Datasets:

---

LLMDH

/

OpenScience

like 0

Follow

LLM - Digital Humanities 16

ERROR: type should be string, got "https://escholarship.org/uc/item/0h66w7fj Journal\nmSystems, 5(3)\nISSN\n2379-5077\nAuthors\nRaghuvanshi, Ruma\nVasco, Karla\nVázquez-Baeza, Yoshiki\net al. Publication Date\n2020-06-01\nDOI\n10.1128/msystems.00292-20\n \nPeer reviewed Title High-Resolution Longitudinal Dynamics of the Cystic Fibrosis Sputum Microbiome and \nMetabolome through Antibiotic Therapy. UC San Diego UC San Diego g\nUC San Diego Previously Published Works Title\nHigh-Resolution Longitudinal Dynamics of the Cystic Fibrosis Sputum Microbiome and \nMetabolome through Antibiotic Therapy. Permalink https://escholarship.org/uc/item/0h66w7fj High-Resolution Longitudinal Dynamics of the Cystic Fibrosis\nSputum Microbiome and Metabolome through Antibiotic\nTherapy Ruma Raghuvanshi,a Karla Vasco,a* Yoshiki Vázquez-Baeza,b,c,d Lingjing Jiang,e James T. Morton,f Danxun Li,f\nAntonio Gonzalez,b Lindsay DeRight Goldasich,b Gregory Humphrey,b Gail Ackermann,b Austin D. Swafford,c,d\nDouglas Conrad,g\nRob Knight,b,c,h Pieter C. Dorrestein,c,i Robert A. Quinna aDepartment of Biochemistry and Molecular Biology, Michigan State University, East Lansing, Michigan, USA\nbDepartment of Pediatrics, University of California San Diego, La Jolla, California, USA\ncCenter for Microbiome Innovation, University of California San Diego, La Jolla, California, USA\ndJacobs School of Engineering, University of California San Diego, La Jolla, California, USA\neDivision of Biostatistics, University of California San Diego, La Jolla, California, USA\nfCenter for Computational Biology, Flatiron Institute, Simons Foundation, New York, New York, USA\ngDepartment of Medicine, University of California San Diego, La Jolla, California, USA\nhDepartment of Computer Science and Engineering and Bioengineering, University of California San Diego, La Jolla, Ca\niSkaggs School of Pharmacy and Pharmaceutical Sciences, University of California San Diego, La Jolla,\nCalifornia, USA Ruma Raghuvanshi and Karla Vasco contributed equally to this work as co-first authors. Author order was determined\nalphabetically by last name. Citation Raghuvanshi R, Vasco K, Vázquez-\nBaeza Y, Jiang L, Morton JT, Li D, Gonzalez A,\nDeRight Goldasich L, Humphrey G, Ackermann\nG, Swafford AD, Conrad D, Knight R, Dorrestein\nPC, Quinn RA. 2020. High-resolution\nlongitudinal dynamics of the cystic fibrosis\nsputum microbiome and metabolome\nthrough antibiotic therapy. mSystems 5:\ne00292-20. https://doi.org/10.1128/mSystems\n.00292-20. ABSTRACT\nMicrobial diversity in the cystic fibrosis (CF) lung decreases over de-\ncades as pathogenic bacteria such as Pseudomonas aeruginosa take over. The dy-\nnamics of the CF microbiome and metabolome over shorter time frames, however,\nremain poorly studied. Here, we analyze paired microbiome and metabolome data\nfrom 594 sputum samples collected over 401 days from six adult CF subjects\n(subject mean \u0002 179 days) through periods of clinical stability and 11 CF pulmo-\nnary exacerbations (CFPE). While microbiome profiles were personalized (permuta-\ntional multivariate analysis of variance [PERMANOVA] r2 \u0002 0.79, P \u0003 0.001), we ob-\nserved significant intraindividual temporal variation that was highest during clinical\nstability (linear mixed-effects [LME] model, P \u0002 0.002). This included periods where\nthe microbiomes of different subjects became highly similar (UniFrac distance,\n\u00030.05). There was a linear increase in the microbiome alpha-diversity and in the log\nratio of anaerobes to pathogens with time (n \u0002 14 days) during the development of\na CFPE (LME P \u0002 0.0045 and P \u0002 0.029, respectively). Powered by the California Digital Library\nUniversity of California eScholarship.org RESEARCH ARTICLE\nClinical Science and Epidemiology RESEARCH ARTICLE\nClinical Science and Epidemiology\ncrossm High-Resolution Longitudinal Dynamics of the Cystic Fibrosis\nSputum Microbiome and Metabolome through Antibiotic\nTherapy Collectively, comparing samples\nacross disease states showed there was a reduction of these two measures during\nantibiotic treatment (LME P \u0002 0.0096 and P \u0002 0.014, respectively), but the stability\ndata and CFPE data were not significantly different from each other. Metabolome\nalpha-diversity was higher during CFPE than during stability (LME P \u0002 0.0085), but\nno consistent metabolite signatures of CFPE across subjects were identified. Virulence-associated metabolites from P. aeruginosa were temporally dynamic but\nwere not associated with any disease state. One subject died during the collection\nperiod, enabling a detailed look at changes in the 194 days prior to death. This sub-\nject had over 90% Pseudomonas in the microbiome at the beginning of sampling,\nand that level gradually increased to over 99% prior to death. This study revealed\nthat the CF microbiome and metabolome of some subjects are dynamic through\ntime. Future work is needed to understand what drives these temporal dynamics\nand if reduction of anaerobes correlate to clinical response to CFPE therapy. Editor Nicholas Chia, Mayo Clinic\nCopyright © 2020 Raghuvanshi et al. This is an\nopen-access article distributed under the terms\nof the Creative Commons Attribution 4.0\nInternational license. Address correspondence to Douglas Conrad\n(clinical aspects), dconrad@ucsd.edu, Rob\nKnight (sequencing), rknight@ucsd.edu, Pieter\nC. Dorrestein (mass spectrometry),\npdorrestein@ucsd.edu, or Robert A. Quinn\n(study design and analysis),\nquinnrob@msu.edu. Address correspondence to Douglas Conrad\n(clinical aspects), dconrad@ucsd.edu, Rob\nKnight (sequencing), rknight@ucsd.edu, Pieter\nC. Dorrestein (mass spectrometry),\nd\nt i @\nd d\nR b t A Q i * Present address: Karla Vasco, Department of\nMicrobiology and Molecular Genetics,\nMichigan State University, East Lansing,\nMichigan, USA. * Present address: Karla Vasco, Department of\nMicrobiology and Molecular Genetics,\nMichigan State University, East Lansing,\nMichigan, USA. The #cysticfibrosis lung #microbiome is\nhighly dynamic through time. Antibiotics have\na large effect on the lung microbiome but\nmost of these effects are unknown by\nclinicians. Paper from @Quinn_Labs\n@Pdorrestein1 @KnightLabNews Received 1 April 2020\nAccepted 4 June 2020\nPublished 23 June 2020 msystems.asm.org\n1 May/June 2020\nVolume 5\nIssue 3\ne00292-20 Raghuvanshi et al. IMPORTANCE Subjects with cystic fibrosis battle polymicrobial lung infections\nthroughout their lifetime. Although antibiotic therapy is a principal treatment for CF\nlung disease, we have little understanding of how antibiotics affect the CF lung mi-\ncrobiome and metabolome and how much the community changes on daily time-\nscales. KEYWORDS cystic fibrosis, microbiome, antibiotics, metabolome KEYWORDS cystic fibrosis, microbiome, antibiotics, metabolome T\nhe respiratory tracts of individuals with cystic fibrosis (CF) are colonized by a polymi-\ncrobial community that impacts the pathology and progression of the disease (1–3). The microbiome of sputum expectorated from the lung includes opportunistic pathogens,\nsuch as Pseudomonas aeruginosa, Staphylococcus aureus, Stenotrophomonas maltophilia,\nBurkholderia cepacia, and Achromobacter xylosoxidans, but a myriad of lesser understood\noral anaerobes are also detected (1–3). It is known that pathogens come to dominate the\ncommunity profiles as subjects age and microbial diversity decreases (1), yet we have a\npoor understanding of these dynamics in shorter longitudinal time frames, such as\nthroughout CF pulmonary exacerbations (CFPEs). Some studies have reported changes\noccurring during CFPE (4–9), primarily via reduction in the relative abundance of rare taxa\nduring treatment, particularly anaerobic bacteria (Prevotella, Veillonella, Gemella, etc.) (4, 5,\n9–11), but whether this represents a dysbiotic shift in the CF microbiome or regular changes\nin microbial dynamics without clinical relevance remains unknown. In other studies, the CF\nmicrobiome was found to remain relatively static through CFPE (8, 12–14), complicating our\nunderstanding of microbiome dynamics. The CF microbiome has also been shown to be\nhighly personalized (1, 11, 15), but it is unknown whether this personalization is maintained\nover shorter longitudinal time frames or if the communities are dynamic. T Recent studies have begun to examine the metabolome of CF sputum, comprised\nof DNA, mucins, surfactant, and a myriad of small molecules from microbial, host, and\nxenobiotic sources that are highly personalized (16) and have important implications\nfor disease pathology (17–20). Virulence-associated metabolites from P. aeruginosa are\nalso detected, as well as fermentation metabolites from streptococci (21). A recent\nstudy showed that as lung function declines, subjects accumulate more peptides and\namino acids in their sputum (17), which are derived from neutrophil elastase activity in\nresponse to microbial infections. The contents of the CF metabolome are highly diverse,\nbut how this chemistry changes through time and around CFPE events is virtually\nunknown. This report presents a high-resolution analysis of the longitudinal dynamics of the\nadult CF sputum metabolome and microbiome. Sputum samples were collected from\nsix subjects in their homes as frequently as possible for a period of 401 days. Eleven\nCFPE events were captured through the sampling period, providing detailed insight\ninto the microbial and chemical changes in sputum through these important clinical\nevents. May/June 2020\nVolume 5\nIssue 3\ne00292-20 High-Resolution Longitudinal Dynamics of the Cystic Fibrosis\nSputum Microbiome and Metabolome through Antibiotic\nTherapy By analyzing 594 longitudinal CF sputum samples from six adult subjects, we\nshow that the sputum microbiome and metabolome are dynamic. Significant\nchanges occur during times of stability and also through pulmonary exacerbations\n(CFPEs). Microbiome alpha-diversity increased as a CFPE developed and then de-\ncreased during treatment in a manner corresponding to the reduction in the log ra-\ntio of anaerobic bacteria to classic pathogens. Levels of metabolites from the patho-\ngen P. aeruginosa were also highly variable through time and were negatively\nassociated with anaerobes. The microbial dynamics observed in this study may have\na significant impact on the outcome of antibiotic therapy for CFPEs and overall sub-\nject health. Longitudinal CF Microbiome Dynamics FIG 1 (a) Bar plots representing the microbiome of sputum samples from the six subjects plotted chronologically through the collection. Gaps in sample\ncollection are not shown. (b) PCoA plot of the weighted UniFrac distances of the microbiome data colored by subject. Inset are the samples colored on a scale\nrepresenting the percentage of the anaerobe or percentage of the pathogen as defined in Data Set S1, sheet 3. Samples where Pseudomonas is the dominant\nASV in the plot are highlighted. (c) PCoA plot of the Bray-Curtis distance of the metabolomic data, including all detected metabolite features colored by subject. (d and e) Within-subject and between-subject distances of microbiome (weighted UniFrac distance) data (d) and metabolome (Bray-Curtis distance) data (e). Significance was tested with an LME model with subject as a fixed effect. FIG 1 (a) Bar plots representing the microbiome of sputum samples from the six subjects plotted chronologically through the collection. Gaps in sample\ncollection are not shown. (b) PCoA plot of the weighted UniFrac distances of the microbiome data colored by subject. Inset are the samples colored on a scale\nrepresenting the percentage of the anaerobe or percentage of the pathogen as defined in Data Set S1, sheet 3. Samples where Pseudomonas is the dominant\nASV in the plot are highlighted. (c) PCoA plot of the Bray-Curtis distance of the metabolomic data, including all detected metabolite features colored by subject. (d and e) Within-subject and between-subject distances of microbiome (weighted UniFrac distance) data (d) and metabolome (Bray-Curtis distance) data (e). Significance was tested with an LME model with subject as a fixed effect. 