Microfluidic cantilever detects bacteria and measures their susceptibility to antibiotics in small confined volumes

In the fight against drug-resistant bacteria, accurate and high-throughput detection is essential. Here, a bimaterial microcantilever with an embedded microfluidic channel with internal surfaces chemically or physically functionalized with receptors selectively captures the bacteria passing through the channel. Bacterial adsorption inside the cantilever results in changes in the resonance frequency (mass) and cantilever deflection (adsorption stress). The excitation of trapped bacteria using infrared radiation (IR) causes the cantilever to deflect in proportion to the infrared absorption of the bacteria, providing a nanomechanical infrared spectrum for selective identification. We demonstrate the in situ detection and discrimination of Listeria monocytogenes at a concentration of single cell per μl. Trapped Escherichia coli in the microchannel shows a distinct nanomechanical response when exposed to antibiotics. This approach, which combines enrichment with three different modes of detection, can serve as a platform for the development of a portable, high-throughput device for use in the real-time detection of bacteria and their response to antibiotics.

The differential nanomechanical deflection induced is plotted against various ligand (AMP or mAb) concentrations in the sample (0.2 mg ml -1 , 0.4 mg ml -1 , 0.6 mg ml -1 , 0.8 mg ml -1 , 1 mg ml -1 , 1.2 mg ml -1 ). The solid line represents the Slogistic calibration fit and error bars represent standard deviations (n = 5). The study suggested density saturation at 0.8 mg ml -1 , which was used subsequently for surface functionalization.      . c and c' shows the nanomechanical motion after exposure to 5% glucose solution by 10 min. The results suggested that E. coli is been killed by ampicillin but it resisted kanamycin.
Removal of antibiotic to introduce 5% glucose to the bacteria enhanced the metabolism of the bacteria exposed to kanamycin to increase the nanomechanical fluctuation. While introducing ampicillin did not show any change of the cantilever fluctuation, indicating that bacteria is been killed.

Bimaterials microchannel cantilever (BMC):
U-shaped microfluidic channel was made-up on top of a silicon nitride microcantilever having a dimension of 32 μm width, 600 μm lengths and a height of 3 μm (Fig 1 in the main contents). assembling. The solution was also kept overnight in the microchannel to certify the functionalization (Fig S1). Prior experiments, the BMC were washed with MQ-water, ethanol and dried under stream of nitrogen gas.

Surface characterization and ligands density measurements:
In We examined the surface functionalization process by observing the changes in the nanomechanical IR readings, resonance frequency and nanodeflection of the cantilever compared to the background spectrum of silicon nitride. The 1 st SAM (ethanolamine layer) was detected by appearance of a distinctive absorption peak at ~1600 cm -1 , suggesting a primary amine absorption peak (Fig S2a). The peak however, nearly vanished subsequent loading of the 2 nd SAM (AMP or/mAb adlayer), signifying the success of the peptide conjugation to the ethanolamine layer and indicating a constant adlayer formation (Fig S2a). The adlayer of the AMP and/or mAb was also defined by the appearance of a strong absorption peak at 1533 cm -1 , which corresponds well to the amide II absorption band 6,7 . Furthermore, the differential analysis of amplitudes of nanomechanical deflections (Fig S2b) and the resonance frequency shifts (Fic S2c) showed the differences in the mass densities of the two adsorbed layers (1 st and 2 nd SAMs), and indicated the attainment of the sensor surface activation.
To ensure a high surface density of the immobilized ligands on the surface of the BMC chip, we performed preliminary tests, where diluted samples of antimicrobial peptide (Leucocin A) and mAb (0.2, 0.4, 0.6, 0.8, and 1 mg ml -1 ) were introduced into the BMC sensor and subjected to nanomechanical readings. The nanomechanical cantilever bending was computed and results were presented as differential deflection against concentration of the peptide in the samples (Fig   S3). Results suggested that a concentration of 0.8 mg mL -1 is an optimum to achieve maximum surface density of both AMP and mAb. Based on the results, we used the highest concentration of 0.8 mg mL -1 for immobilization of the ligands in the BMC microchannel.

