*3.1. Tandem-SERS Methods*

Tandem-SERS refers to conjugating the separation element to the SERS system that can achieve separation and detection simultaneously [90]. The aforementioned functional SERS tag with a separation element (e.g., antibody, aptamer, molecularly-imprinted polymer) is a classical tandem-SERS system. Due to the size of bacterial cells, a sandwich tandem-SERS structure is always developed [15,68] and the detailed illustration is shown in Figure 2b. Antibody is widely used as the recognition element due to its specificity to bacteria via a covalently-bound effect. An antibody conjugated with different SERS nanoprobes such as Ag@silica core-shell nanoparticles [71], popcorn-shaped Au nanoparticles [72], and single walled carbon nanotubes-Au nanoparticles [91] was used to detect normal *Salmonella* or multi-drug-resistant *Salmonella*. High correlation coefficients and LOD of 4 and 5 CFU mL−<sup>1</sup> were obtained using an antibody-SERS employing AuNPs via a sandwich immunoassay for detecting and enumerating *E. coli* (Figure 4b) [89]. The results of testing bacteria in lake and tap water samples were highly consistent with that of the classical plating assay.

Aptamer is another element that can be used and conjugated in tandem-SERS for the recognition, separation, and enrichment of specific bacterial pathogens. Aptamer-based SERS assay was able to

monitor photothermal activity response of MRSA and multi-drug-resistant *Salmonella* DT104 through the change of Raman signal intensity of R6G [32]. Zhang and coauthors reported a simultaneous detection of *S*. Typhimurium and *S. aureus* using Au NPs-aptamer based SERS biosensor (Figure 4c). A high sensitivity with LOD of 35 and 15 CFU/mL for *S. aureus* and *S.* Typhimurium was achieved, respectively [76]. Another format of tandem-SERS was to include SERS sensing in a microfluidic device. A complicated design of the microfluidic device can realize the function of separation of bacterial cells from the sample matrix mainly [49]. Dielectrophoresis is an effective method for concentrating and trapping various types of nanoscale/microscale particles in a microfluidic device, including microorganisms [92]. It is also feasible to conjugate the aforementioned separation elements, such as aptamer, onto the microchannels to form a more comprehensive and effective tandem-SERS platform for simultaneous separation and detection [67]. Lin and co-authors developed a fast single-step SERS detection of *E. coli* O157:H7 at single cell level with speciation capability to sub-species. This was achieved by a multiplexing dual recognition SERS platform that combined specific antibody conjugated SERS tags with a microfluidic dielectrophoresis (Figure 5) [93].

**Figure 5.** The integration of SERS nanoprobes and a microfluidic dielectrophoresis (DEP) device for rapid detection of single bacterium. (**a**) Schematic presentation of using antibody-conjugated nanoaggregate-embedded beads (NAEBs) as SERS nanoprobes for specific detection of bacteria. (**b**) Photograph of the microfluidic DEP device and close-up view of central capturing area with four the quadrupole electrodes. (**c**) The distribution of electric field of four microelectrodes in the microchannel. (**d**) Schematic illustration of the DEP-SERS configuration. Reproduced with permission [93]. Copyright WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim, 2014.

### *3.2. Tandem-SERS Integrated with Multiple Capabilities*

Another major application advantage of such a tandem-SERS platform is to enrich the bacterial cells and subsequently improve the detection sensitivity. Although SERS can theoretically detect a single molecule/cell, its real world application can only detect ~10<sup>3</sup> CFU/mL of bacteria, mainly due to the interference from the sample matrix components [94]. Therefore, a relatively large amount of samples therefore is required for the production of a meaningful SERS signal readout. In a recent study reported by Zhang and others, the SERS-active substrate composed of gold nanoparticles was integrated into the microfluidic device for rapid concentration and detection of *S. aureus* in liquid

samples [95]. The SERS signal intensity of *S. aureus* after concentration in this device was over 100-fold compared to the signal obtained from the raw sample, leading to a LOD of 2 × 102–2 × 10<sup>4</sup> CFU/mL. Hou and colleagues demonstrated a microfluidic system based on a discharge driven vortex technique to concentrate a bacterial suspension of *E. coli* F-amp and *Bacillus subtilis* for SERS detection. The combination of SERS and microfluidic with immunoassay techniques was able to selectively capture the targeted bacterial cells [96]. A SERS-based sandwich immunoassay employing antibody-coated magnetic nanoparticles for *E. coli* enumeration was also reported [97]. The authors accomplished a LOD of 8 CFU/mL by combining bacterial separation with SERS detection using specific SERS labels.

