Remote Query Resonant-Circuit Sensors for Monitoring of Bacteria Growth: Application to Food Quality Control

This paper presents a technique for in-situ remote query monitoring of bacteria growth utilizing a printed thin or thick-film sensor comprised of an inductor-capacitor (LC) resonant circuit. The sensor, which is placed within the biological medium of interest and remotely detected using a loop antenna, measures the complex permittivity of the medium. Since bacteria growth increases the complex permittivity of a biological medium the LC sensor can be used to determine bacteria concentration. This paper presents results on monitoring of three different bacteria strains, Bacillus subtilis, Escherichia coli JM109, and Pseudomonas putida, demonstrating application of the sensor for monitoring bacteria growth in milk, meat, and beer. Due to its low unit cost and remote query detection, the sensor is potentially useful for commercial scale monitoring of food quality.


Introduction
Bacteria are a major source of contamination in water and food supplies resulting in both food poisoning and disease outbreaks [1]. Although constant monitoring of bacteria concentrations can reduce food-related illnesses, counting bacteria in a biological food medium is a time consuming and difficult task due to the complex nature of bacteria and the biological medium. Today most bacteria counts are still determined by standard plate count (SPC) or psychrotrophic bacteria count (PBC) [2], which are labor and time intensive. Microbiological impedance devices [3][4][5][6][7][8] provide a less labor intensive way to estimate bacteria count in a biological medium. Microbiological impedance devices measure the permittivity and/or impedance spectra of the biological medium by immersing two or three electrodes into the medium and measuring the voltage at a constant current. The bacteria concentration or even the strain of the bacteria can then be calculated from the measured permittivity and/or impedance spectra [9][10][11].
One of the major applications of microbiological impedance devices is to determine the initial bacteria count in a food medium through a technique called Impedance Detection Time (IDT) [12][13][14][15][16]. The initial bacteria count obtained from IDT is then used to estimate the useful shelf life of a food product. However, the initial bacteria count does not always correlate well with the real shelf life due to improper storage conditions or post packaging food contamination. Furthermore, direct characterization of bacteria concentrations using impedance devices has some disadvantages such as penetration of the test chamber which introduces the potential for contamination, polarization of the probe electrode, and bubble formation at the electrode surface. An optimal way to monitor food quality is to remotely measure the bacteria population just prior to food consumption.
In this paper, we present a wireless, passive remote-query sensor for in-vivo monitoring of bacteria growth. The sensor, referred to as the LC sensor [17], is a series-connected interdigital capacitor and a spiral inductor printed on a thin plastic or paper substrate (see Fig. 1). A thin polyurethane layer is coated on the sensor to prevent the biological medium, which is electrical conductive, from shorting the capacitor electrodes. The sensor is immersed in the biological medium of interest, and the impedance spectrum of the sensor remotely detected by measuring the impedance across the terminals of a loop antenna used to monitor the sensor, see Fig. 2. The impedance spectra of the antenna is eliminated from the measurement by measuring the antenna impedance when the sensor is absent, and then subtracting that value from the impedance measurement of interest. Fig. 3 is a backgroundsubtracted impedance spectrum of a sensor immersed in water. Two important parameters are determined from the plot: the resonant frequency f 0 which is defined as the frequency at the maximum of the real impedance (resistance), and the zero-reactance frequency f z which is at the frequency where the imaginary impedance (reactance) goes to zero. From f 0 and f z , the effective relative complex permittivity of the biological medium and polyurethane layer, ε eff '− jε eff ", is given as [18]: where ε 0 is the free space permittivity (ε 0 = 8.854 × 10 −12 Farads / meter ), ε S is the relative permittivity of the electrically lossless substrate (that is ε s = ε s '), κ is the cell constant of the interdigital capacitor, and L is the inductance of the spiral inductor in Henry's. The cell constant κ and inductance L are calculated from the sensor geometry using [19,20]: where a, b, , OD, and ID are the dimensions of the sensor defined in Fig. 1, N c is the number of electrode fingers in the capacitor, N L is the number of the inductor turns, and K is the elliptic integral of the first kind. For any given individual sensor, the thickness of the protective polyurethane coating affects the measurement accuracy. Hence to eliminate effects due to variation in the polyurethane coating thickness we first measure the complex permittivity of deionized water using the LC sensor and a strip-line cavity [21]. A correction factor is then calculated by normalizing the LC sensor measurement ε cal '− jε cal " to the strip-line measurement ε ref '− jε ref ". The actual complex permittivity of the biological medium, ε r '− jε r ", is then calculated by multiplying the measured effective permittivity by the correction factor: Experimentally, using the calibration technique we found the LC sensors to have permittivity measurement errors of less than 8% in comparison to stripline cavity measurments. The complex permittivity of a biological medium increases with bacteria concentration causing a change in the capacitance of the interdigital capacitor, and in turn shifting f Z and f 0 . The changes in f Z and f 0 are used to calculate ε r ′ and ε r ′′ to determine the bacteria concentration in the biological medium. In this paper we present the results on Bacillus subtilis, Escherichia coli JM109, Pseudomonas putida grown in Luria Bertani medium. The results are compared to optical density measurements at 600 nm. Due to its low unit cost and wireless detection, the LC sensor is suitable for commercialized food quality monitoring by measuring the bacteria concentration of the food before consumption. In this work we demonstrate application of the LC sensor technology for monitoring the quality of milk, meat, and beer. To measure bacteria concentration in a liquid such as milk or beer, the sensor is fully immersed in the medium, while to measure bacteria concentration in a solid medium such as meat, the sensor is placed next to the object with the interdigital capacitor facing the object. Since the response of the passive LC sensor is remotely measured through a loop antenna, placing an autoclaved sensor within the food package prior to sealing would enable the quality of the food within the package to be determined at any time. Since the sensor is powered by the query field generated by the antenna, it does not require any internal batteries thus avoiding battery lifetime issues. The sensor is a simple resonant circuit consisting of conductor lines printed on a paper or plastic substrate, which allows it to be inexpensively fabricated and used on a disposable basis.

