Article pubs.acs.org/jpr
Broad Identification of Bacterial Type in Urinary Tract Infection Using 1 H NMR Spectroscopy Ashish Gupta,*,†,‡ Mayank Dwivedi,§ Abbas Ali Mahdi,‡ Chunni Lal Khetrapal,† and Mahendra Bhandari∥ †
Centre of Biomedical Magnetic Resonance, Sanjay Gandhi Postgraduate Institute of Medical Sciences, Raebareli Road, Lucknow 226 014, India ‡ Department of Biochemistry, Chhatrapati Shahuji Maharaj Medical University, Lucknow 226003, India § Department of Microbiology, Lady Hardinge Medical College, New Delhi, India ∥ Vattikuti Urology Institute, Henry Ford Hospital System, Detroit, Michigan, United States
ABSTRACT: To address the shortcomings of urine culture for the rapid identification of urinary tract infection (UTI), we applied 1H-nuclear magnetic resonance (NMR) spectroscopy as a surrogate method for fast screening of microorganisms. Study includes 682 urine samples from suspected UTI patients, 50 healthy volunteers, and commercially available standard strains of gram negative bacilli (GNB) (Escherichia coli, Pseudomonas aeruginosa, Klebsiella pneumonia, Enterobacter, Acinetobacter, Proteus mirabilis, Citrobacter f rundii) and gram positive cocci (GPC) (Enterococcus faecalis, Streptococcus group B, Staphylococcus saprophyticus). Acetate, lactate, ethanol, succinate, creatinine, trimethylamine (TMA), citrate, trimethylamin-N-oxide, glycine, urea, and hippurate were measured by 1H NMR spectroscopy. All urine specimens were evaluated with culture method. Multivariate discriminant function analysis (DFA) reveals that acetate, lactate, succinate, and formate were able to differentiate, with high accuracy (99.5%), healthy controls from UTI patients. This statistical analysis was also able to classify GNB to GPC infected urine samples with high accuracy (96%). This technique appears to be a promising, rapid, and noninvasive approach to probing GNB and GPC infected urine specimens with its distinguishing metabolic profile. The determination of infection will be very important for rapidly and efficiently measuring the efficacy of a tailored treatment, leading to prompt and appropriate care of UTI patients. KEYWORDS: NMR spectroscopy, UTI, urine metabolic profile
■
INTRODUCTION Urinary tract infections (UTIs) are among the most common bacterial infections of humans, particularly in women and children.1,2 Although the majority of bacteria causing UTIs among all patient populations are gram negative bacilli (GNB)1,2 and gram positive cocci (GPC) also contribute to a large number of infections.1,3 The identification of bacterial infection is currently achieved by conventional culture method and various biochemical tests. However, intrinsic limitations in these methods cause reporting delays and high chances of contamination, thus making them poor tools for rapid identification of infection and monitoring therapeutic response to treatment.4,5 Among the various causes of inherent reporting delays, the most important reason is that anaerobic bacteria are usually slow growing, because during transportation and processing of urine samples, they are exposed to environmental © 2012 American Chemical Society
oxygen. Moreover, culture plates of anaerobic bacteria are incubated in anaerobic jars which must be opened to examine the plates. In principle, the jars should be opened to room air only after 48−72 h of growth, introducing systematic delay in rapid bacterial identification, which is relatively impractical. Moreover, despite the best techniques, only 20% of samples show bacterial growth on screening.6 An alternative timesaving method for identifying UTI samples would be worthwhile. Many approaches such as microscopic, colorimetric, filtration/staining, photometric, automated method and Fourier transform-infrared spectroscopy6−8 have been reported for bacterial identification and were found to have limited clinical acceptance. Therefore, a rapid, precise, simple, sensitive, and Received: October 27, 2011 Published: January 31, 2012 1844
dx.doi.org/10.1021/pr2010692 | J. Proteome Res. 2012, 11, 1844−1854