190) (see Fig. S1 in the supplemental material; see also Data Set S1, sheets 1 and 2, in\nthe supplemental material). Subjects did not all begin sampling on the same day but\nwere asked to collect as frequently as possible during their collection period, resulting\nin varied sample numbers per subject (Fig. S1; see also Data Set S1, sheets 1 and 2). All\nexperienced at least one period of CFPE and treatment during the study (Data Set S1,\nsheets 1 and 2). Samples were classified as “CFPE” samples if they were collected within\n14 days prior to treatment, as “treatment” samples if they were collected during the\n21 days of treatment, and as “stable” if they were collected outside those periods. May/June 2020\nVolume 5\nIssue 3\ne00292-20 RESULTS Longitudinal sampling and microbiome and metabolome data generation\nfrom CF sputum samples. A total of 594 sputum samples were self-collected by six CF\nsubjects (CF066, CF146, CF176, CF189, CF318, and CF353) with declining lung function\nand health over a total of 401 days (subject mean \u0002 179 sampled days, range \u0002 22 to May/June 2020\nVolume 5\nIssue 3\ne00292-20 msystems.asm.org\n2 variate analysis of variance [PERMANOVA] by subject F \u0002 30.48, r2 \u0002 0.79, P \u0002 0.001;\nBray-Curtis distances for metabolites, PERMANOVA by subject F \u0002 24.81, r2 \u0002 0.372,\nP \u0002 0.001) (Fig. 1b and c). There was greater beta-diversity variation across subjects\nthan within subjects for both data types (Fig. 1d and e). Alpha-diversities of the\nmicrobiome and metabolome were also individualized and were significantly different\nbased on subject source (except CF146 and CF318; Fig. S3a and b; see also Fig. S4). CF176 had the lowest microbiome alpha-diversity (Fig. S3a) but the highest metabo-\nlome alpha-diversity (Fig. S3b), a contrasting phenomenon similar to that reported from\na previous CF sputum multi-omics study (17). Despite the overall personalization, four\nof the subjects developed highly similar microbiomes at times during the dense\nlongitudinal sampling period (CF176, CF146, CF353, and CF189, Fig. 1b and c). Com-\nparing to CF176 as a reference, 23.0% of samples from CF146, 9.7% from CF189, and\n1.3% from CF353 had a weighted UniFrac distance value of less than 0.05 (Fig. S5a; see\nalso Data Set S1, sheet 4), indicating almost identical microbial communities. The\nmetabolome data had no samples where the Bray-Curtis distance value was below 0.05,\nindicating stronger personalization in this data set as a whole; however, CF176 and\nCF353 showed similarity in the PCoA plot with a mean Bray-Curtis distance value of 0.39\n(Fig. 1b and c; see also Fig. S5b). g\ng\nMicrobiome and metabolome of sputum are dynamic in short time frames. To\nbetter characterize intraindividual dynamics, we quantified multi-omic variation as the\npercentage of samples within each subject with values that were greater than 1.5\u0004\ntheir beta-diversity interquartile range (IQR) (analogous to the analysis by Caverly et al. [23]) and the incidence of samples that had a weighted UniFrac or Bray-Curtis distance\nvalue above 0.6 (a cutoff to represent a microbiome that was highly differentiated with\nrespect to 16S or metabolomic data, respectively). To provide a reference frame for\ncomparison, we computed the same metrics for variation in samples from a recently\npublished cross-sectional study (n \u0002 88) (17; Fig. S3c and d; see also Video S1 in the\nsupplemental material). In our cohort, 23.8% of the microbiomes within an individual\nwere outside their IQR (Table 1), a value similar to the 24.4% of samples across\nindividuals seen in the cross-sectional study (Fig. S3c). May/June 2020\nVolume 5\nIssue 3\ne00292-20 All\nsamples were delivered to the clinic by each subject after storage in a home freezer,\nwith the exception of the last 22 samples from one subject (CF176) who died during the\nstudy. Paired 16S rRNA gene sequencing and untargeted metabolomic data were\ngenerated to evaluate the microbial community and chemical composition of these\nsputum samples. As expected, the microbiome contained a mixture of classic CF pathogens and oral\nanaerobic bacteria (classified according to Data Set S1, sheet 3). The metabolomic data\nincluded molecules from host cells, microbial cells, and xenobiotics (Fig. S2). There were\n4,988 unique spectra detected in the sputum samples, with 394 annotations from the\nGlobal Natural Products Social Molecular Networking (GNPS) mass spectral libraries\n(7.9%) (22). The most prevalent known molecules were phospholipids, sphingolipids,\nand antibiotics (Fig. S2). Sputum microbiome and metabolome are largely subject specific. Four of the\nsix subjects had Pseudomonas as the most abundant classic pathogen in their sputa,\nwhile the other two were infected with Stenotrophomonas (CF318) or Escherichia\n(CF066) (Fig. 1a). Principal-coordinate analysis (PCoA) of the beta-diversity between\nsamples showed that both the sputum microbiome and the sputum metabolome were\nhighly individualized (weighted UniFrac distances for microbes, permutational multi- May/June 2020\nVolume 5\nIssue 3\ne00292-20 msystems.asm.org\n3 Raghuvanshi et al. TABLE 1 Microbiome (UniFrac distance) and metabolome (Bray-Curtis distance) variation\nin the different subjects through timea TABLE 1 Microbiome (UniFrac distance) and metabolome (Bray-Curtis distance) variation\nin the different subjects through timea\nCategory and\nsubject ID\n% outside\nIQR\n% above\n0.6\nNo. of\ncomparisons\nMicrobiome\nCF66\n24.512\n8.902\n4,561\nCF146\n33.068\n2.703\n629\nCF176\n10.297\n0.023\n4,370\nCF189\n25.062\n30.53\n2,414\nCF318\n22.96\n8.742\n13,040\nCF353\n27.185\n12.185\n2,700\nAvg\n23.848\n10.514\nCross-sectional\n24.4\n45.6\n10,278\nMetabolome\nCF66\n25.961\n32.822\n5,778\nCF146\n20.509\n20.509\n629\nCF176\n33.37\n2.632\n4,558\nCF189\n30.655\n7.995\n2,414\nCF318\n21.707\n9.87\n13,899\nCF353\n27.509\n2.501\n3,159\nAvg\n26.618\n12.721\nCross-sectional\n28\n56.4\n10,278\naAll samples were compared to all others, and the percentages of comparisons outside the interquartile\nrange (IQR), as well as the number of comparisons with a beta-diversity distance value above 0.6, are\nreported. aAll samples were compared to all others, and the percentages of comparisons outside the interquartile\nrange (IQR), as well as the number of comparisons with a beta-diversity distance value above 0.6, are\nreported. The within-subject weighted\nUniFrac distance values were above 0.6 for 10.5% of the longitudinal comparisons,\ncompared to 45.6% across individuals in the cross-sectional study (Fig. S3c). The msystems.asm.org\n4 Longitudinal CF Microbiome Dynamics incidence of these samples with high beta-diversity were most common in two subjects\nwith Pseudomonas as the pathogen with the highest relative abundance (CF189 [30.5%]\nand CF353 [12.1%]), while the end-stage subject (CF176) had no samples above this\ndistance threshold and the lowest microbial variation through time (Fig. 1b and c; see\nalso Fig. S3c) (Table 1). Similarly, the mean proportion of metabolome sample com-\nparisons with values 1.5\u0004 outside their IQR was 26.6% (28.0% in the cross-sectional\nstudy), with 12.7% having a Bray-Curtis distance above 0.6 (56.4% in the cross-sectional\nstudy) (Table 1; see also Fig. S3d). Two of the Pseudomonas-dominated subjects, CF176\nand CF353, showed little change in their sputum metabolome, with \u00033% of samples\nhaving values above the Bray-Curtis distance value of 0.6 (Fig. S3d). Collectively, the\nalpha- and beta-diversity results demonstrate that although there was strong person-\nalization in the overall microbiome profiles, the communities within five of the six\nindividuals were dynamic and driven by changes in the relative abundances of anaer-\nobes and dominant pathogens, such as Pseudomonas or Stenotrophomonas (Fig. 1a; see\nalso the animated video [24] showing changes through time [see Video S1 in the\nsupplemental material]). Multi-omic variation around exacerbation. To examine whether disease state\n(CFPE, treatment, or stable) was a primary driver of the dynamism seen, the population-\nlevel alpha-diversity and beta-diversity data from the microbial communities and\nmetabolomes across disease states were compared. With subject source accounted for\nas a covariate, the beta-diversity values were significantly different based on disease\nstate for the metabolomic data, though the level of variance explained by this param-\neter was low (PERMANOVA r2 \u0002 0.032, P \u0002 0.001). There was not a significant difference\nin beta-diversity based on disease state for the microbiome data (PERMANOVA\nr2 \u0002 0.072, P \u0002 0.223). Pairwise comparisons of the weighted UniFrac distance values\nwithin disease states from each subject using a linear mixed-effects (LME) model\nshowed that the stable disease state had the highest degree of microbial variability\n(Fig. 2a). Metabolome variability was highest during the treatment period, followed by\nthe stable period, and was lowest during CFPE (Fig. 2b, LME P \u0003 0.001, all pairwise\ncomparisons). Collectively, the alpha-diversity of the microbiome was significantly\nhigher during the stability period than during the treatment period (LME P \u0002 0.001) and\nduring the CFPE period than during the treatment period (LME P \u0002 0.0096), but the\nlevels did not differ significantly between the stable and CFPE states (Fig. 2c). Alpha-\ndiversity of the metabolome was highest during exacerbations, but this was only\nsignificant compared to the stable state (Fig. 2d, LME P \u0002 0.0085). With the identifica-\ntion of lower microbial alpha-diversity during the treatment period, we further inves-\ntigated these changes by comparing the log ratio of anaerobes to pathogens and\nfound that this ratio was also significantly higher during stability and CFPE than during\ntreatment but that the ratios did not differ significantly between the stable and CFPE\nstates (Fig. 2e). Because we had high-resolution longitudinal samples, we took a closer look at the\nmicrobial dynamics that had occurred through time during the