Characterization of bacteria causing acute otitis media using Raman microspectroscopy

Abstract

Otitis media (OM) is a prevalent disease that is the most frequent cause of physician visits and prescription of antibiotics for children. Current methods to diagnose OM and differentiate between the two main types of OM, acute otitis media (AOM) and otitis media with effusion (OME), rely on interpreting symptoms that may overlap between them. Since AOM requires antibiotic treatment and OME does not, there is a clinical need to distinguish between AOM and OME to determine whether antibiotic treatment is necessary and guide future prescriptions. We used an optical spectroscopy technique, Raman spectroscopy (RS), to identify and characterize the biochemical features of the three main pathogens that cause AOM in vitro. A Renishaw inVia confocal Raman microscope at 785 nm was used to spectrally investigate the Raman signatures of Haemophilus influenzae, Moraxella catarrhalis, and Streptococcus pneumoniae. Biochemical features or biomarkers important for classification of each bacterial species were identified and yielded a 97% accuracy of discrimination. To test the effectiveness of Raman-based bacterial classification in a clinical sample, human middle ear effusion (MEE) from patients affected by recurrent AOM was collected, cultured, and measured using RS. The probability of bacterial involvement from each of the three main bacteria that cause AOM was determined from the clinical MEE samples. These results suggest the potential of utilizing RS to aid in accurately diagnosing AOM and providing physicians with bacterial identification to guide treatment.

Introduction

Otitis media (OM), an inflammatory disease of the middle ear, is the leading cause of acute physician visits and prescription of antibiotics for children¹. Worldwide, there are over 700 million cases of acute otitis media (AOM) every year with 51% of these cases occurring in children less than five years of age². The impact of AOM can extend beyond an infection, possibly leading to complications such as mastoiditis, chronic suppurative otitis media (CSOM), and hearing impairment, which may be severely debilitating for child development³. Otitis media with effusion (OME) is one of the two types of OM and is described as asymptomatic inflammation of the middle ear with a build-up of fluid in the middle ear space. Contrary to OME, AOM presents with a rapid onset of signs and symptoms, such as fever, associated with acute infection within the middle ear. AOM is commonly caused by an active bacterial infection in the upper respiratory tract that is refluxed into the middle ear space. Antibiotic therapy is prescribed to manage AOM in children six months and older presenting severe signs and symptoms, while antibiotic treatment is not recommended for children with OME. Currently, clinical diagnosis of AOM is based on visual evaluation of the tympanic membrane (TM) and symptoms caused by the infection. Clinical guidelines issued by the American Academy of Family Physicians (AAFP) and American Academy of Pediatrics (AAP) are based on visual evidence such as bulging of the TM with recent onset of ear pain or erythema to diagnose AOM⁴. These symptoms are further assessed using a pneumatic otoscope, the current standard tool for diagnosing OM. Additionally, diagnosis relies on the assessment of the contour, color, translucency, and mobility of the TM⁵. Pneumatic otoscopy allows the physicians to view the TM and apply pressure to observe its mobility. Pneumatic otoscopy is 70% - 90% sensitive and specific for determining accumulation of middle ear effusion (MEE) in the middle ear, which usually develops post-infection^5–9. Basic otoscopy, which relies only on subtle visual changes of the tympanic membrane, has a sensitivity and specificity of 60% - 70%^7,8. Although pneumatic otoscopy improves visualization of symptomatic changes in the TM, findings are not able to identify or correlate with bacteria causing an infection. Other techniques, though less commonly implemented in routine clinical care include: tympanometry, which measures TM compliance using sound; acoustic reflectometry, which seeks to identify the presence of fluid behind the TM by emitting and detecting the reflected sound; and tympanocentesis, which is an invasive technique used to extract MEE through the tympanic membrane to be cultured for identification of bacteria causing an infection. Pneumatic otoscopy and tympanometry are limited in their performance and do not detect or identify bacteria in ear effusion. Tympanocentesis then, is currently the “gold standard” for identifying bacteria causing an ear infection. In addition to the invasive nature of the procedure, not all fluid may be collected and more importantly not all MEE is easily cultured, delaying identification of bacteria causing an infection. In fact, tympanocentesis is rarely practiced and only performed when antibiotic treatment is repeatedly unsuccessful, which can still result in not identifying the causative microorganisms. This gap of diagnostic information may cause physicians to over-prescribe antibiotics for cases of OME, which are rarely caused by a bacterial infection, or prescription of antibiotics to pathogens that have developed resistance to specific classes of antibiotics in acute infections.