Combination of SERS platform and a filter (e.g., polymer fiber) has been recently used for the identification and detection of bacteria from clinical and environmental samples. For instance, Lin and others demonstrated a filter-like SERS substrate prepared with AuNPs embedded in mesoporous silica for the detection of *Staphylococcus aureus* from the aqueous samples [98]. The targeted cells could be concentrated on the filter-like substrates within a few seconds. Strong SERS signals with good bacterial discrimination were obtained without any need for pre-labeling, and the reproducibility was also significantly improved. More recently, Kami ´nska and colleagues presented a new label-free tandem-SERS platform for rapid detection of *Neisseria meningitidis* [99]. This bacterium is a Gram-negative diplococcus and one of the three major bacteria that cause acute bacterial meningitis. The applied SERS substrate was based on Si/ZnO layers and electrospun polymer mats covered with a thin layer of sputtered gold. A wide range of pore sizes makes the polymer mat an excellent material to filter bacteria from fluids and then immobilize them onto the SERS nanostructures for the collection of Raman signals, enabling the detection of single bacterial cells of *N. meningitidis* present in cerebrospinal fluid samples. A similar approach was developed to detect bacteria from blood plasma [100]. Covering the forcespun polymer mat with Au/Ag alloy turns it into a SERS-active platform, which can be used as a filter to separate the microorganisms from fluids and immobilize them on the surface of the mat during the measurement. *S. aureus*, *Pseudomonas aeruginosa*, and *S.* Typhimurium were successfully detected and identified from blood plasma using the developed platforms. These SERS-active nanostructures based on polymer mats provide the possibility for simultaneous filtration, immobilization, and enhancement of Raman signals in a few seconds, demonstrating a simple and low-cost method to analyze bacterial suspensions in biological fluids with SERS.

In addition, the tandem-SERS platform can achieve multiplex detection of bacteria by integrating several different elements into a single system. By using a systematic evolution of ligands by exponential enrichment (SELEX), different aptamers can be synthesized and each one targets one species of bacteria. By conjugating the aptamers onto a substrate, such as the microchannel in a microfluidic device, the mixture of bacterial cocktails can be individually captured by each aptamer that eventually achieve multiplex detection in a simultaneous manner. For example, *S.* Typhimurium and *S. aureus* were simultaneously identified using different aptamers in a sandwich-type tandem-SERS detection within 3 h [76]. Sandeep and co-workers proposed another simple and robust cross-platform approach using different nanoparticles functionalized with specific capturing ligands and Raman reporter molecules. This multiplex detection platform was applied for simultaneous detection of three different pathogens and offered an LOD ranging between 10<sup>2</sup> and 10<sup>3</sup> CFU/mL with a total detection time less than 45 min [64].

### *3.3. "Two-Step" and "One-Step" SERS*

In comparison to the aforementioned concepts of "direct sensing" and "indirect sensing", "two-step sensing" and "one-step sensing" is another pair of the terminologies that are related to tandem-SERS platform. Once the separation and SERS detection are separate, it refers to "two-step" sensing. An intriguing "two-step" SERS approach based on a sandwich assay for the separation and detection of multiple pathogens in food samples was demonstrated by Wang and co-authors [70]. Figure 6a depicted the key steps of the process. The targeted pathogens in a food matrix were

first captured and separated using silica-coated magnetic nanoparticles functionalized with the corresponding antibodies. Then, AuNPs integrated with a Raman reporter and surface-modified antibodies specific to the pathogen were used to complete the SERS detection. This platform achieved a LOD of 10<sup>3</sup> CFU/mL for multiplex detection of *S*. Typhimurium and *S. aureus* in spinach wash water and peanut butter. "One step" sensing indicates that the separation and detection can occur simultaneously. Once "one-step" sensing is applied, a critical parameter is to ensure that the distance of the separation element is within 10 nm from the SERS-active substrate [90]; otherwise, the SERS effect will be tremendously reduced [101]. Naja and coauthors presented a "one-step" sensing of bacteria using silver nanoparticles functionalized with antibodies (Figure 6b). When the model bacteria attached to the corresponding antibodies absorbed on the protein-A-modified silver nanoparticles, the distance between the bacterium and the nanoparticle surface was 8 nm, thus the SERS signal of the bacterial cell wall would be generated and detected [102]. Further, "one-step" tandem-SERS sensing requires a relatively more complete clean-up of the sample matrices than that of the "two-step" tandem-SERS sensing method [90].