development and\ntreatment of the 11 CFPE events. The alpha-diversity of the microbial community\nsignificantly increased in the 14 days leading up to antibiotic therapy for a CFPE\n(Spearman’s rho \u0002 0.348 with time in days, LME P \u0002 0.0045, Fig. 2f), but there was no\nchange in alpha diversity through time during the 21 days of treatment (Spearman’s\nrho \u0002 \u00050.110, LME P \u0002 0.93, Fig. 2f). There were no significant changes in diversity\nmeasures of the metabolome through time during the CFPE period or the treatment\nperiod (Fig. 2g). The log ratio of anaerobes to pathogens also increased with time\napproaching the start of antibiotic treatment for a CFPE (rho \u0002 0.242, LME P \u0002 0.029)\nbut did not change with time during the treatment period (rho \u0002 \u00050.248, LME\nP \u0002 0.645, Fig. 2h). To further test the changes in microbial diversity that occurred with\ntime through the CFPE and treatment periods, we also used an LME model to describe\nthe temporal trajectories with a linear spline (or broken stick) and found that the May/June 2020\nVolume 5\nIssue 3\ne00292-20 msystems.asm.org\n5 Raghuvanshi et al. CF066\nCF146\nCF176\nCF189\nCF318\nCF353\n0.00\n0.25\n0.50\n0.75\n1.00\nSt\nEx\nTr\nWeighted UniFrac Distance\np=0.002 p=0.004\n0.25\n0.50\n0.75\n1.00\nBray-Curtis Distance\nSt\nEx\nTr\na)\nb)\np<0.001\np<0.001\n−2\n−1\n0\n1\n4\n5\n6\n7\n0\n1\n2\n3\n4\nSt\nEx\nTr\nMicrobiome Shannon Index\np=0.0096\np=0.001\n−0.5\n0.0\n0.5\n1.0\n1.5\n2.0\n−10\n−5\n0\n0\n1\n2\n3\n−10\n−5\n0\nfactor(host_subject_id)\nCF066\nCF146\nCF176\nCF189\nCF318\nCF353\n1\n5\n10\n15\n20\nTime in Days\nLME p = 0.0045\nLME p = 0.93\nc)\nf)\nrho = 0.348\nrho = -0.110\nrho = -0.018\nrho = -0.028\n5.0\n5.5\n6.0\n6.5\n7.0\n7.5\n−10\n−5\n0\nfactor(deidentified_patient_number)\nCF066\nCF146\nCF176\nCF189\nCF318\nCF353\n1\n5\n10\n15\n20\nLME p = 0.057\nLME p = 0.165\nLogRatio Anaerobe/Pathogen\ne)\nLogRatio Anaerobe/Pathogen\nrho = 0.242\n1\n5\n10\n15\n20\nrho = 0.-248\nLME p = 0.645\nLME p = 0.029\nExacerbation\nTreatment\nExacerbation\nTreatment\nExacerbation\nTreatment\np=0.014\np<0.001\nd)\np=0.0085\nh)\nMetabolome Shannon Index\ng)\nSt\nEx\nTr\nSt\nEx\nTr\np<0.001\nMICROBIOME\nMETABOLOME\nMicrobiome Shannon Index\nMetabolome Shannon Index\nAb\nAb\nAb\nTime in Days\nTime in Days\nFIG 2 The microbiome and metabolome variation around CFPEs. (a and b) Notch plots of the (a) microbiome weighted UniFrac distances and (b) metabolome\nBray-Curtis distances between samples classified as CFPE (\u000514 days from antibiotic treatment), treatment (during 21 days of antibiotic treatment), or stable\n(outside these time periods). Statistical significance across the class comparisons was tested using an LME model with subject as random effects and Tukey’s\npost hoc tests. (c and d) Shannon index of microbiome diversity (c) and metabolome diversity (d) in samples collected during different disease states. Statistical\nsignificance across the class comparisons was tested using an LME model with subject as random effects and Tukey’s post hoc tests. (e) Notch plots of the log\nratios of anaerobes to pathogens in samples classified as CFPE, treatment, or stable. (Statistics are presented as described above). (f and g) Shannon index of\nmicrobiome diversity (f) and metabolome diversity (g) through the 14 days prior to a CFPE and the 21 days of treatment. Spearman’s rho and the corresponding\nP value from an LME model are shown for the regression with time in days. (f and g) Shannon index of\nmicrobiome diversity (f) and metabolome diversity (g) through the 14 days prior to a CFPE and the 21 days of treatment. Spearman’s rho and the corresponding\nP value from an LME model are shown for the regression with time in days. (h) Log ratio of anaerobes to pathogens through the 14 days prior to a CFPE and\nthe 21 days of treatment. (Statistics are presented as described for panel f). Antibiotics were administered between day 0 and day 1 (denoted as “Ab” in panels\nf to h). All exacerbations for all subjects are shown in the plots. FIG 2 The microbiome and metabolome variation around CFPEs. (a and b) Notch plots of the (a) microbiome weighted UniFrac distances and (b) metabolome\nBray-Curtis distances between samples classified as CFPE (\u000514 days from antibiotic treatment), treatment (during 21 days of antibiotic treatment), or stable\n(outside these time periods). Statistical significance across the class comparisons was tested using an LME model with subject as random effects and Tukey’s\npost hoc tests. (c and d) Shannon index of microbiome diversity (c) and metabolome diversity (d) in samples collected during different disease states. Statistical\nsignificance across the class comparisons was tested using an LME model with subject as random effects and Tukey’s post hoc tests. (e) Notch plots of the log\nratios of anaerobes to pathogens in samples classified as CFPE, treatment, or stable. (Statistics are presented as described above). (f and g) Shannon index of\nmicrobiome diversity (f) and metabolome diversity (g) through the 14 days prior to a CFPE and the 21 days of treatment. Spearman’s rho and the corresponding\nP value from an LME model are shown for the regression with time in days. (h) Log ratio of anaerobes to pathogens through the 14 days prior to a CFPE and\nthe 21 days of treatment. (Statistics are presented as described for panel f). Antibiotics were administered between day 0 and day 1 (denoted as “Ab” in panels\nf to h). All exacerbations for all subjects are shown in the plots. Shannon diversity slope with time was significantly higher through CFPE than the\ntreatment period (LME P \u0002 0.008, Data Set S1, sheet 5). To identify metabolites changing in the three disease states, we trained a\nrandom forest classification model (25) on data from the disease state to determine\nwhether individual metabolites were altered. (h) Log ratio of anaerobes to pathogens through the 14 days prior to a CFPE and\nthe 21 days of treatment. (Statistics are presented as described for panel f). Antibiotics were administered between day 0 and day 1 (denoted as “Ab” in panels\nf to h). All exacerbations for all subjects are shown in the plots. −0.5\n0.0\n0.5\n1.0\n1.5\n2.0\n−10\n−5\n0\n0\n1\n2\n3\n−10\n−5\n0\nfactor(host_subject_id)\nCF066\nCF146\nCF176\nCF189\nCF318\nCF353\n1\n5\n10\n15\n20\nTime in Days\nLME p = 0.0045\nLME p = 0.93\nf)\nrho = 0.348\nrho = -0.110\nrho = -0.018\nrho = -0.028\n5.0\n5.5\n6.0\n6.5\n7.0\n7.5\n−10\n−5\n0\nfactor(deidentified_patient_number)\nCF066\nCF146\nCF176\nCF189\nCF318\nCF353\n1\n5\n10\n15\n20\nLME p = 0.057\nLME p = 0.165\nLogRatio Anaerobe/Pathogen\nrho = 0.242\n1\n5\n10\n15\n20\nrho = 0.-248\nLME p = 0.645\nLME p = 0.029\nExacerbation\nTreatment\nExacerbation\nTreatment\nExacerbation\nTreatment\nh)\ng)\nMicrobiome Shannon Index\nMetabolome Shannon Index\nAb\nAb\nAb\nTime in Days\nTime in Days\na and b) Notch plots of the (a) microbiome weighted UniFrac distances and (b) metabolome\nys from antibiotic treatment), treatment (during 21 days of antibiotic treatment), or stable\ns comparisons was tested using an LME model with subject as random effects and Tukey’s\nand metabolome diversity (d) in samples collected during different disease states. Statistical\nmodel with subject as random effects and Tukey’s post hoc tests. (e) Notch plots of the log\natment, or stable. (Statistics are presented as described above). (f and g) Shannon index of\n14 days prior to a CFPE and the 21 days of treatment. May/June 2020\nVolume 5\nIssue 3\ne00292-20 Spearman’s rho and the corresponding\ne in days (h) Log ratio of anaerobes to pathogens through the 14 days prior to a CFPE and 0\n1\n2\n3\n4\nSt\nEx\nTr\nMicrobiome Shannon Index\np=0.0096\np=0.001\nc) 0.00\n0.25\n0.50\n0.75\n1.00\nSt\nEx\nTr\nWeighted UniFrac Distance\np=0.002 p=0.004\na)\nMICROBIOME 0\n1\n2\n3\n−10\n−5\n0\nfactor(host_subject_id)\nCF066\nCF146\nCF176\nCF189\nCF318\nCF353\n1\n5\n10\n15\n20\nTime in Days\nLME p = 0.0045\nLME p = 0.93\nf)\nrho = 0.348\nrho = -0.110\nrho = -0.018\nrho = -0.028\n5.0\n5.5\n6.0\n6.5\n7.0\n7.5\n−10\n−5\n0\nfactor(deidentified_patient_number)\nCF066\nCF146\nCF176\nCF189\nCF318\nCF353\n1\n5\n10\n15\n20\nLME p = 0.057\nLME p = 0.165\nExacerbation\nTreatment\nExacerbation\nTreatment\nh)\ng)\nMicrobiome Shannon Index\nMetabolome Shannon Index\nAb\nAb\nTime in Days f) c) a) Microbiome Shannon Index Treatment Microbiome Sha St 0.25\n0.50\n0.75\n1.00\nBray-Curtis Distance\nSt\nEx\nTr\nb)\np<0.001\np<0.001\np<0.001\nMETABOLOME 4\n5\n6\n7\nd)\np=0.0085\nMetabolome Shannon Index\nSt\nEx\nTr b) d) g)\nMetabolome Shannon Index Metabolome Shannon −0.5\n0.0\n0.5\n1.0\n1.5\n2.0\n−10\n−5\n0\n5.0\n5.5\n−10\n−5\n0\n1\n5\n10\n15\n20\nLogRatio Anaerobe/Pathogen\nrho = 0.242\n1\n5\n10\n15\n20\nrho = 0.-248\nLME p = 0.645\nLME p = 0.029\nExacerbation\nTreatment\nh)\nMetab\nAb\nAb\nTime in Days\nTime in Days −2\n−1\n0\n1\nLogRatio Anaerobe/Pathogen\ne)\np=0.014\np<0.001\nSt\nEx\nTr e) 20 FIG 2 The microbiome and metabolome variation around CFPEs. (a and b) Notch plots of the (a) microbiome weighted UniFrac distances and (b) metabolome\nBray-Curtis distances between samples classified as CFPE (\u000514 days from antibiotic treatment), treatment (during 21 days of antibiotic treatment), or stable\n(outside these time periods). Statistical significance across the class comparisons was tested using an LME model with subject as random effects and Tukey’s\npost hoc tests. (c and d) Shannon index of microbiome diversity (c) and metabolome diversity (d) in samples collected during different disease states. Statistical\nsignificance across the class comparisons was tested using an LME model with subject as random effects and Tukey’s post hoc tests. (e) Notch plots of the log\nratios of anaerobes to pathogens in samples classified as CFPE, treatment, or stable. (Statistics are presented as described above). msystems.asm.org\n6 aeruginosa virulence-associated metabolites, including\nphenazines, rhamnolipids, and quinolones, that can have strong effects on other\nmicrobes in the community and host cells in vitro (24, 27, 28). We detected 36 different\nquinolones, eight rhamnolipids, and one siderophore (pyochelin) known to be pro-\nduced by Pseudomonas aeruginosa in this longitudinal data set but detected no\nphenazines. At least one of these metabolites was found in four of the six subjects\n(CF066, CF176, CF353, and CF189, Fig. S7a), and all had Pseudomonas in their micro-\nbiome. Interestingly, none of these metabolites were detected in subject CF146, even\nthough this subject’s microbiome had on average a 77.0% relative abundance of\nPseudomonas through the collection, and CF353 had low to undetectable amounts\nof quinolones and rhamnolipids but an average of 41.0% Pseudomonas. The abundance\nof these metabolites was not associated with any disease state (LME P \u0006 0.05). Using\nthe MMvec method, we found that the quinolone 4-hydroxy-2-nonylquinolone (NHQ)\nhad a high conditional log-probability of association with Pseudomonas in the data\n(1.18 logP, 98th percentile) whereas its related metabolite 4-hydroxy-2-heptylquinolone\n(HHQ) also showed a high level of association (1.035 logP, 96th percentile). These\nconditional probabilities did not correspond to strong linear associations of the relative\nabundance of Pseudomonas with the metabolite abundances (Fig. S7a), highlighting\nthe importance of compositionally coherent approaches such as MMvec (29). However,\nthe metabolites themselves, particularly the two quinolones and the rhamnolipids,\nwere highly correlated with each other (Fig. S7a). Learning associations of metabolites\nwith the other ASVs of interest showed a negative association between the Pseudomo-\nnas quinolones and anaerobes. NHQ and HHQ were highly negatively associated with\nStreptococcus (\u00054.34 logP and 99th percentile and \u00053.80 and 99th percentile, respec-\ntively), Veillonella parvula (\u00054.65 logP and 99th percentile and \u00054.06 and 99th percen-\ntile), and Prevotella melaninogenica (\u00053.51 logP and 99th percentile and \u00053.06 and\n99th percentile) (Fig. S6b; see also Data Set S1, sheet 9). Multi-omics analysis of a CF mortality event. The death of subject CF176 due to\nrespiratory failure presented an opportunity to study the changes that occurred in the\nmicrobiome and metabolome as the fatal CFPE developed and subsequent treatment\nfailed. Samples from CF176 were available for 118 (60%) of the 194 days prior to death\nduring which the subject experienced four separate CFPEs and subsequent treatments\nwith intravenous (i.v.) antibiotics, including the final course prior to death (Fig. 3a). The random forest model did not\nindicate strong overall changes in the metabolome around exacerbation (out-of-\nbag error rate \u0002 27.5%, Data Set S1, sheet 6). Variables of importance to the classifi-\ncation were primarily represented by antibiotics given to the subjects. Nonantibiotic\nmetabolites of importance to the classification included stearoyl-L-carnitine and hemin;\nhowever, the corresponding data were not significantly different between the exacer- May/June 2020\nVolume 5\nIssue 3\ne00292-20 msystems.asm.org\n6 msystems.asm.org\n6 Longitudinal CF Microbiome Dynamics bation and treatment states across subjects using an LME model (false-discovery-rate\n[FDR] corrected q \u0006 0.05). bation and treatment states across subjects using an LME model (false-discovery-rate\n[FDR] corrected q \u0006 0.05). In light of the findings previously reported by Caverly et al. (23), who showed\nchanges in bacterial diversity during times of stability due to maintenance therapies,\nand of the fact that we detected some maintenance antibiotics in the metabolomic\ndata during stable periods, we analyzed the correlations between the abundances of\nazithromycin and trimethoprim (provided as maintenance therapies) and microbiome\nalpha-diversity. We found a negative correlation between microbiome Shannon diver-\nsity and the abundance of trimethoprim in the same samples during stability (linear\nmixed model [LMM] P \u0002 0.0048, Spearman’s rho \u0002 \u00050.39) but not between micro-\nbiome Shannon diversity and the abundance of azithromycin (LMM P \u0002 0.489, Spear-\nman’s rho \u0002 0.35, Fig. S6). [FDR] corrected q \u0006 0.05). In light of the findings previously reported by Caverly et al. (23), who showed\nchanges in bacterial diversity during times of stability due to maintenance therapies,\nand of the fact that we detected some maintenance antibiotics in the metabolomic\ndata during stable periods, we analyzed the correlations between the abundances of\nazithromycin and trimethoprim (provided as maintenance therapies) and microbiome\nalpha-diversity. We found a negative correlation between microbiome Shannon diver-\nsity and the abundance of trimethoprim in the same samples during stability (linear\nmixed model [LMM] P \u0002 0.0048, Spearman’s rho \u0002 \u00050.39) but not between micro-\nbiome Shannon diversity and the abundance of azithromycin (LMM P \u0002 0.489, Spear-\nman’s rho \u0002 0.35, Fig. S6). Metabolite and microbiome associations. We used a method of identifying\nmetabolites associated with microbial amplicon sequence variants (ASVs) called\nmicrobial-metabolite vectors (MMvec; 26) to investigate the relationship of these\nmolecules with a changing microbial community (Data Set S1, sheet 9). We gave\nparticular attention to P. May/June 2020\nVolume 5\nIssue 3\ne00292-20 These four exacerbations were experienced in a short time frame, indicating they may\nhave been related, but each event was treated with a new course of different antibiotic\ncombinations. The microbial community was dominated by Pseudomonas, with a moderate in-\ncrease in relative abundance through time (Pearson’s r \u0002 0.323, P \u0002 0.0004). There were\nspikes in the relative abundances of anaerobes and other oral bacteria observed, but msystems.asm.org\n7 May/June 2020\nVolume 5\nIssue 3\ne00292-20 Raghuvanshi et al. these did not reliably correspond to CFPEs or their treatment in the first 60 days of\n0e+00\n2e+06\n4e+06\n6e+06\n0\n50\n100\n150\nNormDay\n0\n50\n100\n150\nMolecule\nAzithromycin \nCiprofloxacin\nDescladinose Azithromycin\nIvacaftor\nLinezolid\nSulfamethoxazole\nTrimethoprim\nArea Under Curve\nPseudomonas\n100%\n0%\nRelative Abundance\nCipro/\nMeropenem/\nColistin\nAzith Stopped\nCipro/Bactrim\nDays\nLevoquin\nMeropenem/\nColistin/Linezolid\nDeceased\nDay 194\nDay \n183\n1e+03\n1e+04\n1e+05\n1e+06\n0\n50\n100\n150\nMolecule\nArg−Ile/Leu\nCer(d18:1/14:0)\nGlu−Trp\nHNP1\nIle/Leu−Val\nIle/Leu−Trp\nPGK1_peptide\nMS \nBackground\n100%\n0%\nPseudomonas\nRelative Abundance\nDays\nLog Area Under Curve\n0e+00\n2e+05\n4e+05\n6e+05\n0\n50\n100\n150\nMolecule\nHHQ\nNHQ\nPyochelin\nPseudomonas\n100%\n0%\nRelative Abundance\nEx\nTr\nEx\nTr\nEx\nTr\nEx\nTr\nEx\nTr\nEx\nTr\nEx\nTr\nEx\nTr\nArea Under Curve\na)\nb)\nc)\nDays\nFIG 3 Microbial and metabolite changes prior to death of subject CF176. (a) Area under curve abundance of antibiotics detected in\nthe metabolomics data from this subject (left y axis). The right y axis shows the abundance of Pseudomonas in the microbiome data\nplotted as a black line for reference. The x-axis data represent continuous time in days of sample collection for this subject. Black stars\nindicate antibiotics administered to the subject but not detected in the metabolomic data. (b) Area under curve abundance of\nPseudomonas aeruginosa virulence-associated metabolites detected in the metabolomics data (left y axis). The right y axis shows the\nabundance of Pseudomonas in the microbiome data plotted as a black line for reference. (c) Log10 area under curve abundance of\nincreasing levels of metabolites in CF176 through the collection time. The abundances of the metabolites are plotted along with the\nlocally estimated scatterplot smoothing (LOESS) regression line for each molecule. The Pseudomonas relative abundance data are\nagain plotted as a black line on the right y axis for reference. Ex, exacerbation; HNP1, human neutrophil peptide 1; PGK1,\nphosphoglycerate kinase 1; Tr, treatment. Raghuvanshi et al. 0e+00\n2e+06\n4e+06\n6e+06\n0\n50\n100\n150\nNormDay\n0\n50\n100\n150\nMolecule\nAzithromycin \nCiprofloxacin\nDescladinose Azithromycin\nIvacaftor\nLinezolid\nSulfamethoxazole\nTrimethoprim\nArea Under Curve\nPseudomonas\n100%\n0%\nRelative Abundance\nCipro/\nMeropenem/\nAzith Stopped\nCipro/Bactrim\nDays\nLevoquin\nMeropenem/\nColistin/Linezolid\nDeceased\nDay 194\nDay \n183\na) 0e+00\n2e+06\n4e+06\n6e+06\n0\n50\n100\n150\nNormDay\n0\n50\n100\n150\nMolecule\nAzithromycin \nCiprofloxacin\nDescladinose Azithromycin\nIvacaftor\nLinezolid\nSulfamethoxazole\nTrimethoprim\nArea Under Curve\nPseudomonas\n100%\n0%\nRelative Abundance\nCipro/\nMeropenem/\nColistin\nAzith Stopped\nCipro/Bactrim\nDays\nLevoquin\nMeropenem/\nColistin/Linezolid\nDeceased\nDay 194\nDay \n183\n0e+00\n2e+05\n4e+05\n6e+05\n0\n50\n100\n150\nMolecule\nHHQ\nNHQ\nPyochelin\nPseudomonas\n100%\n0%\nRelative Abundance\nEx\nTr\nEx\nTr\nEx\nTr\nEx\nTr\nArea Under Curve\na)\nb)\nc)\nDays a) Meropenem/\nColistin\nCipro/Bactrim\nDays\nMeropenem/\nColistin/Linezolid\nDay 194\n183\n0e+00\n2e+05\n4e+05\n6e+05\n0\n50\n100\n150\nMolecule\nHHQ\nNHQ\nPyochelin\nPseudomonas\n100%\n0%\nRelative Abundance\nEx\nTr\nEx\nTr\nEx\nTr\nEx\nTr\nArea Under Curve\nb)\nc)\nDays b) Relative Abundance 1e+03\n1e+04\n1e+05\n1e+06\n0\n50\n100\n150\nMolecule\nArg−Ile/Leu\nCer(d18:1/14:0)\nGlu−Trp\nHNP1\nIle/Leu−Val\nIle/Leu−Trp\nPGK1_peptide\nMS \nBackground\n100%\n0%\nPseudomonas\nRelative Abundance\nDays\nLog Area Under Curve\nEx\nTr\nEx\nTr\nEx\nTr\nEx\nTr\nEx\nTr\nEx\nTr\nEx\nTr\nEx\nTr\nc)\nDays c) Relative Abundance FIG 3 Microbial and metabolite changes prior to death of subject CF176. (a) Area under curve abundance of antibiotics detected in\nthe metabolomics data from this subject (left y axis). The right y axis shows the abundance of Pseudomonas in the microbiome data\nplotted as a black line for reference. The x-axis data represent continuous time in days of sample collection for this subject. Black stars\nindicate antibiotics administered to the subject but not detected in the metabolomic data. (b) Area under curve abundance of\nPseudomonas aeruginosa virulence-associated metabolites detected in the metabolomics data (left y axis). The right y axis shows the\nabundance of Pseudomonas in the microbiome data plotted as a black line for reference. (c) Log10 area under curve abundance of\nincreasing levels of metabolites in CF176 through the collection time. The abundances of the metabolites are plotted along with the\nlocally estimated scatterplot smoothing (LOESS) regression line for each molecule. The Pseudomonas relative abundance data are\nagain plotted as a black line on the right y axis for reference. May/June 2020\nVolume 5\nIssue 3\ne00292-20 Ex, exacerbation; HNP1, human neutrophil peptide 1; PGK1,\nphosphoglycerate kinase 1; Tr, treatment. these did not reliably correspond to CFPEs or their treatment in the first 60 days of\ncollection (Fig. S8a and b). However, the onset of the third CFPE corresponded to an\nincrease in the relative abundance of Haemophilus parainfluenzae which decreased\nupon treatment with levofloxacin (Levoquin) (Fig. S8a and b). The subject then entered msystems.asm.org\n8 May/June 2020\nVolume 5\nIssue 3\ne00292-20 Longitudinal CF Microbiome Dynamics a brief period of stability before experiencing a final fourth CFPE which featured\nanother spike in the relative abundance of oral microbes (Fig. S8a and b). Treatment of\nthis final CFPE consisted of intravenous (i.v.) meropenem, colistin, and linezolid fol-\nlowed by oral antibiotics in the 3 weeks before death. During this time, the Pseudomo-\nnas data reached a level of over 99.9% of sequenced reads and did not drop below\n99.7% throughout the treatment course (Fig. 3a). No other known CF pathogens were\ndetected in the final days of life (Fig. S8b). There were no increases in the relative\nabundances of anaerobes during the final treatments until 4 days before hospitalization\nat day 183. Unweighted UniFrac analysis suggested that the microbiome was changing\nthrough time (Fig. S8c and d), but weighted UniFrac analysis showed that the high\nrelative abundance of Pseudomonas may have obscured the presence of a new infec-\ntion