of Diliën, because T ₂ is the temperature at the junction itself, inside the thermocouple, while the actual T ₂ of the liquid may even increase. In the article by Canfarotta, the authors used a novel MIP type (so called nanoMIPs) that was synthesized in a solid-phase process around immobilized template molecules [38, 39]. The resulting nanoparticles had a uniform diameter of ± 200 nm, depending on the target. The targets were biotin (also known as vitamin H), the antibiotic vancomycin, a peptide derived from the epithelial growth factor EGFR, and the enzyme trypsin, a protein. For all analytes, the limits of detection were around 5 nM with a useful analytical range up to 500 nM in buffer. As a proof of a medical application, EGFR spiked into saliva could be detected within the same analytical range.

4. Detection of cells and bacteria

4.1. Preparation of surface-imprinted polymers

HTM can also be configured to detect cells and bacteria in a selective way by using surface-imprinted polymers (SIPs) as a biomimetic receptor layer on the chip. The chip is at the underside of the measuring compartment to facilitate the cells reaching the receptors by sedimentation, while the temperature T ₁ is usually kept at a constant value (e.g. 37 °C). The SIP concept goes back to the work of F. L. Dickert and co-workers, who succeeded in synthesizing imprints of, among others, red blood cells and viruses in thin polyurethane (PU) layers [40–42]. Using a quartz-crystal microbalance (QCM) for readout, they demonstrated that the SIPs could indeed distinguish between different virus types, and between red blood cells of different blood groups. In the classic imprinting approach, template particles (cells, bacteria, viruses) are sedimented on a PDMS (silicone rubber) stamp, followed by removing the liquid by spinning to obtain a dry layer of templates. Then, the chip is coated with a thin layer of pre-polymerized PU (polyurethane, thickness ≤ 1 μm) and the stamp is gently pressed on the chip; the templates sink into the viscous PU by about 100–200 nm. After curing the PU to a solid with the templates still in place, the stamp is removed and residues are rinsed off with an anionic detergent. Figure 5a shows the imprint of an individual Escherichia coli bacterium; for bacteria in general one can obtain a surface coverage of 10⁶–10⁷ imprints per cm² of chip surface [43, 44]. For the considerably larger cancer cells (typical diameter of 20 μm), the corresponding value is in the order of 10⁴ per cm² [45]. Besides of soft-lithographic stamping with template removal as proposed by Dickert et al., there are several alternative methods described in the literature, an overview covering the developments up to 2016 can be found in a review by Eersels et al. [9].

Figure 5:

Bacterial detection with surface-imprinted polymers, from concept to dose-response calibration curve. (a) AFM image of an individual E. coli imprint in a 1 μm thick polyurethane layer on a glass chip, the areal density is 10⁶ imprints per cm² [43]. Panel (b) illustrates the idea that binding of target cells to the imprints reduces the heat flow from the chip to the liquid, resulting in an increase of R _th and a decrease of T ₂. (c) Scheme of a meander heater that is placed underneath the chip to provide heating power and keeping T ₁ constant at 37 °C. (d) Dose-response curve for E. coli in cloudy apple juice with a detection limit < 100 CFU/mL and an analytically useful range up to 10⁵ CFU/mL. There are two data points for each concentration, being R _th with the sensing compartment filled with apple juice and after flushing with PBS buffer. Reprinted from ref. [43], Copyright 2019, with permission from Elsevier.

When aiming at low detection limits, which are especially important in the context of food safety, QCM is not the most sensitive detection principle. Furthermore, measurements suffer from high noise levels due to viscous damping exerted by the polymer layer. Therefore, HTM was initially employed with a proof-of-concept application on the human cancer-cell lines MCF-7 (breast cancer) and Jurkat (leukemia) that are expected to block heat flow efficiently due to their large size with 20 μm diameter [45]. The idea is sketched in Figure 5b, where the heat flow to the liquid occurs mainly through the imprinted sites. Binding of cells to the imprints causes heat-flow blockage and increases R _th since the phospholipid bilayers of the cells have a low thermal conductivity, which is estimated to be five times less than water [46].

4.2. Detection of cancer cells

Detecting cancer cells with SIPs is an appealing idea and the technology works well to distinguish between cancer cells of different cells lines [45], or have differences in their glycosylation patterns as shown with Chinese hamster ovarian cells (CHO) by Bers and coworkers [47]. Eersels et al. studied the quality of the breast-cancer cell line ZF-75-1 (adherent growth) during a cultivation experiment with 25 cycles [48]: initially, SIPs were made from a sample of pure ZF-75-1 cells to obtain negative replicas of the native cells. The cells were passaged, after which the daughter cell generations were tested for their ability to bind to the imprints made with the parent cell generation. Based on the HTM signal, it became clear that all cells up to the fifteenth daughter generation were purely ZF-75-1 while the binding signal diminished for later generations. This result was confirmed by genomic profiling, showing that a mutated, descended cell line (ZF-75-1s, growing in suspension) became dominant in the colony. A potential application can be seen in fast and efficient testing of the quality of cell cultures that is important for medicine and pharmacy. As a long-term perspective, one may speculate that it is possible to use SIP-type imprints to detect circulating tumor cells (CTCs) in whole blood as a diagnostic approach, or even clear CTCs from the blood stream of cancer patients. For the time being, this is hypothetical, but it is known that metastases often result from CTCs that are released from tumors and invade elsewhere in the human body [49]. Problems to overcome are the sparsity of CTCs with only 1–10 CTCs per milliliter of whole blood that also contains ca. 10⁹ other cells [50]. Furthermore, cell-membrane composition and morphology may differ between different patients, even for the same type of cancer. For first results on CTCs detection with imprinting strategies, the reader may consult refs. [51, 52]; leukemic cells with their higher abundancy than CTCs are addressed in ref. [53]. Given that this is highly challenging, the first applications that can be expected from combining SIPs with HTM readout aim at the detection of bacteria.

4.3. Detection of bacteria

Combining SIPs with standard HTM resulted in an initial detection limit of ca. 10⁴ CFU/mL for E. coli and allowed for distinguishing between E. coli and Staphylococcus aureus [54]. Both bacteria are similar in size, but the membranes are different: E. coli is Gram-negative and S. aureus Gram-positive. Furthermore, it was observed that imprints made with living template E. coli cells had a stronger binding affinity to alive E. coli targets than for inactivated E. coli bacteria. To reduce the detection limit further, a refined version of HTM was designed, known as TWTA or “Thermal Wave Transport Analysis” [55]. TWTA employs a standard HTM configuration, but the heating power P oscillates at a low frequency such as f = 0.03 Hz. This results in thermal waves penetrating the SIP layer and the thermocouple for measuring T ₂ picks up sinusoidal temperature oscillations. Benefits are: The fact that the T ₂ signal can be measured with a lock-in amplifier for higher accuracy. It is possible to adjust the penetration depth of the thermal wave into the sample by selecting the frequency, and the phase angle φ between T ₂ and P can serve as an additional measuring parameter. This way, spiked E. coli concentrations could be detected in urine samples down to 3 × 10⁴ CFU/mL.

