Rapid Antibiotic Susceptibility Testing on Pathogenic Bacteria in Solid Growth Medium Using a Real-Time and Contactless Planar Microwave Resonator Sensor

Infection diagnosis and antibiotic susceptibility testing (AST) are pertinent clinical microbiology practices that are in dire need of improvement, as current standards are not able to keep up with the mutations and resistance development of certain bacterial strains. This paper presents a novel way to conduct AST which hybridizes disk diffusion AST with microwave resonators for rapid, contactless, non-invasive and high-throughput testing. This work uses Escherichia coli ( E. coli ) cultured on solid agar and places bacteria samples on a microwave split-ring resonator along with antibiotic disks (erythromycin) of various doses to demonstrate the viability of this sensing method in a clinical microbiological setting. The microwave resonator, operating at a 1.76 GHz resonant frequency, boasted a 5 mm 2 sensitive sensing region. A one-port sensor was designed and optimized for detecting dielectric property variations of lossy dielectric materials accurately. This sensor was calibrated to detect uninhibited growth of the bacteria at 0.005 dB/hr, with a maximum change of 0.07 dB over the course of 15 hrs. The transient resonant amplitude change was subsequently dampened for each increasing dosage of antibiotic tested, with 45 µg of erythromycin showing negligible change indicating complete inhibited growth. This AST sensor demonstrated decisive results of antibiotic susceptibility in under 6 hours and shows great promise to further automate the intricate workflow of AST in clinical settings, while providing rapid, sensitive, non-invasive and high-throughput detection capabilities.


Introduction
Bacterial infections have become a public health crisis around the world in recent years. Due to over-prescription of antibiotics and non-compliance of their usage around the world, more families and strains of pathogens have begun to take a route towards further resistance against antibiotics [1][2][3][4] .
Accordingly, the World Health Organization (WHO) has developed a list of 12 pathogen families which are highly resistant to antibiotics and pose a threat to public health on a global scale 3 . The new-found resistance of these families towards antibiotics, partnered with the WHO's bid towards further research into new antibiotics has also created a need for faster and more high-throughput antibiotic susceptibility testing (AST) practices 4,5 .
Current AST methods often suffer from the pitfalls of being extremely expensive, time consuming, labor intensive, prone to cross contamination, and have unstandardized practices 5 . Disk diffusion and broth dilution methods are currently the most widely utilized tests which use various antibiotics in different concentrations against bacterial colonies to test for antibiotic susceptibility or resistance [4][5][6][7][8] . These methods are extremely time-consuming and require upwards of 24 hours of incubation of the bacteria. Furthermore, they display poor performance when analyzing slow growing bacteria and both methods require expensive resources such as large volumes of reagents, and sizeable experimental equipment. Recently, matrixassisted laser desorption ionization-time of flight mass spectrometry (MALDI-TOF MS) has been deemed efficient in differentiating between various bacteria strains, however it is far too expensive to operate and maintain for most clinics globally 9,10 . Additionally, newly emerging genotypic methods have been used, which utilize polymerase chain reaction (PCR) 11,12 , DNA microarrays 13 , or loop-mediated isothermal amplification (LAMP) 14 , however their complex and user intensive protocols make them hard to implement in various clinics and situations 4,11 . Other emerging methods of AST which utilize fluorescence 15 , electrochemical sensors 16 , microfluidics [17][18][19][20][21] , microscopy 22 , and bioluminescence 15 are currently improving scope of AST however, they are far too expensive and complex to develop, require laborious protocols to execute and need various expensive reagents and equipment making testing cumbersome 4 . This creates a crucial need for a more efficient and faster method of AST which is cheap enough to implement and does not require a plethora of resources to operate and maintain.
Microwave sensing is recently developing for clinical microbiology practices [23][24][25][26] . Microwave resonators have shown to be highly sensitive in sensing and monitoring bacteria [25][26][27][28] . Through translating variations of dielectric properties to quantifiable signals such has resonant amplitude and frequency, microwave sensing has shown to be adaptive towards the needs of biosensing and inexpensively implemented without the need to execute laborious protocols 25,27,29,30 . Planar microwave resonators are easily fabricated, can be reused for multiple assays, and have shown highly sensitive, real-time, and highthroughput results [31][32][33][34][35] . Microwave resonators have been coupled with microfluidic chips to simulate broth dilution tests and have already displayed high levels of sensitivity in monitoring bacteria in various conditions and monitoring bacterial growth 25,26,36,37 . Furthermore, differential microwave resonators have been implemented to reduce environmental noise in performing a more robust analysis on bioassays 27,30 .
However, AST has only been proposed through microwave sensing and not yet demonstrated 25 . This lays on the premise in which antibiotics have proven to impact the growth of microorganisms by targeting the protein synthesis, DNA and RNA multiplication, and other metabolic activities to inhibit bacterial growth 38,39 . The change of concentration of charged by products can indirectly aid in monitoring the metabolic activity by measuring conductivity and permittivity changes with resonant profiles overtime 25,26,28 . Therefore, to tackle the need of faster and more efficient AST practices, microwave sensing can address all of the pitfalls suffered by the methods of AST previously discussed.
In this study, a simplified model of AST on Escherichia Coli (E. coli) using microwave split ringresonators is demonstrated through only using high concentrations of antibiotics, which the bacteria were proven to be susceptible to. Among the existing microstrip resonator variations, a frequency variation resonator consisting of a single split ring was employed to monitor and detect the impact of various concentrations of antibiotics on the growth of E. coli. Solid MH agar inoculated with E. coli is cultured on a microwave resonator ring gap and the transient resonant response of the resonator is dictated by the growth of the bacterial cells. Upon introducing various concentrations of the antibiotic erythromycin (E), bacteria growth is inhibited which deviates from the resonant response of this test as opposed to growing bacteria as seen in the decrease of variation in transient resonant amplitude change. This electrical signal is analyzed by a vector network analyzer (VNA) in a real-time, non-invasive and sensitive manner. Each sample was tested for 15 hours, however, decisive signals of growth inhibition was determined as early as 6 hours. This study not only calibrates a microwave resonator for AST on E. coli using erythromycin, but it demonstrates the efficacy of microwave sensors in clinical microbiological practices and has shown to vastly enhance the efficiency of AST.

