Clinical Insights from Metagenomic Analysis of Sputum Samples fromPatients with Cystic Fibrosis

Yan Wei Lim,a Jose S. Evangelista III,b Robert Schmieder,c Barbara Bailey,d Matthew Haynes,a* Mike Furlan,a Heather Maughan,e

Robert Edwards,c,f Forest Rohwer,a Douglas Conradb

Department of Biology, San Diego State University, San Diego, California, USAa; Department of Medicine, University of California San Diego, La Jolla, California, USAb;Computational Science Research Center, San Diego State University, San Diego, California, USAc; Department of Mathematics and Statistics, San Diego State University,San Diego, California, USAd; Ronin Institute, Montclair, New Jersey, USAe; Mathematics and Computer Science Division, Argonne National Laboratory, Argonne, Illinois,USAf

As DNA sequencing becomes faster and cheaper, genomics-based approaches are being explored for their use in personalizeddiagnoses and treatments. Here, we provide a proof of principle for disease monitoring using personal metagenomic sequencingand traditional clinical microbiology by focusing on three adults with cystic fibrosis (CF). The CF lung is a dynamic environmentthat hosts a complex ecosystem composed of bacteria, viruses, and fungi that can vary in space and time. Not surprisingly, themicrobiome data from the induced sputum samples we collected revealed a significant amount of species diversity not seen inroutine clinical laboratory cultures. The relative abundances of several species changed as clinical treatment was altered, en-abling the identification of the climax and attack communities that were proposed in an earlier work. All patient microbiomesencoded a diversity of mechanisms to resist antibiotics, consistent with the characteristics of multidrug-resistant microbial com-munities that are commonly observed in CF patients. The metabolic potentials of these communities differed by the health sta-tus and recovery route of each patient. Thus, this pilot study provides an example of how metagenomic data might be used withclinical assessments for the development of treatments tailored to individual patients.

Adecade of advancements in sequencing technology and bioin-formatics applications is shuttling in a new era of personal-

ized medicine. Pathogens can be rapidly characterized during out-breaks (1, 2), and cancer patients can receive personalizeddiagnoses and treatments (3, 4). Despite these significant ad-vances, newer technologies have yet to be used in routine clinicalmicrobiology practice. Treating polymicrobial infections will alsorequire a personalized approach, because the taxonomic identitiesand functional characteristics of microbial communities are oftenpatient specific (5). Here, we moved toward this goal by usingmetagenomics to monitor complex pulmonary infections in pa-tients with cystic fibrosis (CF).

Cystic fibrosis (CF) is a genetic disease affecting 70,000 indi-viduals worldwide (per the Cystic Fibrosis Foundation [see www.cff.org]) and results from mutations in the gene that encodes thecystic fibrosis transmembrane conductance regulator (CFTR) (6).These mutations result in altered transepithelial ion transport,leading to a dysfunctional mucus layer overlying epithelial cells inthe respiratory and gastrointestinal tracts (7). In the lungs, themucociliary clearance mechanism is impaired, resulting inchronic polymicrobial infections of the airway. The associatedacute inflammatory and adaptive immune responses lead to abreakdown in the integrity of the airway wall, progressive gas ex-change abnormalities, and respiratory failure in many patients.Early in life, culture-based assessments indicate that patients areusually infected with Staphylococcus aureus, Haemophilus influen-zae, and Pseudomonas aeruginosa. In the more advanced stages ofdisease, P. aeruginosa dominates, along with Staphylococcus spp.,Stenotrophomonas spp., and Achromobacter spp. Current treat-ments focus on controlling inflammation, the frequent use ofbroad-spectrum antibiotics, and physically clearing the airwaybiofilm by augmenting airway clearance.

Our working model describes two functional microbial com-

munities in CF patients, the climax and attack communities (8).Climax communities are typically bacterial and fungal popula-tions that are stable over time and are inherently resistant to anti-biotic therapy. They elicit prolonged innate and adaptive immuneresponses and are adapted to particular niches. The attack com-munities are predominantly newly acquired bacterial and viralpopulations that elicit strong innate immune responses and fre-quently trigger acute pulmonary exacerbations and are thus tar-gets of therapy. Attack communities dominate earlier in CF airwaydisease, when airway remodeling and damage are minimal (8). Inthe advanced stages of disease, the predominant climax popula-tions are thought to persist, while the attack communities fluctu-ate in accordance with exacerbation and treatment events. Theclimax and attack communities need not differ taxonomically,because it is the functional capabilities of a community that deter-mine how it affects the health of a patient. We hypothesize thatcommunity metabolic functions respond to perturbations and

Received 15 August 2013 Returned for modification 18 September 2013Accepted 3 November 2013

Published ahead of print 20 November 2013

Editor: B. A. Forbes

Address correspondence to Yan Wei Lim, ylim@rohan.sdsu.edu.

* Present address: Matthew Haynes, DOE Joint Genome Institute, Walnut Creek,California, USA.

Supplemental material for this article may be found at http://dx.doi.org/10.1128/JCM.02204-13.

Copyright © 2014, American Society for Microbiology. All Rights Reserved.

doi:10.1128/JCM.02204-13

The authors have paid a fee to allow immediate free access to this article.

February 2014 Volume 52 Number 2 Journal of Clinical Microbiology p. 425– 437 jcm.asm.org 425

on August 18, 2018 by guest

http://jcm.asm

.org/D

ownloaded from

that these responses can be used to identify the attack and climaxcommunities within each patient.

Metagenomics sequences total community DNA from a par-ticular source (e.g., sputum) to identify the taxonomic makeupand functional capabilities of the resident populations. It differsfrom community analyses that sequence only the 16S rRNA genebecause all genes are sequenced, not just those used for making ataxonomic identification. Thus, information on communityfunction is typically only obtainable with metagenomic sequenc-ing, although some functions can be predicted based on taxonomyalone (e.g., methanogenesis). Neither approach requires culturingbacteria, and when applied to CF patients, both methods haverevealed the microbial diversity and community complexity inairways to be unexpectedly high (5, 9). Furthermore, these se-quence-based technologies have demonstrated that bacterial di-versity decreases during treatment with broad-spectrum antibiot-ics and as the disease advances over longer periods of time (10, 11).Although microbial communities can differ across the regions ofthe lung and between simultaneous samplings, even one finding ofa gene encoding antibiotic resistance can have important implica-tions for treatment.

Here, we report a pilot study that focused on obtaining a largeamount of sequence data from microbial communities sampledlongitudinally from the lungs of a few CF patients. The goal was todetermine the types of information that can be obtained fromsuch sequence data and to frame this information in the context ofclinical treatments used and measured antibiotic resistance. AsDNA sequencing becomes more affordable and sequence analysisbecomes more efficient, this approach can be further developed toassist clinical decision making and formulation of personalizedtherapies. Here, we demonstrate an early attempt to use a meta-genomic approach for probing changes in microbial communityfunctions in individual patients over time to identify the candidatechanges that most drastically affect the patient. We discuss theseresults in the context of climax and attack communities for a thor-ough understanding of CF disease ecology.

MATERIALS AND METHODSTen sputum samples and clinical data were collected from three adult CFpatients at the Adult Cystic Fibrosis Clinic at the University of CaliforniaSan Diego Medical Center. The collections were made in accordance withthe University of California institutional review board (HRPP no.081500) and the San Diego State University institutional review board(SDSU IRB no. 2121).

The study subjects were selected based on eligibility criteria that in-cluded all of the following: (i) a diagnosis of CF, i.e., two known mutationsin the CFTR gene and/or an abnormally elevated sweat chloride test result,(ii) an increase in respiratory symptoms associated with CF pulmonaryexacerbations (see the supplemental material), and (iii) a drop in theforced expiratory volume in 1 s (FEV1) of �15% compared to the pa-tient’s best FEV1 in the past 12 months. Using these criteria, 15 patientswere initially recruited and screened for inclusion in this study, resultingin the collection and processing of 54 total samples. However, samplesfrom the patients who dropped out during the study period were notsequenced because we preferred to focus our limited resources on patientsfor whom we had more complete longitudinal sampling and clinical in-formation. Of those patients who remained, the three patients discussedhere are representative of different levels of underlying lung function andresponses to treatment.