Optical spectroscopy has in recent years received significant attention for disease diagnosis. Optical methods that have been explored for detecting OM include diffuse reflectance spectroscopy, fluorescence spectroscopy, and optical coherence tomography (OCT). Diffuse reflectance spectroscopy utilizing a coupled fiber-optic bundle with an otoscope has been used to distinguish the color of the tympanic membrane for diagnosis of AOM in 15 normal and 15 AOM patients¹⁰. While this group was able to distinguish between OM with mucous versus serous effusion, the performance of the technique to differentiate between AOM and OME was limited since it relied primarily on detecting the inflammatory state of the TM. Fluorescence spectroscopy has also been used in vitro to characterize the main bacteria that cause OM and to create a library of fluorescence features of these pathogens¹¹. In a subsequent publication, fluorescence was measured from 12 chinchilla AOM models in vivo with limited success¹². OCT, an optical imaging method that provides high-resolution real-time in vivo images of tissue microstructures, has been used to measure the thickness of the human TM at different infection states in vivo¹³. Researchers of this study were able to classify normal, acute, and chronic states of OM in adult patients based on TM thickness and biofilm formation for chronic cases. Performance accuracy of 70–80% was achieved due in part to the lack of consistency in biofilm growth across the TM and in all patients. Although all three optical methods were researched with the goal of in vivo application, these approaches are limited by their poor specificity and inability to detect and identify bacteria that cause AOM.

Raman spectroscopy (RS) is an optical technique that uses inelastically scattered light to provide biochemical information of a particular sample. This technique is sensitive to biochemical features such as nucleic acids, lipids, proteins, and carbohydrates and is able to provide a biochemical profile without the need of added contrast agents. RS has been used for many years to probe the biochemistry of various biological molecules¹⁴ and more recently for disease detection^15–17. More specifically, RS has been applied to characterize and identify bacteria in vitro as a proof of concept design. One example includes utilizing RS to characterize bacterial signatures in microbial colonies with the goal of detecting their presence in a shorter incubation time¹⁸. Another research group used a benchtop confocal Raman microscopy to identify bacteria within a mixed bacteria biofilm model¹⁹. Furthermore, Raman microspectroscopy has been used for Mycoplasma pneumoniae strain typing to distinguish between multiple clusters of strains²⁰. The feasibility of implementing a fiber-optic probe-based Raman system to characterize spectral signatures of bacterial colonies has also been shown and used to determine biochemical features important for distinguishing between Gram-positive and Gram-negative bacteria²¹. Interrogation of bacterial components such as surface wall features have also been investigated using surface-enhanced RS (SERS), which involves the addition of nanoparticles to the sample to enhance the Raman signal of targeted biomarkers²². Although these studies have shown the potential of RS for bacterial detection and identification, no studies to date have investigated bacteria that cause AOM and none have focused on the potential development for in vivo application. Currently, there is no tool available to rapidly and non-invasively detect the presence and identity of bacteria causing a middle ear infection. The goal of this study is to determine the feasibility of discriminating between the three main bacteria that cause AOM. Successful classification of these species was accomplished by spectrally characterizing their biochemical composition. We present the ability to classify bacteria causing AOM using Raman microspectroscopy and assess the feasibility of developing this technique for the diagnosis of AOM.

Materials and methods

Selection of agar growth media

Two of the most common agar types for bacterial culture were tested to determine their ability to grow all three bacteria while having the least spectral interference. Chocolate agar medium (Thermo Fisher Scientific, Waltham, MA), which is derived from lysed red blood cells and mainly used for fastidious organisms, was purchased in prepared 85 mm monoplates to culture bacteria. Chocolate agar was compared with Mueller-Hinton (MH) agar, which is a non-selective, non-differential microbiological growth medium that contains basic nutrients and no additives, was prepared by suspending 11 g of MH (BD, Franklin Lakes, NJ) powder and 7.5 g (15% agar/L) of agar (Thermo Fisher Scientific, Waltham, MA) in 500 mL of distilled water while heating (180°F) and stirring. The mixture was then autoclaved at 121 °C for 10 minutes. Bacteria were streaked separately on both MH agar and chocolate agar plates for comparison of agar and subsequent spectroscopic analysis of the bacterial strains.