Electromagnetic Field Analysis of the Sensor
A planar microstrip ring resonator structure was modelled in High Frequency Structure Simulator (HFSS) to study the operation of the resonator. The resonator was designed to operate at a frequency of 1.76 GHz due to low effective loss of the solid MH agar media. The resonant structure consists of a halfwavelength split ring resonator and a feed line as shown in Figure 1 Where l is the length of the split ring resonator (m), c is the speed of light (3×10 8 m/s), f is the resonant frequency (1.76 GHz), and is the relative permittivity of the substrate (2.2). Based on the equation (1), the length of the split ring resonator was calculated to be 57.5 mm. However, resonant frequency is also governed by the capacitance of the resonant circuit 40 . Since, the ring gap of the resonator was modified to monitor an area of 5 mm 2 , the length of the resonator was modified to 73.8 mm to compensate the increase in the capacitive area and achieve the desired resonant frequency. The coupling gap width was selected as 0.8 mm to achieve maximum magnetic coupling between the feed line and the split ring as shown in Figure   1(b). The feedline was matched to the electromagnetic (EM) source impedance i.e. 50 Ohm using LineCalc (by Advanced Design Systems 2020) to allow maximum power transfer from source to resonator structure.
The dimensions of the feed line and the split ring resonator is shown in Figure 1(a).
The resonant structure was designed as a one-port device to reduce the complexity of the readout circuit. One port devices are widely used for accurately detecting and characterizing conductive dielectric materials 41 . As shown in Figure 1(a), one end of the feedline was connected to an EM source and the other end was terminated to ground using a high impedance quarter wavelength microstrip stub. The stub was added to discharge any weak capacitances formed between the feedline and the sample MH agar petri dish.
The high impedance of the stub prevented loading of the matched feed line, while simultaneously acting as an open circuit to generate a microwave signal at the resonance frequency and short circuit at DC frequencies. An alignment marker was added in the design to strategically align the sample petri dish with the sensor. The simulated response of the bare sensor in the absence of test material is presented in Figure   1(d), shown in black, in which the resonant frequency and amplitude were found to be 1.76 GHz and -11.58 dB, respectively.
To determine the sensitive region of the resonator, a full 3D electromagnetic field simulation was