**Figure 6.** Representative "two-step" (**a**) and "one-step" (**b**) tandem-SERS sensing methods. (**a**) Tandem-SERS platform composed of the magnetic-based separation and SERS detection for multiple pathogens in food matrices. Pathogens were first captured with silica-coated magnetic probes, and then pathogen specific SERS probes (gold nanoparticles integrated with a Raman reporter and corresponding antibodies) were deployed to complete the following detection. (**b**) Schematic diagram for SERS-based detection of *E. coli* using silver nanoparticles conjugated with antibodies. Reproduced with permission [70]. Copyright Springer-Verlag, 2010. Reproduced with permission [102]. Copyright Royal Society of Chemistry, 2007.

### **4. Elucidating Antibiotic Resistant Mechanism of Bacteria Using SERS and Chemometrics**

Besides the detection of antibiotic resistant bacteria either in a simple matrix or a complicated environmental, agri-food or clinical sample matrix, another major research direction of using SERS is to study the working mode and mechanism of antibiotics to inactivate bacteria. Bacterial cells can develop various strategies to resist to the antibiotic treatment as the pinnacle of evolution. Although new antibiotic resistance has been continuously emerging and spreading globally, bacteria use is one of two leading genetic strategies to deal with antibiotic treatment, namely mutation in genes associated with the action of antibiotic compounds and the acquisition of external DNA for the resistance determinants through horizontal gene transfer [103]. These genetic variations will lead to the change in biochemical composition of the bacterial cells. For example, three different biochemical

routes can arise, fluoroquinolone resistance, including over-expression of efflux pumps to extrude the antibiotics from the bacterial cells, mutations in genes encoding DNA gyrase and topoisomerase, and generating specific proteins to protect the targeted site of fluoroquinolone [104,105].

### *4.1. Characterization of Antibiotic Resistance of Bacteria Using SERS*

As a three-dimensional complex surrounding the bacterial cells, peptidoglycan is the major component of the bacterial cell wall [106]. Since a relatively large amount of antibiotics is designed to target the bacterial cell wall, the biochemical compositions of the bacterial cell wall are expected to change along with the treatment of these antibiotics. Because SERS can record the macromolecular fingerprints of the bacterial cell membrane and cell wall, it can be applied to determine the effectiveness of antibiotic treatment as well as the antibiotic resistance patterns of the bacterial cells [15]. Although conventional Raman spectroscopy has been widely applied to profile the phenotypic response of bacteria to the antibiotic treatment, it requires a high concentration of bacterial culture for the collection of Raman signal [107]. Therefore, a relatively long time for bacterial cultivation and enrichment is necessary. By applying SERS for characterization, the antibiotic-resistant pattern of a single bacterial cell can be achieved. In addition, it will be critical to study the variations in responses among individual cells to the antibiotic treatment.

Antibiotic susceptibility testing (AST) is used to evaluate the effectiveness of antibiotic treatment against the pathogen infections. SERS-based AST could reduce the time by avoiding the need for overnight culture in MIC determination through the conventional AST methods. Liu and coauthors used an SERS-active substrate made of AgNPs imbedded in AAO to determine the antibiotic sensitivity of *E. coli* and *S. aureus* at the single-bacterium level [108]. Antibiotic-sensitive bacteria could be differentiated from antibiotic-resistant ones within 1 h after antibiotic exposure by monitoring the characteristic changes in SERS spectral profile. This approach demonstrated that SERS has the potential for direct detection and characterization of antibiotic resistance in real world samples instead of pure bacterial culture. Another study employed SERS-active AuNPs to study the antibiotic susceptibility of 12 urinary tract infection-causing bacteria [109]. Strain-specific identification was achieved with analytical sensitivity >95% and specificity >99%. The time for positive identification and AST was reduced to less than one hour.