The treating physicians determined the selection and duration of an-tibiotic therapy and the frequency of follow-up examinations. Sputumsamples were obtained at the following clinical time points: onset of an

exacerbation prior to the initiation of antibiotic therapy (Ex), within 24 hprior to a change in antibiotic therapy (Tr), within 12 h posttreatment,i.e., cessation of antibiotic therapy (Pt), and about 4 weeks after complet-ing antibiotics when the patient was in the stable state (St), defined asachieving a maintenance level of respiratory symptoms without the needto alter the outpatient therapies. The Cystic Fibrosis Questionnaire-Re-vised (CFQR) evaluation and the UCSD Shortness of Breath (SOB) ques-tionnaire were used to obtain additional data at the time of each collec-tion.

During sample collection, a sterile saline solution (60 ml) was used asa mouth rinse to minimize contamination by oral microbes. The sputumsamples were then collected over a thirty-minute time period after theinhalation of four milliliters of 7% hypertonic saline via a Pari LC Plusnebulizer. The samples were processed immediately, as described previ-ously (5) (Fig. 1).

In brief, the sputum samples were homogenized with a syringe andthen aliquoted for the separate isolation of viral particles and microbialcells. Viral samples (i.e., the virome) were diluted and treated with dithio-threitol (DTT) to dissociate the mucus and then passed through a0.45-�m filter to remove large particles. The viral particles were thenisolated and concentrated using cesium chloride density ultracentrif-ugation. Viral DNA was extracted using the formamide-cetyltrimeth-ylammonium bromide (CTAB)/phenol-chloroform method (5). Micro-bial samples (i.e., the microbiome) were treated with �-mercaptoethanolto dissociate the mucus. The cells were repeatedly washed with sterilizeddeionized water to lyse the human cells and then treated with DNase toremove extracellular DNA (e.g., human DNA and that associated withbiofilms [5]). Microbial DNA was extracted using the NucleoSpin tissuekit with the Gram-positive variation, which includes a lysozyme treat-ment.

All samples were sequenced using Roche/454 pyrosequencing withGS-FLX Titanium chemistry. All the data sets were preprocessed as pre-viously described (5). Duplicate reads and reads of low quality were re-moved using PrinSeq (12). DeconSeq (13) was used to screen for andexclude human-derived sequences from the microbiome data. The taxon-omies of the resultant sequences were identified using a BLASTn searchagainst the NCBI nucleotide (NT) database. Sequences were removed ifthey were assigned to the phylum Chordata or any synthetic/vector se-quence (see Table S1 in the supplemental material). The viromes werefurther annotated using a tBLASTx search against an in-house viral data-base. The abundance values were normalized based on the total number ofreads per metagenome. All metagenomes were additionally annotatedusing the KEGG database (14) and analyzed using the HUMAnN pipeline(15). The normalized relative abundance values were used for subsequentprincipal component analysis (PCA). The antibiotic resistance potential

FIG 1 Work flow for the preparation of CF sputum samples for microbiomeand virome sequencing. VLP, virus-like particle; euk, eukaryotic.

Lim et al.

426 jcm.asm.org Journal of Clinical Microbiology

on August 18, 2018 by guest

http://jcm.asm

.org/D

ownloaded from

of each microbiome was identified by comparing the data with (i) theantibiotic resistance database (ARDB) (16), which contains 23,137 anti-biotic resistance-associated sequences, using BLASTx with a threshold of40% identity over �60% of the query sequence, and (ii) an up-to-datemanually curated macrolide and aminoglycoside resistance database fromthe UncovAR pipeline (17). The abundance of each resistance annotationwas normalized by the number of reads in each metagenome andweighted by the length of each gene and the total number of base pairs inthe respective database.

For additional details on database generation, content, and BLASTparameters, see the supplemental material.

Nucleotide sequence accession number. All sequence data can beretrieved from the NCBI Sequence Read Archive (SRA) under acces-sion no. SRP009392 (see http://trace.ddbj.nig.ac.jp/DRASearch/study?acc�SRP009392).

RESULTSSample and data collection. Sputum was induced and collectedfrom three adult CF patients (median age, 36 years) followingmouth rinsing with a sterile saline solution (Table 1). BaselineFEV1 values, defined by the best FEV1 value obtained for eachpatient in the 12 months preceding the first sample collection,ranged from 1.16 liters (27% predicted) to 4.15 liters (89% pre-dicted). The first samples were collected when each patient pre-sented with a severe CF pulmonary exacerbation. The patientsreceived standard care for an acute pulmonary exacerbation, in-cluding bronchodilators, chest physiotherapy, and systemic anti-biotics. The attending physician selected the antibiotic type andduration of intravenous treatment. Up to four sputum sampleswere collected from each patient based on the clinical course thatfollowed the initiation of their treatment, resulting in a total of 10samples with matched exacerbation/posttreatment sample pairsfrom all three patients.

Clinical data were collected from the UCSD adult CF clinic(Tables 1 and 2). These data included culture-based microbiologyassessments conducted 100 days before and after collection of thefirst sample. The patient-reported outcome surveys were collectedat the time of sampling (see Materials and Methods). For eachpatient, these clinical data were combined with the metagenomedata to study individual cases; the data from all patients were alsocombined for a comparative analysis.

Metabolism and resistance in CF microbiomes. Microbialmetabolism and antibiotic resistance are compelling indicators ofmicrobial community function because they reveal which path-ways facilitate microbial colonization and persistence in the CFlung. The metabolic pathways present in each microbiome wereinvestigated with the goal of using this information to eventuallydevelop tools for identifying important biomarkers of the climaxand attack communities. Because these communities are likely tobe patient specific and therefore have important implications forpersonalized diagnosis and treatment, the differences in commu-nity metabolism were examined. A total of 222 metabolic path-ways were identified from all patient data sets using BLASTx com-parison against the KEGG database, as described in Materials andMethods. For pattern exploration purposes, principal componentanalysis (PCA) was used for dimension reduction. Twenty meta-bolic pathways with the greatest variance between patient micro-biomes were ordered (see Materials and Methods) and used toidentify the functions that varied most across the microbiomes(Fig. 2). The metabolic profiles are shown separately for each pa-tient in Fig. S1 to S3 in the supplemental material.

As resistance against antibiotic treatment remains one of themain challenges in the treatment of CF pulmonary infections, theabundances of genes whose products or mutations are known toconfer resistance to antibiotics were also determined (Fig. 3A andB; see also Table S3 in the supplemental material). One contribu-tion of this pilot study is the examination of whether predictedantibiotic resistance profiles fluctuate through time, or whether aconsistent increase (or decrease) in community resistance is ob-served. For clinicians to find metagenome data useful, it will beimportant to understand how quickly antibiotic resistance mightchange in the microbial community.

Because the majority of viruses found in the CF lung are bac-teriophages (18, 19), which are known to transfer genes betweenmicrobial hosts, the potential for the exchange of antibiotic resis-tance genes was also evaluated using the viromes corresponding toeach microbiome (Fig. 3C; see also Table S4 supplemental mate-rial). Bacterial contaminating sequences in the viromes were min-imal, based on the low abundance of 16S rRNA genes in theviromes (�0.05%).

TABLE 1 Information on patient samples

Patient no. (gender,age [yr])

CFTRgenotype

Samplename

Patient health statusat time of sampling

Time scale(day)a

FEV1 (liters[% predicted])b Treatment received

CF6 (female, 39) �F508/Q372Q CF6-A-Ex Onset of exacerbation 0 1.91 (57) Colistin, ceftazidimeCF6-B-Tr Treatment 12 2.03 (60) Ciprofloxacin, aztreonamCF6-C-Pt Posttreatment 17 2.06 (61) Stopped all antibioticsCF6-D-St Stable 46 2.07 (61) Not on antibiotics

CF7 (male, 36) �F508/�F508 CF7-A-Ex Onset of exacerbation 0 0.87 (21) Tobramycin, ceftazidime,trimethoprim-sulfamethoxazole

CF7-B-Tr Treatment 20 0.80 (19) Piperacillin-tazobactam,trimethoprim-sulfamethoxazole

CF7 C-Tr Treatment change 27 0.82 (19) Piperacillin-tazobactam,trimethoprim-sulfamethoxazole

CF7-D-Pt Posttreatment 37 0.92 (22) Stopped antibiotic treatment, prednisonewas added on day 33

CF8 (male, 26) �F508/�F508 CF8-A-Ex Onset of exacerbation 0 3.39 (73) Meropenem, tobramycinCF8-B-Pt Posttreatment 17 4.15 (89) Meropenem, tobramycin

a Time scale starts at day 0 during the onset of an exacerbation, which is when the first sample was collected. The subsequent numbers indicate the days after the first sample wascollected.b FEV1 is measured as the forced expiratory volume in 1 s (% predicted).