Another modification to HTM was designed by Cornelis et al. with the explicit purpose to considerably reduce the detection limit for bacteria in real samples [41]. The method was denoted as M-HTM or modified heat-transfer method: the first modification was to replace the heater and thermometer for T ₁ by an on-chip meander, see Figure 5c, that is placed underneath the chip with the SIP coating. In this configuration, ca. 75 % of the heating power transmits through the chip-liquid interface and the resulting baselines of R _th are higher because only 25 % of P is dissipated. Therefore, one can keep the total input power low. Second, the top lid of the measuring compartment consisted of titanium instead of glass to support heat removal in the vertical direction. Furthermore, a calibrated Pt100 resistor was embedded in the titanium lid for more accurate T ₂ measurements. The M-HTM concept makes detection limits for E. coli below 100 CFU/mL possible with an analytically useful range up to 10⁵ CFU/mL not only in buffer, but also in a complex sample such as non-filtered apple-juice, see Figure 5d. The LoD is lower than the allowed concentration for non-pasteurized fruit juices, meaning that the sensor can distinguish whether specimens comply with the European regulations or not. For comparison, the combination of SIP-type receptors with impedance spectroscopy as transducer principle still reaches lower detection limits of only 30 CFU/mL, also in a complex matrix [44]. In a veterinary application, M-HTM was employed for the quantitative detection of Campylobacter species in chicken droppings that were dissolved in a 1 to 10 ratio in buffer to liquefy the samples [56]. The receptors did not respond to the native gut flora of chicken and the LoD was in the order of 5000 CFU/mL. The high LoD is unproblematic because, in the case of a Campylobacter infection, the bacterial load goes up to 10⁹ cells per gram of dropping. This type of sensor can be used to detect Campylobacter in the food chain at an early stage, e.g. at a poultry farm or in a slaughterhouse. Fast and early detection is desirable since Campylobacteriosis is a widespread food intoxicant (the most frequent one in Europe) and the existing cultivation tests in analytical laboratories are hampered by several days of incubation and the impossibility to perform the analysis on spot [57]. The reader will find additional information on bacterial detection in food-safety context by thermometric principles in refs. [58, 59].

4.4. Selectivity aspects

The selectivity of SIPs for their targets is still a topic of continuous optimization. There is the mechanical complementarity between imprints and targets in the sense of “an egg falls into an eggcup”, but some kind of chemical recognition is necessary as well. In a cross-selectivity study on different equally-sized human cancer cell lines, Eersels et al. found that all cells fitted into the imprints but, after mild washing, only the real targets remained in the imprints and the “lookalikes” were rinsed out of these recognition sites [45]. This observation was utilized in repeated exposure experiments in which the sample passes several times over the chip, leading to a gradual enrichment of the targets on the chip and continuous removal of competitors. Hence, it is possible to extract targets at low concentration from samples containing a multitude of competitors, which was demonstrated for cancer cells as well as for bacteria [47, 54]. To unravel the chemical contribution to recognition, Yongabi et al. analyzed yeast-imprinted chips in a complementary way by scanning-electron microscopy, Fourier-transform infrared spectroscopy, and X-ray photoemission spectroscopy [60]. There was clear evidence that the imprinted cavities featured nano-scale phospholipid fragments that are transferred during the imprinting step. These fragments seem to act as anchoring points for target cells and a potential contribution by membrane proteins was ruled out. At least for microbial detection, the selectivity topic is now well studied and one can conclude that the imprints are selective at species level, i.e. able to distinguish between bacteria belonging to different species, and inclusive at strain level: making imprints with template cells belonging to a certain species (e.g. a given E. coli strain), the imprints will bind all or at least most other strains belonging to the same species [43, 54, 56].

5. Detection of proteins and virus particles

5.1. Protein detection with aptamers and nanoMIPs

Antibodies are the best-known receptor for proteins, but there is only one publication on antibody-based protein detection with heat-transfer redout, see further below. The difficulty lies in the fact that antibodies form a soft layer on the chip that is 5–10 nm thick. Binding proteins would mean that there is a second organic layer on the chip, resembling the thermal transport properties of the receptors, so that there is insufficient “contrast” in thermal material properties. To address this topic, Peeters et al. developed a HTM protein sensor in which the chip (gold on silicon) was functionalized with aptamers, being oligonucleotides with a length of 80 bases, able to bind the protein Ara h 1 [61]. Ara h 1 is present in peanuts and peanut butter, causing allergic reactions in sensitized individuals. The sensor had a detection limit of 5 nM in binding buffer and showed an R _th increase of 10 % in presence of a peanut-butter extract that was spiked intentionally with an additional 100 nM of Ara h 1. It was not possible to determine the absolute Ara h 1 concentration in peanut-butter extract by HTM, but a control measurement with the quartz-crystal microbalance and the same type of aptamers revealed a frequency drop by – 5 Hz upon exposure to non-spiked peanut-butter extract.

Regarding the detection of diagnostically relevant protein markers, Crapnell and coworkers developed a sensor for the cardiac biomarkers H-FABP (heart-fatty acid binding protein) and ST2, which is a subtype of interleukin receptors [62]. Both markers are released from heart tissue to the blood stream at the onset of acute coronary syndrome that can eventually lead to heart failures, such as myocardial infarction. The receptors were again nanoMIPs (see Section 3.2), synthesized by a solid phase approach, with hydrodynamic diameters of 200–300 nm. By using surface-plasmon resonance (SPR), the authors determined dissociation constants K _d of 4 nM for H-FABP and 14 nM in case of ST2. K _d is the inverse of the better known affinity constant K _a ; the given values are on par with commercially available antibodies, meaning that MIPs have now reached the level of highly selective receptors. The sensor setup employed the “inverted HTM concept” (see Figure 4a) with three thermocouples placed together in a 40 μL flow cell: one thermocouple was equipped with H-FABP nanoMIPs, the second one with ST2-MIPs, and the third one was left blank as a reference channel. In spiked buffers, the LoD for both proteins was a few ng/mL (in the order of 100 pM), which is within the physiological range, and cross-selectivity was minimal. First tests with spiked fetal bovine serum (FBS, mimicking human blood serum) demonstrated the feasibility of detecting both markers directly in the relevant body fluid. For the time being, commercially available H-FABP- and ST2 assays still reach lower LoD’s (around or below 1 ng/mL), but the HTM approach is fast and low cost. Furthermore, multiplex detection of both markers has not been demonstrated before to best of our knowledge and the required sample volume is extremely small.

