MDR and Antimicrobial-Susceptible Bacterial Strain Procurement:
Six MDR and six antimicrobial-susceptible strains were selected for this study. These 12 strains represented six different species from the ESKAPEE group – the acronym for a group of seven MDR species on World Health Organization Critical Priority I and II Pathogens lists: Staphylococcus aureus, Klebsiella pneumoniae, Acinetobacter baumannii, Pseudomonas aeruginosa, Enterobacter cloacae and Escherichia coli. The strain ID and susceptibility profile to six antimicrobials (one antimicrobial per species) are summarized in the table in Figure 1. All 12 strains were obtained from ATCC. These strains were cultured, harvested and stored according to ATCC guideline in our BSL-2 facility. We used the antimicrobial susceptibility profile and MIC50 values reported by ATCC, from the CLSI-M100 tables and literature for the strain without further verification. Bacterial samples emanated from the virgin ATCC samples were used in all the experiments (i.e. subcultured sample were not used).
Sterile, disposable, untreated cell culture flasks with 0.22 µm filter cap (Culture Area: 75 cm2, VWR®) with 60-75 mL broth were used to rehydrate the sample. Each stain was emanated for 18-24 hours with constant agitation in tryptic soy broth at 37 °C, except for Klebsiella pneumoniae 13883 which required nutrient broth. After incubation, the bacterial cells were harvest by transferring the flask content to multiple sterile, 50 mL conical tubes and centrifuge at 8000 rpm for 10 minutes to obtain cell pellet. The cell pellet is aliquoted to multiple 1.5 mL cryogenic vials with 20% skim milk as cryoprotectant. The vials were stored in the vapor phase of a liquid nitrogen Dewar.
Before centrifugation, 1 mL was sampled from the flask to determine the number of viable cells by quantitative cell culture. The procedure is following. The approximate cell concentration in the sample was estimated based on the cloudiness. Based on the estimation the 1 mL sample was serially diluted with tryptic soy broth or nutrient broth until reaching 10-10 to 10-12 (undiluted sample = 100) level. Minimally 5 most diluted sample in the serial dilution were plated. The averaged number from these five plates was reported as the number of viable cells in the sample, typically in the 1011-1013 cfu/mL range.
ATR-FTIR Measurement Setup:
We used a Nicolet® iS10 FTIR spectrometer (ThermoFisher®) with a horizontal ATR accessory (HATR, PIKE Technologies®) to obtain all ATR spectra in this report. The enclosure of the HATR accessory was removed to allow greater access to the ATR element. Both the HATR accessory and the FTIR spectrometer were under constant nitrogen purging to minimize absorption due to atmospheric water and CO2. Both HATR accessory and the ATR element are disinfected by 70% ethanol/water between each measurement.
A multi-bounce silicon 45° trapezoidal ATR element (80 x 10 x 2 mm L x W x H, PIKE Technologies®) was used in all ATR measurements. We selected Si because of its hardness and chemical resistance to 70% ethanol/water compared to other ATR element materials. An air background spectrum was collected before each ATR measurement. The bacterial ATR spectra were acquired and analyzed using OMNIC® Software (ThermoFisher®) with the following setting:
Spectral region recorded: 4000 – 1500 cm-1. Spectral Resolution: 4 cm-1. Atmospheric suppression: On. # of scans: 256.
Bacterial Sample Preparation:
An 0.25 mL aliquot sample was removed from the liquid nitrogen storage and thawed at 37 °C. The content was transferred to two 0.5 mL microcentrifuge tubes, one would be used for acquiring spectra before antimicrobial incubation and the other one would be for after antimicrobial incubation. Cell pellet was collected by centrifuging at 10,000g for 5 minutes. The skim milk supernatant was removed and 80 – 100 μL 5% dextrose/saline (freshly prepared and double-filtered by 0.22 μm filter) were used to wash the cell pellet 2-3 times.
To obtain spectra before antimicrobial incubation, immediately after the last wash the cell pellet was dispersed in 4 μL 5% dextrose/saline. All liquid contents were transferred to the Si ATR crystal surface using 1 μL pipette tips in 7-8 evenly distributed droplets along the length of the crystal. The added 4 μL liquid helped to lift the cell pellet off the microcentrifuge tube wall. The total sample volume (4 μL + cell pellet) should be 8 – 10 μL. An identical Si ATR element was placed over the sample to form a “sandwich”. The sample was smeared into a thin film of liquid with the top crystal. The smeared sample would dry up in air in 1 -2 minutes, forming a thin, uniform layer of dried bacterial cells. The ATR element was inserted into the HATR accessory and a set of five ATR spectra was acquired.
To obtain spectra after antimicrobial incubation, antimicrobial stock solution at 2x concentration was added to the bacterial sample by mixing equal volume of the stock solution with the bacteria sample. The antimicrobial stock solution was prepared within 30 minutes before experiment in 5% dextrose saline and stored at 2-8 °C until use. Then the mixture was incubated at 35 °C for 4 hours without agitation. After incubation, the cell pellet was collected, washed. ATR spectra were then acquired following the procedure described before.
ATR Spectra Preprocessing, Building SVMDA models and Classification In Unknown Samples:
Each ATR spectrum was subjected to spectral preprocessing before compiling into a spectral reference library. Spectral preprocessing can help to improve classification model sensitivity and specificity by enhancing inherent differences between classes while reducing variation within the sample class. The library was used to train classification models.
The ATR range cutoff at low wavenumber for the Si HATR element is 1500 cm-1 in Mid-IR. Thus the spectral region between1500 – 1200 cm-1 was recorded as shown Figure 1, but not considered in the spectral analysis. Each ATR spectrum was baselined by an “Automatic Baseline Correct” algorithm in the OMNIC software so that spectra from different strains and from before vs. after antimicrobial incubation could be quantitatively compared.
To prepare for training classification model to distinguish the six different bacterial species, the spectra were min-max normalized to ensure the model would “focus” on finding species-specific characteristics instead of the absolute peak intensity differences within the dataset.
We utilized the PLS_Toolbox® from Eigenvector Research, Inc (Mason, MA), a chemometric toolbox on MATLAB platform, to build our classification model. We first built and optimized a partial least square discriminant analysis (PLSDA) pre-model. PLSDA is a “supervised” version of principal component analysis, i.e. class ID are requested from the user and utilized to decide class boundary. We examined the pre-model to ensure genuine differences existed between classes. Then a sum vector machine discriminant analysis (SVMDA) classification model was built based on the PLSDA results. SVMDA is a model optimization technique; the class boundary SVMDA is non-linear to allow better classification for data points that are too close to the boundary in the PLSDA pre-model. The curviness of boundary is governed by the data points closest to the boundary (i.e. “support vector”). The SVMDA model parameters were fine-tuned and compiled, and the analytical sensitivity and specificity were calculated.
To test the performance of the SVMDA model, ATR Spectra from 4 different unknown samples were acquired and baseline/normalized by methods described above. The model performance was assessed by determining the % of data from these unknow sample that were correctly classified the model.