Clinical Insights from Metagenomics

February 2014 Volume 52 Number 2 jcm.asm.org 427

on August 18, 2018 by guest

http://jcm.asm

.org/D

ownloaded from

Individual case studies. (i) Patient CF6. The clinical and sam-ple information for patient CF6 are presented in Table 1 and Fig. 4.The patient presented with dyspnea, increased cough, sputumproduction, and a 19% drop in her FEV1 during her outpatientvisit. The baseline FEV1 was recorded as 2.36 liters (69% pre-dicted). She initially had intravenous therapy that included 11days of colistin and ceftazidime, but due to a suboptimal response,her therapy was changed to 5 days of ciprofloxacin and aztreonam.

The first sample (CF6-A-Ex) was collected on day 0, before theadministration of intravenous antibiotics. The second sample(CF6-B-Tr) was collected on day 12, before the change in antibi-otics (Table 1). The third sample (CF6-C-Pt) was taken 5 dayslater (day 17), when the patient’s presenting symptoms resolvedand her score on the UCSD Shortness of Breath questionnaireimproved. At this point, her FEV1 improved to 2.06 liters (61%predicted), which was approximately 8% lower than her baselineFEV1 value. A month following the end of therapy, the patient wasclinically stable when the fourth sample (CF6-D-St) was collected.At this time, the patient-reported outcome measures reflected inthe UCSD SOB and CFQR cumulative scores showed significantimprovement even though the FEV1 was essentially unchangedand remained below the baseline at 2.07 liters (61% predicted)(see Table S2 in the supplemental material). Importantly, therewas a significant improvement in the respiratory domain of theCFQR evaluation (see Table S2 in the supplemental material). Thepatient was clinically defined as an intermediate responder basedon her recovery and responses to therapy.

Clinical culturing revealed growth of the fungi Candida albi-

cans and Scedosporium apiospermum, as well as extended-spec-trum �-lactamase (ESBL) Escherichia coli that exhibited sensitivi-ties to various antibiotics during the study period (Table 2).Metagenomic analyses of the microbial communities (micro-biomes) were consistent with the clinical culturing, showing ahigh abundance of E coli and a low abundance of P. aeruginosa(Fig. 4). However, the microbiomes also included species that arenot typically incorporated into clinical culturing protocols, suchas Streptococcus spp. and Rothia mucilaginosa, largely because theyare considered to be oral contaminants or benign respiratory mi-crobes (Fig. 4).

The relative abundance of E. coli decreased during the course ofantibiotic therapy and then increased slightly when the patient’shealth was stable. In contrast, the relative abundances of Strepto-coccus spp. and R. mucilaginosa increased during the course oftherapy. One month following the completion of intravenous an-tibiotic therapy, R. mucilaginosa and E. coli were the dominantspecies in microbiome (see Fig. 4, time point D). E. coli, Strepto-coccus spp., and R. mucilaginosa were the numerically dominantbacteria in this patient, with P. aeruginosa being detectable but atlower abundances at all time points (�0.1% at A, B, and C and1.25% in D).

Historically and during the course of sampling, this patient wasextensively exposed to a diversity of antibiotics, including amino-glycosides, �-lactams, fluoroquinolones, macrolides, and variouspolypeptides (Table 1). Hence, the microbial community was ex-pected to have evolved resistances to these major groups of anti-biotics. This was confirmed by clinical lab tests (Table 2 and Fig. 4)

TABLE 2 Bacterial culture data from the hospital’s clinical lab

Patient no. (gender)Time scale(day)a Culture-positive organism Antibiotic resistance profile

CF6 (female) �57 Escherichia coli (ESBL) Ampicillin, ampicillin-sulbactam, cefazolin, ceftazidime, cefotaxime,cefepime, cefuroxime, ciprofloxacin, gentamicin, piperacillin-tazobactam, tobramycin

�6 E. coli (ESBL) Ampicillin, ampicillin-sulbactam, cefazolin, ceftazidime, cefotaxime,cefepime, cefuroxime, ciprofloxacin, gentamicin, piperacillin-tazobactam, trimethoprim-sulfamethoxazole

�82 E. coli (ESBL) Ampicillin, ampicillin-sulbactam, cefazolin, ceftazidime, cefotaxime,cefepime, cefuroxime, ciprofloxacin, gentamicin, piperacillin-tazobactam, trimethoprim-sulfamethoxazole

CF7 (male) �38 Pseudomonas aeruginosa (mucoid) Amikacin, ciprofloxacin, meropenem, tobramycinStenotrophomonas maltophilia Amikacin, ciprofloxacin, gentamicin, tobramycin

�3 P. aeruginosa (mucoid) Amikacin, ciprofloxacin, gentamicin, trimethoprim-sulfamethoxazoleS. maltophilia Amikacin, ciprofloxacin, gentamicin, sulfamethoxazole-trimethoprim,

meropenem, piperacillin, tobramycin, ceftazidime�13 P. aeruginosa Ciprofloxacin, sulfamethoxazole-trimethoprim, meropenem�86 S. maltophilia Ceftazidime

P. aeruginosa Amikacin, ciprofloxacin, gentamicin, tobramycin,trimethoprim-sulfamethoxazole

CF8 (male) �36 Streptococcus group C NAb

P. aeruginosa (mucoid), strain 1P. aeruginosa (mucoid), strain 2 Amikacin, gentamicin, trimethoprim-sulfamethoxazole

0 Streptococcus group C NAP. aeruginosa (mucoid) Amikacin, ciprofloxacin, gentamicin, tobramycin,

trimethoprim-sulfamethoxazole10 S. maltophilia Ceftazidime

P. aeruginosa (mucoid) Amikacin, ciprofloxacin, gentamicin, tobramycin,trimethoprim-sulfamethoxazole

a The timeline corresponds to the date (day 0) when the first sputum sample was taken during an exacerbation of symptoms.b NA, not applicable.

Lim et al.

428 jcm.asm.org Journal of Clinical Microbiology

on August 18, 2018 by guest

http://jcm.asm

.org/D

ownloaded from

and predicted by the metagenomics analysis (Fig. 3A; see also Ta-ble S3 in the supplemental material).

The metagenomic data suggest that the microbes in this com-munity encode a plethora of antibiotic resistance mechanisms:multidrug resistance efflux pumps, �-lactamases, and various en-zymes that confer resistance (see Table S3 in the supplementalmaterial). The abundances of two known resistance genes, arnAand bla, increased following treatment with colistin and ceftazi-dime, respectively. Two antifolate genes were present: dfrA, whichencodes group A drug-insensitive dihydrofolate reductase for tri-methoprim resistance, and sul, which encodes sulfonamide-resis-tant dihydropteroate synthase for sulfonamide (sulfamethoxa-zole) resistance. Although the patient was not treated with anymacrolides within the year prior to this study, various macrolideresistance mechanisms were found throughout the microbiome(Fig. 3A; see also Table S3 in the supplemental material).

The PCA based on the potential microbial metabolic func-

tions showed that the characteristics of patient CF6 differ con-siderably from those of the other patients, especially during theexacerbation and stable states (Fig. 2). The separation of theexacerbation sample (A) was largely driven by the presence ofgenes encoding streptomycin biosynthesis (KEGG accessionno. ko00521 [http://www.genome.jp/kegg/]) and phospho-transferases (ko02060), whereas separation of the stable sample(D) was driven by the presence of genes that encode taurine andhypotaurine metabolism (ko00430), folate biosynthesis (ko00790), asulfur relay system (ko04122), D-glutamine and D-glutamate me-tabolism (ko00471), and valine, leucine, and isoleucine biosyn-thesis (ko00290). The first and second principal components forthe treatment and posttreatment samples were very similar for allpatients, indicating that these microbial communities had similarmetabolic potentials at the two time points.