in the final days of life (Fig. S8c and d). The metabolomic data did not reveal strong changes in the months before death,\nthough the effects of treatment were evident. For example, the level of i.v. antibiotics\ndetected in the sputum corresponded to time of treatment, suggesting that the drugs\nwere successfully diffusing to the target community. Metabolites from P. aeruginosa\nwere also detected in this subject through time, but only sporadically, and did not\ncoincide with CFPE, changes in Pseudomonas relative abundance, antibiotics, or death. HHQ, NHQ, and pyochelin were detected primarily early on in the sample collection and\nnot in the final days of life (Fig. 3b). Molecules whose levels increased with time toward\ndeath were primarily peptides, human neutrophil protein 1 (HNP1), and a ceramide\n(Fig. 3c). The levels of these molecules increased gradually throughout the collection,\nwith HNP1 increasing in the final 50 days of life. Overall, the multi-omics analysis in this subject did not reveal any new CF pathogen\ninfections, increased production of metabolites from Pseudomonas, or large changes in\nthe overall metabolome/microbiome that could explain the subject’s death (Fig. 3; see\nalso Fig. S8). Instead, the data reported here support a scenario where this subject had\nhad a gradual increase in the relative abundance of Pseudomonas in sputum corre-\nsponding to increasing levels of metabolites associated with pulmonary inflammation\nleading up to death. May/June 2020\nVolume 5\nIssue 3\ne00292-20 DISCUSSION This longitudinal study of six CF subjects provides further support for the notion of\nthe presence of a dynamic microbiome and metabolome in CF sputum, as described by\nCaverly et al. (23), and provides new evidence for changes through the development\nand subsequent treatment of CFPEs. Collectively comparing across disease states, the\nlevel of microbial alpha-diversity and the log ratio of anaerobes to pathogens were\ndecreased during antibiotic treatment for CFPEs but not between the stable and CFPE\nstates. Comparing both of these measures with time showed that there was a linear\nincrease approaching a CFPE treatment event, indicating that the microbiome changed\nas a CFPE developed. There was then a decrease in alpha-diversity upon initiation of\ntreatment, and this remained lower without a temporal change through the treatment\nperiod. Linear changes with time around CFPE were not identified in the metabolome,\nlikely due to the strong personalization in this data set. The signal for a reduction in\nmicrobial diversity and anaerobe abundance during antibiotic treatment supports\npreviously described changes in a large cross-sectional study (5) and in vitro experi-\nments (4). Thus, there is mounting evidence for changes between classic pathogens\nand anaerobic bacteria around CFPE events (4, 5, 13). However, our data also showed\nmicrobiome and metabolome dynamics during times of stability which was, at least in\npart, driven by maintenance antibiotics. Thus, future work is needed to determine if\nthere are clinically relevant consequences of a changing microbiome during stability\nand if reduction of anaerobes during CFPE treatment corresponds to clinical response. This work will require extensive in vitro and in vivo experiments supported by mathe-\nmatical models and clinical insights (23). There are several important caveats for this study. First, at least a portion of the\nanaerobes detected in our samples might have been contaminants from the oral cavity msystems.asm.org\n9 Raghuvanshi et al. during sputum expectoration (25). It is therefore possible that antibiotics were impact-\ning the oral microbiota, which would then be reflected in contaminated sputum (26). The microbiome of subject CF189, however, where a single Prevotella ASV came to\ndominate the sputum microbiome for months, demonstrated that the anaerobe dy-\nnamics were unlikely to have stemmed from saliva contamination during expectora-\ntion, which would be expected to produce a more diverse and consistent assemblage\nof oral bacteria through time. DISCUSSION Second, the grouping of samples into stable, CFPE, and\ntreatment periods is not without its limitations. Subjects often receive oral antibiotics\nprior to CFPEs, and many cycle antibiotics as maintenance therapies, including the\nsubjects in this study. Caverly et al. (23) identified maintenance antibiotics as a\ncontributing factor to microbiome changes during stability. This study found a negative\nrelationship between the abundance of the maintenance antibiotic trimethoprim in the\nmetabolomic data and microbiome Shannon diversity during times of stability. This\nsupports the notion that maintenance antibiotics may have contributed to dynamics\nduring stability, although this relationship was not found with azithromycin and we\ncannot detect all maintenance drugs with our mass spectrometry methods. Third,\nsputum may be produced from different parts of the lung, which are known to have\ndiffering populations of microbiota (20). Nevertheless, even though sputum samples\nare not completely representative of the lung microbiome and may contain differing\ndegrees of contamination from the oral cavity, the ease and low invasiveness of sputum\ncollection enabled the large sample size in this study, which would have been impos-\nsible with more-invasive sample techniques, such as bronchoalveolar lavage, that more\ndirectly target lung microbiota. Another important point is that the subjects in this\nstudy were in different stages of disease progression, a factor which has been shown\nto impact the microbiome dynamics during CFPE therapy (5). There is evidence for this\nin our study as well, as late-stage subject CF176 had the most stable microbiome\nthrough time. The collection of paired microbiome and metabolome data in our study enabled\nfurther investigation into the association between metabolites produced by Pseudomo-\nnas aeruginosa and ASVs of interest in the microbiome profiles. As reported previously\nfrom a cross-sectional study (17), in many samples that had large amounts of P. aeruginosa in the microbiome, there were small amounts or no metabolites from the\nbacterium detected. This may reflect dynamic changes in the growth rate and metab-\nolite production of the bacterium in a subject through time, but this is only speculative,\nas many biological phenomena could explain this disparity, including disproportionate\nproduction of these compounds from different P. aeruginosa strains. Because micro-\nbiome/metabolite correlations are complicated by compositionality (30, 31), we applied\na recently published method of computing metabolite-microbe conditional probabili-\nties and did find an association between the Pseudomonas ASV and its quinolones. This\nmethod also identified negative associations between the P. May/June 2020\nVolume 5\nIssue 3\ne00292-20 msystems.asm.org\n10 MATERIALS AND METHODS Sample collection and clinical information. The first sample collection comprised a total of 572\nsputum samples from six CF subjects (see Data Set S1, sheet 2, in the supplemental material). These\nsubjects were targeted due to recently poor health measures and lung function decline. A single Midea\nWHS-129C1 single-door chest freezer (3.5 cubic feet) was sent to the homes of each subject after consent\nto the study under institutional review board (IRB) research protocol no. 160078 (University of California\n[UC] San Diego). Subjects were asked to collect samples daily or as frequently as possible at their own\ndiscretion. One subject (CF146) collected twice daily for 15 of the 22 collection days, and data from these\nsamples were averaged to represent a single sputum sample from that date. Samples were expectorated\ninto 50-ml conical tubes that were then labeled by the subjects and stored in the freezers. At their\nconvenience, subjects collected their samples and brought them on ice to the adult CF clinic at UC San\nDiego for permanent storage at –80°C. Samples were thawed once for aliquoting into cryovials and\nfrozen again prior to multi-omics analysis. After this initial sample collection, a secondary set of 22 samples was collected from subject 176 after\nthis subject died. These samples were collected from the freezer retrospectively after death and shipped\novernight on ice to the UC San Diego research laboratory for processing. Due to their priority, they were\nprocessed in the same manner as all others in the collection except that DNA and metabolites were\nextracted in triplicate wells of a 96-well plate and analyzed in triplicate. All subsequent plots and\nstatistical analysis were done using means of data from the three triplicates for both the microbiome and\nmetabolome. Samples from this secondary collection were integrated with the first set from the same\nsubject as a case study of this single individual. However, these additional samples were not included in\nthe statistical analyses describing the microbiome and metabolome dynamics from the initial collection,\ndue to the potential for batch effects between runs that especially affect beta-diversity measures. The samples were classified as exacerbation, treatment, or stable samples according to the clinical\ndata obtained from the attending physician (Data Set S1, sheets 1 and 2). DISCUSSION aeruginosa quinolones and\ncertain anaerobes, supporting the notion of the mutually exclusive dynamic between\nanaerobes and this bacterium that has been described previously (10). This negative\nassociation may represent antagonism between anaerobes and P. aeruginosa or con-\ntrasting niche occupancies, but future experiments are needed to identify any causal\nrelationships behind these associations. An important limitation of our metabolomics\nmethods is that the liquid chromatography-tandem mass spectrometry (LC-MS/MS)\nand extraction protocols used sample only a portion of the metabolome; additional\nextraction and mass spectrometry approaches will be needed to more comprehensively\nassess the sputum metabolomic makeup. It is also of note that other organisms can also\nproduce some of these secondary metabolites; for example, Burkholderia spp. can\nproduce rhamnolipids, pyochelin, and quinolones (32–34). The unfortunate death of subject CF176 during this study provided an opportunity\nto study the changes in the microbial community and metabolome of sputum that\noccurred in the final stages of this disease. While a Pseudomonas ASV was the most\nhighly abundant organism for most of the final 182 days of life, there were periods\npunctuated by increases in other microbes, particularly Haemophilus. There was no msystems.asm.org\n10 Longitudinal CF Microbiome Dynamics evidence of a new pathogen infection, a particularly marked microbiome change, or a\ndramatic increase in inflammation that might have explained this subject’s mortality. Instead, the data showed a progressive increase in the relative abundance of Pseu-\ndomonas, possibly driven by antibiotics administered for CFPEs, and a progressive\nincrease in the relative abundances of metabolites and peptides associated with\npulmonary inflammation. This type of