Bacterial species

The three main bacteria that cause acute otitis media (AOM) were purchased from American Type Culture Collection (ATCC): nontypeable Haemophilus influenzae (ATCC #49766), Moraxella catarrhalis (ATCC #49143), and Streptococcus pneumoniae (ATCC #6301). Propagation methods as recommended by ATCC were used for each strain in preparation for bacteria cultures. Each bacterial species was streaked separately onto MH agar and chocolate agar plates. H. influenzae is a fastidious organism that requires lysed red blood cells not found in MH agar, therefore it is commonly grown on chocolate agar. To be effectively grown on MH agar, hemin and nicotinamide adenine dinucleotide (NAD)-rich disks (Hardy Diagnostics, Santa Maria, CA) were added to MH agar plates using steel tweezers that were disinfected between the additions of disks. Chocolate and MH agar plates were cultured for 24 hours at 37 °C with 5% CO₂.

Human middle ear effusion samples

As a proof of concept model to determine the ability of Raman microspectroscopy to identify bacteria derived from clinical samples, de-identified middle ear effusion (MEE) was collected from patients scheduled for myringotomy with tympanostomy at Monroe Carell Jr. Children’s Hospital at Vanderbilt. A protocol for collection of discarded, de-identified MEE specimens was approved by the Vanderbilt University Institutional Review Board (IRB# 130960) as non-human subjects research. To test the feasibility of our approach, three MEE samples were captured using a sterile Juhn Tym-Tap middle ear fluid device (Medtronic Inc., Minneapolis, MN) with an aspirator. A swab was used to spread MEE on MH agar plates with added hemin and NAD-rich disks and allowed to incubate for 72 hours at 37 °C with 5% CO₂. Viable colonies were then collected using sterile loop and streaked on a new MH agar plate with hemin and NAD-rich disks for a subsequent 24 hour culture at 37 °C with 5% CO₂. Evaluation of colony morphology was used as the standard for bacterial identification from MEE samples.

Raman microspectroscopy

Raman spectra were acquired using a confocal Raman microscope (inVia Raman Microscope, Renishaw plc, Gloucestershire, UK) with a 785 nm laser diode (Renishaw plc, Gloucestershire, UK). A 100X (N PLAN EPI, NA=0.85, Leica, Weltzlar, Germany) objective was used to focus a ~1 μm laser spot onto the bacterial colony on the agar surface at 27 mW. Raman scattered light was epi-detected through the same objective, then passed through a 35 μm slit and dispersed by a holographic grating (1200 lines/mm) onto a thermoelectrically cooled (−70 °C) deep-depleted, CCD that provided a 1 cm⁻¹ spectral resolution. The theoretical spatial resolution of the confocal Raman microscope system is ~0.6 μm. System alignment and light throughput to the sample was confirmed before and after experimental measurements with an internal silicon standard intensity at 520 cm⁻¹ and laser power at the sample.

Raman microspectroscopy was used to investigate the bacteria of interest since it provides high resolution for each wavenumber, an important feature to accurately characterize the spectral signature of the pathogens. Spectral measurements of pure bacteria included three acquisitions per spot, three spots per colony, and three colonies per bacteria, which presented an optimal standard deviation for each bacterial strain. Spectral acquisition parameters included a 30-second photobleach followed by a 15-second exposure with 7 accumulations from 700–1800 cm⁻¹. Cosmic ray removal from collected Raman spectra was performed using a custom MATLAB script (Mathworks, Natick, MA). Raman spectra were then processed to remove background fluorescence using a least squares modified polynomial fitting algorithm²³ and smoothed for noise with a second-order Savitzky-Golay filter²⁴. Post-processed spectra were mean normalized to each individual Raman spectrum for comparative analysis.