5.2. Virus detection of SARS-Cov-2

Fast antigen tests for Covid-19 utilize antibodies that bind in a selective way to the protein-based receptor binding domain (RBD) located at the spikes of Corona viruses, see Figure 6a and refs. [63, 64]. While these tests are reasonably cheap and deliver the result within ca. 15 min, they entail a risk of false-negative outcomes for asymptomatic patients with low viral load. In addition, they have a limited shelf life, need storing at room temperature, and cannot be sterilized by autoclavation, since this would denature the antibodies. In recent work, McClements et al. introduced a test for SARS-Cov-2 that combines nanoMIPs as receptors with a HTM platform in standard geometry (see Figure 1), now adapted to sample volumes of just 100 μL [65]. The nanoMIPs were synthesized around short fragments of the RBD region, consisting of only 10 amino acids; an approach known in literature as epitope imprinting [66]. The idea is that the MIP binds its complementary part within the spike protein, and therefore, also the spike protein itself. Hence, such layers should also bind entire virus particles carrying these spike proteins. The advantage of epitope imprinting is that the epitopes are both cheap and do not present any health risk, which would be the case for imprinting complete, contagious viruses. The chip was a screen-printed carbon electrode to which the nanoMIPs were grafted electrochemically, see Figure 6b. The sensor had a turn-around time of only 15 min per analysis (similar to commercial fast tests) and it could detect the spike proteins (without virus particles) at concentrations below 10 fg/mL for the alpha- and delta variants of this protein. This LoD, at the sub-femtomolar level, is several orders of magnitude lower than the corresponding LoD for spike proteins found with commercially available antigen tests. For comparison, also anti-RBD antibodies were employed in the HTM setup, resulting in much high detection limits than in case of the nanoMIPs and low changes of the R _th signal.

Figure 6:

Detection of Covid virus particles in patients' samples using nanoMIPs on screen-printed carbon layers. (a) Illustration of the SARS-Cov-2 virus with its spike proteins and the receptor-binding domain RBD at the terminus of each spike [63, 64]. (b) The nanoMIP particles with a diameter of 70 nm are bound covalently to screen-printed carbon electrodes using 4-aminobenzoic acid as a linker molecule [65]. (c) Analysis of clinical samples obtained from Covid-positive and negative individuals. Each dot represents one sample and there is hardly any change of the heat-transfer resistance for the negative samples. The positive samples cause an R _th increase of 0.3 °C/W on average and the spreading of the data may correlate with the actual virus load. Panel (a) is adapted from ref. [61] with permission, Copyright 2021, Springer Nature. The panels (b) and (c) are reprinted from ref. [65] under Creative Commons CC BY 4.0 license.

Most importantly, the sensor based on nanoMIPs has undergone its first clinical evaluation with samples consisting of nose- and throat swaps of seven Covid-negative and seven positive patients, the status of the patients (negative or positive) had been tested before by gold-standard clinical tests. As shown in Figure 6c, the signal change ΔR _th has a median value of 0 °C/W for the negative samples while it was ≈ + 0.3 °C/W in case of the patients who had tested positive. The results may even allow to estimate the virus load of an individual. In case that new strains of SARS-Cov-2 emerge, the nanoMIP technology allows to adapt the receptors quickly by changing the peptide sequence of the epitope accordingly. For further information, we refer to ref. [65].

6. HTM study on self-assembling monolayers and lipid vesicles

6.1. Formation monitoring on thiol SAMs

It also has been demonstrated by HTM that the R _th parameter responds sensitively to the formation of self-assembling monolayers (SAMs) by the example of thiols on gold-coated chips, see Figure 7a [67]. The monolayers were formed from solutions of different thiol concentrations in ethanol (EtOH) as presented in Figure 7b. The R _th signatures obtained for the concentration series clearly correlated with the mass-loading signal Δm/A and the dissipation signal ΔD that were determined by reference measurements with a dissipation-sensitive quartz-crystal microbalance as shown in Figure 7c. The design of this experiment excluded formation of thiol multilayers so that the R _th increase could be solely attributed to the presence of a mono-molecular layer at the solid-to-liquid interface. In the cited study, two different thiols were assessed, being 1-dodecane thiol with a terminal –CH₃ group and 11-mercaptoundecanoic acid (11-MUA thiol), terminated with a –COOH group. In the case of a 11-MUA thiol monolayer, the R _th increase was stronger than with 1-dodecane thiol. This was interpreted in the sense that the methyl group of 1-dodecane thiol has strong vibrational overlap with the supernatant EtOH (as seen in FTIR spectra) while there are hardly matching vibration frequencies for 11-MUA with its carboxyl group. This indicates that heat transfer from the solid, where energy is present in the form of phonon modes, to the liquid (molecular vibrations) through a molecular thiol brush depends on matching vibration frequencies in the monolayer and in the liquid. Putting these observations together, one may conclude that any ligand (or larger particle) at an interface, whose vibrational modes and frequencies neither match the frequencies in the solid nor those in the liquid, will increase thermal interface resistance. As shown above in Figure 2b, silane SAMs also induce a small increase of R _th (red curve in Figure 2b as compared to the black curve), in this case for the interface between synthetic sapphire Al₂O₃ and the supernatant PBS buffer.

Figure 7:

Heat-transfer signature upon self-assembly of thiol monolayers on gold surfaces. (a) Increase of the thermal interface resistance upon formation of different self-assembling thiol layers on gold-coated silicon chips in contact with ethanol as supernatant medium [67]. The black line is a reference measurement with thiol-free EtOH. The heat-flow attenuation is more pronounced for 11-MUA (blue) than for 1-dodecane thiol (red), which has a terminal methyl group in common with EtOH. (b) Increase in R _th for different 11-MUA concentrations after 3 h of incubation. The fit curve shown as a green line was calculated with a dose-response model. The R _th data correlate with the mass-loading signal Δm/A and the dissipation signal ΔD obtained with a dissipation-sensitive quartz-crystal microbalance under the same temperature and concentration conditions, see panel (c). Reprinted from ref. [67], Copyright 2020, with permission from Elsevier.

