Thermal measurements in PBS
At first, it was necessary to determine the required liquid volume required in the addition-type cell in order to achieve stable measurement signals. To this end, water was pipetted into the resin reservoir with different volumes. When a volume of 90 μL was used, within 30 min, the resistance of air was measured rather than of the liquid, suggesting this is not sufficient to conduct temporal measurements. A stable baseline level was achieved when using sample volumes of 100 μL, which was further used for measurements with filament and resin-printed reservoirs (Figure S-6). A subsequent experiment focused on determining the optimal stabilisation time in addition to the impact on the noise on the signal when moving away from the original flow cell design. Instead of water, PBS was used since it mimics the ionic strength and pH (7.4) of blood.
The results are summarised in Table 1. Experiments were conducted at different environmental temperatures and PID settings. These experiments were conducted with the resin-printed and copper measurement cell, as initial results for the measurement cell produced by fused filament deposition had aforementioned issues with voiding, leading to sample leakage and high rate of evaporation. Even though the copper measurement cell had some issues in this respect since it relies on a similar production principle, this could easily be combatted by applying a coat of varnish on the exterior of the cells which was not possible with the filament measurement cells. Furthermore, the printer has an additional feature to adjust size and account for shrinkage, making voidage less likely.
Figure S-7 provides an overview of reported relative noise signal in the literature of SLA-printed flow cells and flow cells manufactured (drilled) from Perspex. While these experiments were conducted with different substrates, it is expected that the origin of the noise tends to be electronic and therefore does not have a significant impact on the average standard deviation. In general, literature reports that the average standard deviation on the signal tends to be between 0.7 and 1%. From Table 1, the average deviation on the signal can be calculated which for the resin is ~ 0.9% at 32 °C (P = 1, I = 10, D = 0.3) and could be further optimised to ~ 0.6% by adjusting the PID settings to P = 1, I = 15, D = 0. For the copper flow cell, there is no considerable difference between the PID settings and it remains around 1%. Therefore, it was demonstrated that both the measurement cells exhibited a stable signal and are suitable for thermal measurements and have potential to improve the signal to noise ratio compared to the original flow cell design.
The influence of the ambient temperature was also studied. In general, there was a trend towards lower Rth values with increasing environmental temperatures. This can be explained by the fact that a smaller temperature difference with the environment leads to less heat loss and therefore a lower overall temperature gradient (T1 – T2), resulting in a reduction of the Rth. Since we report the signal as a relative increase to the baseline, it would be beneficial to have lower Rth values. However, when the incubator was kept at the same temperature as the measurement cell, this led to overcompensation by the PID feedback loop, which therefore made it not suitable for measurements. Pre-heating of the samples therefore did not improve response time; thus, a balance needs to be struck between the incubator temperature and keeping a sufficient temperature gradient for the PID feedback to rapidly correct the measured temperature to the set value.
The PID parameters did not have a significant impact on noise. However, it is known they do have a considerable impact on start-up times and therefore we further investigated how the material characteristics and feedback loop affect the time until stabilisation. This is an important consideration for the commercial prospect of the sensor platform considering measurement times should ideally not exceed 5 min for point-of-care applications.
Table 2 summarises the time to reach steady state for the different measurement cells, which was defined as the time when the average reading of the thermocouple varied by less than 0.1ºC degree over a 1-min period. It was not possible to reach a stable signal for SLA cell 2 at 37 °C since the PID loop needs a temperature gradient in order to stabilise the signal.
In a previous research, stabilisation times of at least 30 min (1800s) were considered prior to injecting samples, which limited the point-of-care application of the sensor. Modelling studies confirmed this was due to the velocity of the liquid, which leads to disturbance in the signal and therefore prolongs the stabilisation period . When samples are simply pipetted in, the velocity of the liquid is virtually zero and in all scenarios, the stabilisation time does not exceed 20 min. While it was anticipated that the copper flow cells would stabilise faster due to having a higher thermal mass, this was not the case. A likely explanation for this is that directing the heat flow via one copper sink is faster compared to heating being distributed all over the cell.
While the impact of the PID parameters was not very clear with regard to reducing the overall noise, the I (integral) component played a vital role in the response time. For instance, we determined that by changing the I from 10 to 15 led to improving the time until steady state from 900 to 450 s. Therefore, it is key to adapt the PID settings depending on the material and design of the measurement cell used prior to each measurement in order to optimal sensor performance.
Computational fluid modelling
Figure 2 shows the temperature of the fluid in contact with the thermocouple tip vs time, and as such gives an indication of the expected settling time of the sensor measurement. It was demonstrated that there is negligible difference between the copper and resin cells, which was largely in agreement with the experimental results. A lower ambient temperature led to a lower temperature of the liquid, which was in line with the performed measurements. Lower ambient temperatures were expected to lead to slightly faster settling times, but experimentally, no clear trend was found.