Comparing different time points for patient CF6 (see Fig. S1 inthe supplemental material), drug metabolism was one of the main

FIG 2 A comparison of metabolic pathways between microbiomes using principal component analysis (PCA) of the 20 metabolic pathways that varied the mostbetween microbiomes. The bottom panel presents a closeup view of the squared region in the upper panel. Patient CF6 is represented by circles, patient CF7 isrepresented by squares, and patient CF8 is represented by triangles. The colors inside the shapes represent the health status of each patient as shown in the key.

Clinical Insights from Metagenomics

February 2014 Volume 52 Number 2 jcm.asm.org 429

on August 18, 2018 by guest

http://jcm.asm

.org/D

ownloaded from

FIG 3 Abundances of antibiotic resistance genes based on the Antibiotic Resistance Database (ARDB) (A) and the program UncovAR, which predicts resistanceto aminoglycoside and macrolide antibiotics (B). (C) The antibiotic resistance gene profiles of the viromes based on BLASTx comparison against the ARDB. Allabundances were normalized by metagenome and gene size and were weighted by database size.

Lim et al.

430 jcm.asm.org Journal of Clinical Microbiology

on August 18, 2018 by guest

http://jcm.asm

.org/D

ownloaded from

metabolic pathways that drove the separation of the treatment (B)and posttreatment (C) samples. Separation of the stable samplewas driven by several core metabolic pathways, which may indi-cate a stable climax community in the lungs of this patient.

(ii) Patient CF7. The clinical and sample information for pa-tient CF7 is presented in Table 1 and Fig. 5. The baseline FEV1 wasrecorded as 1.16 liters (27% predicted). This patient presentedwith increased dyspnea and sputum production as well as a 25%drop in his FEV1, which was recorded as 0.87 liter (21% pre-dicted), prompting treatment with different combinations of in-travenous antibiotics, including tobramycin, ceftazidime, pipera-cillin-tazobactam, and trimethoprim-sulfamethoxazole. The firstsample (CF7-A-Ex) was collected before the administration oftherapy. Sample CF7-B-Tr was collected 20 days after the begin-ning of treatment with tobramycin and ceftazidime, and sampleCF7-C-Tr was collected 7 days after stopping ceftazidime andprior to the initiation of piperacillin-tazobactam and tri-methoprim-sulfamethoxazole therapy due to the lack of clinicaland physiological response. The fourth sample (CF7-D-Pt) wastaken 10 days later upon the completion of treatment. However,the patient had not completely improved physiologically, as heshowed no resolution of the initial respiratory symptoms and aworsening of shortness of breath (see Table S2 in the supplementalmaterial). The FEV1 improved to 0.92 liter (22% predicted) butremained 20% lower than the baseline FEV1 value. The respira-

tory domain on the CFQR improved by a value greater than theminimal clinically important difference (MCID) of 5 (see Table S2in the supplemental material), suggesting that the patient mighthave improved slightly. The patient underwent a lung transplantabout 3 months following the last sample we collected.

Clinical culturing revealed growth of the fungi C. albicans andAspergillus fumigatus. S. maltophilia and mucoid P. aeruginosawere also cultured, and all had various patterns of antibiotic sus-ceptibility (Table 2 and Fig. 5). Metagenomic analysis at all timepoints showed an overall high relative abundance of S. maltophiliathat ranged from 41% to 90%, whereas P. aeruginosa was relativelyrare (�1%). R. mucilaginosa, Rothia dentocariosa, Streptococcusspp., and Prevotella melaninogenica were highly abundant duringthe onset of symptom exacerbation and following the first roundof unresponsive therapy. However, 1 month following the onset ofsymptom exacerbation, during which the patient underwent in-tense treatment with a combination of antibiotics, S. maltophiliarepopulated the lung community, and the overall bacterial diver-sity decreased. S. maltophilia was the key player within the micro-bial community in this patient, and clinical testing indicated thatit was highly resistant to all major groups of antibiotics, includingaminoglycosides, macrolides, �-lactamases, and fluoroquinolo-nes (Fig. 3A and 5; see also Table S3 in the supplemental material).The metagenome data suggest that the mechanisms of antibioticresistance in this microbial community were multidrug resistance

FIG 4 An overview of clinical and metagenomic data for patient CF6. This patient was clinically defined as an intermediate responder. The FEV1% is illustratedacross a 550-day period, and a red diamond indicates the baseline FEV1%. The resistance profile line shows the results from laboratory resistance tests on clinicallycultured microbes, obtained at the time point immediately below their placement. The first sample was collected during the onset of a clinically definedexacerbation, time point 0. The medications prescribed during each chronic and acute round of therapy are shown in the time line. The length of the black linecorresponds to the duration of therapy. Acute therapy is represented as follows: M, meropenem; Ctx, ceftriaxone; Ctz, ceftazidime; PT, piperacillin-tazobactam;Cp, ciprofloxacin; T, tobramycin; G, gentamicin; Pred, prednisone; Imp, imipenem; Azt, aztreonam; Lz, linezolid; Col, colistin; Ampho, inhaled amphotericin;and Cayston, inhaled aztreonam.

Clinical Insights from Metagenomics

February 2014 Volume 52 Number 2 jcm.asm.org 431

on August 18, 2018 by guest

http://jcm.asm

.org/D

ownloaded from

efflux pumps, a protein that prevents tetracycline from inhibitingthe ribosome, and various enzymes and transporters that conferresistance (Fig. 3). As seen in patient CF6, even though macrolideswere not reported in his recent medical history, various macro-lide-specific resistance mechanisms were found (Fig. 3A; see alsoTable S3 in the supplemental material). Further comparison of thedata with aminoglycoside and macrolide resistance genes in Un-covAR revealed that S. maltophilia likely relies on efflux pumps(e.g., gene acr) to purge antibiotics from the cell (Fig. 3B).

PCA of metabolic pathways showed that the metabolic profilesof the microbial communities in patient CF7 were similar overtime (Fig. 2; see also Table S3 in the supplemental material). Thislimited change in the patient’s microbial taxonomical and func-tional profiles was consistent with the patient’s unresponsive clin-ical status. These data also suggest that a particular set of metabolicfunctions (Fig. 3) may have been responsible for the persistence ofhis unresponsive climax microbial community. These metabolicfunctions include the synthesis and degradation of ketone bodies(ko00072), carbon fixation pathways in prokaryotes (ko00720),drug metabolism pathways (ko00983), and riboflavin metabolism(ko00740). Further examination within the drug metabolismpathway revealed the presence of the arylamine N-acetyltrans-ferase (NAT) gene involved in isoniazid metabolism, which isknown to occur in E. coli (20, 21). Isoniazid is commonly used totreat tuberculosis, but it was not used in this patient, according tohis medical history. It is not known whether NAT is capable ofmetabolizing any drugs that were prescribed to patient CF7.

(iii) Patient CF8. The clinical and sample information for pa-tient CF8 is presented in Table 1 and Fig. 6. The baseline FEV1 was4.15 liters (89% predicted). The patient was admitted to the hos-pital for increased cough, dyspnea, and sputum production, andhe had an 18% drop in his FEV1 (Table 1). The patient was startedon a combination of tobramycin and meropenem for a total of 16days (Table 1). The first sample (CF8-A-Ex) was collected beforethe administration of therapy, and the second sample (CF8-B-Pt)was collected 17 days later when the patient completed treatment.At the end of therapy, the patient reported resolution of his initialrespiratory symptoms, which was confirmed in his reported out-comes assessed by both the CFQR respiratory and Shortness ofBreath scores (Fig. 6; see also Table S2 in the supplemental mate-rial). The FEV1 improved to the baseline of 4.15 liters (89% pre-dicted). Based on the patient’s recovery and responses to therapy,patient CF8 was considered a responder.