Reference [67] also provides an overview on typical ΔR _th amplitudes found in a broad variety of bioanalytical applications. We mention here that R _th baselines often depend on the respective device due to parasitic heat loss that can be more or less pronounced, depending on design details. The changes however are typically in the order of ΔR _th ≈ 0.2–2 °C/W and independent of the size of the targets such as small molecules, proteins or entire cells, pointing together to the fact that the R _th increase is an interface effect. In addition to the data included in ref. [67], all examples discussed within the present review article also go along with an increase of R _th, supporting the assumption that the effect has the mismatch of vibrational frequencies at interfaces as a common denominator.

6.2. Phase transitions in lipid vesicles

Lipid vesicles can be considered organelle-free model systems for cells and they are formed by self-assembly of lipids in aqueous buffers, allowing to control their composition and selecting them with specific diameters [68]. While these tailored vesicles are used in basic research, it is worth mentioning that extracellular vesicles (EVs), excreted e.g. by cancer cells, are a promising route for mutation analysis towards personalized therapies [69, 70]. In our case, the vesicles with a diameter of 82 ± 30 nm, comprised of dipalmitoylphosphatidycholine (DPPC). According to literature, these vesicles display two structural phase transitions from the gel phase (L _β ) to the ripple phase (P _β ) at 34 °C and the so-called main phase transition from the ripple phase to the liquid-disordered phase L _α at 41.5 °C. To trace the main phase transition with HTM, the vesicles were adsorbed on hydrogenated nanocrystalline diamond chips using HEPES buffer as supernatant, see Figure 8a. Ramping the temperature T ₁ from 36 to 46 °C and back at a slow rate of 0.2 °C/min reveals step-like changes of the R _th signal at T ₁ ≈ 41 °C while the hysteretic behavior indicates a first-order phase transition as shown in Figure 8b. For comparison with a reference technique, the authors of ref. [66] performed also heat-capacity measurements using pASC (Peltier-element-based adiabatic scanning calorimetry) with the vesicles dissolved in HEPES buffer. After subtracting the heat capacity of the fluid, Figure 8c shows the isobaric heat capacity c _p (T) of the vesicles with phase transitions from the gel phase (L _β ) to the ripple phase (P _β ) at 34 °C and the main phase transition from the ripple phase to the liquid-disordered phase L _α at 41.5 °C. Especially the latter transition causes a sharp local maximum in the c _p (T) signal. The HTM data are considerably more noisy than the pASC results, but we wish to point out that the R _th signal detects the phase transition as such and, in addition, indicates that the thermal resistance of the vesicles in their liquid-disordered phase is higher than in the partially ordered ripple phase. The R _th change is persistent and reversible while pASC detects only the phase transition itself, albeit with high accuracy. Furthermore, HTM can be performed faster than calorimetry while the instrumental complexity of HTM is much lower than that of calorimeter devices.

Figure 8:

Monitoring the main phase transition of lipid vesicles by HTM and reference data obtained by calorimetry. (a) For studying the main phase transition of DPPC lipid vesicles from the gel phase to the ripple phase (P _β → L _α ) with the heat-transfer method, the vesicles were adsorbed on a synthetic, hydrogenated diamond chip (H-NCD). Panel (b) shows an hysteric jump in the R _th signal at 41 °C during heating (red) and cooling back at the same rate (blue). The liquid disordered phase has a higher R _th than the partially ordered ripple phase, the difference in molecular organization is symbolized in the inserts. For comparison, panel (c) displays the isobaric heat capacity c _p (T) of these vesicles, as measured with an adiabatic scanning calorimeter. There are two maxima at the transition from the gel phase to the ripple phase (L _β → P _β ) at 34 °C and at the main phase transition to the liquid disordered phase at 41.5 °C, in agreement with the HTM data [68]. Reprinted with permission from ref. [68]. Copyright 2014 John Wiley & Sons.

7. The hot-wire concept

Given the low noise level of M-HTM, in which an on-chip meander served as heater and temperature sensor for T ₁ together, Khorshid and coworkers designed a hot-wire sensor based on the 3ω principle [71]. The general idea is that reducing the number of physical sensor components will reduce the noise level by avoiding that uncertainties in e.g. T ₁ and T ₂ sum up cumulatively. The wire, more specifically a microwire, is the only sensing probe that emits thermal waves upon triggering with an AC current. The 3ω technique was originally proposed by Cahill and Pohl for highly accurate thermal conductivity measurements on solids [72, 73]. It has widespread use in solid-state physics and in measuring the concentration and identity of gases, see e.g. refs. [74, 75]. For gases, freestanding metal wires are the optimal configuration while, in case of solids, a thin line heater is deposited on the sample under study. In a biological application, Clausen and coworkers used a platinum meander on a ceramic chip to monitor the formation of bacterial biofilms in growth medium [76]. The method borrows its name from the fact that an AC current with frequency ω, imposed on the wire, causes Joule heating with the frequency 2ω and, as a result, a voltage U _3ω along the wire at the third harmonic of the triggering frequency. U _3ω is related to periodic Joule heating that induces also periodic oscillations of the wire resistance due to its temperature coefficient of resistance, denoted as TCR or β. The voltage U _3ω can be calculated through the following equation, its derivation is provided in ref. [71]:

$$\kappa =\frac{\beta \cdot R^2\cdot I_0^3}{8\pi l\cdot U_{3\omega }}\cdot \frac{K_0\left(q\cdot r\right)}{q\cdot r\cdot K_1\left(q\cdot r\right)}\Rightarrow U_{3\omega }\propto \beta \cdot \frac{I_0^3}{\kappa }$$ (3)