In addition, SERS-active substrate can be employed as a means to establish MICs for various bacteria. Liu and colleagues demonstrated that SERS could monitor the reduction of specific bacterial biomarkers along with the treatment of antibiotics within two hours [110]. Clinical isolates of MRSA were exposed to vancomycin, while *E. coli*, *A. baumannii*, and *K. pneumoniae* were exposed to imipenem at the incremental concentrations. The isolates were determined as susceptible, intermediate, and resistant based on the change of the characteristic bands in SERS signals at a very early stage of antibiotic treatment, and the SERS MIC results were in excellent agreemen<sup>t</sup> with the standardized plate dilution methods that took upward of 24 h to complete. In a recent study, Cui and coauthors developed a homogeneous vacuum filtration-based method to improve SERS signal reproducibility and illustrated that the existence of heavy metal arsenic could increase the MIC of bacteria to the treatment of tetracycline. The authors claimed that SERS has the potential for culture-free characterization of resistome in a real microbiota system at the single cell sensitivity level [111].

Furthermore, monitoring the characteristic bacteria cell wall bands in the SERS spectra allowed for a further understanding of the antibiotic degradation mechanisms. The antibiotic response of *Lactococcus latis* was investigated using SERS-active AuNPs [112]. Antibiotic-induced spectral changes from ampicillin and ciprofloxacin were observed at 60 min after exposure to both antibiotics. However, ciprofloxacin induced only minor changes while ampicillin induced large SERS spectral changes. This was possibly because the inactivation mechanism of ciprofloxacin is to disrupt DNA synthesis, therefore the cell wall integrity was maintained for extended time periods and the cell wall signatures remained stable in the SERS spectra. While ampicillin interrupts the cell wall synthesis, which was directly detected by the SERS-active AuNPs. In another study, the SERS signals of *E. coli* were tracked upon antibiotic exposure to chloramphenicol, trimethoprim, polymyxin B, ampicillin, and formalin [113]. No spectral changes were observed after exposure to formalin although in vitro tests, which confirmed the cells were not viable. The authors noted that it was most likely due to the mechanism of formalin to crosslink membrane proteins but not degrade the cell wall. Similar results were observed with chloramphenicol and trimethoprim, which inactivate bacteria by inhibiting protein and DNA synthesis, respectively. The SERS signals remained unchanged after 2h exposure, which is possibly attributable to the sustained cell wall integrity. In contrast, SERS spectra changed within 5 min after antibiotic exposure to polymyxin B and ampicillin. They both aggressively degraded the bacterial cell wall, which released the SERS-active AgNPs and drastically reduced the SERS intensities. The technique could be used to further understand the fundamental mechanisms of microbial inactivation.

### *4.2. Chemometrics Used with SERS*

Chemometric statistical analyses are usually required to decipher Raman spectral patterns so that minor variations in the spectral features of different biological samples can be distinguished. Multidimensional information of SERS spectra can be reduced into a few independent latent variables (called principal components) that account for the most variability of the original dataset by multivariate statistical analyses [114]. These principal components can then be used to segregate and quantify analytes based upon specific calibration models [115]. Chemometric methods include both unsupervised and supervised algorithms [116]. Among the spectroscopic-based pattern recognition methods, unsupervised principal component analysis (PCA) and hierarchical cluster analysis (HCA) are commonly used to provide either cluster plots or dendrogram structures for segregation and discrimination based upon the minor differences in Raman spectra [117]. Supervised chemometric models are generally used with some known answer from existing knowledge of the sample. Discriminant function analysis (DFA), partial least squares regression (PLSR), and soft independent modeling of class analog (SIMCA) are some of the most widely used models for the interpretation of SERS results [114]. For instance, a discriminant analysis is divided into two steps: to build a model using Raman spectra of bacterial cultures exposed to antibiotics of known class assignments, and to classify a new Raman spectrum of an antibiotic-exposed culture based on the distance to the multivariate mean of the closest class [118].