Clinical culturing revealed growth of mucoid P. aeruginosa,Streptococcus group C, and S. maltophilia during the period whenthe samples were collected (Table 2). P. aeruginosa demonstratedvarious antibiotic susceptibility patterns (Fig. 6 and Table 2).Streptococcus was considered an oral contaminant in the clinicallab, and therefore its antibiotic susceptibilities were not tested.Metagenomic analysis showed a high diversity of bacteria in themicrobiome of patient CF8, particularly in sample A (Fig. 6). Themost abundant bacteria were P. aeruginosa, Streptococcus spp.,Rothia spp., and the anaerobes P. melaninogenica, Veillonella par-vula, and Fusobacterium nucleatum. In sample B, which followed

FIG 5 An overview of clinical and metagenomic data for patient CF7. This patient was clinically defined as a nonresponder. The FEV1% is illustrated across a520-day period, and a red diamond indicates the baseline FEV1%; the spike at the end of the FEV1% graph indicates the effect of lung transplantation. Theresistance profile line shows the results from laboratory resistance tests on clinically cultured microbes, obtained at the time point immediately below theirplacement. The first sample was collected during the onset of a clinically defined exacerbation, time point 0. The medications prescribed during each chronic andacute round of therapy are shown in the time line. The length of the black line corresponds to the duration of therapy. Acute therapy is represented as follows:M, meropenem; Ctx, ceftriaxone; Ctz, ceftazidime; PT, piperacillin-tazobactam; Cp, ciprofloxacin; T, tobramycin; G, gentamicin; Pred, prednisone; Imp,imipenem; Azt, aztreonam; Lz, linezolid; Col, colistin; Ampho, inhaled amphotericin; and Cayston, inhaled aztreonam.

Lim et al.

432 jcm.asm.org Journal of Clinical Microbiology

on August 18, 2018 by guest

http://jcm.asm

.org/D

ownloaded from

antibiotic treatment, most of the Streptococcus spp. and anaerobeswere eliminated, and the microbiome was dominated by P. aerugi-nosa, R. mucilaginosa, and Lactobacillus spp.

The patient was treated with an aminoglycoside (tobramycin)and a �-lactam (meropenem) during the course of sampling (Ta-ble 1). Treatment with macrolides or fluoroquinolones was notreported during the 600-day medical history. However, the met-agenomic data predicted resistance to several groups of antibiotics(Fig. 3A and 6; see also Table S3 in the supplemental material),including aminoglycosides, �-lactams, fluoroquinolones, andmacrolides. The antibiotic resistance mechanisms detected in themicrobiome data included �-lactamases, multidrug efflux pumps,the same ribosomal protection protein identified in patient CF7,and various enzymes that confer resistance (see Table S3 in thesupplemental material). A high abundance of genes that conferresistance to tetracycline and macrolides were detected in the ex-acerbation sample (CF8-A-Ex), but their abundances decreasedupon treatment with meropenem and tobramycin. However, theabundances of the �-lactamase genes (conferring resistance to�-lactams) and those encoding multidrug resistance efflux pumpsincreased posttreatment (Fig. 3C).

PCA indicated that the metabolic pathways present in the mi-crobial communities found in the two samples were quite similar.Interestingly, the exacerbation sample was closer to the diseasestate sample from patient CF7, whereas the posttreatment samplewas similar to the posttreatment sample from patient CF6, who wasan intermediate responder. This posttreatment community wascharacterized by C5-branched dibasic acid metabolism, protein ex-

port, and selenocompound metabolism (Fig. 2). The C5-brancheddibasic acid metabolism belongs to the carbohydrate metabolismsuperclass and is known to provide an alternative source of carbonand energy. A comparison of the top 20 representative metabolicpathways that differed between the two samples from patient CF8(see Fig. S3 in the supplemental material) indicated that folatebiosynthesis, glycan degradation, and naphthalene and dioxindegradation were responsible for distinguishing the metagenomesfrom these samples.

DISCUSSION

The CF lung is a complex ecosystem hosting a wide range of bac-teria, viruses, and fungi that interact and collectively alter hostimmune responses. This dynamic ecosystem drives the short- andlong-term clinical outcomes of CF patients. To survive, these air-way microbes must adapt their intermediary metabolisms to theavailable resources to resist therapy and the host immune re-sponses.

The combination of clinical information with metagenomicanalysis (Fig. 4 to 6) provided valuable insights into the potentialuse of sequencing in clinical settings. The patients were chosenfrom a larger cohort based on their distinct responses to treat-ments and underlying levels of lung function. Patient CF6 wascharacterized by moderate lung disease but responded to the ther-apeutic plan, patient CF7 was characterized by severe end-stagedisease and did not respond to therapy, and patient CF8 was char-acterized by mild pulmonary disease and completely responded to

FIG 6 An overview of clinical and metagenomic data in patient CF8. This patient was clinically defined as a responder. The FEV1% was illustrated across a600-day period, and a red diamond indicates the baseline FEV1%. The resistance profile line shows the results from laboratory resistance tests on clinicallycultured microbes, obtained at the time point immediately below their placement. The first sample was collected during the onset of a clinically definedexacerbation at time point 0. The medications prescribed during each chronic and acute round of therapy are shown in the time line. The length of the black linecorresponds to the duration of therapy. Acute therapy is represented as follows: M, meropenem; Ctx, ceftriaxone; Ctz, ceftazidime; PT, piperacillin-tazobactam;Cp, ciprofloxacin; T, tobramycin; G, gentamicin; Pred, prednisone; Imp, imipenem; Azt, aztreonam; Lz, linezolid; Col, colistin; Ampho, inhaled amphotericin;and Cayston, inhaled aztreonam.

Clinical Insights from Metagenomics

February 2014 Volume 52 Number 2 jcm.asm.org 433

on August 18, 2018 by guest

http://jcm.asm

.org/D

ownloaded from

therapy. In a case report format, we specifically demonstrated thefollowing.

(i) Each patient hosts a unique polymicrobial community.Metagenomics detected a high level of species diversity and com-munity dynamicity in our patients that is not seen in routinecultures. Semiquantitative measurements of individual speciesshowed that their relative abundances fluctuated temporally.Whether these fluctuations are due to sampling or communitydynamics remains to be determined.

(ii) Microbial community metabolism differed between pa-tients and within a patient over time. Predictions of communitymetabolism suggested that this too is dynamic, changing over timeand variable between patients that differed in their health statuses.Although we cannot attribute these fluctuations to patient char-acteristics alone, as they may be due to sampling, it is clear thatthere is abundant variation for investigation in subsequentstudies.

(iii) An unappreciated diversity of genes encoding antibioticresistance pathways was detected from the metagenomes in allpatients. The fact that these genes are often found in bacterio-phage genomes makes it likely that they can be transferred hori-zontally between community members. Clinical decision makingmay benefit from such information in order to understandwhether it is best to target individual bacteria or individual genesand functions.

The following is a discussion of each of these points.Unique polymicrobial communities. The results showed that

the numerically dominant bacterial species varied considerablybetween patients. The microbiome of patient CF6 was representedmainly by E. coli, R. mucilaginosa, Streptococcus parasanguinis, andKlebsiella pneumoniae, that of patient CF7 was represented by S.maltophilia, R. mucilaginosa, and Streptococcus spp., and that ofpatient CF8 was represented by P. aeruginosa and R. mucilaginosa.A previous study showed that lung bacteria are most likely ac-quired from a patient’s living environment and that the microbialcommunity fluctuates in respond to therapeutic perturbations(5). The results presented here are consistent with these previousfindings and extend the list of microbes known to be associatedwith the CF lung. For example, patient CF8 had a high abundanceof Lactobacillus spp., which are commonly found in the oral andgastrointestinal (GI) tract; its presence in the lung of immuno-compromised individuals has been associated with life-threaten-ing pulmonary cases (22). Lactobacillus rhamnosus can be intro-duced into the GI tract through the consumption of yogurt andother dairy products (23). Even though it is beneficial in mostcases, it has also been associated with endocarditis (24), pulmo-nary abscess, and pleuritis (25). Lactobacillus casei was found tohave a protective role in the lung of a mouse model during Strep-tococcus pneumoniae infection (26). In patient CF8, the presenceof Lactobacillus spp. and the significant reduction of Streptococcusspp. following treatment may indicate a protective role for Lacto-bacillus spp. in Streptococcus infection. This is not surprising, as asmall study by Bruzzese et al. (27) found that treatment with theprobiotic L. rhamnosus strain GG decreased the level of intestinalinflammation markers and rectal nitric oxide production in chil-dren with CF compared to those in a placebo-controlled group. Aseparate study further showed that children with CF treated withL. rhamnosus GG showed a reduction in pulmonary exacerbations(28).