Journal of Proteome Research
Article
were also prepared under similar conditions. The person determining the bacterial counts by NMR and the microbiologist culturing the urine samples were unaware of sample origin until the end of the study. Serial urine samples were collected from UTI patients (n = 10) either after immediate admission or within 1−2 days of antibiotics administration in the hospital and continued for several days post-treatment of antibiotics for longitudinal study. In vitro Study. Midstream urine samples from age-matched healthy volunteers were obtained, and NMR spectra were recorded to obtain information about the metabolic profile of urine samples with no infection. In the current study, the GNB strains including E. coli NCTC-10418, ATCC- 25923, MTCC41; K. pneumoniae NCTC-9633, ATCC-13883, MTCC-39; P. aeruginosa NCTC-10662, ATCC-25922; Pr. mirabilis ATCC49565, MTCC-743; E. aerogenes ATCC-13048, MTCC-111; A. baumanii ATCC-19606, MTCC-1425; C. f rundii ATCC-8090, MTCC-1658, and GPC strains including E. faecalis ATCC19433, MTCC-439; Streptococcus gp B ATCC-13813 and Staph. saprophyticus ATCC 15305 were used. ATCC strains were obtained from Manassas, VA, NCTC strains from Central Public Health Laboratory in London, U.K. and MTCC strains from the microbial type culture collection at IMTECH in Chandigarh, India. For the in vitro study, the artificial growth medium was prepared using urine samples from healthy volunteers, sterilized by filtration and supplemented with standard glucose (2 mg/ mL). Bacterial suspensions for all the listed bacterial species were prepared from freshly grown colonies in normal saline solution (5 mL) and were adjusted to a turbidity equivalent to that of 1.0 McFarland unit standard suspension (bacterial count of ∼3.0 × 108 colony-forming units/mL (CFU/mL). For qualitative estimation, each bacterial species was prepared in 1.0 mL artificial growth media, by adding of 0.1 mL bacterial suspension to 0.9 mL of growth medium, resulting in a viable bacterial count of ∼3.0 × 107 CFU/mL. This final suspension with each bacterial species was incubated at 37 °C for 6 h. After incubation, the bacterial suspensions were removed by centrifugation for 5 min at 4 °C; 10000 rpm The supernatant parts were decanted and subjected to NMR experiments. Control solutions without bacteria were also prepared under similar conditions.
comprehensive bacterial identification method is required. The purpose of rapid bacterial identification is 2-fold: (i) to eliminate negative specimens rapidly, allowing the microbiologists to spend more time on positive specimens, which leads to improved efficiency and cost effectiveness, and (ii) to provide accurate information to the physicians in a timely manner, which leads to prompt care of patients. This, in turn, may contribute to a more rapid recovery from the infection, less sequelae, and a shortened hospital stay resulting in appreciable cost savings to the medical care system. Proton nuclear magnetic resonance (1H NMR) spectroscopy has been widely used to characterize the endogenous biochemistry and metabolic profile of urine within a reasonable time (10−12 min).9−11 Bacterial metabolic end-products (acetate, formate, β-hydroxybutyrate, lactate and trimethylamine) were observed in contaminated urine samples.11,12 Previously, we not only identified and quantified P. aeruginosa, K. pneumoniae, E. coli and Pr. mirabilis with corresponding specific substrate metabolism13−15 but also differentiated the infected, contaminated and sterile urine samples15 and observed the inhibition of adherence of multidrug resistant E. coli by proanthocyanidin using NMR spectroscopy.16 In this study we have extended the scope of the NMR method not only to distinguishing between GNB and GPC uropathogens, which differ in their morphology and bioenergetic metabolism, but also investigating the patient’s recovery from infection through trajectory metabolic profile to a normal urine metabolic profile during treatment.