Data analysis

To quantify the spectral analysis, a Bayesian machine learning algorithm, sparse multinomial logistic regression (SMLR), was implemented to classify collected Raman spectra as H. influenzae, M. catarrhalis, or S. pneumoniae. SMLR is a supervised learning algorithm that reduces high dimensional multiclass data into features needed for distinguishing between classes²⁵. SMLR calculates a weight value for each spectral feature in a given spectral range based on its ability to separate classes within a given training data set. The statistical model also outputs how often (frequency) spectral features are utilized from the training data to determine classification across all cross-validations. SMLR was selected for our application since it provides the tools to classify multiclass data and identify spectral biomarkers important for discrimination. Both of these features were important for characterizing the three main pathogens that cause AOM.

To evaluate the importance of spectral features used for classification, a scaled version (from 0 to 1) of both the weight and how often spectral features were found from SMLR was utilized. The product of these values is used to calculate the SMLR feature importance, which is a quantitative metric that considers both the biochemical differences across the three bacteria characterized in this study and spectral heterogeneity among the same bacteria²⁶. The sparsity (λ) for SMLR, which controls the capacity for the number of spectral features used for classification, was adjusted to minimize data overfitting. SMLR feature importance was calculated for H. influenzae, M. catarrhalis, and S. pneumoniae using 77 features (λ=1.0) (Fig. 2). From the total spectral features available to use, about 8% were used for classification. A total of 917 spectral features from each Raman measurement were available for evaluation. Classification was based on implementing a leave-one-colony-out cross-validation approach. To accomplish this, a k-fold cross-validation was implemented, which separates the original data into k equally sized partitions called subsamples. This cross-validation technique retains one of the k subsamples and uses it to test the model while the remaining k-1 subsamples are utilized as the training data set. A 9-fold cross-validation was used for Raman spectral data analysis. This approach translates to classifying a bacterial colony belonging to a specific bacteria and would more accurately evaluate a predictive model.

Figure 2:

(A) Mean ± standard deviation Raman spectra of H. influenzae, M. catarrhalis, and S. pneumoniae grown on MH agar. Gray bands represent spectral features used for SMLR classification of each bacteria type by using a sparsity value of λ=1.0 for the SMLR input. The band gradient was based on SMLR feature importance. (B) Posterior probability of class membership from SMLR classification for each bacteria based on leave-one-colony-out cross-validation.

Results

Figure 1 shows the average Raman spectra collected from the three main bacteria that cause AOM, H. influenzae, M. catarrhalis, and S. pneumoniae, after being cultured on chocolate agar and MH agar. A qualitative analysis of bacteria cultured in chocolate agar shows many broad spectral regions with higher standard deviations compared to bacteria cultured on MH agar. Spectral regions that were challenging to discern in chocolate agar are indicated in Figure 1A, C, and E. with a dashed vertical line. Raman peaks in this same spectral region for bacteria cultured on MH agar were identified as indicated. A 10-fold reduction in spectral noise was calculated for Raman spectra of bacteria grown in MH agar compared to chocolate agar by using the standard deviation of the mean normalized intensity between 1500 cm⁻¹ and 1504 cm⁻¹, which contained minimal Raman features. Spectral analysis of bacteria grown in MH agar resulted in identifiable, reproducible peaks for the three bacteria under investigation that were originally not possible in chocolate agar as shown with arrows in Figure 1B, D, and F. Raman features included 827 cm⁻¹ (Tyrosine), 1298 cm⁻¹ (lipid), and 1447 cm⁻¹ (CH₂ and CH₃ deformations in proteins) for H. influenzae, 852 cm⁻¹ (CCH aromatic) and 1339 cm⁻¹ (CH₂ and CH₃ fatty acids and proteins) for M. catarrhalis, and 783 cm⁻¹(Cytosine, uracil) and 1317 cm⁻¹ (Guanine) for S. pneumoniae. The signal base line of Raman spectra and spectral peaks highlighted above from MH agar cultures were not affected by the addition of hemin and NAD disks, which were required for growth of H. influenzae. From these findings, MH agar was selected as the agar of choice for growing bacteria that cause AOM based on its minimal spectral interference and reduction in noise compared to chocolate agar.