BMC measurements; bacterial detection/sensitivity and selectivity:
BMC fabrication, instrumentation for data acquisition and software for data analysis are described in our previous report 8 . Here, by passing the fluid through the channel, the device is not only detecting the change in the total mass density of the cantilever, but it also identifies the molecular fingerprint of the present molecules in the sample by providing the nanomechanical IR spectra of the entire delivery. Initially, the measurements was performed to characterize the surface of the channel and to endorse the surface functionalization processes, as illustrated above.
In the bacterial detection experiments, artificially contaminated samples with L. monocytogenes at 10 3 cfu mL -1 (100 cells /100 µl were conceded through the BMC sensors, incubated and subjected to the nanomechanical readings. Three different readings, IR signatures, magnitude of nanomechanical cantilever deflections and resonance frequency shifts were measured. To estimate the device sensitivity and limit of detection, diluted samples with bacterial cells suspended in water at a range of 10 3 -10 6 cfu ml -1 were subjected to the sensors readings.
Various strains of bacteria were also exposed to the BMC sensors in order to determine the sensors selectivity. Each experiment was repeated at least five times under same conditions and at different time sets. Signals of the readings were plotted with respect to the wavenumber of IR light that generates nanomechanical IR spectra of the analytes inside the BMC. The IR spectral features are often overlapped. Thus, some data preprocessing were performed to analyze the data, such as binning, smoothing, and second derivative transformation analysis. Binning reduces the number of data points in a spectrum, smoothing eliminates noise by averaging neighboring data points. Second-derivative transformation separates overlapping absorption bands and removes baseline offsets.

Confocal microscopy:
A Stock solution of the CyQUANT dye (a green color probe) was made by following the manufacture protocol. Briefly, CyQUANT probe reagent (0.8 μL) was dissolved in HBSS buffer Inc., Guelph, Canada) through a magnification20×/1.4. All captured images were recorded using a Quorum digital camera and were analyzed using a velocity three-dimensional image analysis software.

Sensor re-usability:
Effective regeneration is a key for successful sensor assays. Therefore, a valuable investment would be establishing a suitable re-generation condition that allows a number of recycling with maintaining a sufficient activity and efficient performance. The BMC chip of both, AMPcoated and mAb-coated sensors, was simply re-generated via two steps. First, the sensors were or more with best regeneration achieved at pH 2.5 and 3.0. Accordingly, the regeneration can be performed confidently at pH 3.0 or 2.5. In contrast to the AMP sensor, the repeated usage of the mAb-coated BMC sensor after its regenerations at harsh environment (pH 2.0 and 1.5) had resulted in dramatic loss of its binding activity to more than 50%, which indicate that the immobilized mAb may undergo unfolding and denaturation (Fig S6). However, the sensor was sufficiently stable at milder conditions (pH 2.5 and 3.0) where a restored response reaches ~80%.
The results suggest that mAb-coated BMC can be regenerated at pH 2.5 -3.0 (or possibly higher), but it loses significant activity at lower pH. Steps for BMC sensor regeneration may need further optimizations using further reagents since milder conditions showed to preserve the sensor performance.

Bacteria drug resistance
All chemicals, culture media, reagents and antibiotics (ampicillin and kanamycin), with analytical grade, were obtained from Sigma-Aldrich. The ampicillin is a β-lactam containing antibiotics, penicillin alike, that act by inhibition of bacterial cell-wall synthesis through its interference with the peptidoglycan biosynthesis. The kanamycin; on the other hand, is an aminoglycoside subtype that kills bacteria by causing a membrane-damage and inhibiting DNA and RNA synthesis. E. coli DH5 is a well-known strain with its sensitivity to ampicillin and its resistance to kanamycin. The AMP (Leucocin A) was also applied in this study in order to verify applicability of the sensor to detect various drug-resistances and to explore response of the bacteria to antibiotics and antimicrobial peptides. Leucocin A is very unique class IIa bacteriocin peptide, with very strong activity against L. monocytogenes. The peptide acts by targeting specific membrane allocated receptor found on the surface membrane of bacterial cells that is known as mannose phosphotransferase 5 . Interestingly, some bacterial cells express this receptor and others do not; some cells have higher expression level of this receptor than others and some cells develop resistance gene to modify this targeted receptor. In order to identify bacterial resistance to this AMP, we have used two different strains of bacteria, E. coli DH5α and L.
monocytogenes. While L. monocytogenes is very sensitive to Leucocin A, DH5α-strain is unsusceptible to it 5 .