where I ₀ is the amplitude of the triggering current and the amplitude of U _3ω is inversely proportional to the thermal conductivity, κ at the wire-to-liquid interface. K ₀ and K ₁ are modified Bessel functions of the second kind, q is the thermal wave number, R is the resistance of the wire at ambient temperature, and l and r are the length and radius of the cylindrical wire. In practice, U _3ω is small at the mV level; hence, accurate measurements require a four-point configuration and a lock-in amplifier. Furthermore, metals with a high TCR are preferable such as gold (β = 3.4 × 10⁻³ °C⁻¹), aluminum (3.8 × 10⁻³ °C⁻¹), or tungsten (4.5 × 10⁻³ °C⁻¹) [77, 78]. Since we opted for the geometry with freestanding wires, we chose aluminum microwires that are protected by a native oxide layer of ca. 5 nm thickness, at least in solutions with neutral pH. Gold and tungsten are stable in a wider pH range; Au due to its chemical inertness and high charge-transfer resistance in contact with H₂O and W for its strong, biocompatible oxide layer. Figure 9 shows the basic geometry with two Al wires and an active, immersed length of 10 mm. All other parts are low-Ohmic to focus the Joule heat on the wire itself. The wires were used in blank version (only with the native oxide layer), with a linker layer of silanes (N-[3-trimethoxysilyl)propyl]-ethylenediamine triacetic acid), with single-stranded probe DNA attached to the silanes, and with double stranded DNA. Calculating κ for the blank wire by using Eq. (3) rendered correctly the literature value of water (κ = 0.61 Wm⁻¹ K⁻¹) while we found a lower κ value in case of the silane layer. This suggests that the terminal carboxyl (–COOH) group of the silanes promotes the transfer of thermal energy because the O–H vibration of this group has the same frequency as the corresponding vibration in H₂O molecules, resulting in a resonance effect. Furthermore, ss-DNA reduces the efficiency of heat transfer, see Section 2 above, while the heat-flow blocking of ds-DNA is again less pronounced as compared to the single-stranded DNA.

Figure 9:

The 3ω thermal-wave principle probes the thermal conductivity at the interface between an aqueous solution and microwires with different surface layers. (a) Schematic drawing of the sensing compartment with two freestanding aluminum microwires contacted in 4-point geometry [71]. (b) Schematic cross section of the micro-wires featuring i) a native oxide layer, ii) a silane linker layer, iii) single-stranded probe DNA (36 nucleotides), and v) double-stranded DNA. (c) The 3ω voltages render the efficiency of heat-transfer from the wire to the liquid in a way that is consistent with Figure 2. The error bars are standard deviations from three independent measurement, showing the high reproducibility of the method. Reprinted from ref. [71], Copyright 2021, with permission from Elsevier.

While the hot-wire technique has not yet been demonstrated for other bioanalytical applications beyond the proof-of-concept described, it offers several intrinsic advantages: Microwires for wire bonding are inexpensive mass products in the electronic industry. In addition, a few centimeters are sufficient, and the wire has triple functionality as receptor-immobilization platform, heat source, and temperature sensor. By design, there is no parasitic heat loss to the environment and the entire heating power transmits the interface of interest. The wire can also be wound to a tiny coil, enabling measurements in small sample volumes. The output signal U _3ω is typically stable after less than 60 s, making the response time extremely short and, without triggering the wire with an AC current, it will automatically attain the temperature of the surrounding sample. Furthermore, the principle is only sensitive to the temperature difference between the wire and the environment, but not to absolute temperatures that require stringent control in case of the original HTM technique. Furthermore, there are no limits to hot-wire measurements in streaming liquids, which cannot be studied in HTM due to its (quasi-) static temperature gradient. In summary, the hot-wire concept deserves further attention. Current drawbacks, such as the lack of multiplexing, appear to be technically solvable.

8. Monitoring of cell proliferation and metabolic activity

8.1. Standard HTM assays

Studying the proliferation of cell cultures, often in presence of prospective medical drugs or toxic substances, is an important field in cell biology and pharmacology. Assessing the response to such substances at cell-culture level provides a first indication of the efficacy and tolerability of the toxin or drug when it is applied to humans at a later stage. This is the idea behind cell-based biosensors that often probe the growth of cell cultures by impedance spectroscopy and their metabolic activity by measuring the oxygen uptake and acidification rate [79–81]. A comparison between several commercial cell-based biosensor systems is provided in ref. [82]. Also HTM should allow to follow proliferation over time, with the potential advantage that the cells will not be exposed to electrical voltages.

The first article on HTM for monitoring the growth of yeast cell cultures (Saccharomyces cerevisiae, wild type) was published in 2018 by Betlem et al. [83]. They utilized a 3D-printed HTM device in the conventional configuration of Figure 1a, although with an opening in the top lid to let metabolically formed CO₂ gas escape. The medium, containing nutrients, was continuously administered at a low flow rate. This is worth mentioning because the vast majority of HTM experiments takes place at no-flow conditions with stagnant liquid to facilitate building up a natural temperature gradient between the chip and the supernatant liquid. The chip material was gold-coated, doped silicon using a thin chromium layer as adhesive. The measurement revealed an R _th increase by + 0.7 °C/W over an observation period of 42 h. Furthermore, the growth rate was determined for chip temperatures T ₁ ranging from 30 to 51 °C. The strongest R _th change (over a 1 h period) was found at T ₁ = 38.0 °C, meaning that this temperature provides optimal growth conditions. The study by Betlem et al. also gave other interesting results, such as proving the elimination of yeast by heat treatment and inactivation by a low dose of cytotoxic CuSO₄. The concept was furthermore applied to study the growth of S. aureus bacteria in buffer solutions and digestate samples obtained from bioreactors [84]. S. aureus can cause infections of the skin and the respiratory tract that are usually treated by antibiotics; however, the MRSA strain (methicillin-resistant S. aureus) plays a notorious role in hospital-acquired infectious diseases [85]. The results of ref. [83] suggest that HTM is also useful in highly complex samples (digestate) to quantify their total microbial load.

8.2. Thermal waves and heat pulses

Already in 2014, Reyes-Romero et al. developed a thermal-wave sensing principle to study the formation of Enterococcus faecalis Symbioflor 1 biofilms directly on a passivated chip in the presence of cell-culture medium [86]. The chip is shown in Figure 10a and consists of a meander-shape chromium heater that surrounds a thermistor made of germanium, which can determine temperature changes with an accuracy of better than 1 mK. The setup was placed in an incubator to ensure a constant background temperature of 37 °C. For monitoring the growth of the biofilm, the heater was triggered with an AC current at a frequency of 20 Hz, resulting in a thermal wave of 40 Hz so that we can regard this a 2ω technique. The thermal wave transmits energy in all directions; the fraction arriving at the thermistor causes local temperature oscillations with the same 2ω frequency, an amplitude ΔT, and a phase shift with respect to the phase of the triggering current. The bacterial layer growing over the heater and thermistor reduced the transmission of heat to the supernatant medium, resulting in an increase in both ΔT and the phase shift as shown in Figure 10b and c. Interestingly, the technique was also able to demonstrate that the antibiotic ampicillin stops cell proliferation as visible in the phase angle and the temperature amplitude. While the changes of T ₂ in HTM are typically in the order of 1 °C, the changes of ΔT were in the order of only a few mK, but the thermistor was sensitive enough to record this shift with high resolution. Furthermore, the heat flow was not “through the chip to the liquid” as in HTM, essentially a thermal wave was monitored as it propagates along a direction parallel to the chip surface.