Figure 2.
The general experimental setup for bacteria monitoring. An impedance analyzer is used to measure the response of the sensor. The coordinate system used to analyze sensor performance with orientation and location is also shown. The resonant frequency f 0 is defined as the maximum of the real impedance, while the zeroreactance frequency f Z is the zero of the imaginary impedance. The complex permittivity of the medium is calculated from the measured values of f 0 and f Z .

Bacteria Growth Monitoring
The growth of three bacteria strains, Bacillus subtilis, Escherichia coli JM109, and Pseudomonas putida, were monitored at room temperature (25 °C) as a function of time. The bacteria culture was continuously stirred at a constant temperature using a water bath throughout the duration of the experiment (see Fig. 4). The culture medium, sensor, and the container used for the experiment were autoclaved at 120 °C for 15 min prior to the experiment. All three bacteria strains were grown in Luria Bertani (LB) medium, obtained from Difco Laboratories (Detroit, MI), at a constant temperature of 25°C. The sensor was 4 cm × 4 cm and protected by 150 µm thick polyurethane layer. The antenna was a 6-turn loop antenna of 9 cm in diameter, placed at 8 cm away from the sensor with both the sensor and antenna basal planes in parallel. The impedance of the sensor is measured with a HP 4192A Impedance Analyzer. While the polyurethane coating protects the electrical integrity of the sensor, when immersing the sensor in a liquid medium water molecules are initially absorbed into the polyurethane layer increasing its complex permittivity. However, this absorption process stops within a few hours after the polyurethane pores become saturated. Hence to eliminate the effect of water absorption on liquid immersed sensor experiments the sensors were initially soaked in water for four hours prior to use.
Optical density measurements were taken while the permittivity of the medium was measured. Fig. 5 shows a linear correlation between the optical density and the change in the permittivity magnitude |∆ε r | for both E. coli and Pseudomonas cultures. The optical density measurements require more than one hour between two consecutive measurements to detect any change. In contrast, the LC sensor can resolve changes in |ε r | of less than 0.01, corresponding to a resolvable continuous monitoring time period of approximately 5 s.   6a and 6b plot the changes in medium complex permittivity, ∆ε r ' and ∆ε r ", of E. coli, Pseudomonas, and Bacillus cultures over a 48 hr period, with the upward trend in permittivity indicating culture growth. The initial complex permittivities of the three bacteria are different, with ε r ' ranging from 118 to 137 and ε r " ranging from 19 to 37, and so for comparison starting values have been normalized to zero. As can be seen from Fig. 6 E. coli has the fastest growth rate with an increase of 13.1 in ε r ' and 2.0 in ε r " over 48 hr, while Bacillus has the slowest growth rate with a 5.1 increase in ε r ' and 0.8 in ε r " over 48 hr. Also notice a well-defined growth step in the E. coli ∆ε r ' spectral indicating synchronized growth, whereas the ∆ε r ' spectra of both Bacillus and Pseudomonas are continuous.