The dynamics of the microbial communities within each pa-

tient support the climax and attack model previously described inan ecological view of the CF airways (8). The main players (theclimax community) persist across time even though their abun-dances change with perturbations, while the attack community istransient and dynamic. Every patient presented with a complexlung microbial ecosystem consisting of distinct climax and attackcommunities.

Recently, 16S rRNA gene surveys have been suggested for rou-tine clinical use (29). However, 16S sequence data have a limitedability to resolve taxonomy identification to the species level (30)and may introduce biases during the primer-binding step of PCR(31). For example, different species of Pseudomonas or Streptococ-cus were distinguished using this method, and Staphylococcus au-reus that was detected by culturing was not detected by sequencing(29, 30).

Community metabolism. Functional information gleanedfrom the metagenomic data show that the metabolic potentials ofthese microbial communities were distinct, which helps deter-mine whether a community should be labeled as climax or attack.The PCA of the top 20 most variable functions provided a prelim-inary view on the metabolic potentials associated with attack com-munities (Fig. 2, quadrants 1 and 3) and the metabolic potentialsassociated with climax communities that render them resistant totreatment and enable persistence through perturbations (Fig. 2,quadrants 2, 3, and 4). It is important to note that the groupings ofthese samples are not mutually exclusive. The positive loadings ofthe first and second principal components may represent meta-bolic potentials that are important for both recovery and responseto treatment. The taxonomical and functional profiles of patientCF7 did not considerably change over time, which is consistentwith this patient’s unresponsiveness to treatment and his un-changed health status. This suggests the metabolic functions of theclimax community were associated with persistence and thatchanges in the attack community were associated with exacerba-tion and decline in lung function.

Antibiotic resistance. Of the many challenges facing the CFcommunity, the evolution of antibiotic resistance is one of themost pressing concerns. The presence of antibiotic resistancegenes that encode resistance against major groups of antibiotics(e.g., aminoglycosides, macrolides, �-lactamases, and fluoro-quinolones) suggests these microbial communities may be capa-ble of rapid genetic adaptation to resist perturbations and stressesimposed by treatment. This rapid adaptation is at least partiallyfueled by horizontal gene transfer via phages and plasmids. Activemultidrug efflux mechanisms are known to be one of the majordeterminants of antibiotic resistance in many CF pathogens, in-cluding P. aeruginosa, Burkholderia cepacia, and S. maltophilia(32–34). Our data (Fig. 3A) support this observation, as 50% ofthe antibiotic resistance genes identified were predicted to encodeefflux-mediated resistance mechanisms. �-Lactamase genes werethe most abundant genes identified. Their high abundance in theviromes also suggested that phage-mediated spread of �-lactama-ses might occur within the community (Fig. 3C). Similar to P.aeruginosa, S. maltophilia is highly resistant to antibiotics due tothe presence of various intrinsic and acquired resistance mecha-nisms that include �-lactamases, penicillinase, cephalosporinase,aminoglycoside acetyl-transferase (aac), efflux pumps, and bio-film formation (35–37). An up-to-date database containing com-prehensive annotations for aminoglycoside and macrolide resis-tance mechanisms facilitated the detection of resistance genes

Lim et al.

434 jcm.asm.org Journal of Clinical Microbiology

on August 18, 2018 by guest

http://jcm.asm

.org/D

ownloaded from

present in the metagenomic data, especially across the S. malto-philia-rich microbiomes in patient CF7. With this manually cu-rated database and data analysis framework, we have demon-strated the potential uses of metagenomics for identifying andmonitoring of antibiotic resistance in clinical microbiology(17, 38).

The antibiotic resistance profiles of each patient were dynamic.In several cases, microbial community members appeared to loseresistance to particular antibiotics, which was reflected in bothclinical and metagenomic measurements. In the case of patientCF6, E. coli appeared to lose its resistance to tobramycin, andStreptococcus and Rothia were predicted to be less resistant to ami-noglycosides and macrolides, respectively (Fig. 4). It remains to bedetermined whether these “losses” are due to fluctuations in thenumber of cells sampled or that are truly present or to the transferof these resistance genes between members of the microbial com-munity. For example, a decrease in the number of Streptococcuscells sampled might explain the predicted loss in the presence ofStreptococcus-associated antibiotic resistance genes in the later sta-ble sample (time point D). But such a direct relationship does notexplain the decrease in Rothia-associated macrolide resistancegenes cooccurring with an increase in Rothia abundance. In thiscase, it is possible that these genes are monitoring a separate non-Rothia community member.

The sequencing of the virome portion of CF lungs also suggeststhat antibiotic resistance is not likely confined to one bacterialspecies, for the genes conferring such resistances can be shuttledback and forth between microbes. Therefore, having informationon the predicted resistance of the whole microbial community isperhaps one of the most useful pieces of information extractedfrom metagenome sequencing. If the lung metagenome of a pa-tient suggests resistance to macrolides, despite his not having beenprescribed such antibiotics, this can prove to be vital informationfor clinicians to prescribe the appropriate antibiotic therapy. Ourdata provide a minimal estimate of the resistance potential for thecommunity, and additional sequence coverage can be used toidentify resistance genes present at lower abundances. Such raregenes are also important because they might dominate as the com-munity composition is altered by antibiotic treatment.

Study limitations. This study illustrates the use of meta-genomics for monitoring microbial communities in the clinicalsetting, which may eventually help clinicians in their daily effortsto improve the lives of CF patients. Because this metagenomicapproach is at an early stage of development, the numbers of pa-tients (n � 3) and samples (10 sputum samples) presented in thisstudy are relatively small due to the amount of effort required foreach sample. This limited sampling restricts our ability to deter-mine whether the observed fluctuations in community composi-tion, metabolism, and antibiotic resistance truly occur over timeor whether they are due to variability associated with sputum sam-pling, which may not consistently originate from the same regionof the lung but from different regions that harbor their own per-sistent microbial communities. Although larger studies will beneeded to sort out these matters, this study highlights some of themost important issues to be solved prior to the introduction ofpersonalized metagenomics into the clinic setting.

Another potential limitation concerns the ongoing contro-versy surrounding the extent of oropharyngeal contamination ofsputum samples. Washing and rinsing of the oral cavity using asterile saline solution prior to sputum induction is a National

Institutes of Health (NIH)-recommended standard protocol forobtaining a minimally contaminated sputum sample (39). Previ-ous studies showed significant evidence that induced CF sputumsamples are strongly indicative of the lung environment and areonly minimally contaminated with mouth microbes (40–42). Al-though we cannot rule out that some of our sputum samples werecontaminated with oral microbes, the presence of such microbesin the oral cavity suggests they too might colonize CF lungs andshould therefore be considered members of the community. Thisis particularly important for tracking antibiotic resistance genes,because even if an oral microbe has little chance of surviving in thelung environment, its genes may be transferred to those microbesthat thrive in the lung environment.

Concluding remarks and study significance. The primary sig-nificance of this study is the combined use of metagenomic se-quencing and clinical microbiology for monitoring polymicrobialinfections in individual patients. This shotgun metagenomics ap-proach not only provides accurate species-level (sometimesstrain-level) taxonomic assignments, it also provides functionalinformation at the gene level, e.g., the presence of potential anti-biotic resistance genes and mutation-induced resistance mecha-nisms (38). In addition, the reconstruction of whole genomes ispossible (43), and this can potentially provide important molecu-lar information that is necessary for infection control (44). Vali-dation and normalization of the metagenomic data also improvesthe quantification of microbes and the downstream clinical inter-pretation and therapeutic strategies. Other concerns and specificexamples were reviewed in Dunne et al. (44). Sample preparation,methodology, and bioinformatics will continue to improve theseefforts, eventually leading to real-time monitoring of microbialcommunities in CF patients.

