Pneumococcal Pneumonia: Potential for Diagnosis through a Urinary Metabolic Profile Carolyn M. Slupsky,*,†,¶ Kathryn N. Rankin,‡ Hao Fu,‡ David Chang,§ Brian H. Rowe,‡,| Patrick G. P. Charles,∇ Allison McGeer,O Donald Low,O Richard Long,† Dennis Kunimoto,† Michael B. Sawyer,# Richard N. Fedorak,† Darryl J. Adamko,‡,⊥ Erik J. Saude,⊥ Sirish L. Shah,§ and Thomas J. Marrie† Department of Medicine, Magnetic Resonance Diagnostic Centre, Department of Chemical and Materials Engineering, Department of Emergency Medicine, Department of Pediatrics, and Department of Oncology, University of Alberta, Edmonton, Alberta, Canada T6G 2S2, Department of Infectious Diseases, Austin Health, Heidelberg, Australia, and Mount Sinai Hospital, Department of Microbiology, University of Toronto, Toronto, Ontario, M5G 1X5 Received July 20, 2009
Pneumonia, an infection of the lower respiratory tract, is caused by any of a number of different microbial organisms including bacteria, viruses, fungi, and parasites. Community-acquired pneumonia (CAP) causes a significant number of deaths worldwide, and is the sixth leading cause of death in the United States. However, the pathogen(s) responsible for CAP can be difficult to identify, often leading to delays in appropriate antimicrobial therapies. In the present study, we use nuclear magnetic resonance spectroscopy to quantitatively measure the profile of metabolites excreted in the urine of patients with pneumonia caused by Streptococcus pneumoniae and other microbes. We found that the urinary metabolomic profile for pneumococcal pneumonia was significantly different from the profiles for viral and other bacterial forms of pneumonia. These data demonstrate that urinary metabolomic profiles may be useful for the effective diagnosis of CAP. Keywords: NMR • metabolomics • metabonomics • infection • targeted profiling • pneumonia • urine • Streptococcus pneumoniae
Introduction Pneumonia is an acute, occasionally chronic, infection of the lower respiratory tract caused by a variety of pathogens including bacteria, viruses, fungi and parasites.1,2 A variety of noninfectious processes such as congestive heart failure, pulmonary infarction, vasculitis, and drug reactions can mimic pneumonia.3 In addition, chronic respiratory conditions such as asthma and chronic obstructive pulmonary disease (COPD) are frequently complicated by pneumonia. Despite progress in the development of antibiotics, diagnostic imaging, critical care medicine, decision aids, and clinical practice guidelines to assist clinicians with empiric therapy, pneumonia remains the leading cause of death from infection in developed coun* Corresponding author: Carolyn Slupsky, Departments of Nutrition and Food Science & Technology, University of California, Davis, Davis, California, 95616. Phone, (530) 219 - 5757; fax, (530) 752 - 8966. † Department of Medicine, University of Alberta. ¶ Present address: Departments of Nutrition and Food Science & Technology, UC Davis, Davis, CA 95616. ‡ Magnetic Resonance Diagnostic Centre, University of Alberta. § Department of Chemical and Materials Engineering, University of Alberta. | Department of Emergency Medicine, University of Alberta. ∇ Department of Infectious Diseases, Austin Health. O Department of Microbiology, University of Toronto. # Department of Oncology, University of Alberta. ⊥ Department of Pediatrics, University of Alberta.