mortality event may be common in late-stage CF,\nbecause many individuals exhibit a pathogen-dominated microbiome at this stage of\ndisease progression (2, 17), but other multi-omics data surrounding mortality events\nbesides this n \u0002 1 case study are scarce. One report that is available on death associ-\nated with a severe CFPE implicated a new infection from Escherichia coli as a possible\ncause, though the conclusions were only speculative (35). As demonstrated here,\nmulti-omic analyses enable insights into the complex interactions between host,\nmicrobe, and drugs through these tragic events of chronic disease. Perhaps most\nsignificantly, microbiome and metabolome data can now be generated in clinically\nrelevant time frames (36), so this approach is a feasible route for future investigations\nin the clinic. DISCUSSION In conclusion, this study showed that the CF sputum microbiome is highly dynamic\nin some subjects through time, including during periods of clinical stability. During the\ndevelopment of a pulmonary exacerbation, there was an increase in microbial diversity\ncorresponding to a relative increase of the ratio of anaerobes to pathogens, which then\ndecreased during treatment. Thus, dynamics between classic pathogens and anaerobic\nbacteria around CFPE events may be important for therapeutic outcomes (4, 5, 10). Future studies that target CFPE therapy in a systematic manner are needed, particularly\nwhen the same antibiotic is repeatedly provided, to determine if any of the observed\ndynamics are predictable and if the reduction in anaerobe abundance during treatment\ncorresponds to positive clinical outcomes. If so, this could lead to more precisely\ntargeted and efficacious treatments for CF and improvements in subject quality and\nduration of life. May/June 2020\nVolume 5\nIssue 3\ne00292-20 MATERIALS AND METHODS Exacerbations were defined as\nan increase in pulmonary symptoms associated with CF disease and the decision to administer intrave-\nnous or oral antibiotics to treat these symptoms for 21 days. The specific antibiotics administered for a\nCFPE were recorded as well as the start and end dates of each CFPE treatment course (see Data Set S1,\nsheets 1 and 2). The maintenance therapies given to each subject were also recorded, but the dates when\nthese drugs were taken are not known (note that some maintenance therapies are detected in the\nmetabolomics data, aiding interpretation of administration date; see the supplemental material). For\nanalysis of antibiotic therapy effects on the microbiome and metabolome, samples were classified as\n“exacerbation” samples if they were collected within 14 days of an exacerbation diagnosis or as\n“treatment” samples if they were collected during the 21-day treatment course. If there was a change in msystems.asm.org\n11 Raghuvanshi et al. the antibiotic chosen during the 21 days, this did not affect sample classification. Whether subjects were\nprescribed routine oral antibiotics prior to an exacerbation diagnosis was not considered in the\nclassification, but to be included in the analysis as a CFPE, there had to be at least four samples collected\nprior to treatment. DNA extraction and 16S rRNA gene PCR. A 200-\u0002l aliquot of each sputum sample from a cryovial\nwas added to a Thermo Scientific 96-well deep-well plate after thawing. The six plates containing these\nsputum samples were then subjected to DNA extraction performed with a Qiagen PowerSoil DNA\nextraction kit in 96-well format. The DNA extraction, PCR amplification, and barcoding of the V4 region\nof the bacterial 16S rRNA gene were completed according to the protocols for the earth microbiome\nproject (37) described elsewhere (http://press.igsb.anl.gov/earthmicrobiome/protocols-and-standards/\n16s/). These protocols contain blank (no template) control samples that are used to identify background\nsequences in reagents or other contaminants. Microbiome data processing and analysis. Raw sequence data were processed using Qiita (38) and\nwere quality filtered following filtering recommendations (39) and processed by Deblur (40) to generate\namplicon sequence variants (ASVs). Sequences were aligned in QIIME2 version 1.9.1 (41) using MAFFT in\norder to construct a phylogenetic tree using fasttree2. Taxonomy was assigned using q2-feature-classifier\n(42) against the 99% GreenGenes 16S rRNA reference database (version 13-8). MATERIALS AND METHODS Samples with fewer than\n500 reads were removed, and the data were rarefied to 500 reads per sample, leaving 552 sputum\nsamples for analysis. Information about the sequencing depth is available in the supplemental material\n(Data Set S1, sheet 7; see also Fig. S2 in the supplemental material). For analysis, QIIME2 version 2019.4.0\nwas used throughout. Core diversity metrics were computed using core metrics phylogenetic analysis for\nalpha- and beta-diversity indices. Based on their assigned taxonomy, ASVs were classified as either\npathogens or anaerobes (Data Set S1, sheet 3). This classification was performed based on which highly\nabundant ASVs correspond to known classic CF pathogens targeted for antibiotic susceptibility in clinical\nlaboratories and on other highly abundant ASVs in the entire data set that are known to represent\nobligate or aerotolerant anaerobes. The classified organisms collectively comprised 94.4% of total\nsequence reads in the data set. Organisms that did not fall into the classic pathogen or anaerobe\ncategories (i.e., those that are not commonly considered CF pathogens but are not known anaerobes)\nwere not included in calculations of pathogen/anaerobe ratios. The ASV assigned to the Pseudomon-\nadaceae family was searched against the NCBI database with BLAST and verified to have 100% sequence\nsimilarity to Pseudomonas aeruginosa and other pseudomonads. It is therefore referred to as Pseudomo-\nnas throughout the manuscript. Similarly, the ASV assigned to Escherichia had 100% identity to Esche-\nrichia spp., but the species was not identified due to high levels of similarity in the 16S rRNA gene V4\nregion within this group. Metabolite extraction and metabolomics. Metabolites were extracted using a modified version of\nthe 96-well plate extraction procedure described previously by Quinn et al. (19). Briefly, a 200-\u0002l aliquot\nof each sputum sample was added to a Thermo Scientific 96-well deep-well plate for metabolite\nextraction. First, 300 \u0002l of ethyl acetate was added to the sputum, subjected to vortex mixing, and\nallowed to extract at room temperature overnight. The plate was then spun at 2,000 \u0004 g in a tabletop\ncentrifuge, and then 200 \u0002l of the ethyl acetate layer was removed and dried in the plate overnight. Next,\n300 \u0002l of methanol was added to the remaining sputum, subjected to vortex mixing, and then extracted\novernight at 4°C. The plate was spun again to separate particulates from the methanol extract, and 200 \u0002l\nwas added to the dried ethyl acetate extract. May/June 2020\nVolume 5\nIssue 3\ne00292-20 MATERIALS AND METHODS The same respective distance measures were used to test for significance of within-subject and\nbetween-subject beta-diversity by computing the mean distances of each sample from a subject to all\nsamples from other subjects (between distances) and comparing the results to the distances of each\nsample from a subject to all others from the same subject (within distances). Comparisons within and\nbetween subjects for both microbiome and metabolome were tested for significance using a linear\nmixed-effects model (LME) from the lmer4 package in R with subject as a random effect. The alpha- and\nbeta-diversities of microbiome and metabolome samples were also compared across subjects using a\nKruskall-Wallis test for significant differences. A post hoc Dunn test adjusted for multiple comparisons\nperformed with the Benjamini-Hochberg method was used to compare pairwise significance data\ncorresponding to microbiome and metabolome beta-diversity between individuals. To determine whether the levels of microbiome and metabolome beta-diversity differed between\ndisease states, all samples from each subject were compared to each other within each disease state in\na pairwise manner using the weighted UniFrac distance (fixed effect). These comparisons were limited,\nhowever, to samples collected within 21 days of each other, to minimize the large variations that might\nhave occurred through time and to normalize the time across disease state classes (i.e., not to compare\nexacerbations or stable samples collected months apart). These UniFrac distance values were then\ncompared using an LME model with subject source as a random effect. The post hoc pairwise compar-\nisons across disease states were then tested with a Tukey’s test on the mixed model using the\nSimultaneous Inference of General Parametric Models (multcomp) package in R. Alpha-diversity changes\n(Shannon index) and the log ratio of anaerobes to pathogens across different disease states were also\ntested with the same LME method. Changes in total bacterial load in the three disease states were tested\nwith the same LME model. For longitudinal comparisons of changes in the microbiome and metabolome with time through\nCFPE development and treatment, LME models were used with subjects set as random effects to account\nfor the different sample numbers from the six subjects. All models were run with the lmer4 package in\nR statistical software. Statistical significance was computed using Satterthwaite’s degrees of freedom\nmethod from the lmerTest package in R. MATERIALS AND METHODS The chromatograms were built with the following parameters: MS1 noise level of 5000 counts, MS2\nnoise level of 200 counts, minimum time span for chromatograms of 0.01, minimum height of 10,000, msystems.asm.org\n12 Longitudinal CF Microbiome Dynamics and a 10-ppm mass tolerance. Chromatograms were deconvoluted, and the isotope peaks were grouped\nto remove redundancy. The data were aligned with a retention time tolerance of 0.2 min and an m/z\ntolerance of 0.03 or 10 ppm. Metabolites detected in blanks and background controls were removed\nprior to statistical analysis. Statistical analysis. The weighted UniFrac (45) distance matrix was computed in QIIME2 to project\nthe sample similarities of the microbiome data in a three-dimensional principal-coordinate analysis\n(PCoA) plot using EMPeror (46). Similarly, the Bray-Curtis distance matrix was computed in R on the\nmetabolomics feature table and visualized with a PCoA plot (the UniFrac distance cannot currently be\ncalculated on metabolome data). PERMANOVA testing on the beta-diversity measures was done with\nsubject source and disease state classifiers in R using the vegan package. Testing with respect to disease\nstate accounted for subject source as a covariate using the “strata” function. The microbiome data were\nalso plotted in PCoA space with the third axis as time in days, to visualize the changing microbiome\nthrough time (see Video S1 in the supplemental material). The UniFrac and Bray-Curtis distance values\nwere quantified within each subject through time using a method similar to that