■
MATERIALS AND METHODS
Patients and Sample Collection
The study was performed in two steps: ex vivo followed by in vitro study. We studied 682 urine samples from patients with suspected UTI (20−45 years old women) and from age matched 50 healthy women volunteers. Commercially available standard strains (American, National, and Microbial Type Culture Collections, ATCC, NCTC and MTCC) of Escherichia coli, Klebsiella pneumonia, Pseudomonas aeruginosa, Enterobacter aerogenes, Acinobacter baumanii, Proteus mirabilis, Citrobacter f rundii, Enterococcus faecalis, Streptococcus group B, and Staphylococcus saprophyticus were included for in vitro study. Ex vivo Study. Urine samples were obtained from suspected UTI patients attending the outpatient department as well as patients admitted in the wards of Sanjay Gandhi Post Graduate Institute of Medical Sciences Lucknow, (a 600 bed tertiary care hospital) and Chhatrapati Shahuji Maharaj Medical University, Lucknow, India (a large community hospital). Inclusion criteria includes urge to urinate frequently and need to urinate at night, painful burning sensation when urinating, discomfort or pressure in the lower abdomen, cloudy appearance and strong smell in urine, fever (typically lasting more than 2 days), impaired immune systems, or a history of relapsing or recurring UTIs, pain in the flank (pain that runs along the back at about waist level), vomiting and nausea. The samples were stored at 4 °C until (maximum 5−6 h) they were used for NMR experiments and for clinical microbiology laboratory tests. Each urine sample was divided into two parts of 1.0 mL each; one part was used for standard biochemical tests and viable bacterial count; the other was centrifuged for 5 min at 4 °C and 9168× g to remove all cell debris and other contaminants. The supernatant parts were decanted and subjected to 1H NMR experiments. Control urine samples
■
NMR EXPERIMENTS
The NMR spectra were obtained with Avance 400 MHz spectrometer with 5 mm Broad band Inverse probe (Bruker Biospin, India). The 0.6 mL of supernatants and control urine samples were transferred to NMR tubes. A sealed capillary, containing precalibrated 25 μL of 0.75% trimethyl silyl propionic acid (TSP) deuterated at the CH2 groups and dissolved in deuterium oxide, was inserted into the NMR tube. While TSP served as a chemical-shift reference as well as the quantitative standard for estimating metabolites,17 deuterium oxide served as the “field-frequency-lock”. For all specimens, the one-dimensional 1H NMR experiments were done at 22 °C by suppression of water resonance by presaturation. The parameters used were: spectral width, 8000 Hz; time domain points, 32 K; relaxation delay, 3s; pulse angle, 45°; number of scans, 64; spectrum size, 32 K and line broadening, 0.3 Hz. The Xwinnmr software 3.5 was used for the baseline correction. The complete scheme for the analysis of suspected patient urine specimens is shown in Figure 1. 1845
dx.doi.org/10.1021/pr2010692 | J. Proteome Res. 2012, 11, 1844−1854
Journal of Proteome Research
Article
Figure 1. Scheme for the analysis of suspected patient urine specimens.
Statistical Analysis
between the groups. Statistical significance of the differences between the ROC curves was evaluated by comparing the area under the curves. The observed predictors were assessed further by resubstitution and prospective test data methods that provide the sensitivity and specificity of the performance of the DFA model between the chosen groups. The resubstitution method carried out as part of the external validation process, involved feeding the same infection and control data into the DFA model to check the correct classifications based on the discriminant function weights obtained for each variable. The prospective test data methods comprised a reassessment of the prediction possibility by treating 75% of each group of infectious data as training sets and examining the remaining 25% of the data as test sets, which were obtained based on the Random Numbers Table by Fisher and Yates. This method provides a validation of the model build based on DFA comprising four separate training sets, viz. (1) control vs GNB + GPC, (2) control vs GNB, (3) control vs GPC, and (4) GNB vs GPC and checking the percentage classifications of the remaining 25% of the patients data treated as test sets.