Figure 1:

Mean ± standard deviation Raman spectra of bacteria that cause AOM grown on chocolate agar (left column) and MH agar (right column). (A-B) Haemophilus influenzae cultured on chocolate agar (A) and MH agar (B); (C-D) Moraxella catarrhalis cultured on chocolate agar (C) and MH agar (D); (E-F) Streptococcus pneumoniae cultured on chocolate agar (E) and MH agar (F). Vertical dashed lines on the left column represent spectral features from bacteria cultured on chocolate agar that are not discernable, while the arrows in the right column identify these features from the same bacteria cultured on MH agar. A 10x reduction in noise in MH agar compared to chocolate agar was calculated by using the standard deviation of the mean normalized intensity between 1500 cm⁻¹ and 1504 cm⁻¹.

Raman spectra from the three main pathogens that cause AOM were characterized to identify possible biochemical features that may be important in classifying these bacteria. Features of interest based on different peak intensities from mean normalized spectra included cytosine and uracil (ring stretching) at 783 cm⁻¹, tyrosine at 828 cm⁻¹, tryptophan and exopolysaccharide at 1555 cm⁻¹, and adenine, guanine (ring stretching), and C-O vibration modes of peptidoglycan at 1574 cm⁻¹ (Fig. 1). These spectral features presented visual differences and were representative of biochemical components of bacteria. Since traditional differences in peak intensities may not capture all of the information found in spectra and informative spectral changes between bacteria types, multivariate statistical analysis was utilized for feature selection and bacterial classification.

Figure 2 highlights wavenumbers or spectral features that were most important in classification of each bacteria denoted with gray vertical bands on the Raman spectra. The gradient of the vertical gray band in Figure 2A represents the SMLR feature importance, where darker bands indicate spectral features that were both strongly weighted from their regression coefficients and identified frequently for successful classification. The following peaks were most important in classification of bacteria that cause AOM as determined by SMLR: H. influenzae – 783 cm⁻¹ (Cytosine, uracil ring stretching), M. catarrhalis – 1431 cm⁻¹ (symmetric CH₂ bending and wagging), and S. pneumoniae – 840 cm⁻¹ (pyranose in peptidoglycan). Furthermore, the positive and negative slope of the 1449 cm⁻¹ (CH₂/CH₃ deformations in lipids/proteins) peak was consistent in classifying each of the three main bacteria that causes AOM. The predicted probability of class membership for each bacteria type is shown in Figure 2B. This SMLR classification was based on 77 spectral features (λ, sparsity =1.0) after implementing SMLR analysis on a total of 81 spectra collected from the three main otopathogens that cause AOM (27 spectra from each bacteria type) shown in Figure 2B. Table 1 presents the classification results as a confusion matrix, which describes the performance of a classification model based on the actual and predicted values. Sensitivity and specificity were also calculated based on the classification using a 50% threshold probability for class membership (Table 2). From the 81 total spectral measurements across all bacteria, less than 5% were misclassified as seen in M. catarrhalis. To our knowledge, this is the first report that characterizes the three main bacteria that cause AOM using Raman spectroscopy.

Table 1:

Classification of H. influenzae, M. catarrhalis, and S. pneumoniae based on SMLR for each bacteria using 77 spectral features.

λ=1.0, 50% Threshold	H. influenzae	M. catarrhalis	S. pneumoniae
H. influenzae	27	0	0
M. catarrhalis	0	24	3
S. pneumoniae	0	0	27

Table 2:

Sensitivity and specificity for each bacterial type.

λ=1.0, 50% Threshold	Sensitivity	Specificity
H. influenzae	100%	100%
M. catarrhalis	89%	100%
S. pneumoniae	100%	89%

Clinical MEE samples were also analyzed based on the spectral characterization of H. influenzae, M. catarrhalis, and S. pneumoniae (Figure 3). As shown in Figure 3A, mean normalized Raman spectra with standard deviation of MEE samples presented distinct biochemical features used to identify bacteria involved in a MEE sample. After culturing the MEE samples, only one bacterial colony grew from MEE sample #1. For MEE sample #2 and #3, 27 spectra were collected from each across three bacterial colonies. The classification of a bacterial colony belonging to one or more of the three main pathogens that cause AOM was based on spectral characterization of these bacteria, which was utilized for SMLR analysis as shown in Figure 2A. The probability of a MEE sample spectrum belonging to one or more bacteria was analyzed using a posterior probability plot (Fig. 3B) and summarized in Table 3. The first two Raman spectra of MEE sample #1 were not classified since their classification probability was below the threshold of 50%. All three MEE samples showed high probability of belonging to M. catarrhalis according to both Raman spectroscopy (Table 3) and based on features identified from colony morphology using the standard hockey puck test²⁷ and light microscopy. These findings show the importance of characterizing the three main bacteria that cause AOM and implementing a proof of concept model to non-destructively identify bacteria in cultured MEE specimens using Raman microspectroscopy.