Figure 10:

Probing the proliferation of bacteria and the impact of antibiotics with an on-chip heater and thermistor. (a) Design of a silicon-nitride chip used for monitoring the formation of biofilms [86]. The chip features a planar heating meander made of chromium and a germanium-based thermistor as temperature sensor. Triggering the heater with AC current causes thermal waves and the thermistor records temperature oscillations ΔT and a phase shift. The panels (b) and (c) show the amplitude and phase change during the biofilm growth of E. faecalis bacteria with a starting number of 6 × 10⁴ cells (indicated by red lines). The blue lines demonstrate that addition of ampicillin (Amp.) at 2.8 h inhibits the biofilm formation while the black lines render a cell-free control measurement. Reprinted from ref. [86], Copyright 2014, with permission from Elsevier.

In a recent work, Oudebrouckx et al. presented a thermal sensor for cell proliferation and metabolic-heat monitoring that combines miniaturized transient plane sources (TPS) with a 96-well plate, that enables parallelized measurements [87]. Such well plates are industry standard in the pharmaceutical sector for high-throughput drug screening. The active element of each TPS was a double-spiral copper meander at the underside of a thin polyimide foil while the sample, consisting of yeast cells in cell-culture medium, was placed in microwells with the foils at their bottom. The TPS spirals were contacted in 4-point geometry allowing for accurate temperature measurements when taking the TCR of copper into account. A low probe current is sufficient to measure heat released by digestion of nutrients. By sending current as block pulses through the spirals, their temperature increases over time according to Eq. (4), for details see ref. [87]:

$$\mathrm{\Delta }T\left(t\right)=s\cdot \sqrt{t}\text{with}s\propto e^{-1}\text{and}e=\sqrt{\kappa \rho c_p}$$ (4)

where s is the slope of the temperature increase when plotting ΔT as a function of the square root of the time t. The slope s is inversely proportional to the thermal effusivity e of the material(s) on top of the TPS with κ being again the thermal conductivity, ρ the mass density, and c _p the isobaric heat capacity. In brief, the effusivity describes the ability of a material to absorb or release heat and the material in the present case was a triple layer consisting of the polyimide foil, a yeast layer, and the cell-culture medium on top. While it is possible to determine absolute e values by calibrating with reference fluids), the authors were able to determine the changes of e upon the proliferation of a yeast-cell culture due to the increase of cell mass and layer thickness. Furthermore, also the base temperature at the start of each heat pulse was found to increase upon proliferation and digestion of nutrients, which can be interpreted as direct proof of metabolic activity. The increase in s due to colony growth corresponds to a decrease in e, which can easily be understood by the low thermal conductivity κ of cell material in agreement with the findings reported in Section 4.

9. Spontaneous detachment of eukaryotic cells

9.1. Phenomenological observations

The original HTM technique without any further modifications can be employed to characterize eukaryotic cells and their response to nutrients and (bio-) chemical agents [25]. Hereby, it is not necessary to use any receptors: after injecting cells such as yeast strains and cancer cells in 1 × PBS buffer at high concentrations (e.g. 10⁶ cells/mL) into the HTM compartment, they sediment on the chip, which causes an increase of the R _th value. Surprisingly, the cells detach collectively and without an external trigger at a detachment time t _d50 and the R _th signal drops back to its baseline. The effect does not depend on the chip material and, in the examples shown here, the authors used polished aluminum platelets with a 1 μm thick coating of polyurethane. Figure 11a illustrates this behavior for baker’s yeast (S. cerevisiae) and a chip temperature of T ₁ = 33 °C with t _d50 = 41.8 min, the effect is absent for the cell-free reference liquid. The initial sharp rise of the R _th signal is due to injecting liquid at room temperature into the setup, followed by a plateau that correspondsto thermal insulation by the sedimented cells, and eventually the sudden decrease of R _th to its starting value.

Figure 11:

Spontaneous collective detachment effect of yeast- and cancer cells in presence of a temperature gradient. (a) Time dependence of R _th during injection, sedimentation, and spontaneous detachment of yeast cells at a moment t _d50 after injection into the measuring compartment [25]. The red line shows a control experiment with cell-free buffer. The t _d50 parameter is highly reproducible and detachment can either be directly or mediated by bleb-type protrusions. (b) Dependence of the detachment time t _d50 on the chip temperature T ₁ for S. cerevisiae. The red line illustrates the exponential scaling of Eq. (5) in which the two parameters θ and t ₀ are cell-type specific; all data points were measured at least in triplicate. (c) Spontaneous detachment of the human cancer-cell line HeLa for different chip temperatures with characteristic oscillations of R _th at T ₁ = 27 °C, possibly related to glycolytic oscillations. (d) Acceleration of the detachment of S. cerevisiae due to presence of sucrose for different chip temperatures (black curves: no sucrose, red curves: 1 mg/mL sucrose). Reprinted with permission from ref. [25] under Creative Commons CC BY license.

When plotting the detachment time as a function of T ₁, we found that there is an exponential scaling behavior between t _d50 and T ₁, according to Eq. (5), see Figure 11b. This equation is empirical and it holds at least within the experimentally accessible temperature range. For temperatures T ₁ lower than 25 °C, the heating power P is too low to determine it accurately. Furthermore, we avoided temperatures above 37 °C in view of the more sensitive cancer cells.

$$t_{d50}=t_0+A\cdot \text{exp}\left(-\frac{T_1}{\theta }\right)$$ (5)

In this equation, t _d50 is the time between injection and the moment where R _th is halfway between the plateau- and the baseline value. The parameter t ₀ is the horizontal asymptote in the limit of high chip temperatures, T ₁ is the chip temperature in degree centigrade, and θ (also in °C) is a scaling parameters. Measurements for different cell concentrations revealed that t ₀ is a constant while θ very weakly depends on the cell concentration. While the physical meaning of the amplitude A has not yet been explored in detail, we can state that the two parameters t ₀ and θ together uniquely identify of the cell type under study. We point out that for a given cell type and chip temperature, t _d50 is sharply defined and highly reproducible: repeating the experiment with S. cerevisiae and T ₁ = 33 °C for 10 times gave an average value of t _d = 42.0 min with a standard deviation as low as ± 0.5 min. Furthermore, the effect was probed with seven different chip materials and coatings to induce a broad range of hydrophilic properties, ranging from UV-treated glass with a water contact angle (CA) of 13° up to silanized glass with CA = 124°. In all cases, t _d50 was within 42 ± 1 min. The mechanism behind spontaneous detachment is not yet clarified with certainty, but two factors may play a role.