Initial Antibiotic Susceptibility Disk Diffusion Calibration Discussion
Prior to utilizing the microwave resonator to monitor the impact of antibiotics on the growth of microorganisms, a preliminary disk diffusion test was performed. In this test, the growth of E. coli was

Microwave and Image-based Detection of Bacterial Growth against Antibiotics
A study to determine the impact of a hydrated blank paper disk (without antibiotics) and a 45 µg erythromycin disk on the response of the resonator was performed to establish a baseline response to distinguish between detectable and undetectable regions of microbial growth (Figure 3(a)  In the absence of erythromycin, a maximum resonant amplitude variation of 0.07 dB was observed, for the control plate (E-0 µg), due to unrestricted bacterial growth around the sensitive region of the sensor (Figure 3(c)). However, in the presence of erythromycin, the protein synthesis during bacterial metabolism was hindered, thereby impacting bacterial growth. As evident from Figure 3(b), the rate of change of resonant amplitude, which can be associated with the growth rate of bacteria decreased as the concentration of erythromycin increased, which is in positive correlation with the microscopic images captured at constant time intervals (Figure 3(c)-(f)). The slope of the measured resonant amplitude variation in the detectable bacterial growth region was calculated and found to be 0.005, 0.003, 0.002 dB/hr with a coefficient of determination (R 2 ) of 0.999, 0.998, and 0.997 for 0, 7.5, and 30 µg of erythromycin, respectively.
Furthermore, an insignificant change in the Δamplitude of 0.005 dB at 45 µg erythromycin concentration indicated a complete inhibition of bacteria growth and was supported by microscopic images captured at constant time intervals (Figure 3(f)).
Bacterial susceptibility to antibiotics has shown to be concentration-dependent 48,49 . Low doses of erythromycin decrease protein synthesis in bacteria, thereby reducing the growth rate rather than inhibiting bacterial growth completely. This is in accordance with the bacteriostatic nature of erythromycin. However, higher concentrations of erythromycin completely inhibited bacteria growth leading to the formation of zones of inhibition as shown in Figure 2 in the previous experiement [48][49][50] . It is evident from the figure that a high concentration of erythromycin leads to a large zone of inhibition, and vice versa. Correlating these results with Figure 3, this explains not only the greater zone of inhibition for E-45 compared to E-30 and E-7.5, but also the difference of rates of change in the differential resonant amplitude over the course of the 15 hours measured by the microwave resonator. A noticeable difference can be seen in Figure 3(b), when comparing the differential responses of E-30 and E-7.5. Although subtle, E-30 showed a more dampened response of resonant amplitude change over time and had a lesser overall change than E-7.5, which is attributed to the reduced growth or metabolic activity of the bacteria. Therefore, in can be concluded that as the concentration of the erythromycin increased, the diameter of the zone of inhibition increased, overlapping the sensitive region of the microwave sensor. Consequently, the bacteria growth in the sensitive region decreased, which is in positive correlation with the measured microwave sensor's response and is supported by microscope images captured at constant time intervals (Figure 3(c) -(f)). Thus, it can be concluded that placing the antibiotic disks closer to the resonator will also impact the time at which decisive AST can be conclusive. However, the distance between the antibiotic disk and the resonator gap can be optimized as low concentrations marginally inhibiting growth can lead to inaccuracies for susceptibility in different concentrations.
A noteworthy outcome of this work was the sensor's ability to successfully distinguish the impact of different concentrations of erythromycin on the growth of E. coli before any visible cues. The microwave sensor was able to clearly distinguish the impact of erythromycin on E. coli growth within the first 6 hours of the experiment, which is far lower than the technical capacity of conventional microbiological disk diffusion studies. Early detection of the response of the microbial growth to antibiotics can be beneficial in developing treatment strategies against infections. Furthermore, the contactless, high-throughput, noninvasive, portable, inexpensive, and reusable nature of the planar microwave resonator makes it a promising tool in the field of microbiology.
In summary, a planar microwave resonator sensor was implemented to monitor the impact of