5550 Journal of Proteome Research 2009, 8, 5550–5558 Published on Web 10/09/2009
tries. This is mainly due to the difficulty in promptly identifying the etiologic agent,4,5 although age also plays a critical role. In practice, less than 20% of patients admitted to hospital with pneumonia are appropriately diagnosed.4,6 Streptococcus pneumoniae (S. pneumoniae), the major causative pathogen of community-acquired pneumonia,7 is a transient commensal organism of the throat and upper respiratory tract8 present in approximately 10-80% of the population depending on age and location.9 However, S. pneumoniae frequently becomes virulent, especially for those at the extremes of age, immunocompromised individuals, those with chronic diseases, and tobacco smokers.6 Each year, millions of people in the United States develop infections from these bacteria, with more serious illness resulting in 500 000 hospitalizations.10 Indeed, invasive S. pneumoniae is a leading cause of death worldwide, with a mortality rate of up to 25%.11 Furthermore, with widespread antibiotic use, microbial resistance of S. pneumoniae to common agents is a growing concern.12 Rapid identification of this pathogen as a cause of infection would be important clinically, and could save many lives. Currently, diagnosis of pneumococcal pneumonia is made by isolation of the causative organism from blood, sputum, pleural fluid, or bronchoalveolar lavage (BAL) in the presence of a compatible clinical picture. Blood and sputum cultures 10.1021/pr9006427 CCC: $40.75
2009 American Chemical Society
Pneumococcal Pneumonia yield S. pneumoniae in 6-10% and ∼11% of patients, respectively, who have been hospitalized for CAP.13 However, results are rarely available within 36 h, and a positive sputum culture for S. pneumoniae only indicates that this pathogen is possible or at best a probable cause of pneumonia.10 To combat lengthy diagnostic times, other tests have been developed such as the NOW assay (Binax, Inc.) which detects S. pneumoniae Cpolysaccharide in urine. The usefulness of this test is limited, because while it is positive in 70-90% of bacteremic patients,14-17 a 65% false positive rate is seen in children who carry this microorganism in their nasopharynx,17 and a 59% false positive rate is seen with vaccinated children.18 Metabolomics is an emerging science dedicated to the global study of metabolites; their composition, dynamics, and responses to disease or environmental changes in cells, tissues and biofluids. Studies involving metabolic profiling have reported differences between diseased and healthy states.19-21 With respect to the diagnosis of pneumococcal pneumonia, metabolomics has the potential to be an ideal tool, as it has been previously shown that lung infection by different microorganisms in a mouse model induces unique metabolite patterns in urine.22 In this study, we have applied metabolomics to study pneumococcal infection in humans. We analyzed urine samples from patients with different types of pulmonary diseases and infection, and found that infection with S. pneumoniae produces a distinct pattern.
Experimental Section Populations. Written informed consent was obtained from each subject before entering this study, and the institutional ethics committees approved the protocols outlined below. Patients were recruited through the Toronto Invasive Bacterial Diseases Network (150), 6 hospitals in the Edmonton area (Royal Alexandra Hospital, University of Alberta Hospital, Gray Nuns Hospital, Misericordia Community Hospital, Sturgeon Community Hospital, and Leduc Community Hospital) (558), the QE II Health Sciences center in Halifax, Nova Scotia (35), and 6 hospitals in Australia (Austin Health, Alfred Hospital, Monash Medical Centre, Princess Alexandra Hospital, Royal Perth Hospital, and West Gippsland Hospital) (85). While the majority of samples were obtained from individuals of age 50 and above, it is nearly impossible to get all populations exactly age and gender matched when collecting samples from multiple centers. Patients with Pneumococcal Disease. Pneumonia was categorized as definite pneumococcal pneumonia, positive blood culture for S. pneumoniae; or possible pneumococcal pneumonia, positive sputum or endotracheal tube culture for S. pneumoniae, or positive urinary antigen test. All patients had a chest radiograph read as pneumonia by a radiologist. Two of the blood positive patients had pneumococcal peritonitis (S. pneumoniae isolated from peritoneal fluid) and 2 of the blood-positive patients had meningitis (S. pneumoniae isolated from cerebrospinal fluid). S. pneumoniae was identified in microbiology laboratories of the participating hospitals using standard criteria. The entire group consisted of n ) 62 subjects; mean age, 53 ( 23; range, 6 days to 88 years. Eight had diabetes mellitus, and three were pediatric patients. Healthy Volunteers. The group consisted of n ) 115 subjects (45 male, 70 female); mean age, 59 ( 14 years; range, 19-87 years. Three had diabetes. Noninfectious Metabolic Stress. Patients in this category were diagnosed with (1) myocardial infarction, n ) 12 (10 male,