First, in order to detach, cells must loosen their contact with the chip. Oyama et al. heated individual cancer cells (HeLa) in solution locally with a focused infrared laser and observed reversible protrusions (so called “blebs”) of their membrane at the heated site [88]. The temperature gradient was reported as ≥ 0.065 °C/μm, meaning that the temperature difference was ≈ 1.3 °C between the center of the cell and its membrane. Translated to our situation, we suppose that the cells have a higher temperature at the contact side with the chip, and a lower temperature at the top side, which faces the colder supernatant liquid. If this temperature gradient induces directional bleb formation, it would be a likely mechanisms for weakening the cell-chip contact. Performing the detachment experiment on fluorescently-labeled HeLa cells with simultaneous live-cell imaging with a confocal microscope, we indeed saw that the cells start moving laterally on the chip at t _d50 while some cells rotate around their vertical axes [25]. We did not observe rolling, which indicates that cells indeed lost contact to the chip or, in some cases, might rotate with a bleb serving as a pivot. Second, regarding the collective nature of detachment, it is known that cells have channels of intercellular communication: under specific circumstances, yeasts, E. coli bacteria, myocytes, and cancer cells display glycolytic oscillations in which their metabolism alternates between an aerobic and an anaerobic pathway [17]. The oscillations occur at cell-colony level and it is believed that exchange of the co-enzyme NADH (nicotinamide-adenine-dinucleotide) molecules mediates the “entrainment” of individual cells to collective behavior [89].

9.2. Towards potential applications

While spontaneous detachment is a recent observation, we anticipate applications along the lines of discriminating between related cell types without genomic profiling or other high-end methods, and studying the effect of nutrients and (novel) cytotoxic agents at cell-culture level. Similarly to Figure 11a and b, the detachment was studied on the laboratory yeast strain S. cerevisiae S288C and S. pastorianus W34/70, which is also known as “Weihenstephan yeast”. For all chip temperatures, W34/80 detached clearly faster than W34/70 [25]. The fit parameters θ and t ₀ also differed from those of bakers’ yeast shown in Figure 11b above. Studying the detachment behavior on the two human cancer-cell lines MCF-7 and HeLa, see Figure 11c for HeLa, displayed also clear differences in t _d50, while the concentration of both cell types was kept strictly identical. At the relatively low chip temperature T ₁ = 27 °C, we observed for HeLa reproducible oscillations of R _th after initial detachment (see Figure 11c), which might either point to glycolytic oscillations, that go along with oscillations of metabolic heat released from the cells [17], or to micromechanical “hopping” of the cells on the chip. The R _th oscillations during detachment were also observed in case of S. cerevisiae for chip temperature T ₁ < 30 °C and in presence of sucrose, see the original work in ref. [25] for more details.

There is another interesting point: nutrients and other compounds modulate the time t _d50 depending on their concentration, which was studied so far only for yeast cells; for the original data we refer the reader to ref. [25]. Adding sucrose as an energy source reduces t _d50 considerably, pointing to a relationship between detachment and metabolic activity; this can be seen for S. cerevisiae in Figure 11d. Vice versa, HeLa cells that were starved on purpose for 10 h did not detach anymore, while the effect is visible with freshly cultured HeLa. Furthermore, we treated yeast cells with DMSO (dimethyl sulfoxide), which is a mildly cytotoxic organic solvent that increases porosity of cell membranes and causes oxidative stress [90]. Due to the effect on porosity, DMSO in low doses serves as a preserving additive when storing cells by freezing. Cells incubated with DMSO showed systematically longer t _d50 from less than 40 min (no DMSO) up to more than 80 min (50 % v/v DMSO). We found a similarly retarding effect by administering different concentrations of the drug Blebbistatin, which is a therapeutic agent for certain diseases by selectively inhibiting the activity of the motor protein non-muscle myosin II in the cytoskeleton [91]. As per current knowledge, we can summarize that cells require energy to detach (e.g. heat or nutrients) and that agents acting on the membrane integrity or on motor proteins inside the cell can delay the moment of detachment. Future research will show which types of agents affect the detachment and in which way towards the goal of a novel, fast, and cost-efficient platform for studying the impact of prospective drugs and potentially harmful substances in a controlled cell-culture setting.

10. Summary and outlook

In this overview article, we have introduced the heat-transfer method HTM with a variety of bioanalytical applications ranging from small targets such as neurotransmitters to larger ones in the sense of bacteria and cancer cells. Whether or not a specific target can be detected by HTM depends on the availability of suitable receptors in first place. This has been demonstrated with single-stranded probe DNA, nucleic-acids aptamers, surface-imprinted polymers and molecularly-imprinted polymers, the latter prepared via various synthesis routes. Especially the nanoMIPs have led to strong progress and their affinity now reaches values equivalent to those of antibodies. Antibody-type receptors as such do not bring about substantial changes of the heat-transfer resistance while HTM combined with enzymes has not yet been studied. Since HTM is sensitive to temperature changes, it is likely that an HTM setup can also capture the reaction enthalpies going along with enzymatic decomposition reactions. A basic HTM setup, such as introduced in Section 2, requires not more than two thermometers for T ₁ and T ₂, conveniently thermocouples or thin-film resistors, and a controllable power source to keep T ₁ constant or alter it at a predefined rate. This concept in its standard configuration allows for detecting the concentration of small molecules and proteins in the order below 100 nM in all cases and, when nanoMIPs are used as receptors, even down to the (sub-)picomolar regime, see the Sections 3 and 5. As a telling example in Section 5, we mention here the detection of SARS-Cov-2 virus particles in nasopharyngeal swabs, which takes only 15 min by combining standard HTM with nanoMIPs sensitive for the spike protein of the virus. Such low detection limits are typically only possible with ELISA tests (enzyme-linked immunosorbent assays) and by surface plasmon resonance. Inverted HTM, with MIP-functionalized thermocouples, has currently still higher LoD values, ranging from a few nanomolar to slightly below 10 μM depending on the application. Regarding the detection of bacteria in complex media, standard HTM in combination with surface-imprinted polymers reaches detection limits of 10⁴ CFU/mL, but there is rapid progress and the modified heat-transfer method already reaches detection limits below 100 cells per mL as shown in Section 4. This detection limit is only slightly higher than what can be achieved with impedance spectroscopy in its non-Faradaic variant, and lower than the detection limits that quartz-crystal microbalances can reach.