The results obtained for the NMR-based quantified metabolites (μg/mL) in urine samples of control, GNB and GPC infected patient groups are expressed as median and (range) ± SD. Univariate Analysis
The statistical significance for the quantified metabolites were determined by one-way ANOVA followed by a post hoc Student− Newman−Keuls multiple comparisons test, carried out with the Graph Pad INSTAT 3.0 software. A probability p-value of less than 0.05 was taken to indicate statistical significance. Multivariate Analysis
The data were subjected to multivariate linear discriminant function analysis (DFA) to define important variables for differentiation of UTI patients from controls, followed by discrimination of the GNB and GPC infected urine specimens. For DFA, 12 NMR observed metabolites were used, and the analyses were carried out by SPSS version 11.0 (SPSS Inc. Chicago, IL). The discriminant functional scores were obtained by summing the independent variable value and its corresponding discriminant function-based coefficient/weights value. To develop a comprehensive DFA for rapid identification and classification of UTI causing bacteria, a stepwise forward variable selection procedure was adapted to determine the best set of predictors of infection using DFA-based discriminant scores. Leave-one-out cross validation (jack-knife) method of Lachenbruch18 was applied in all calculations to derive optimized discriminant scores. Jack-knife method provides a relatively unbiased estimate of the error rate of DFA.18 The receiveroperating characteristic (ROC) curves were also created by using the discriminant function score to determine the cutoff values
■
RESULTS
Figures 2−5 show various typical 1H NMR spectra of urine samples with ex vivo and in vitro conditions, revealing a diverse metabolic profile and chemical shift assignments of different resonances. On the basis of microbiological culture reports, 1H NMR derived ex vivo urine samples data were classified into two groups: GNB and GPC (Table 1). Resonance signals of various metabolites in urine samples were assigned using known chemical shift and coupling constant parameters.9−11 Among 1846
dx.doi.org/10.1021/pr2010692 | J. Proteome Res. 2012, 11, 1844−1854
Journal of Proteome Research
Article
biomarkers for the classification of each group. The correlationmatrix observations suggest the intrarelationship of NMRderived metabolites for infected groups in comparison with the control, and this correlation allowed selection of a set of a few variables for correct classification of each group. Table 3 reveals the classification of UTI as compared to the respective control groups. Step 1: on the basis of DFA, when GNB+GPC (all UTI) samples were compared with control, 12 metabolites could successfully classify 97% of cases with 84% sensitivity and 99% specificity (Wilks’ Lambda, 0.571; p < 0.001). Step 2: of these 12 metabolites, when only acetate, lactate, succinate, and formate were chosen based on their discriminant function coefficients, and DFA was performed again, overall 99.5% of UTI cases (GNB+GPC) were effectively classified with 99.3% sensitivity and 99.5% specificity (Wilks’ Lambda, 0.632; p < 0.0001). On the basis of DFA, for GNB cases when pooled with the control cases, the 12 metabolites could successfully classify 98% of GNB cases with a sensitivity of 98% for the GNB and a specificity of 98% for the control (Wilks’ Lambda, 0.472; p < 0.0001). Among these metabolites when only acetate, lactate, ethanol, succinate, and formate were chosen based on their discriminant function coefficients and the DFA was carried out again with the chosen metabolites, and the result exhibited 98.3% cases could be successfully classified, with a sensitivity of 99.5% and a specificity of 98% (Wilke’s Lambda, 0.485; p < 0.0001). The observations recommend that only five NMRmeasured metabolites were sufficient for successful classification of a GNB group from the control. Similarly, when GPC infected cases were compared with the control cases, all NMR-measured metabolites could successfully classify 94% of GPC infected cases with a sensitivity of 96% and a specificity of 94% (Wilke’s Lambda, 0.391; p < 0.0001). Acetate, lactate and formate were chosen based on their discriminant function coefficients. DFA was performed again, and the results exhibited that 95% of GPC cases were classified with 99.5% sensitivity and 92% specificity (Wilke’s Lambda, 0.451; p < 0.0001). Thus, the chosen metabolites played key roles in differentiating GPC cases from control. In the same way, when GNB cases were compared with GPC, all NMR-measured metabolites could successfully classify 94% of cases with 96% sensitivity and 91% specificity (Wilke’s Lambda, 0.596; p < 0.0001). On the basis of discriminant function coefficient values, when succinate, lactate, and ethanol were chosen and DFA was performed again, the result exhibited that 96% of GPC cases were classified with 96% sensitivity and 96% specificity (Wilke’s Lambda, 0.607; p < 0.0001). Thus, the chosen metabolites are important for differentiating the GNB from GPC infected urine samples. Figure 6 shows the ROC curves of discriminant predicted probability with sensitivity and 1-specificity of different groups with their respective controls. The area under the ROC curves of discriminant predicted probability of NMR variables for each classification is compared in Table 4. To further confirm the ex vivo findings of UTI patient samples from DFA, we performed in vitro study with standard strains of GNB and GPC and used healthy human urine sample as growth medium. This growth medium supplied with glucose for energy metabolism of microorganism. Figure 2 shows the one-dimensional spectra of incubated bacterial media for E. coli ATCC-25923, K. pneumonia ATCC13883, E. aerogenes ATCC-13048, A. baumanii ATCC-19606, Pr. mirabilis ATCC-4956, C. f rundii ATCC-8090, P. aeruginosa
Figure 2. Proton NMR spectra of in vitro study of showing glucose metabolism in urine samples through GNB (E. coli ATCC-25923, K. pneumonia ATCC-13883, E. aerogenes ATCC-13048, A. baumanii ATCC-19606, Pr. mirabilis ATCC-4956, C. frundii ATCC-8090, P. aeruginosa ATCC-25922) along with control urine and urine with glucose. Key: E, ethanol; L, lactate; A, acetate; S, succinate.
