The whole SERS-AST process could be completed within 4 hours. By directly assaying positive blood cultures without additional subcultures, the time needed for a SERS-AST is significantly shorter than that for VITEK 2, which took 9 ± 1.3 hours to complete an AST directly from blood cultures in the study of Hogan et al. (43) and is comparable to that for the EUCAST rapid disc diffusion method, which takes 4–8 hours (44).
In SERS-AST, bacterial response to antibiotics is determined by changes in SERS peak height, while conventional ASTs are mainly based on changes in bacterial number, which may be determined by OD600 readings. In this study, we found that approximately 30% of the results determined by OD600 values of antibiotic-treated bacterial cultures did not agree with those of SERS-AST or VITEK 2. Most of the disagreements were of Gram-negative bacteria treated with ꞵ-lactam antibiotics, such as E. cloacae-CTX tests (Fig. 4). It has been shown that the initial response of E. cloacae to treatment with a ꞵ-lactam antibiotic is cell elongation, instead of decreased number of cells (45). Therefore, the OD600 value of the treated culture may not change within 2 hours of treatment. Such morphological change is postulated to be a repair process for survival (46). In contrast, SERS-AST measures the amounts of secreted purines and their derivatives in response to antibiotic treatment. Such measurements are less affected by changes in cell morphology as in the case of our E. cloacae-CTX tests.
Several modifications of the SERS-AST protocol were done to obtain reliable results from certain bacteria, e.g., A. baumannii, which failed to generate recognizable SERS signals even when the incubation time of antibiotic treatments was extended to 3 hours. This problem may be due to the low permeability (approximately 1/100 that of E. coli) of its outer membrane to small molecules, thus hindering the secretion of xanthine (47, 48). To determine the optimal condition for A. baumannii SERS-AST, the antibiotic-treated cell suspension in water was incubated in a shaking water bath at 25, 37, or 50°C for 30, 60, or 90 minutes. As the result showed that an additional 30-minute incubation in a shaking water bath at 37°C rendered a 2.2-fold increase in SERS signal, this condition was used for all subsequent SERS-AST tests for A. baumannii. We postulate that this signal improvement is mainly due to the impact of placing bacteria in a nutrient-deficient environment as this starvation process has been shown to stimulate more secretion of purines and their derivatives (49–51).
In this study, 14 SERS-AST tests gave results that did not agree with those of VITEK 2, including 3 each of E. faecalis-LVX and K. pneumoniae-CAZ tests, and one each of S. aureus-OXA, S. aureus-LVX, S. epidermidis-LVX, E. faecium-LVX, E. cloacae-CTX, E. cloacae-CAZ, K. pneumoniae-CTX, and K. pneumoniae-LVX tests (Table 1, 2). Seven (50%) of these 14 cases were LVX tests with mostly Gram-positive bacteria (Table 1). Among the seven antibiotics tested, LVX is the only one not acting on cell wall synthesis. It is a quinolone antibiotic that inhibits gyrase and topoisomerase IV leading to impaired DNA replication, repair, and recombination (52). It has been shown that gyrase is the primary target of quinolones in Gram-negative bacteria, while topoisomerase IV is the main target of quinolones in Gram-positive bacteria (52). In DNA replication, inhibition occurs within minutes when the antibiotic acts on gyrase (53) but takes place later if it targets topoisomerase IV (54). However, bacterial responses to quinolones have been shown to vary on a species-by-species and drug-by-drug basis (55). It is likely that a longer time is required for LVX to inhibit the growth of slow-growing Gram-positive bacteria, such as E. faecalis. In an attempt to optimize the condition for E. faecium-LVX tests, we obtained more clear-cut SERS-AST results when the antibiotic treatment time was extended to 3 or 4 hours. Based on these results, we recommend that the antibiotic treatment time be extended to 3 hours for all LVX tests in future studies.
Five (36%) of the aforementioned 14 cases were of K. pneumoniae tested with CAZ, CTX, or LVX (Table 2). K. pneumoniae is known to produce a pronounced polysaccharide capsule covering the entire bacterial surface resulting in a mucoid phenotype (56) with reduced ability to secrete purines and their derivatives, thus yielding weak SERS signals. K. pneumoniae is also known to produce the CTX M β-lactamase, which is an extended-spectrum β-lactamase (ESBL) and can degrade third-generation cephalosporin antibiotics at different rates (57). Compared to CTX, which is the preferred target of CTX M β-lactamase, CAZ is relatively resistant to that enzyme and may require more than 2 hours to be degraded. Therefore, some CAZ-resistant bacteria may be determined by laboratory testing as susceptible, which is inconsistent with clinical manifestations. The Advanced Expert System (AES) of VITEK 2 can identify bacteria with ESBL by special software and modify the primary laboratory results accordingly (58). In this study, there were five AES-revised K. pneumoniae-CAZ results. Three of them that were determined as susceptible by SERS-AST after the 2-hour antibiotic treatment were interpreted by AES as resistant.
The SERS technology has also been used by Tien et al. (59) to detect antibiotic-resistant bacteria in urine samples with the aid of principal component analysis (PCA). By using magnetic separation and SERS technology with AgNP colloid, Li et al. (60) accurately identified antibiotic-resistant strains of S. aureus, A. baumannii, and P. aeruginosa from 77 clinical blood samples. In the study of Novelli-Rousseau et al. (61), Raman spectrometry coupled with PCA and support vector machine (SVM) algorism was used to determine the MIC of gentamicin, ciprofloxacin, and amoxicillin against E. coli strains. Ho et al. (62) have generated an extensive dataset of bacterial Raman spectra and employed deep learning with convolutional neural network (CNN) algorism to train the computer to identify 30 common bacterial pathogens. With this approach, they achieved 99.0 ± 1.9% accuracy in the identification of 25 clinical isolates from 50 patients and 89.1 ± 0.1% accuracy in the differentiation between methicillin-resistant and methicillin-susceptible S. aureus. Our results are consistent with these observations.
This study has several strengths. With optimized specimen processing protocols and effective SERS device, reproducible bacterial SERS spectra were acquired for AST. This study also successfully developed methods, such as starvation, to perform SERS-AST on bacteria (i.e., A. baumannii) that were previously not assayable. The finding that extending the LVX treatment time to 3 hours, especially for Gram-positive bacteria such as E. faecalis that grows slowly, can generate satisfactory SERS-AST results is very significant.
To move the SERS-AST forward for clinical applications, further improvements are needed. As different bacteria emit different SERS signals, methods for discriminating mixed SERS spectra from samples of patients with polymicrobial infections remain to be developed. The impact of patient treatment on SERS-AST also remains to be investigated. Although P. aeruginosa is a major causative organism of sepsis, we have not been able to perform SERS-AST on it because of interference by its fluorescent pigments. As SERS-AST is based on changes in bacterial metabolism due to antibiotic treatment, it is possible to modulate the metabolic activity of bacteria by altering their growth environments with substances such as culture media and cations (e.g., Mg2+, Ca2+, and Na+) (63) or with pH adjustment (46). A limitation in the performance of SERS-AST is that the instrument is homemade and is currently not commercially available. Since the SERS-AST instrument that we have used is a prototype, there is room for improvements to increase its sensitivity and specificity.