described previously by\nCaverly et al. (23). All pairwise comparisons of each sample to all others from the same subject were\nplotted as notch plots, and the percentages of samples outside 1.5\u0004 the interquartile range (IQR)\nwere calculated. In addition, the percentages of samples within a subject with a distance value above 0.6\nwere reported to assess those that were highly different from the rest. To provide relevant comparisons\nof these numbers, the same calculations were then done on a publicly available cross-sectional CF\nsputum data set analyzed with microbiome and metabolome methods published previously (17) sputum data set analyzed with microbiome and metabolome methods published previously (17). May/June 2020\nVolume 5\nIssue 3\ne00292-20 MATERIALS AND METHODS The extracted metabolites were then diluted 1:2 in\nmethanol spiked with 2 \u0002M ampicillin as an internal standard. Control blank samples also went through\nthe entire extraction process but without sputum to allow for removal of background signals from the\nsolvents and mass spectrometer. This extract was analyzed by injection into a Bruker Daltonics Maxis\nImpact II LC-MS/MS system according to the mass spectrometry protocols described previously by Quinn\net al. (17). Briefly, a 20-\u0002l injection volume was separated using a Kinetex 1.7-\u0002m-pore-size C18\nultraperformance liquid chromatography (UPLC) column (50 by 2.10 mm) with a linear gradient of 2:98\nwater to acetonitrile progressing to 98:2 acetonitrile to water for a 14-min run. The data were then\nconverted to the .mzXML format for metabolite quantitation and annotation with GNPS. Volatile\nmetabolites, those that are strongly nonpolar, and those that ionize only in negative mode are not likely\nto be detected with this method. GNPS analysis and metabolite feature finding. Data from the mass spectrometer expressed in\nBruker .d format were first converted to the .mzXML format and then uploaded to the GNPS database\nand data analysis server (gnps.ucsd.edu). The data are publicly available as MassIVE data set number\nMSV000082667. Molecular networks were built on GNPS with the following parameters: mass tolerances\nof 0.03 Da, cosine score of 0.65, minimum number of matched fragment ions of 4, and minimum cluster\nsize of 3 spectra. Library searching parameters were the same with a cosine score of 0.65 and a minimum\nnumber of matched peaks of 4 (a list of library hits is available in Data Set S1, sheet 4; these are level 2\naccording to the metabolomics standards initiative [43]). Metabolites that were annotated without direct\nGNPS library hits were identified through propagating annotations through the GNPS networks and\ninspection of MS/MS spectral patterns (Data Set S1, sheet 7; level 2 according to a previously described\nclassification system [43]). The molecular network used for analysis of the sputum data in this project is\navailable\nat\nhttps://gnps.ucsd.edu/ProteoSAFe/status.jsp?task\u0002e9e9002371794bedbf8faf38e632a3f4. The secondary data set had the same parameters used, but the molecular network is available at\nhttps://gnps.ucsd.edu/ProteoSAFe/status.jsp?task\u000270f4483662db4489821a3773783f9641. Area under the curve abundances of metabolite features were quantified using mzMine2 software\n(44). REFERENCES profiling reveal specific alterations in bacterial community structure and\nenvironment in the cystic fibrosis airway during exacerbation. PLoS One\n8:e82432. https://doi.org/10.1371/journal.pone.0082432. profiling reveal specific alterations in bacterial community structure and\nenvironment in the cystic fibrosis airway during exacerbation. PLoS One\n8:e82432. https://doi.org/10.1371/journal.pone.0082432. 1. Zhao J, Schloss PD, Kalikin LM, Carmody LA, Foster BK, Petrosino JF,\nCavalcoli JD, VanDevanter DR, Murray S, Li JZ, Young VB, LiPuma JJ. 2012. Decade-long bacterial community dynamics in cystic fibrosis air-\nways. Proc Natl Acad Sci U S A 109:5809–5814. https://doi.org/10.1073/\npnas.1120577109. 1. Zhao J, Schloss PD, Kalikin LM, Carmody LA, Foster BK, Petrosino JF,\nCavalcoli JD, VanDevanter DR, Murray S, Li JZ, Young VB, LiPuma JJ. 2012. Decade-long bacterial community dynamics in cystic fibrosis air-\nways. Proc Natl Acad Sci U S A 109:5809–5814. https://doi.org/10.1073/\npnas.1120577109. 10. Quinn RA, Whiteson K, Lim YW, Zhao J, Conrad D, Lipuma JJ, Rohwer F,\nWidder S. 2016. Ecological networking of cystic fibrosis lung infections. NPJ\nBiofilms Microbiomes 2:4. https://doi.org/10.1038/s41522-016-0002-1. Biofilms Microbiomes 2:4. https://doi.org/10.1038/s41522-016-0 2. Coburn B, Wang PW, Diaz Caballero J, Clark ST, Brahma V, Donaldson S,\nZhang Y, Surendra A, Gong Y, Elizabeth Tullis D, Yau YCW, Waters VJ, Hwang\nDM, Guttman DS. 2015. Lung microbiota across age and disease stage in\ncystic fibrosis. Sci Rep 5:10241. https://doi.org/10.1038/srep10241. 11. Price KE, Hampton TH, Gifford AH, Dolben EL, Hogan DA, Morrison HG,\nSogin ML, Toole G. 2013. Unique microbial communities persist in\nindividual cystic fibrosis patients throughout a clinical exacerbation. Microbiome 1:27. https://doi.org/10.1186/2049-2618-1-27. 3. Caverly LJ, Zhao J, LiPuma JJ. 2015. Cystic fibrosis lung microbiome:\nopportunities to reconsider management of airway infection. Pediatr\nPulmonol 50(Suppl 4):S31–S38. https://doi.org/10.1002/ppul.23243. Microbiome 1:27. https://doi.org/10.1186/2049-2618-1-27. 12. Carmody LA, Zhao J, Kalikin LM, LeBar W, Simon RH, Venkataraman A,\nSchmidt TM, Abdo Z, Schloss PD, LiPuma JJ. 2015. The daily dynamics of\ncystic fibrosis airway microbiota during clinical stability and at exacer-\nbation. Microbiome 3:12. https://doi.org/10.1186/s40168-015-0074-9. 4. Quinn RA, Whiteson K, Lim Y-W, Salamon P, Bailey B, Mienardi S, Sanchez\nSE, Blake D, Conrad D, Rohwer F. 2015. A Winogradsky-based culture\nsystem shows an association between microbial fermentation and cystic\nfibrosis exacerbation. ISME J 9:1052–1052. https://doi.org/10.1038/ismej\n.2014.266. 13. Comstock WJ, Huh E, Weekes R, Watson C, Xu T, Dorrestein PC, Quinn RA. 2017. The winCF model - an inexpensive and tractable microcosm of a\nmucus plugged bronchiole to study the microbiology of lung infections. J Vis Exp 2017 https://doi.org/10.3791/55532. 5. MATERIALS AND METHODS The linear trend in Shannon diversity (fixed effect) and the log\nratio of the relative abundance of anaerobes compared to pathogens (fixed effect) were modeled using\nthis method. For further support, the trend in Shannon diversity through time during CFPE treatment was\nalso modeled and tested with mixed-effects models with linear splines by the use of lme in the nlme R\npackages. The temporal trajectories were assessed with a linear spline (or broken stick) model and then\ntested using “Eigen” and S4 (lme4) to determine whether there was an effect of subject status on the\nShannon diversity (fixed effects). Subjects and days relative to CFPE over time were also considered\nrandom effects in this model. The same linear mixed-model approach was used to test the correlation\nbetween the abundance of a maintenance antibiotic from the metabolomic data and Shannon diversity\nfrom the microbiome data in the same samples. A random forests classification model was used to identify metabolites that separated the stable,\nCFPE, and treatment groups. This model was run using the randomForest package in R with 5,000 trees\nand 59 variables tried at each split and with stratification due to differential sample numbers in each\ndisease class. Microbe-metabolite cooccurrence probabilities were calculated using MMvec, a neural network\napproach trained to predict metabolite abundances given the presence of a single microbe (29). This\nmodel was trained using three principal axes with a batch size of 10,000 and 10,000 epochs. MMvec\nperforms cross-validation by leaving out samples and evaluating how well the metabolites can be\npredicted solely from the microbe abundances in the hold-out samples. The cross-validation error values\nconverged, suggesting little overfitting. We first preprocessed the data by removing features/ASVs that\nappeared in fewer than 10 samples. Five samples were selected for hold-out testing, where the model\npredicted the metabolite abundances from the microbe abundances in these samples. msystems.asm.org\n13 Raghuvanshi et al. Raghuvanshi et al. Data accessibility. The microbiome data were deposited in the Qiita (38) database as project\nnumber 11400, and the second data set was deposited under accession no. 11433. The data are also\npublicly available at the European Bioinformatics Institute under accession no. ERP119164. ACKNOWLEDGMENTS We acknowledge funding from the Cystic Fibrosis Research Innovation award\nprovided by Vertex Pharmaceuticals to R.A.Q. We also acknowledge funding provided\nto R.A.Q. from the NIH (R01AI145925). Y.V.-B. is funded by the Janssen Human Micro-\nbiome Initiative through the Center for Microbiome Innovation at UC San Diego. P.C.D., R.K., D.C., and R.A.Q. designed the study. L.D.G., G.H., G.A., A.D.S., and R.A.Q. generated data. R.R., K.V., Y.V.-B., L.J., A.G., J.T.M., D.L., and R.A.Q. analyzed data. R.R., K.V.,\nL.J., A.D.S., and R.A.Q. wrote the paper. SUPPLEMENTAL MATERIAL Supplemental material is available online only. FIG S1, PDF file, 0.3 MB. FIG S2, PDF file, 0.4 MB. FIG S3, PDF file, 0.7 MB. FIG S4, PDF file, 0.6 MB. FIG S5, PDF file, 9.1 MB. FIG S6, PDF file, 0.3 MB. FIG S7, PDF file, 7.3 MB. FIG S8, PDF file, 5.4 MB. DATA SET S1, XLSX file, 0.9 MB. VIDEO S1, MOV file, 10.1 MB. REFERENCES Carmody LA, Caverly LJ, Foster BK, Rogers MAM, Kalikin LM, Simon RH,\nVanDevanter DR, LiPuma JJ. 2018. Fluctuations in airway bacterial com-\nmunities associated with clinical states and disease stages in cystic\nfibrosis. PLoS One 13:e0194060. https://doi.org/10.1371/journal.pone\n.0194060. 14. Fodor AA, Klem ER, Gilpin DF, Elborn JS, Boucher RC, Tunney MM,\nWolfgang MC. 2012. The adult cystic fibrosis airway microbiota is stable\nover time and infection type, and highly resilient to antibiotic treatment\nof exacerbations. PLoS One 7:e45001. https://doi.org/10.1371/journal\n.pone.0045001. 6. Layeghifard M, Li H, Wang PW, Donaldson SL, Coburn B, Clark ST,\nCaballero JD, Zhang Y, Tullis DE, Yau YCW, Waters V, Hwang DM,\nGuttman DS. 2019. Microbiome networks and change-point analysis\nreveal key community changes associated with cystic fibrosis pulmonary\nexacerbations. NPJ Biofilms Microbiomes 5:4. https://doi.org/10.1038/\ns41522-018-0077-y. 15. Lim YW, Evangelista JS, Schmieder R, Bailey B, Haynes M, Furlan M,\nMaughan H, Edwards R, Rohwer F, Conrad D. 2014. Clinical insights from\nmetagenomic analysis of cystic fibrosis sputum. J Clin Microbiol 52:\n425–457. https://doi.org/10.1128/JCM.02204-13. 16. Quinn RA, Lim YW, Mak TD, Whiteson K, Furlan M, Conrad D, Rohwer F,\nDorrestein P. 2016. Metabolomics of pulmonary exacerbations reveals\nthe personalized nature of cystic fibrosis disease. PeerJ 4:e2174. https://\ndoi.org/10.7717/peerj.2174. 7. Smith DJ, Badrick AC, Zakrzewski M, Krause L, Bell SC, Anderson GJ, Reid\nDW. 2014. Pyrosequencing reveals transient cystic fibrosis lung micro-\nbiome changes with intravenous antibiotics. Eur Respir J 44:922–930. https://doi.org/10.1183/09031936.00203013. 