Besides of a high target affinity of the receptors, the key to low detection limits is a high signal-to-noise ratio, meaning that T ₁ and T ₂ should be measured with high accuracy and without drift effects. To this end, HTM devices can be encapsulated with thermal insulation, air flow around the device should be avoided, and it is recommended to perform HTM measurements in a temperature-stabilized environment: changes of the ambient temperature affect the magnitude of the heat flux through the interface between the chip and the liquid. However, this can be compensated for by increasing T ₁ in a warmer environment to keep the temperature gradient at a given level. Furthermore, increasing T ₁ is beneficial to lower the detection limit as the heat flux will increase and potential noise on the T ₁ and T ₂ data will decrease compared to their absolute values. Focusing the thermal flux through the interface of interest is a point of ongoing improvements; still, accurate measurements are possible even if a large heat fraction dissipates to the environment without passing the interface. Besides reaching low detection limits, biosensors should exhibit short response times: For bacterial detection, measuring a sample still takes 30–60 min, which is mainly due to sedimentation that takes longer than temperature stabilization. Still, this is considerably faster than classical cell-cultivation tests, taking several days. For molecular detection, assay times as low as 5 min are sufficient when looking at MIP-functionalized thermocouples and the hot-wire technique in Sections 3.2 and 7. In summary, a low thermal mass of the sensitive element is highly beneficial to speed up the response. Another category of applications is not about detection, but about monitoring the proliferation of cell cultures over extended times, often focusing on studying the impact of e.g. antimicrobial drugs: This was not only tested by HTM, but also several thermal-wave methods fit this purpose as explained in Section 8.

An advantage of HTM lies in the fact that heat flux can permeate all kinds of solids and liquids, irrespective of their electrical conductivity. Electrochemical methods, including impedance spectroscopy, rely on conducting electrode materials; low-conductivity samples need to be supplemented with electrolytes. This is not the case with HTM for which we have reviewed applications with e.g. synthetic-sapphire chips (Section 2), or receptor layers made of insulating polyurethane (Section 4), and measurements in pure ethanol as described in Section 6.1. Hence, HTM can be applied in situations where electrochemical method reach their limits. Still, it should be mentioned that thermal currents are more difficult to isolate and confine than their electrical counterparts, which makes multiplexing more tedious. HTM has so far been multiplexed to measure four miniaturized sensing compartments in parallel using four individual thermocouples, but this is not yet comparable to multi-electrode arrays. Replacing the thermocouples by transient-plane sources, functioning simultaneously as heater and thermometer, multiplexing was achieved to eight separate compartments, see Section 8.2. It seems only a question of time until thermal methods will be upgraded towards the industry standard of 96-well plates. For fundamental research, one may also turn to scanning thermal microscopy to probe the local temperature above differently functionalized sensing spots on a single chip. We would also like to mention that HTM lends itself for multiparametric analyses: The combination with impedance spectroscopy and fluorescence microscopy has already been demonstrated while integration with surface-plasmon resonance and a microbalance functionality should be feasible as well. Since high-end biosensors usually come with active temperature control, adding on a thermal-resistance measurement would only need adding a thermocouple for T ₂ in the sample compartment. Such a combination of methods would provide multi-faceted information without crosstalk, since the transducer principles are distinctly different.

Last but not least, we anticipate that the spontaneous cell-detachment effect, published in 2022 and discussed in Section 9, opens up a completely new category of applications for HTM. This is not about the detection of molecules or bacteria, but a kind of a cell-based biosensor that measures the effect of drugs and other chemicals at cell-culture level. High-throughput drug testing is an important issue for the pharmaceutical sector and a stringent requirement that precedes tests on laboratory animals and human volunteers. The effect as such is a fundamentally novel. For the time being, we can state that the effect allows to measure the impact of drugs that affect the metabolism of eukaryotic cells. Further research will show whether the method can also chart the impact of drugs on other damage pathways than cell metabolism and whether the effect also occurs with prokaryotic cells such as bacteria. In conclusion, the heat-transfer concept opens up a plethora of research possibilities and development aspects that are worthwhile exploring.

Biographies

Soroush Bakhshi Sichani obtained his MSc in micro- and nano-electromechanical systems engineering (MEMS & NEMS) from the University of Tehran, Iran. Currently, he is a PhD student at the Laboratory of Soft Matter and Biophysics, KU Leuven, Belgium. His research focuses on developing a multi-parametric label free sensing platform based on electrochemical impedance spectroscopy (EIS), quartz crystal microbalance (QCM), and heat transfer method (HTM).

Mehran Khorshid obtained his Doctorate of Veterinary Medicine (DVM) from the Azad University, Karaj branch, Iran in 2009. Afterwards, he obtained a Master of Biomedical Sciences at Hasselt University in 2014. Since completing his PhD in Science (2018) from KU Leuven, Belgium, he has been working as a post-doctoral researcher at the Laboratory for Soft Matter and Biophysics, KU Leuven. His research interest covers engineering of biosensors for medical and biomedical applications, as well as study of complex biophysical systems.

Peter A. Lieberzeit obtained his PhD in Chemistry in 1999 at the University of Vienna and his postdoctoral lecture qualification (habilitation) in 2007. Since 2011 he is a Full Professor at the Faculty of Chemistry of the University of Vienna. His research focuses on generating biomimetic recognition systems for sensing chemical and biological analytes, mainly based on molecularly imprinted polymers and mass-sensitive sensors. He is a member of the editorial board of Sensors and Actuators B: Chemical, of the editorial advisory board of Analytical and Bioanalytical chemistry and chairman of the International Steering Committee of IMCS conferences.

Patricia Losada-Pérez obtained her PhD in physics in 2009 at the University of Vigo. She has been a postdoctoral researcher at the Laboratory of Soft Matter and Biophysics at the KU Leuven and a research associate at the Institute of Materials Research of Hasselt University. She is currently an associate professor in the Experimental Soft Matter and Thermal Physics group of the Université libre de Bruxelles, Belgium. Her research interests include soft-condensed matter physics, thermodynamics, lipid biophysics, and biosensors.

Derick Yongabi obtained his PhD in Physics in 2021 from KU Leuven, Belgium. Before then, he received a Master of Science in Research Methods in 2011 from the University of Leeds, UK, and a Master of Biomedical Science – Bioelectronics and Nanotechnology from the University of Hasselt, Belgium (2015). He also holds a Postgraduate Certificate in Advanced Medical Imaging from KU Leuven (2018). He is currently a postdoctoral researcher at KU Leuven, laboratory for Soft Matter and Biophysics. His research focuses on unraveling fundamental cell-material interactions towards biomedical and biosensor applications, as well as development of synthetic receptors and bio(mimetic) sensors.