Figure 3:

(A) Mean ± standard deviation Raman spectra of bacterial colonies from clinical MEE samples cultured on MH agar. (B) Posterior probability of class membership of clinical MEE samples to each of the three main pathogens that cause AOM.

Table 3:

Probability of each clinical MEE sample involving one or more of the three main bacteria that cause AOM.

λ=1.0, 50% Threshold	Sample #1	Sample #2	Sample #3
Probability of H. influenzae	0%	0%	0%
Probability of M. catarrhalis	67%	100%	100%
Probability of S. pneumoniae	11%	0%	0%

Discussion

Current methods to diagnose OM rely primarily on visual assessment and focus on predicting the presence of fluid in the middle ear space. The challenge is distinguishing whether there are active bacteria causing an acute infection (AOM) or only effusion, which is rarely caused by a bacterial infection (OME). Antibiotic treatment should only be prescribed for patients with AOM and not OME since they target a broad range of active bacteria. The inability to determine the presence and identity of bacteria causing AOM has led to an over prescription of antibiotics, leading to antibiotic-resistant bacteria²⁸. These antibiotic-resistant bacteria along with development of biofilms in the middle ear mucosa lead to the development of chronic OM infections. Action to investigate the middle ear effusion for bacterial identification is rarely practiced, serving as a last resort, and may be misleading due to obtaining negative cultures^29–31. A method that can characterize and classify bacteria that cause AOM will provide physicians with information on bacteria involved in an ear infection, allowing them to prescribe more targeted antibiotics and reducing antibiotic resistance. This paper focuses on determining the feasibility of using RS to discriminate between the three main bacteria that cause AOM, H. influenzae, M. catarrhalis, and S. pneumoniae by characterizing their biochemical signatures. Preliminary findings show promise for implementing this technique as an in vivo diagnostic tool.

Prior to investigating bacteria using RS, it was important to first select a culture agar medium that would be able to grow all three of the main bacteria that cause AOM while minimizing agar spectral contribution within the typical fingerprint window (700–1800 cm⁻¹). Although chocolate agar is one of the most common agar types to use for culturing bacteria, it absorbs light much more strongly as an opaque medium compared to a translucent medium such as MH agar. This resulted in less photons being Raman scattered and therefore Raman spectra with higher noise, making it more challenging to discern spectral peaks as shown in Figure 1. While M. catarrhalis and S. pneumoniae can grow in other agar types, H. influenzae, a fastidious organism, requires hemin and NAD to grow, which is released from the lysed RBCs as part of the chocolate agar media. Therefore, these factors were added to MH agar to grow H. influenzae. Spectral features of bacteria grown in MH agar as shown in Figure 1 can be easily identified with lower spectral noise compared to the same bacteria grown in chocolate agar. Spectral contribution of MH agar was minimal compared to the spectral features found in the main bacteria that cause AOM. Signal from underlying culture media has been investigated with the goal of minimizing incubation time while still obtaining spectral features from bacteria of interest. Although this was not the goal of this paper, Maquelin et al. investigated the potential of identifying bacteria in agar within 6 hours post-culture¹⁸. Since colonies from that study were ~10–100 μm in diameter and limited in thickness, there was an overwhelming signal from the underlying culture medium interfering with strain identification. Therefore, they developed and applied a vector correction algorithm on first derivative spectra to remove signal contributions from culture medium in bacterial microcolonies. Although this method may be applied for known bacteria, clinical samples may take more than 6 hours to culture and involve polymicrobial infections, which may limit the application of this algorithm.