An indication of signal noise for each flow cell can be obtained by studying temperature contour plots and the average velocity in the cell: if the fluid temperature in the cell significantly varies and a velocity is present, the signal will be affected by fluid of varying temperature flowing across the thermocouple. Figure 3 shows the steady-state temperature contours for both the resin and copper cells at various ambient temperatures on a consistent temperature scale. The ambient temperature had a significant influence on the temperature within the cell while the material of construction had no significant influence. For both materials, the fluid temperature within the flow cell (central section) varies by around 4 °C for an ambient of 17 °C, 1–2 °C for an ambient of 27 °C and with no variation at all for the 37 °C ambient case (due to the environment and heater block being at a constant temperature). Hence, it was expected that measurement cells situated within an ambient environment similar to that of the heater block (in this case 37 °C) would encounter less noise than those in colder environments. This was further supported by the cell velocity magnitude data shown in Figure S-8. As natural convection flows are driven by the temperature difference between source and sink (heat block and ambient temperature), the velocity magnitude reduces with an increase in the ambient temperature, to 0 mm/s for the case of a 37 °C ambient. A lower velocity in the cell should correlate to decreased noise, as there is less refresh of the fluid in contact with the thermocouple. In general, lower noise was found in measurements at 32ºC compared to 27ºC, which was in agreement with the modelling. However, when the incubator was set at the same temperature as the programmed temperature of the heat sink, the feedback loop is not able to control the signal leading to high noise on the signal. Thus, it was found that 32ºC was the optimal ambient temperature for measurements.
The computational fluid modelling results show that lower ambient temperatures resulted in faster settling times, while higher ambient temperatures would lead to less noise on the signal. However, the experimental results do not follow such clear patterns suggesting that other aspects of the flow cell design (e.g. the temperature controller) have a greater impact on the signal stability compared to the underlying fluid dynamics and heat flow in the cell.
Thermal measurements of MIP-functionalised electrodes.
MIP layers were functionalised onto the glass slides as described in 2.3. Scanning electron microscopy analysis (SEM) confirmed there was full and uniform coverage of the surface, as shown in Figure S-4. These MIP-modified glass slides were inserted into the measurement cells and stabilised in PBS for approximately 30 min after which a solution of 100-nM amoxicillin in PBS was injected. This process was repeated for all measurement cells including the original flow-through design, as shown in Fig. 4. To assess the specificity of the sensor, measurements were performed with a reference NIP, which results are shown in Figure S-9 and indicate that no increase in thermal resistance was found when NIPs were exposed to solutions spiked with amoxicillin. The copper measurement had a lower thermal resistance compared to the other cells, which can be explained by the higher conductivity of the material. This will ultimately lead to a smaller temperature gradient and minimal changes in signal when samples are added. There was also difficulty in repeating measurements with this design as the lower level of heat loss to the environment led to a higher rate of evaporation, which was observed in some cases as a continuous upwards drift of the signal.
Figure 4 clearly demonstrates the difference in stabilisation time between the addition-type set up and the flow-cell design. All addition cells are stabilised between 450 and 1000 s, whereas for the flow-cell design at least 1500 s was needed. In fact, in general at least two injections with PBS are performed (Fig. 3c shows signal after second PBS injection) to ensure the signal was stable before the measurement was amoxicillin was conducted. The addition cells also required a considerably shorter time (750 s) to stabilise after the samples are added via pipetting, thus having a much faster measurement time compared to the flow cell that needed at least 1500 s or more to record a reliable value. Increasing the speed of injection will not help in that respect as a certain residence time is needed to achieve binding between target and polymer.
While one example was shown for the FFF PLA cell, it was very difficult to repeat this measurement, and the reported increase was not found to be significant. We also encountered issues with the FFF metal cell; whilst it was possible to combat leakage by applying varnish to the outside of the cell, a coat of varnish might also need to be applied to the inside. Another option is to add slightly more liquid to the measurement chamber (200 vs 100 μL) since some of the liquid will be absorbed by the pores inside the flow chamber. Despite repeating these measurements several times, no significant response was found with copper as material for the flow cell. A possible explanation for this could be that a temperature gradient is necessary to analyse heat flow at the interface, whereas the FFF metals have an even heat distribution across the cell walls. Table 3 summarises the results for the four different designs, comparing the differences in measurements. The measurement with the cells printed with SLA was performed in triplicate and had error bars of around 1% for the relative response to amoxicillin and ± 100 s for the stabilisation time.
The SLA resin addition cell and SLA flow cell are user-friendly, circumvent issues with voiding, and are both suitable for thermal analysis of antibiotics with MIP-based sensors. The addition-type measurement cell demonstrated a similar response at 100 nM compared to previous research where MIP-modified screen-printed electrodes were used . These designs can be applied in different areas; for instance, an addition-type design due to its low sample volume and shorter stabilisation time will have its merit in measuring clinical samples. However, the flow cell design will be advantageous for high throughput of samples or if one is interested in studying dynamic systems.
To establish proof-of-application of the system, further measurements were conducted with the SLA cell which demonstrated the highest potential for future applications (see Fig. 5). The results for both MIP (Fig. 5a) and NIP (Fig. 5b) are shown, with a comparison of the relative Rth response upon exposure to PBS solutions spiked with amoxicillin (Fig. 5c). The measurements were repeated in triplicate with freshly prepared MIP-functionalised electrodes with a standard deviation of ~ 1% at respective concentrations.
There were some minor differences in absolute Rth values of MIP and NIP, which is likely linked to changes in structure (MIP structure is more porous) and film thickness. Once the thermal response was normalised, it was clear that only for the MIP a significant increase in Rth was observed which can be attributed to binding to the polymer layer. Using the three sigma method, the limit of detection was estimated to be ~ 10 nM, which is well below the tolerated maximum residue level of amoxicillin. In the future, dose–response curves will need to be established for determining amoxicillin content in water and environmental samples.