Food Quality Monitoring
The LC sensor has been used for monitoring the quality of milk, meat, and beer by determining the bacteria populations as a function of time. All sensors in our experiment were 4 × 4 × 0.05 cm in size except the sensor used in the meat package, which was 2.5 × 2.5 × 0.05 cm. A polyurethane layer, approximately 150 µm-thick, was sprayed on the sensor, and the sensor baked at 120°C for 2 hrs. Before starting the experiment, all sensors were autoclaved at 120 °C for 15 mins. The LC sensor measurements were correlated with standard plate counts (SPC) for the milk and meat monitoring experiments.

Milk Quality Monitoring
The milk sample used in our experiment was 1%-lowfat skim milk purchased from a local grocery store. A portion of the milk (500 mL) was poured into a beaker and then autoclaved at 120 °C for 10 mins and used as a control sample. Another 500 mL of milk was poured into another beaker. A sensor was inserted into each beaker; the autoclaved milk was aseptically sealed, and the regular milk loosely covered with a plastic wrap. The experimental setup of the milk is similar to the one shown in Fig. 2, except that two single-turn 8 cm-diameter loop antennas were used to simultaneously measure both milk samples. Both samples were kept at room temperature and monitored for 24 hours. To correlate the complex permittivity with the bacteria count samples were taken from the open milk culture every 2 hrs with the bacteria population determined using standard plate count (SPC). The autoclaved milk sample remained unopened until the termination of the experiment.   7a plots the change in complex permittivity over time for the sensors immersed in the autoclaved milk and in the regular 1%-lowfat milk. As can be seen from the plot, both the real and imaginary permittivity values of the sensor within the autoclaved milk are almost constant over time, indicating no significant bacteria growth. Conversely, the real permittivity of the 1%-skim milk increases by 16 in 24 hrs while the imaginary permittivity increases by 3. Fig. 7b plots the change in the real and imaginary permittivity of the skim milk versus the bacteria count. Correlating Fig. 7a to Fig. 7b, the complex permittivity can be readily used to indicate if the milk is spoiled. For example, if a bacteria count of 1.2 × 10 7 cfu/ml is considered unsafe for human consumption, the milk in our experiment should be discarded when the ∆ε r ' reaches 12 and ∆ε r " reaches 2.5, which is about 21 hrs after the experiment began.

Meat Quality Monitoring
A 90%-lean 0.5 kg ground beef package was purchased from a local grocery store. The sensor was inserted between the beef and the Styrofoam container, as shown in Fig. 8, with the package then covered, but not hermetically sealed, with a plastic wrap. The complex permittivity of the meat, kept at room temperature, was measured for 1.5 days with the results plotted in Fig. 9. Plate counts were not performed on the meat sample due to an inadequate volume of liquid inside the meat package, instead another experiment was conducted to correlate the changes in the meat permittivity to bacteria growth. In this experiment, 100 mL of Luria Bertani (LB) media were inoculated with 30 µL of an overnight bacterial culture grown from 2 g of meat. The LB-meat mixture was used to represent the bacteria populations typically found in ground beef. The sensor was presoaked in the LB media for 8 hours before the culture was added, and the complex permittivity of the LB culture monitored as a function of time. Samples were taken from the LB culture every 4 hrs to perform SPC. Fig. 10a plots the changes in the LB complex permittivity that had been inoculated with bacteria from a meat sample. Fig 10b plots the correlation between the complex permittivity of the LB-meat medium and the plate count of the medium, showing that the increasing trend of the complex permittivity in Fig. 10a is due to bacteria growth in the medium.

Beer Fermentation Monitoring
The LC sensor can also be used to monitor the fermentation processes in-situ, without opening the container, thus preventing unwanted contamination. Here, we demonstrate the monitoring of the fermentation of a beer culture consisting of 1/2 cup of sugar, 1 cup of oatmeal, 2 cups of carbonfiltered tap water, and 7g of Fleischman's Rapidrise TM yeast. The beer container was sealed, and the response of the sensor was monitored for 4 days. The complex permittivity of the yeast culture, calculated from the sensor f 0 and f z , was determined as a function of time and plotted in Fig. 11. The plot in Fig. 11 shows a step-wise change in ∆ε r ' and a continuous change in ∆ε r " due to the yeast growth.

Effects of Temperature and Coating Thickness
In our experiments the effects of the polyurethane coating were calibrated using Eq. (4), with the calibration process specific to the polyurethane coating thickness. Temperature also has a significant effect on the response of the sensor. The resonant frequency of a 4 cm square sensor made on a polystyrene substrate shows a change of 6.4 kHz change in resonant frequency for every degree °C change, which causes a decrease of about 0.7 -j 0.4 in the measured complex permittivity of LB medium. Hence for practical applications the LC sensor has to be made upon a more thermally stable substrate, or used in conjunction with a second LC sensor having a different temperature dependency to enable elimination of temperature effects through cross-correlation.