We have reported the characterization and identification of the three main pathogens that cause AOM using Raman microspectroscopy. As can be seen in Figure 2A, our SMLR feature importance (SMLR-FI) algorithm extracted specific spectral features critical for identification. A threshold of at least 25% importance was set to present more important biomarkers used for classification. The nontypeable H. influenzae (NTHi) strain showed 100% sensitivity and specificity. The biomarker with the highest SMLR-FI used for identification of H. influenzae was at 783 cm⁻¹ (Cytosine, uracil ring stretching). Identification of M. catarrhalis in MH agar was found with 89% sensitivity and 100% specificity. As can be seen from Figure 2B, three spectral measurements fell below 50% for the probability of belonging to specific bacteria class (H. influenzae, M. catarrhalis, and S. pneumoniae). This may be due to phase variation in bacteria, which alters protein expression in different regions of a bacterial population. One example of phase variation commonly seen in M. catarrhalis is the UspA1 protein, which affects adherence factors that facilitate adhesion to other cells and surfaces³². This type of phase variation has been shown to occur in an in vitro environment when individual colonies were tagged using monoclonal antibodies for the UspA1 protein³³. Although there may have been phase variation between M. catarrhalis colonies, multiple spectral features were identified for classifying this bacteria. The spectral feature identified to be the most important for classification of M. catarrhalis using SMLR-FI was 1431 cm⁻¹ (symmetric CH₂ bending and wagging). Classification for the third main bacteria that causes AOM, S. pneumoniae, presented a 100% sensitivity and 89% specificity. As can be seen from Figure 2A, the most important spectral marker for discrimination of S. pneumoniae was at 840 cm⁻¹, which is tentatively assigned as pyranose, a sugar commonly found in the cell wall structure of bacteria³⁴. This sugar may be found more predominantly in peptidoglycan from Gram-positive bacteria, such as S. pneumoniae, and may be important for determining bacterial susceptibility³⁴.

The spectral characterization of the main bacteria that cause AOM was used to identify those same bacteria involved in MEE from patients suffering from recurrent OM. This proof of concept approach was able to identify bacteria from cultured MEE samples. For MEE sample #1, only 9 spectra were collected since only one bacterial colony grew post-culture from this sample. Although two of the spectra collected from sample #1 had a probability of less than 50% for belonging to a specific type of bacteria, the remaining 7 spectra had at least a 50% chance of belonging to M. catarrhalis. For MEE samples #2 and #3, 100% of the 27 spectra collected were categorized as M. catarrhalis. Overall, nearly 80% of Raman spectra collected across all clinical MEE samples had an 80% or above probability of belonging to M. catarrhalis. These results were also supported by a hockey puck test²⁷, which uses a sterile wooden stick to push the colonies across the MH agar plate. The bacterial colonies easily slid across the agar plate, which indicated a positive outcome for M. catarrhalis. A major challenge for bacterial identification from clinical samples is the difficulty associated with culturing bacteria. This is more frequently presented with bacteria immersed in a biofilm environment, which limits the ability to culture particular clinical samples. Ultimately, the inability to culture bacteria in a biofilm state may limit our diagnosis of bacteria involved in chronic infections. This drawback highlights the importance of being able to detect the presence and identity of bacteria directly in a biofilm without the need to culture the bacteria. The potential impact of this solution may increase bacterial identification accuracy and decrease diagnostic time and cost.

Our findings from characterizing the biochemical features of the three main otopathogens that cause AOM and accurately identifying them show the potential application of RS as a diagnostic tool for patients suffering from OM. While additional bacteria species and isogenic variants that cause AOM will need to be interrogated, non-destructive spectral identification and classification of the three main bacteria that cause AOM is a critical first step for developing a diverse spectral database to accurately detect and identify bacteria causing AOM. This work sets the stage for other applications of RS where bacterial identification may also be utilized as a research tool to investigate bacterial growth patterns, antibiotic susceptibility, or characterize biochemical changes in mutant forms of bacteria. Spectral results from these experiments may serve to create a better understanding of the microbial pathogenesis of other clinical bacterial infections. These studies provide insight into the biochemical changes occurring at the micro-scale and portends to the global application of this technique for the development of targeted antibiotics for susceptible and antibiotic-resistant bacteria. Numerous reports have been published recently describing the effects of over-prescription of broad-spectrum antibiotics and prescriptions of antibiotics for pathogens causing AOM that are no longer susceptible to them^28,35–38. This is a major problem that has led to antibiotic resistance in many bacteria and even multi-drug resistant (MDR) microorganisms. Providing a rapid technique that accurately detects and identifies pathogens causing AOM will aid in OM diagnostic efforts and inform physicians on proper treatment.