Bacterial Samples Preparation
Mueller Hinton Agar (MH agar) was used to prepare growth media used for the experiment.
1.91 mg of MH agar was thoroughly mixed with 50 ml of deionized water and sterilized for 15 minutes at 121 ℃ using an autoclave. The media was cooled down to 50 ℃ and 3 ml of the sterilized media was poured into a Petri dish to achieve an agar thickness of 1 mm. The plates were cooled down to room temperature and 3 µL of E.coli with OD600 = 1.5 was inoculated at the predesignated position.
The antibiotic disks were prepared by soaking a blank paper disk with 30 µL of the desired erythromycin concentration. The disk was transferred to the inoculated plate and placed at a constant distance of 10 mm from the inoculated E. coli. The Petri dish was sealed using a parafilm to avoid changes in the Petri dish due to external ambient factors.

Fabrication of Microwave Resonator
A microstrip planar ring resonator sensor was fabricated on a Rogers RO5880 laminate from Rogers Corporation Pvt. Ltd. with dielectric thickness of 0.79 mm, copper cladding thickness of 35 µm, permittivity of 2.2, and loss tangent of 0.0009. The resonator pattern was etched following previously used protocols 37 . Briefly, the split ring design was designed and simulated in ANSYS HFSS to optimize parameters and operating resonant profiles. Subsequently, the layout of the resonator was patterned using a laminator by incorporating several heating and pressing cycles. The patterned laminate was etched using ammonium persulfate. One end of the feedline was connected to a SMA connector and the other end was grounded. The modelled resonator structure and the simulation and fabrication results are discussed in detail in the above sections.

Initial Antibiotic Susceptibility Disk Diffusion Calibration
To demonstrate the impact of 0, 7.5, 30, and 45 µl of erythromycin on the growth of E. coli, MH agar plates were inoculated using streak plate method. Blank paper disks were soaked in 30 µl of RO water with 7.5, 30, and 45 µg of erythromycin and tapped on the surface of the growth medium. The petri dish was sealed using parafilm to avoid precipitation and fungal contamination inside the petri dish. The petri dish was incubated at 22℃ for 36 hours and the diameters of zone of inhibition was measured. The captured visual microbial growth at the end of this study are discussed in detail in the above sections.

Experimental Setup of the Sensor and VNA
The fabricated sensor arrangement shown in Figure 4(a) was enclosed in a thermally insulated and mechanically stabilized Styrofoam enclosure equipped with a LED light source, a digital microscope, and a temperature probe. The temperature probe was employed to monitor the variations in the ambient temperature. The ambient temperature was recorded to be 23.25 ± 0.25℃. The bacterial growth images were captured at fixed time intervals using a 10x Celestron digital microscope. The experimental setup is demonstrated in Figure 4(b). The inoculated plate was strategically positioned (Figure 4(a)) and firmly secured on the sensor such that the distance between the paper disk and the ring gap was 1 cm. The sensor was connected to a ZNB20 Vector Network Analyzer (VNA) by Rohde and Schwarz. The VNA was calibrated every 15 hours to minimize the impact of drift in the measurement instrument on the response of the resonator. The VNA was operated between 1.55 GHz to 1.8 GHz. The intermediate frequency (IF) bandwidth was set to 300 Hz to minimize the broadband noise, and the number of points was selected as 6401 to increase the sampling resolution. The output power of the VNA was set to 0 dB i.e. 1 mW which was sufficiently low to have no impact the growth of E. coli through adverse affects just as joule heating 51,52 . The VNA was triggered to measure the reflection coefficient (S11 dB) every two minutes using an automated LabVIEW program made in house.
To test the ability of the sensor to monitor the susceptibility of E.coli to various concentrations of the antibiotic, erythromycin (7.5, 30, and 45 µg), the reflection gain (S11 (dB)) of the sensor was measured and recorded using VNA and LabVIEW, respectively. The measured S11 (dB) was normalized by subtracting from the initially measured value due to day-to-day variations of ambient conditions. Furthermore, the data pertaining to the initial 90 minutes were excluded to account for stabilization of the measurement apparatus.