Bacteria preparations
As described previously, frozen stocks of bacteria, stored at -80°C in glycerol-supplemented media, were initially streaked in agar growth media and few bacterial colonies were collected afterward and incubated overnight at 37°C in 1 ml of broth media (LB for E. coli DH5 and TSBYE for L. monocytogenes 43256). After incubation, the bacterial culture was centrifuged; bacteria were precipitated and re-suspended in a phosphate buffered saline -pH 7.4.

BMC sensor preparation, calibration and detection of bacterial-drug resistance
Our home-made silicon nitride microchannel cantilevers coated from bottom with a thin film of gold layer (300 nm) and having dimensions of 32 μm width, 600 μm lengths and a microchannel height of 3 μm were embedded on it. Initially, the BMC was treated with (3-aminopropyl) triethoxysilane (APTES)a linker molecule that promotes adherence of bacterial cells to the cantilever surface. The linker provides loose attachment of the cells to the cantilever surface without affecting its metabolic and viable activities 9 . Specifically, the BMC was subjected to a 0.2% solution of APTES in MQ water for approximately 3-5 min and then rinsed with ultrapure water. The BMC sensor was introduced into the sensor chamber for analysis and calibration. The calibration was performed by injecting buffer solution free from bacteria, and taking its nanomechanical reading as a baseline for measuring the subsequent experiments. Bacteria cells either E. coli, in case of (ampicillin and kanamycin or Listeria monocytogenes, in case of the AMP Leucocin A, were diluted at 10 -5 and introduced into the BMC. The cells were left to incubate for ~10 min at room temperature and were then rinsed gently with PBS to remove any floating bacterial cells. Standard LB media or LB media containing antibiotics were injecting individually to the BMC sensor and data of the resonance frequency, cantilever deflection and IR signatures were measured simultaneously after each step. The measurements were performed at 5 min from the injection and after 30 min from the injection. The measurement was performed also after the antibiotics were removed and re-introduced LB media again. In addition, in order to enhance the metabolism of the bacteria, we introduced 5% glucose solution to bacteria after exposure to antibiotics and measured the sensor response 10 min later. The experiments are intended to elaborate the viability of the cells and its susceptibility to the treatments. Due to overlapping, the IR spectral preprocessing such as binning, smoothing, and second derivative transformation analysis were performed to analyze the data. Binning reduces the number of data points in a spectrum, smoothing eliminates noise by averaging neighboring data points. Secondderivative transformation separates overlapping absorption bands and removes baseline offsets. In addition, in order to differentiate intact from dead bacteria. IR Multivariate analysis, analogous to principal component analysis (PSA), was performed to differentiation life from dead bacteria. The analysis involved applying a stepwise variable selection to decrease the multidimensionality of the data into its most significant scores as described previously.

Bacteria viability assay (Microscopy)
The viability of bacterial cells attached to the inner walls of the cantilever was evaluated using a live/dead Bacterial Viability Kit (Life Technologies Inc., Burlington, ON, Canada). Live/dead bacterial viability stains includes CyQUANT green and propidium iodide (PI). The two dyes were prepared separately by dilution in MQ-water (1:10) and then mixed together in equivalent ratio (1:1 vol/vol). The mixed live/dead solution (~ 10 µL) was introduced into BMC contains bacteria had exposed either to ampicillin or kanamycin. The BMC left for 10 min in darkness at room temperature prior to analysis. The captured bacteria were examined using a Quorum WaveFX spinning disk confocal microscopy (Quorum Technologies Inc., Guelph, Canada) through a magnification20×/1.4. All captured images were recorded using a Quorum digital camera and were analyzed using a velocity three-dimensional image analysis software.

Statistical analysis:
All nanomechanical measurements were averaged and each experiment was performed at least five times. Data are presented as mean ± SD throughout the manuscript. The statistical difference was tested either using the unpaired t-test or the one way ANOVA test. In all statistical analysis the significance level (P value) was sat at as 0.05.