several resonances, 12 metabolites (acetate, lactate, ethanol, succinate, formate, creatinine, trimethyl-amine (TMA), citrate, trimethyl-amine-N-oxide (TMAO), glycine, urea and hippurate) were unequivocally with well-resolved resonances present in urine samples, and their quantities were estimated from their respective resonances (viz., 1.91 (s), 1.33 (d), 1.17 (t), 2.41 (s), 8.46 (s), 3.01 (s), 2.88 (s), 2.55 (d), 3.27 (s), 3.56 (s), 5.78 (s), and 7.84 (d) ppm, s; singlet, d; doublet, t; triplet) and subjected to univariate and multivariate statistical analysis. Univariate Statistical Analysis
Table 2 reveals that the concentrations of acetate, lactate, ethanol, succinate, creatinine, citrate, urea, and formate were significantly different in the urinary tract infected urine samples than in the control group. Lactate, succinate, hippurate and formate were significantly different in GNB with compared to GPC infected urine samples. Since, univariate analysis lacks the ability to provide expedient, sensitive, specific, and discriminant predictive probability for differential diagnosis of each class of UTI, multivariate statistical analysis was also performed. Multivariate Statistical Analysis
To do the multivariate analysis, data shown in Table 2 were considered to see their impact and revealing the key descriptor 1847
dx.doi.org/10.1021/pr2010692 | J. Proteome Res. 2012, 11, 1844−1854
Journal of Proteome Research
Article
Figure 5. Proton NMR spectra of ex vivo urine samples obtained from GPC infected UTI patients. Key: E, ethanol; L, lactate; A, acetate; T, Trimethyl-amine; Cr, Creatinine; TM, Trimethyl-amine-N-oxide; U, urea; C, citrate; G, glycine, F, formate; H, hippurate. Expanded view of resonances from 6.5 to 9.0 ppm with 4× multiplication is also demonstrated.
Table 1. Suspected UTI Patients (n = 682) Urine Samples Investigated by Conventional Culture Method and NMR Method
Figure 3. Proton NMR spectra of the ex vivo urine samples obtained from GNB infected UTI patients. Key: E, ethanol; L, lactate; A, acetate; S, succinate; T, Trimethyl-amine; Cr, Creatinine; TM, Trimethyl-amine-N-oxide; U, urea; C, citrate; G, glycine, F, formate; H, hippurate. Expanded view of resonances from 6.5 to 9.0 ppm with 4× multiplication is also demonstrated.
bacterial species
identified by conventional method
GNB E. coli K. pneumoniae P. aeruginosa Pr. mirabilis E. aerogenes A. baumanii C. f rundii GPC E. faecalis Streptococcus gp B Staph. Saprophyticus Sterile
472 256 110 60 10 14 12 10 120 42 38 40 90
identified by NMR spectroscopy 440 240 101 58 09 12 11 09 111 38 37 36 131 (sterile + microbes with bacterial strength