Medical diagnosis is a multidimensional process that includesphysical assessment of the patient by physicians and nurses, non-specific screening tests, monitoring of the efficacy of selectedtreatments, and the collection of specimens for biomedical labo-ratory processing. Nowadays, it is increasingly possible to comple-ment this information with sequencing data. Real-time pathogensequencing has been suggested as a way to help control pathogenoutbreaks, as current methods are slow and offer limited resolu-tions (44). As a proof of principle, this study presents the value ofcoupling metagenomics with clinical findings, helping to move uscloser to molecular diagnoses. Diagrams, such as those in Fig. 4 to6, are instrumental in condensing vast amounts of data into clin-ically useful tools for tracking patient disease progression, corre-sponding treatments, and microbial community responses tothose treatments.

The advancement in sequencing technologies and their de-creasing cost are bringing us closer to diagnoses and treatmentsthat are augmented by genomic technologies. To be relevant forclinical applications, the workflow is only possible with the aid ofrobotics and automation, and the turnaround times can be scaledto within 48 h, which is over two times faster than a conventionalculture-based procedure that takes 3 to 5 days for CF samples.Such timely information will affect clinical management of CFpatients. Of course, the quality of the data is very dependent on thechoice of sequencing technology and the data analysis pipeline.However, the optimization of upstream robotics, the further de-velopment of bioinformatics tools, and increasing computingpower will continually move the field toward this goal. Althoughthe implementation of metagenomic analysis as a clinical diagnos-

Clinical Insights from Metagenomics

February 2014 Volume 52 Number 2 jcm.asm.org 435

on August 18, 2018 by guest

http://jcm.asm

.org/D

ownloaded from

tic tool would be accompanied by the challenges of data interpre-tation by health care professionals, the consistency and accuracyof the technologies, and navigation of the complexities of admin-istrative policy, we see invaluable therapeutic potential in the real-time monitoring of microbial communities and their capabilitiesto resist treatment efforts.

REFERENCES1. Köser CU, Holden MT, Ellington MJ, Cartwright EJ, Brown NM,

Ogilvy-Stuart AL, Hsu LY, Chewapreecha C, Croucher NJ, Harris SR,Sanders M, Enright MC, Dougan G, Bentley SD, Parkhill J, Fraser LJ,Betley JR, Schulz-Trieglaff OB, Smith GP, Peacock SJ. 2012. Rapidwhole-genome sequencing for investigation of a neonatal MRSA out-break. N. Engl. J. Med. 366:2267–2275. http://dx.doi.org/10.1056/NEJMoa1109910.

2. Underwood AP, Dallman T, Thomson NR, Williams M, Harker K,Perry N, Adak B, Willshaw G, Cheasty T, Green J, Dougan G, ParkhillJ, Wain J. 2013. Public health value of next-generation DNA sequencingof enterohemorrhagic Escherichia coli isolates from an outbreak. J. Clin.Microbiol. 51:232–237. http://dx.doi.org/10.1128/JCM.01696-12.

3. Tran B, Brown AM, Bedard PL, Winquist E, Goss GD, Hotte SJ, WelchSA, Hirte HW, Zhang T, Stein LD, Ferretti V, Watt S, Jiao W, Ng K,Ghai S, Shaw P, Petrocelli T, Hudson TJ, Neel BG, Onetto N, Siu LL,McPherson JD, Kamel-Reid S, Dancey JE. 2013. Feasibility of real timenext generation sequencing of cancer genes linked to drug response: re-sults from a clinical trial. Int. J. Cancer 132:1547–1555. http://dx.doi.org/10.1002/ijc.27817.

4. Ross JS, Ali SM, Wang K, Palmer G, Yelensky R, Lipson D, Miller VA,Zajchowski D, Shawver LK, Stephens PJ. 2013. Comprehensive genomicprofiling of epithelial ovarian cancer by next generation sequencing-baseddiagnostic assay reveals new routes to targeted therapies. Gynecol. Oncol.130:554 –559. http://dx.doi.org/10.1016/j.ygyno.2013.06.019.

5. Lim YW, Schmieder R, Haynes M, Willner D, Furlan M, Youle M,Abbott K, Edwards R, Evangelista J, Conrad D, Rohwer F. 2012.Metagenomics and metatranscriptomics: windows on CF-associated viraland microbial communities. J. Cyst. Fibros. 12:154 –164. http://dx.doi.org/10.1016/j.jcf.2012.07.009.

6. Kerem B, Rommens JM, Buchanan JA, Markiewicz D, Cox TK, Chakra-varti A, Buchwald M, Tsui LC. 1989. Identification of the Cystic Fibrosisgene: genetic analysis. Science 245:1073–1080. http://dx.doi.org/10.1126/science.2570460.

7. Quinton PM. 2010. Role of epithelial HCO3� transport in mucin secre-tion: lessons from cystic fibrosis. Am. J. Physiol. Cell Physiol. 299:C1222–1233. http://dx.doi.org/10.1152/ajpcell.00362.2010.

8. Conrad D, Haynes M, Salamon P, Rainey PB, Youle M, Rohwer F.2013. Cystic fibrosis therapy: a community ecology perspective. Am. J.Respir. Cell Mol. Biol. 48:150 –156. http://dx.doi.org/10.1165/rcmb.2012-0059PS.

9. LiPuma JJ. 2010. The changing microbial epidemiology in cystic fibrosis.Clin. Microbiol. Rev. 23:299 –323. http://dx.doi.org/10.1128/CMR.00068-09.

10. Cox MJ, Allgaier M, Taylor B, Baek MS, Huang YJ, Daly RA, Karaoz U,Andersen GL, Brown R, Fujimura KE, Wu B, Tran D, Koff J, KleinhenzME, Nielson D, Brodie EL, Lynch SV. 2010. Airway microbiota andpathogen abundance in age-stratified cystic fibrosis patients. PLoS One5:e11044. http://dx.doi.org/10.1371/journal.pone.0011044.

11. Zhao J, Schloss PD, Kalikin LM, Carmody LA, Foster BK, Petrosino JF,Cavalcoli JD, VanDevanter DR, Murray S, Li JZ, Young VB, LiPuma JJ.2012. Decade-long bacterial community dynamics in cystic fibrosis air-ways. Proc. Natl. Acad. Sci. U. S. A. 109:5809 –5814. http://dx.doi.org/10.1073/pnas.1120577109.

12. Schmieder R, Edwards R. 2011. Quality control and preprocessing ofmetagenomic datasets. Bioinformatics 27:863– 864. http://dx.doi.org/10.1093/bioinformatics/btr026.

13. Schmieder R, Edwards R. 2011. Fast identification and removal of se-quence contamination from genomic and metagenomic datasets. PLoSOne 6:e17288. http://dx.doi.org/10.1371/journal.pone.0017288.

14. Kanehisa M, Goto S. 2000. KEGG: Kyoto Encyclopedia of Genes andGenomes. Nucleic Acids Res. 28:27–30. http://dx.doi.org/10.1093/nar/28.1.27.

15. Abubucker S, Segata N, Goll J, Schubert A, Izard J, Cantarel BL,

Rodriguez-Mueller B, Zucker J, Thiagarajan M, Henrissat B, White O,Kelley ST, Methé, Scholss PD, Gevers D, Mitreva M, Huttenhower C.2012. Metabolic reconstruction for metagenomic data and its applicationto the human microbiome. PLoS Comput. Biol. 8:e1002358. http://dx.doi.org/10.1371/journal.pcbi.1002358.

16. Liu B, Pop M. 2009. ARDB–Antibiotic Resistance Genes Database. Nu-cleic Acids Res. 37:D443–D447. http://dx.doi.org/10.1093/nar/gkn656.

17. Schmieder RA. 2012. A framework for identifying antibiotic resistance inthe human microbiome. Ph.D. dissertation. Claremont Graduate Univer-sity, Claremont, CA, and San Diego State University, San Diego, CA.

18. Willner D, Furlan M, Haynes M, Schmieder R, Angly FE, Silva J,Tammadoni S, Nosrat B, Conrad D, Rohwer F. 2009. Metagenomicanalysis of respiratory tract DNA viral communities in cystic fibrosis andnon-cystic fibrosis individuals. PLoS One 4:e7370. http://dx.doi.org/10.1371/journal.pone.0007370.

19. Willner D, Furlan M. 2010. Deciphering the role of phage in the cysticfibrosis airway. Virulence 1:309 –313. http://dx.doi.org/10.4161/viru.1.4.12071.