17. Quinn RA, Adem S, Mills RH, Comstock W, DeRight Goldasich L, Humphrey\nG, Aksenov AA, Melnik AV, da Silva R, Ackermann G, Bandeira N, Gonzalez\nDJ, Conrad D, O’Donoghue AJ, Knight R, Dorrestein PC. 2019. Neutrophilic\nproteolysis in the cystic fibrosis lung correlates with a pathogenic micro-\nbiome. Microbiome 7:23. https://doi.org/10.1186/s40168-019-0636-3. 8. Carmody LA, Zhao J, Schloss PD, Petrosino JF, Murray S, Young VB, Li JZ,\nLiPuma JJ. 2013. Changes in cystic fibrosis airway microbiota at pulmo-\nnary exacerbation. Ann Am Thorac Soc 10:179–187. https://doi.org/10\n.1513/AnnalsATS.201211-107OC. 9. Twomey KB, Alston M, An S-Q, O’Connell OJ, McCarthy Y, Swarbreck D,\nFebrer M, Dow JM, Plant BJ, Ryan RP. 2013. Microbiota and metabolite 9. Twomey KB, Alston M, An S-Q, O’Connell OJ, McCarthy Y, Swarbreck D,\nFebrer M, Dow JM, Plant BJ, Ryan RP. 2013. Microbiota and metabolite 18. Eiserich JP, Yang J, Morrissey BM, Hammock BD, Cross CE. 2012. .org/10.1128/mBio.00431-19. 36. Quinn RA, Navas-Molina JA, Hyde ER, Song SJ, Vázquez-Baeza Y, Hum-\nphrey G, Gaffney J, Minich JJ, Melnik AV, Herschend J, DeReus J, Durant\nA, Dutton RJ, Khosroheidari M, Green C, da Silva R, Dorrestein PC, Knight\nR. 2016. From sample to multi-omics conclusions in under 48 hours. mSystems 1:e00038-16. https://doi.org/10.1128/mSystems.00038-16. 21. Whiteson KL, Meinardi S, Lim YW, Schmieder R, Maughan H, Quinn R,\nBlake DR, Conrad D, Rohwer F. 2014. Breath gas metabolites and bac-\nterial metagenomes from cystic fibrosis airways indicate active pH neu-\ntral 2,3-butanedione fermentation. ISME J 8:1247–1258. https://doi.org/\n10.1038/ismej.2013.229. 37. Thompson LR, Sanders JG, McDonald D, Amir A, Ladau J, Locey KJ, Prill\nRJ, Tripathi A, Gibbons SM, Ackermann G, Navas-Molina JA, Janssen S,\nKopylova E, Vázquez-Baeza Y, González A, Morton JT, Mirarab S, Zech Xu\nZ, Jiang L, Haroon MF, Kanbar J, Zhu Q, Jin Song S, Kosciolek T, Bokulich\nNA, Lefler J, Brislawn CJ, Humphrey G, Owens SM, Hampton-Marcell J,\nBerg-Lyons D, McKenzie V, Fierer N, Fuhrman JA, Clauset A, Stevens RL,\nShade A, Pollard KS, Goodwin KD, Jansson JK, Gilbert JA, Knight R, Rivera\nJLA, Al-Moosawi L, Alverdy J, Amato KR, Andras J, Angenent LT, Anto-\nnopoulos DA, et al. 2017. A communal catalogue reveals Earth’s multi-\nscale microbial diversity. Nature 551:457–463. https://doi.org/10.1038/\nnature24621. 22. Wang M, Carver JJ, Phelan VV, Sanchez LM, Garg N, Peng Y, Nguyen DD,\nWatrous J, Kapono CA, Luzzatto-Knaan T, Porto C, Bouslimani A, Melnik\nAV, Meehan MJ, Liu W-T, Crüsemann M, Boudreau PD, Esquenazi E,\nSandoval-Calderón M, Kersten RD, Pace LA, Quinn RA, Duncan KR, Hsu\nC-C, Floros DJ, Gavilan RG, Kleigrewe K, Northen T, Dutton RJ, Parrot D,\nCarlson EE, Aigle B, Michelsen CF, Jelsbak L, Sohlenkamp C, Pevzner P,\nEdlund A, McLean J, Piel J, Murphy BT, Gerwick L, Liaw C-C, Yang Y-L,\nHumpf H-U, Maansson M, Keyzers RA, Sims AC, Johnson AR, Sidebottom\nAM, Sedio BE, et al. 2016. Sharing and community curation of mass\nspectrometry data with global natural products social molecular net-\nworking. Nat Biotechnol 34:828–837. https://doi.org/10.1038/nbt.3597. 38. Gonzalez A, Navas-Molina JA, Kosciolek T, McDonald D, Vázquez-Baeza\nY, Ackermann G, DeReus J, Janssen S, Swafford AD, Orchanian SB,\nSanders JG, Shorenstein J, Holste H, Petrus S, Robbins-Pianka A, Brislawn\nCJ, Wang M, Rideout JR, Bolyen E, Dillon M, Caporaso JG, Dorrestein PC,\nKnight R. 2018. Qiita: rapid, web-enabled microbiome meta-analysis. Nat\nMethods 15:796–798. https://doi.org/10.1038/s41592-018-0141-9. 23. .org/10.1128/mBio.00431-19. Caverly LJ, Lu J, Carmody LA, Kalikin LM, Shedden K, Opron K, Azar M,\nCahalan S, Foster B, VanDevanter DR, Simon RH, LiPuma JJ. 2019. Measures of cystic fibrosis airway microbiota during periods of clinical\nstability. Ann Am Thorac Soc https://doi.org/10.1513/AnnalsATS.201903\n-270OC. 24. Usher LR, Lawson RA, Geary I, Taylor CJ, Bingle CD, Taylor GW, Whyte M. 2002. Induction of neutrophil apoptosis by the Pseudomonas aeruginosa\nexotoxin pyocyanin: a potential mechanism of persistent infection. J Immu-\nnol 168:1861–1868. https://doi.org/10.4049/jimmunol.168.4.1861. 39. Bokulich NA, Subramanian S, Faith JJ, Gevers D, Gordon JI, Knight R, Mills\nDA, Caporaso JG. 2013. Quality-filtering vastly improves diversity esti-\nmates from Illumina amplicon sequencing. Nat Methods 10:57–59. https://doi.org/10.1038/nmeth.2276. 25. Goddard AF, Staudinger BJ, Dowd SE, Joshi-Datar A, Wolcott RD, Aitken\nML, Fligner CL, Singh PK. 2012. Direct sampling of cystic fibrosis lungs\nindicates that DNA-based analyses of upper-airway specimens can mis-\nrepresent lung microbiota. Proc Natl Acad Sci U S A 109:13769–13774. https://doi.org/10.1073/pnas.1107435109. 40. Amir A, McDonald D, Navas-Molina JA, Kopylova E, Morton JT, Zech Xu\nZ, Kightley EP, Thompson LR, Hyde ER, Gonzalez A, Knight R. 2017. Deblur rapidly resolves single-nucleotide community sequence patterns. mSystems 2:e00191-16. https://doi.org/10.1128/mSystems.00191-16. y\ng\ny\n41. Bolyen E, Rideout JR, Dillon MR, Bokulich NA, Abnet C, Al Ghalith GA,\nAlexander H, Alm EJ, Arumugam M, Bai Y, Bisanz JE, Bittinger K, Brejnrod\nA, Colin J, Brown CT, Callahan BJ, Mauricio A, Rodríguez C, Chase J, Cope\nE, Da Silva R, Dorrestein PC, Douglas GM, Duvallet C, Edwardson CF, Ernst\nM, Fouquier J, Gauglitz JM, Gibson DL, Gonzalez A, Huttley GA, Janssen\nS, Jarmusch AK, Kaehler BD, Bin Kang K, Keefe CR, Keim P, Kelley ST, Ley\nR, Loftfield E, Marotz C, Martin B, Mcdonald D, Mciver LJ, Alexey V,\nMetcalf JL, Morgan SC, Morton JT, Naimey AT, et al. 2018. QIIME 2:\nreproducible, interactive, scalable, and extensible microbiome data sci-\nence. PeerJ https://peerj.com/preprints/27295/. 26. Hagenfeld D, Koch R, Jünemann S, Prior K, Harks I, Eickholz P, Hoffmann\nT, Kim T-S, Kocher T, Meyle J, Kaner D, Schlagenhauf U, Ehmke B,\nHarmsen D. 2018. Do we treat our patients or rather periodontal mi-\ncrobes with adjunctive antibiotics in periodontal therapy? A 16S rDNA\nmicrobial community analysis. PLoS One 13:e0195534. https://doi.org/\n10.1371/journal.pone.0195534. 27. Mitchell G, Séguin DL, Asselin A-E, Déziel E, Cantin AM, Frost EH, Mi-\nchaud S, Malouin F. 2010. REFERENCES Omics\napproaches in cystic fibrosis research: a focus on oxylipin profiling in May/June 2020\nVolume 5\nIssue 3\ne00292-20 14 msystems.asm.org\n14 msystems.asm.org\n14 Longitudinal CF Microbiome Dynamics Quinolones: from antibiotics to autoinducers. FEMS Microbiol Rev 35:\n247–274. https://doi.org/10.1111/j.1574-6976.2010.00247.x. airway secretions. Ann N Y Acad Sci 1259:1–9. https://doi.org/10.1111/j\n.1749-6632.2012.06580.x. airway secretions. Ann N Y Acad Sci 1259:1–9. https://doi.org/10.1111/j\n.1749-6632.2012.06580.x. 19. Quinn RA, Phelan VV, Whiteson KL, Garg N, Bailey BA, Lim YW, Conrad\nDJ, Dorrestein PC, Rohwer FL. 2016. Microbial, host and xenobiotic\ndiversity in the cystic fibrosis sputum metabolome. ISME J 10:\n1483–1498. https://doi.org/10.1038/ismej.2015.207. 34. Farmer KL, Thomas MS. 2004. Isolation and characterization of Burkhold-\neria cenocepacia mutants deficient in pyochelin production: pyochelin\nbiosynthesis is sensitive to sulfur availability. J Bacteriol 186:270–277. https://doi.org/10.1128/jb.186.2.270-277.2004. 20. Garg N, Wang M, Hyde E, da Silva RR, Melnik AV, Protsyuk I, Bouslimani\nA, Lim YW, Wong R, Humphrey G, Ackermann G, Spivey T, Brouha SS,\nBandeira N, Lin GY, Rohwer F, Conrad DJ, Alexandrov T, Knight R,\nDorrestein PC. 2017. Three-dimensional microbiome and metabolome\ncartography of a diseased human lung. Cell Host Microbe 22:705–716.e4. https://doi.org/10.1016/j.chom.2017.10.001. 35. Cobián Güemes AG, Lim YW, Quinn RA, Conrad DJ, Benler S, Maughan H,\nEdwards R, Brettin T, Cantú VA, Cuevas D, Hamidi R, Dorrestein P,\nRohwer F. 2019. Cystic fibrosis rapid response: translating multi-omics\ndata into clinically relevant information. mBio 10:e00431-19. https://doi\n/10 1128/\nBi 00431 19 .org/10.1128/mBio.00431-19. Staphylococcus aureus sigma B-dependent\nemergence of small-colony variants and biofilm production following\nexposure to Pseudomonas aeruginosa 4-hydroxy-2-heptylquinoline-N-\noxide. BMC Microbiol 10:33. https://doi.org/10.1186/1471-2180-10-33. 42. Bokulich NA, Kaehler BD, Rideout JR, Dillon M, Bolyen E, Knight R, Huttley\nGA, Gregory Caporaso J. 2018. Optimizing taxonomic classification of\nmarker-gene amplicon sequences with QIIME 2’s q2-feature-classifier\nplugin. Microbiome 6:90. https://doi.org/10.1186/s40168-018-0470-z. 28. Irie Y, O’Toole GA, Yuk MH. 2005. Pseudomonas aeruginosa rhamnolip-\nids disperse Bordetella bronchiseptica biofilms. FEMS Microbiol Lett\n250:237–243. https://doi.org/10.1016/j.femsle.2005.07.012. 43. Sumner LW, Amberg A, Barrett D, Beale MH, Beger R, Daykin CA, Fan\nT-M, Fiehn O, Goodacre R, Griffin JL, Hankemeier T, Hardy N, Harnly J,\nHigashi R, Kopka J, Lane AN, Lindon JC, Marriott P, Nicholls AW, Reily MD,\nThaden JJ, Viant MR. 2007. Proposed minimum reporting standards for\nchemical analysis Chemical Analysis Working Group (CAWG) Metabolo-\nmics Standards Initiative (MSI). Metabolomics 3:211–221. https://doi.org/\n10.1007/s11306-007-0082-2. 29. Morton JT, Aksenov AA, Nothias LF, Foulds JR, Quinn RA, Badri MH,\nSwenson TL, Van Goethem MW, Northen TR, Vazquez-Baeza Y, Wang M,\nBokulich NA, Watters A, Song SJ, Bonneau R, Dorrestein PC, Knight R. 2019. Learning representations of microbe–metabolite interactions. Nat\nMethods 16:1306–1314. https://doi.org/10.1038/s41592-019-0616-3. 30. Gloor GB, Macklaim JM, Pawlowsky-Glahn V, Egozcue JJ. 2017. Micro-\nbiome datasets are compositional: and this is not optional. Front Micro-\nbiol 8:2224. https://doi.org/10.3389/fmicb.2017.02224. 10.1007/s11306-007-0082-2. 44. Pluskal T, Castillo S, Villar-Briones A, Oresic M. 2010. MZmine 2: modular\nframework for processing, visualizing, and analyzing mass spectrometry-\nbased molecular profile data. BMC Bioinformatics 11:395. https://doi\n.org/10.1186/1471-2105-11-395. 31. Morton JT, Sanders J, Quinn RA, McDonald D, Gonzalez A, Vázquez-Baeza Y,\nNavas-Molina JA, Song SJ, Metcalf JL, Hyde ER, Lladser M, Dorrestein PC,\nKnight R. 2017. Balance trees reveal microbial niche differentiation. mSys-\ntems 2:e00162-16. https://doi.org/10.1128/mSystems.00162-16. 45. Lozupone C, Knight R. 2005. UniFrac: a new phylogenetic method for\ncomparing microbial communities. Appl Environ Microbiol 71:\n8228–8235. https://doi.org/10.1128/AEM.71.12.8228-8235.2005. 32. Funston SJ, Tsaousi K, Rudden M, Smyth TJ, Stevenson PS, Marchant R,\nBanat IM. 2016. Characterising rhamnolipid production in Burkholderia\nthailandensis E264, a non-pathogenic producer. Appl Microbiol Biotech-\nnol 100:7945–7956. https://doi.org/10.1007/s00253-016-7564-y. 46. Vázquez-Baeza Y, Pirrung M, Gonzalez A, Knight R. 2013. EMPeror: a tool\nfor visualizing high-throughput microbial community data. Gigascience\n2:16. https://doi.org/10.1186/2047-217X-2-16. 33. Heeb S, Fletcher MP, Chhabra SR, Diggle SP, Williams P, Cámara M. 2011. May/June 2020\nVolume 5\nIssue 3\ne00292-20 msystems.asm.org\n15"