Sample collection and clinical design.
The objectives of this study were to determine if the microbiome and the metabolome of sputum from pwCF on ETI therapy (n = 7) changed through time within the first 300 days of starting therapy, but after the previously reported rapid change at 1 month [8] and if these dynamics were different from those not on ETI. As a control group, our longitudinal data was compared to sputum samples similarly collected in home freezers from those not taking ETI (n = 9). Six of the non-ETI subject’s samples and data were previously published in a longitudinal study of microbial and metabolite dynamics of CF [18], and three additional subject’s collections were added for this study (Fig. 1a). There is no overlap of subjects between each group. Clinical parameters, medical treatments, and patient demographic information are presented in Table 1 and Table S1. All subjects in both groups were asked to produce sputum samples ad libitum at home and store in home freezers provided by the study team. The ETI group was asked to provide a sample at least weekly, but this was not always possible due to the reduction in sputum production in this group and some subject collected more often. Most of the collections were performed during the COVID-19 pandemic, which may have an unknown impact on our results due to social distancing or other factors, but the home study design facilitated collection of samples for this study when routine clinical visits were greatly reduced. However, challenges with delivering freezers and consenting patients during the pandemic were encountered, therefore not all subjects began sample collection at the same time after taking ETI. The average collection period for subjects on ETI was 267 days (SD = 106), the average start of collection days after taking ETI was 236 (SD = 87) while the average number of samples collected from subjects on/off ETI are 23.14 (SD = 10.28) and 73.44 (SD = 58.42), respectively.
Table 1
Clinical, demographic and sample characteristics of pwCF (n = 16) on/off ETI therapy (n = 7/9). Subjects id’s information on italic was obtained from [18]. ETI treatment status as well as the time (days) on treatment since started until the last sample collection, body mass index (BMI), gender, highest predicted lung function (ppFEV1 and FVC), and pathogen cultures within sputum are presented. Pathogen results are abbreviated as follows: Pseudomonas aeruginosa (Pa), Staphylococcus aureus (Sa), Burkholderia cepacia (Bc), Achromobacter sp. (Ac), Stenotrophomonas sp. (Steno) and Streptococcus sp. (Strep).
Subject ID | ETI | Days on ETI | Period (Days) between first sample and ETI | Samples Collected | BMI | Gender | ppFEV1 PP (%) | FVC PP (%) | Pathogen Cultures (Sputum) |
CF066 | No | - | | 104 | 18.9 | F | 54.4 | 73.8 | - |
CF146 | No | - | | 36 | 27.5 | M | 54.8 | 73.1 | - |
CF176 | No | - | | 114 | 23.0 | M | 38.7 | 48.3 | Pa, Sa |
CF189 | No | - | | 75 | 20.6 | F | 52.9 | 63.4 | Pa, Ac |
CF318 | No | - | | 190 | 29.0 | M | 69.6 | 91.1 | Steno |
CF353 | No | - | | 90 | 18.5 | F | 52.0 | 78.4 | - |
P292 | No | - | | 17 | 30.2 | F | 50.0 | 60.0 | Pa, Sa, Ac, Strep |
P246 | No | - | | 20 | 21.5 | F | 52.0 | 70.0 | Pa, Sa |
P76 | No | - | | 15 | 22.1 | M | 40.0 | 66.0 | Pa |
P18 | Yes | 537 | 277 | 18 | 27.3 | F | 111.0 | 114.0 | Pa, Sa |
P239 | Yes | 286 | 163 | 40 | 27.9 | M | 79.0 | 100.0 | Pa, Sa, Steno, Strep |
P262 | Yes | 454 | 226 | 20 | 20.2 | F | 67.0 | 84.0 | Pa, Sa, Strep |
P299 | Yes | 453 | 345 | 13 | 23.3 | F | 72.0 | 75.0 | Pa, Sa, Bc, |
P3 | Yes | 227 | 103 | 35 | 20.7 | F | 53.0 | - | Bc |
P399 | Yes | 426 | 326 | 15 | 22.4 | F | 52.0 | 70.0 | Pa |
P415 | Yes | 377 | 210 | 21 | 23.2 | F | 101.0 | 107.0 | Pa, Sa |
Because the effects of ETI therapy from clinical trials and early clinical observation was a reduction in sputum production, we first aimed to determine if our sputum-producing (n = 7) group of ETI volunteers had a different clinical response to treatment measured by the percent predicted forced expiratory volume in 1 second (ppFEV1), than other consented group of pwCF taking ETI in the same study clinic. We compared the highest ppFEV1 predicted for each subject within a year pre- and post-ETI treatment and found a significant improvement post-treatment in both the CF-consented population (DM t-test, p = 3.1E-06) and our longitudinal sputum-producing group (DM t-test, p = 0.011) (Fig. 1b). Comparing absolute ΔppFEV1 improvement between the two populations was not significantly different (Welch’s t-test, p = 0.14), indicating that the longitudinal study subjects had similar responses to ETI as the clinic’s population, though their improvement trended lower (Fig. 1b). We also evaluated the lung function of the longitudinal subjects since starting ETI and found that 5 subjects (P18, P239, P262, P299 and P3) displayed significant gain in the ppFEV1 during the collection period (Fig. 1c).