research articles 2 female); mean age, 59 ( 14 years; range, 41-76; (2) congestive heart failure, n ) 12 (7 male, 5 female); mean age, 78 ( 9 years; range, 59-91; (3) trauma (fractures), n ) 17 (11 male, 6 female); mean age, 55 ( 14 years; range, 22-76; and (4) trauma (lacerations), n ) 14 (10 male, 4 female); mean age, 32 ( 13 years; range, 19-57. In all instances, the patient’s attending physician made diagnoses of the above conditions. Patients in groups 1-3 had no obvious evidence of infection. Fasting Individuals. Patients presenting for routine colonoscopy who were fasting for at least 1 day were recruited (n ) 70). Longitudinal Study. Serial urine study: Patients presenting with bacteremic pneumococcal pneumonia (n ) 8) had samples collected within 4 days of receiving antibiotics in hospital, and several days postadmission after treatment with antibiotics. Comparison to Other Lung Infections. Patients with Legionella pneumophila (Legionnaires’ disease and Pontiac Fever) (n ) 62), Mycobacterium tuberculosis (tuberculosis) (n ) 65), Staphylococcus aureus (n ) 27), Coxiella burnetii (n ) 15), Haemophilus influenzae (n ) 11), Mycoplasma pneumoniae (n ) 9), Escherichia coli (n ) 7), Enterococcus faecalis (n ) 3), Moraxella catarrhalis (n ) 4), Streptococcus viridans (n ) 2), Streptococcus anginosus (n ) 2), influenza A (n ) 16), picornavirus (n ) 12), respiratory syncycial virus (RSV) (n ) 11), parainfluenza viruses (n ) 8), coronavirus (n ) 6), human metapneumovirus (hMPV) (n ) 4), and hantavirus (n ) 1) were collected from Toronto, Edmonton and Australia. Comparison to Other Lung Diseases. Patients with asthma (n ) 31) or COPD exacerbations (n ) 44) were collected from the Emergency Department (ED) of the University of Alberta Hospital in Edmonton, Alberta, Canada. Patients were seen and assessed in the ED by treating physicians and a formal interview was completed with an ED chart review. Blinded Study. One of us (T.J.M.) assembled a set of urine samples from patients not part of the original learning set with the following features: bacteremic pneumococcal pneumonia n ) 35; healthy n ) 42; noninfectious stress n ) 9; COPD ) 6; Asthma n ) 8; Tuberculosis n ) 24; Legionnaires’ disease n ) 1; C. burnetii (Q-fever) n ) 20. The etiological diagnoses were unknown to C.M.S. who analyzed the samples and provided a diagnosis from metabolite concentrations before the code was broken. Data Collection. Urine samples were obtained from volunteers, and transferred into urine cups containing sodium azide dried on the bottom (such that the final concentration would be approximately 0.02% with the addition of 100 mL of urine) to prevent bacterial growth. Urine samples were subsequently placed in a refrigerator at 4 °C, or a freezer at -20 °C, and transferred within 24 h of collection to -80 °C. Within 24 h of data acquisition, internal standard containing DSS-d6, NaN3 and D2O (65 µL) was added to the urine samples in a 1:10 ratio, and the pH was adjusted to 6.8 ( 0.1. NMR spectra were acquired as previously described.23 Analysis of these data was accomplished using targeted profiling through the use of Chenomx NMR Suite 4.6 (Chenomx, Inc., Edmonton, Canada) and concentrations determined as previously described.23 Data Analysis. Multivariate data analysis (PCA and OPLSDA) was performed on log10-transformed metabolite concentrations using SIMCA-P (version 11, Umetrics, Umeå, Sweden) with mean centering and unit variance scaling applied. Metabolite concentrations were log10-transformed prior to application of mean centering and unit variance scaling to account for the non-normal distribution of the concentration Journal of Proteome Research • Vol. 8, No. 12, 2009 5551
research articles
Slupsky et al.
Figure 1. Urinary metabolite profiles derived from patients with pneumonia caused by S. pneumoniae are different from healthy subjects, subjects with noninfectious metabolic stress, fasting subjects, and subjects with liver dysfunction. (a) PCA model (based on 61 measured metabolites) of age- and gender-matched “healthy” subjects versus those with pneumococcal pneumonia. “Healthy” subjects (9, n ) 47); bacteremic pneumococcal pneumonia (b (red), n ) 32); sputum or endotracheal tube positive S. pneumoniae cultures ([ (blue), n ) 15). (b) PCA model as in panel a with removal of diabetics (8 pneumonia patients, and 3 “healthy” subjects) from the data set. (c) OPLS-DA model based on 61 measured metabolites using all “healthy” subjects (n ) 118 (9)) and S. pneumoniae infected patients (n ) 62 (b (red))) (R2 ) 0.902; Q2 ) 0.820). (d) Loadings plot derived from OPLS-DA plot in panel c. (e) OPLS-DA prediction of two patients (yellow triangles indicated with *) with positive sputum culture, but no other evidence of lung infection. (f) OPLS-DA model based on 61 measured metabolites of an S. pneumoniae infected group (n ) 62 (9)), and noninfectious metabolic stress (n ) 56 (b (red))) (R2 ) 0.828; Q2 ) 0.655). (g) OPLS-DA model based on 61 measured metabolites of individuals with pneumococcal pneumonia (infected) (n ) 62 (9)), and a group of fasting individuals (n ) 70 (b (red))) (R2 ) 0.877; Q2 ) 0.842). (h) OPLS-DA model based on 61 measured metabolites of individuals with pneumococcal pneumonia (infected) (n ) 62 (9)), and a group with liver disease (Hepatitis C and cirrhosis) (n ) 16 (b (red))) (R2 ) 0.936; Q2 ) 0.899).