Effects of Sensor Location
The frequency response of the sensor experiences a small shift when the location and orientation of the sensor vary. However the sensor can operate with small error tolerance within a well-defined region. To determine the sensor operating region the zero-reactance frequency f Z of a 4 cm sensor interrogated with an 8 cm-diameter loop antenna was measured as a function of the sensor displacement in the x and z directions, and the tilt angle θ between the basal planes of the sensor and the antenna (see the coordinate system defined in Fig. 2). The results, shown in Fig. 12, indicate that within a region (0 cm < x < 3 cm, 3 cm < z < 6 cm) the variation in f Z was less than 5 kHz, or about 0.25% error in the measurement. To maintain the 5 kHz error tolerance, the tilt angle θ must also be less than 30°. There is a minimum separation distance in the z direction between the sensor and the antenna due to the near-field inductance coupling between the sensor inductor and the antenna that causes a change in f 0 and f z . The maximum separation distance between the sensor and the antenna is due to the errors from the electronic instrumental noise when the sensor signal becomes too small to be detectable at large distance.

Effects of Water Absorption in the Sensor Surface
When the LC sensor is immersed in a liquid, water molecules are absorbed into the polyurethane layer, decreasing f 0 and f z and causing a shift in measured complex permittivity. However, the absorption process stops in a few hours after the polyurethane layer saturates. Fig. 13 shows f z of a sensor coated with 150 µm-thick polyurethane layer, subsequently baked at 120 °C for 2 hrs, and then immersed in water. As can be seen from Fig. 13, the saturation time is ≈ 4 hrs. The saturation time of the water absorption process depends upon the polyurethane thickness, and the pore size which is a function of the curing temperature. Experimentally we found that the saturation time increases to 15 hrs when the curing temperature is 80°C, and 28 hrs when the curing temperature is 60°C. The saturation time also increases from 4 hrs to 14 hrs when the polyurethane thickness increases from 150 µm to 400 µm. In summary, the saturation time decreases with increasing baking time and temperature, and decreasing coating thickness.  Long-term exposure to a bacteria laden medium will damage the protective polyurethane coating. The surface of an as-made sensor is shown in Fig 14(a), and the surface of a sensor immersed in distilled water for 4 days is shown in Fig 14(b). Fig. 14(c) shows the sensor surface after immersion in a bacteria culture for 2 days, and 14(d) the sensor surface after immersion in a bacteria for culture for 4 days. The lifetime of a polyurethane-coated sensor in bacteria laden solution is about one week, after which the polyurethane layer peels away from the sensor.

Conclusions
A wireless, passive remote-query sensor technology is presented for monitoring bacteria growth. The sensor is a planar, printed or photolithographically defined resonant circuit, the response of which is remotely detected by a loop antenna. The substrates upon which the resonant circuits are defined can be flexible (e.g. paper or plastic film) or rigid (e.g. ceramic, plastic, Plexiglas) depending upon the desired application. The sensor is immersed in the medium of interest, and the complex permittivity of the medium calculated from the electromagnetic response of the sensor. Our results show the sensor can measure complex permittivity changes in bacteria cultures consisting of E. coli, Bacillus, and Pseudomonas with greater discernable measurement sensitivity than optical density measurements.
The application of the LC sensor technology to monitoring milk, meat, and beer quality was illustrated. Experimental results show the complex permittivity of the autoclaved milk was constant over time, while the complex permittivity of the 1%-fat skim milk increased over time as the bacteria population increased. The sensor technology was also used to measure the increase in complex permittivity of a meat sample, with time, as the meat spoiled. By correlating the complex permittivity shift with the bacteria count the sensor can be used to indicate if a food product is safe for human consumption. In the case of a dry food product, such as cereal or grain, the presence of humidity due to package failure would readily change the measured impedance of the sensor. Additionally, the LC sensor was used to monitor the fermentation process of a beer culture during fermentation.
Since the LC-sensor monitoring electronics could be reduced to a relatively inexpensive small-scale portable package, the LC sensor technology presented here shows promising potential for commercial food quality monitoring in a grocery store or even at home. Each sensor costs a fraction of a penny, so it can readily be used on a disposable basis. Furthermore the LC sensor is relatively insensitive to location and orientation within its operation range so untrained users can operate it.