Microbiome and Metabolome Diversity Dynamics With and Without ETI Therapy
We measured the microbiome and metabolome alpha- and beta-diversity change per day from the sputum samples and compared those that were off ETI to those that were on treatment. Here we found that the degree of daily increase in ΔShannon index was higher for those on ETI in both the microbiome and metabolome (Wilcoxon test, p = 0.011 and p = 0.039, respectively). Calculation of the beta-diversity change normalized for the time between samples (ΔUniFrac for microbiome or ΔBray-Curtis for metabolome) showed that the microbiome and metabolome of those on ETI was also changing more rapidly (Wilcoxon test, p = 0.011 and p = 0.042, respectively) (Fig. 2a, b). This data supports that the microbial community and chemical constituents of sputum were more dynamic in those taking ETI compared to our control subjects.
We then used a machine learning approach to determine if these changes had a linear association with time since ETI which would support that the data was progressively changing in a predictable manner while on therapy. RF regression analysis was performed by subject to determine if the algorithm could predict the time since starting the drug for each sample based on the omics data (Table S2). We found that data from 5/7 subjects on ETI had a significant linear relationship in both their metabolome (P239, P262, P299, P3, P399) and microbiome (P18, P239, P262, P399, P415) with time since treatment started. This indicates that these subjects have a progressively changing microbiome and metabolome since taking the drug, however, some subjects showed no linear association with time indicating that their microbiome and metabolome dynamics were more static during the study period (Fig. 2c, d).
CF Pathogen Dynamics in Sputum of Subjects on ETI
The genera resembling classic CF pathogens, including Pseudomonas, Burkholderia, and Staphylococcus, were identified in the microbiomes of those on ETI as well as oral anaerobes such as Streptococcus, Prevotella, and Veillonella. We referenced the clinical culture record during the time of sample collection and found that our sputum-producing subjects on ETI had positive cultures of P. aeruginosa (6/7 subjects) and S. aureus (5/7 subjects) at different time points during the treatment period (Fig. 3a, Table 1). We tested whether the relative abundance of these pathogens was decreasing with time on ETI therapy within each individual subject. To account for the compositional nature of the microbiome data and the different pathogens in each subject, we binned the organisms into classic ‘pathogens’ or ‘anaerobes’ according to Raghuvanshi et al. 2020 and compared the log-ratio of pathogens/anaerobes through time on ETI. We did not find significant differences in the pathogen/anaerobe log-ratio within subjects on ETI over time except for patient P399, which saw an increase in this ratio (R = 0.57, p = 0.026) (Fig. 3b). Additionally, the total bacterial load (measured by the rRNA gene copy number) did not change significantly across all subjects on ETI through time, however, P239 displayed a significant longitudinal decrease (R = -0.42, p = 0.01) (Fig. 3c). This data demonstrates that some subjects on ETI (4 of the 7 studied here) still have pathogens in their sputum that persisted until the end of the sample collection period.
Metabolome Changes in Subjects on ETI
We used CANOPUS to determine if different molecular families were changing across the cohort on ETI and RF variable importance plots to identify metabolites across the study that were changing with time. We found that the chemical composition of the sputum from the overall subjects on ETI was mainly composed of glycerophospholipids (GPLs) and small peptides (Fig. 4a). We therefore averaged the abundance of all GPLs and small peptides and compared their compositional log-ratio change with time. These log-ratios revealed a positive relationship with time on ETI in 4/7 subjects with one reaching statistical significance of the linear regression at an alpha-level of 0.05 and two others nearing significance (p = 0.052 and p = 0.056; Fig. 4b). RF analysis on molecular families changing with time (64.27% variance explained by time on ETI) revealed macrolides (Azithromycin) and amino acids had the strongest association with time on ETI (Figure S1). Due to the personalization within the metabolome, there were no individual metabolites universally changing with time on ETI across subjects.
Because of the importance of P. aeruginosa to CF and our ability to detect its specialized metabolites in our metabolomic data, we explored the presence and dynamics of its various small molecule virulence factors in subjects taking ETI. By searching our metabolomics data against the GNPS mass spectral libraries based on their MS/MS patterns, we identified pyochelin, 2-nonylquinolin-4(1H)-one (NHQ) and 2-(undec-1-en-1yl)quinoline-4-ol. These molecules were detected only in subjects P239 and P399 (Fig. 4c), with only P239 showing a significantly positive correlation with the time on ETI (R = 0.61; p = 3.5E-05, Fig. 4d; and Figure S2), however, the production of Pseudomonas metabolites in P239 does not exhibit a discernible pattern in relation to the changing abundance of Pseudomonas over time.
Microbiome Dynamics Become More Neutral After ETI therapy.
It has been reported that the healthy lung microbiota displayed neutral community dynamics, i.e., microbial abundances were explained by immigration from adjacent body sites and local replacement [38, 39]. This raised the question whether the observed variability under ETI treatment could be caused by changed dispersal limitations for bacteria immigrating to the lung microenvironment. To investigate this, we implemented a simplified neutrality model in parallel with a stochastic binomial model and compared fits using Akaike information criterion (AIC, Fig. 5a) [37]. We found that a simplified neutral model reflected microbial abundances better than a stochastic distribution without dispersal (Wilcoxon, p < 2e-16). Next, we tested whether ETI therapy changed community neutrality. Indeed, modulator therapy was associated with a better fit (Wilcoxon test on negative log likelihood p = 3.7E-13, generalized R2 p < 2E-16, RSME p < 6E-7, Fig. 5b-d) and the model predicted increased immigration (Wilcoxon, p < 2E-16, Fig. 6e). However, a linear mixed model relating immigration and therapy duration correcting for subjects as random effects estimated that immigration rates decreased with treatment duration (LMM, k = -7.8E-4, p = 7.9E-2, Fig. 5f). This may indicate that the original increase of community turnover after therapy start can reduce with time.