data and reduce the chance of skewed variables. All PCA and OPLS-DA models were 2-component models. Significance tests using Wilcoxon’s rank-sum test was performed using GraphPad Prism version 4.0c for MacIntosh (GraphPad Software, San Diego, CA). Significance was determined after Bonferroni correction and set at R ) 0.0082. Receiver operating curves (ROC) were generated, and area under the curve (AUC) was calculated using SPSS software version 10 (SPSS Inc., Chicago, IL).
Results Comparison of 61 metabolite concentrations measured in urine from age- and gender- matched S. pneumoniae infected (n ) 47) and noninfected (n ) 47) subjects revealed complete class distinction (R2 ) 0.582; Q2 ) 0.364) using principal components analysis (PCA) (Figure 1a). No distinction was observed between those with bacteremia (bacteria present in the blood) (n ) 32) and those with S. pneumoniae-positive sputum or respiratory secretions obtained via endotracheal tube culture (n ) 15) (Figure 1a). Removal of eight individuals with diabetes from the pneumococcal group, and three diabetics from the “healthy” group did not affect the distribution of the PCA plots (R2 ) 0.508; Q2 ) 0.376) (Figure 1b). Application of orthogonal partial least-squares-discriminant analysis (OPLSDA) to the entire data set to optimize intergroup variation resulted in clear distinction between pneumococcal patients and “healthy” subjects (Figure 1c). The three pediatric patients with pneumococcal pneumonia were equally distributed within the S. pneumoniae cohort on the OPLS-DA plot. Severity of disease and symptoms did not appear to affect the metabolite pattern in any discernible way. Both cohorts included subjects with a variety of comorbidities including asthma and chronic 5552
Journal of Proteome Research • Vol. 8, No. 12, 2009
obstructive pulmonary disease (COPD). The model parameters for the explained variation, R2, and the predictive capability, Q2, were significantly high (R2 ) 0.902; Q2 ) 0.820), indicating an excellent model. Out of a total of 61 quantified metabolites, 6 significantly decreased in concentration, and 27 significantly increased when comparing subjects infected with S. pneumoniae to uninfected subjects (Table 1). Of the 6 metabolites that decreased significantly, two are TCA cycle intermediates (citrate, and succinate), and one is involved with nicotinamide metabolistm (1-methylnicotinamide). Other metabolites that decreased in concentration are associated with food intake (levoglucosan, and trigonelline) and protein catabolism (1-methylhistidine). Metabolites that increased in concentration included amino acids (alanine, asparagine, isoleucine, leucine, lysine, serine, threonine, tryptophan, tyrosine, and valine), those involved with glycolysis (glucose, lactate), fatty acid oxidation (3-hydroxybutyrate, acetone, carnitine, acetylcarnitine), inflammation (hypoxanthine, fucose), osmolytes (myo-inositol, taurine), acetate, quinolinate, adipate, dimethylamine, and creatine. Of interest, the TCA cycle intermediates 2-oxoglutarate and fumarate appeared to increase upon pneumococcal infection. Metabolites that did not change with pneumococcal infection included creatinine, some amino acids (glycine, glutamine, histidine and pyroglutamate), 3-methylhistidine, aconitate (trans and cis), metabolites related to gut microflora (3-indoxylsulfate, 4-hydroxyphenylacetate, hippurate, formate and TMAO (trimethylamine-N-oxide)), dietary metabolites (mannitol, propylene glycol, sucrose, tartrate), and others. PLS-DA class prediction was performed on two patients with S. pneumoniae isolated from sputum, but normal chest radiographs and otherwise no evidence of infection. Both patients
research articles
Pneumococcal Pneumonia
Table 1. Metabolite Changes in Human Urine Induced by S. pneumoniae Lung Infection When Compared to Healthy Group metabolitea
% changeb
p-valuec
rankd
metabolitea
% changeb
p-valuec
rankd
Carnitine Acetylcarnitine myo-Inositol 3-Hydroxybutyrate Taurine Acetone Glucose Fumarate Acetate Leucine Hypoxanthine 2-Oxoglutarate Valine Tryptophan Alanine Lactate Isoleucine
+925 +705 +437 +315 +291 +267 +259 +248 +168 +155 +147 +135 +127 +125 +119 +116 +114