20. Chang FC, Chung JG. 1998. Evidence for arylamine N-acetyltransferaseactivity in the Escherichia coli. Curr. Microbiol. 36:125–130. http://dx.doi.org/10.1007/PL00006755.

21. Schomburg PD, Stephan D. 2006. Arylamine N-acetyltransferase, p 243–258. In Schomburg D, Schomburg A (ed), Springer handbook of enzymes.Springer-Verlag, Heidelberg, Germany.

22. Jones SD, Fullerton DA, Zamora MR, Badesch DB, Campbell DN,Grover FL. 1994. Transmission of Lactobacillus pneumonia by a trans-planted lung. Ann. Thorac. Surg. 58:887– 889. http://dx.doi.org/10.1016/0003-4975(94)90779-X.

23. Holzapfel WH, Haberer P, Snel J, Schillinger U, Huis in’t Veld JH.1998. Overview of gut flora and probiotics. Int. J. Food Microbiol. 41:85–101. http://dx.doi.org/10.1016/S0168-1605(98)00044-0.

24. Avlami A, Kordossis T, Vrizidis N, Sipsas NV. 2001. Lactobacillusrhamnosus endocarditis complicating colonoscopy. J. Infect. 42:283–285.http://dx.doi.org/10.1053/jinf.2001.0793.

25. Shoji H, Yoshida K, Niki Y. 2010. Lung abscess and pleuritis caused byLactobacillus rhamnosus in an immunocompetent patient. J. Infect. Che-mother. 16:45– 48. http://dx.doi.org/10.1007/s10156-009-0004-5.

26. Haro C, Villena J, Zelaya H, Alvarez S, Agüero G. 2009. Lactobacilluscasei modulates the inflammation-coagulation interaction in a pneumo-coccal pneumonia experimental model. J. Inflamm. 6:28. http://dx.doi.org/10.1186/1476-9255-6-28.

27. Bruzzese E, Raia V, Gaudiello G, Polito G, Buccigrossi V, Formicola V,Guarino A. 2004. Intestinal inflammation is a frequent feature of cysticfibrosis and is reduced by probiotic administration. Aliment. Pharmacol.Ther. 20:813– 819. http://dx.doi.org/10.1111/j.1365-2036.2004.02174.x.

28. Bruzzese E, Raia V, Spagnuolo MI, Volpicelli M, De Marco G, MaiuriL, Guarino A. 2007. Effect of Lactobacillus GG supplementation on pul-monary exacerbations in patients with cystic fibrosis: a pilot study. Clin.Nutr. 26:322–328. http://dx.doi.org/10.1016/j.clnu.2007.01.004.

29. Salipante SJ, Sengupta DJ, Rosenthal C, Costa G, Spangler J, Sims EH,Jacobs MA, Miller SI, Hoogestraat DR, Cookson BT, McCoy C, MatsenFA, Shendure J, Lee CC, Harkins TT, Hoffman NG. 2013. Rapid 16SrRNA next-generation sequencing of polymicrobial clinical samples fordiagnosis of complex bacterial infections. PLoS One 8:e65226. http://dx.doi.org/10.1371/journal.pone.0065226.

30. Filkins LM, Hampton TH, Gifford AH, Gross MJ, Hogan DA, SoginML, Morrison HG, Paster BJ, O’Toole GA. 2012. Prevalence of strepto-cocci and increased polymicrobial diversity associated with cystic fibrosispatient stability. J. Bacteriol. 194:4709 – 4717. http://dx.doi.org/10.1128/JB.00566-12.

31. Cai L, Ye L, Tong AH, Lok S, Zhang T. 2013. Biased diversity metricsrevealed by bacterial 16S pyrotags derived from different primer sets. PLoSOne 8:e53649. http://dx.doi.org/10.1371/journal.pone.0053649.

32. Nikaido H. 1996. Multidrug efflux pumps of gram-negative bacteria. J.Bacteriol. 178:5853–5859.

33. Alonso A, Martínez JL. 1997. Multiple antibiotic resistance in Stenotroph-omonas maltophilia. Antimicrob. Agents Chemother. 41:1140 –1142.

34. Zhang L, Li XZ, Poole K. 2000. Multiple antibiotic resistance inStenotrophomonas maltophilia: involvement of a multidrug efflux system.Antimicrob. Agents Chemother. 44:287–293. http://dx.doi.org/10.1128/AAC.44.2.287-293.2000.

35. Avison MB, Higgins CS, von Heldreich CJ, Bennett PM, Walsh TR.2001. Plasmid location and molecular heterogeneity of the L1 and L2

Lim et al.

436 jcm.asm.org Journal of Clinical Microbiology

on August 18, 2018 by guest

http://jcm.asm

.org/D

ownloaded from

beta-lactamase genes of Stenotrophomonas maltophilia. Antimicrob.Agents Chemother. 45:413– 419. http://dx.doi.org/10.1128/AAC.45.2.413-419.2001.

36. Di Bonaventura G, Spedicato I, D’Antonio D, Robuffo I, PiccolominiR. 2004. Biofilm formation by Stenotrophomonas maltophilia: modulationby quinolones, trimethoprim-sulfamethoxazole, and ceftazidime. Anti-microb. Agents Chemother. 48:151–160. http://dx.doi.org/10.1128/AAC.48.1.151-160.2004.

37. Falagas ME, Kastoris AC, Vouloumanou EK, Rafailidis PI, KapaskelisAM, Dimopoulos G. 2009. Attributable mortality of Stenotrophomonasmaltophilia infections: a systematic review of the literature. Future Micro-biol. 4:1103–1109. http://dx.doi.org/10.2217/fmb.09.84.

38. Schmieder R, Edwards R. 2012. Insights into antibiotic resistancethrough metagenomic approaches. Future Microbiol. 7:73– 89. http://dx.doi.org/10.2217/fmb.11.135.

39. National Institutes of Health Critical Care Medicine Department. 2001.Critical care therapy and respiratory care section: sputum induction/oralwash. National Institutes of Health, Bethesda, MD. http://clinicalcenter.nih.gov/ccmd/cctrcs/pdf_docs/Diagnostics/05A-Sputum_Induct_Oral_Wash.pdf.

40. Rogers GB, Carroll MP, Serisier DJ, Hockey PM, Jones G, Kehagia V,Connett GJ, Bruce KD. 2006. Use of 16S rRNA gene profiling by terminalrestriction fragment length polymorphism analysis to compare bacterial

communities in sputum and mouthwash samples from patients with cys-tic fibrosis. J. Clin. Microbiol. 44:2601–2604. http://dx.doi.org/10.1128/JCM.02282-05.

41. Goddard AF, Staudinger BJ, Dowd SE, Joshi-Datar A, Wolcott RD,Aitken ML, Fligner CL, Singh PK. 2012. Direct sampling of cystic fibrosislungs indicates that DNA-based analyses of upper-airway specimens canmisrepresent lung microbiota. Proc. Natl. Acad. Sci. U. S. A. 109:13769 –13774. http://dx.doi.org/10.1073/pnas.1107435109.

42. Fodor AA, Klem ER, Gilpin DF, Elborn JS, Boucher RC, Tunney MM,Wolfgang MC. 2012. The adult cystic fibrosis airway microbiota is stableover time and infection type, and highly resilient to antibiotic treatment ofexacerbations. PLoS One 7:e45001. http://dx.doi.org/10.1371/journal.pone.0045001.

43. Lim YW, Schmieder R, Haynes M, Furlan M, Matthews TD, WhitesonK, Poole SJ, Hayes CS, Low DA, Maughan H, Edwards R, Conrad D,Rohwer F. 2013. Mechanistic model of Rothia mucilaginosa adaptationtoward persistence in the CF lung, based on a genome reconstructed frommetagenomic data. PLoS One 8:e64285. http://dx.doi.org/10.1371/journal.pone.0064285.

44. Dunne WM, Jr, Westblade LF, Ford B. 2012. Next-generation andwhole-genome sequencing in the diagnostic clinical microbiology labora-tory. Eur. J. Clin. Microbiol. Infect. Dis. 31:1719 –1726. http://dx.doi.org/10.1007/s10096-012-1641-7.

Clinical Insights from Metagenomics

February 2014 Volume 52 Number 2 jcm.asm.org 437

on August 18, 2018 by guest

http://jcm.asm

.org/D

ownloaded from