Different bacterial species or strains can be segregated into distinct groups based upon different biochemical compositions reflected by the major latent variables. For example, *E. coli*, *S. epidermidis* and four *Salmonella* strains exhibiting antibiotic resistance to the common therapeutics were detected and differentiated using SERS coupled with PCA [69]. In another work, SERS spectra of *P. mirabillis* and *Enterococcus* were quite similar despite having different cell wall structures. DFA was employed to analyze the subtle differences of SERS spectra from 6 strains of clinical urinary tract infection isolates for identification at genus-level [35]. Chemometric analysis play an important role in the determination of antibiotic resistance by SERS-based methods. Spectral differentiation of antibiotic resistant and sensitive strains can be demonstrated by chemometric models. For instance, Tien and others applied PCA for Raman spectra from MSSA and MRSA. MRSA cluster and MSSA cluster were segregated that can be used to differentiate MRSA from MSSA [119]. A SERS-based PLSR model was used to accurately determine the concentration of an MRSA strain in a mixture containing MSSA [54]. One recent study applied a three level chemometric model based on PLSR in combination with linear discriminant analysis (LDA) to extract those molecular changes and distinguish vancomycin-resistant and sensitive *Enterococci*. In addition, antibiotic-induced spectral changes from ampicillin and ciprofloxacin were monitored and statistically analyzed using PCA to understand the different working mechanisms of these antibiotics [112].

### **5. Conclusions and Future Direction**

Raman spectroscopy and SERS have been validated for their potential in bacterial detection, typing, and characterization for almost three decades. Compared to the application of MALDI-TOF

mass spectrometry for bacterial characterization, the use of Raman spectroscopy and SERS by industry is still in its infancy. This is mainly due to the relatively poor spectral reproducibility by using different types of the manufactured SERS substrates. As indicated in numerous review papers related to SERS bacterial study, to develop a stable SERS-active substrate for consistent and global use in a commercial manner is highly critical to promote this versatile technology to environmental, agri-food and clinical applications. Another major challenge is the relatively high cost of the confocal micro-Raman spectroscopic system. Although very little cost is required for purchasing consumables and instrumental maintenance compared to MALDI-TOF mass spectrometry, industries are still reluctant to purchase a bench-top Raman spectroscopic system. Therefore, a portable/handheld Raman instrument might be more affordable even though the resolution of the collected SERS spectra is relatively low. A more user-friendly software is also required for the convenient spectral interpretation as well as chemometric analyses. Several vendors have developed their own software for spectral processing and chemometrics, but a major doubt is how reliable such software for spectral analysis can be. By only clicking each "black-box" in the software, the researchers may not fully understand how each algorithm will affect the performance of the chemometric models. A standardized protocol for SERS spectral analyses and chemometric analyses therefore is critical to achieve inter-laboratory validation of the results for bacterial characterization, such as the characterization of bacterial antibiotic resistance.

Albeit these aforementioned challenges and potential limitations, SERS is definitely a very promising candidate for the determination of bacterial antibiotic resistance in a high-throughput, multiplex, and ultrafast manner. We sugges<sup>t</sup> that industries use SERS for the detection and characterization of bacterial antibiotic resistance as an innovative fast screening alternative that can couple with the conventional methods for a further confirmation. Along with the further advancement in optical instrumentation and machine learning, the new version of the Raman spectroscopic system will be more user-friendly and cost-effective. We also envision that SERS can be used to further illustrate the modes of antibiotic and antimicrobial resistance of bacteria. This may contribute to the design of more effective antimicrobial treatment. Although SERS itself can be regarded as the core technology for an individual project, such as the detection of antibiotic resistance bacteria in a clinical specimen, we also believe it can be integrated as part of a more complicated study to drive very fundamental scientific research questions related to bacterial antibiotic resistance.

**Author Contributions:** K.W. is the leading author to draft this review paper along with additional contributions from other coauthors. K.W. and X.L. developed the structure of this review paper. S.L. and M.P. provided critical feedback and helped revise the paper. S.W. contributed to the final version of the manuscript.

**Funding:** This research was funded by Discovery Grant from the National Sciences and Engineering Research Council of Canada (NSERC RGPIN-2014-05487).

**Acknowledgments:** Financial support to X.L. in the form of a Discovery Grant from the National Sciences and Engineering Research Council of Canada (NSERC RGPIN-2014-05487) is gratefully acknowledged.

**Conflicts of Interest:** The authors declare no conflict of interest.
