PwCF attending the Calgary Adult Cystic Fibrosis Clinic (CACFC) are seen quarterly and during periods of PEx at which times they contribute to the CACFC Sputum Biobank – a prospective repository of over 22,000 sputum samples linked to clinical outcomes, approved by the regional ethics board (REB15-0854). In real-time, sputum samples were processed by the clinical laboratory following established procedures to identify CF pathogens. Sputum-derived pathogens were prospectively frozen in glycerol and transferred to the CACFC Strain Biobank. All sputum samples were maintained at −80°C for future analysis. These sampling procedures were consistent with previous sputum microbiome studies (Acosta et al., 2017, 2018; Heirali et al., 2017, 2019, 2020; Thornton et al., 2022).

Participants were identified based on a clinical history of S. maltophilia infection. Inclusion criteria for participants with S. maltophilia infection were: adults aged ≥18 years with a diagnosis of CF, the occurrence of incident infection while attending the CACFC between 2010 and 2018, and the availability of at least one sputum sample in the biobank from the year before/at the first identification of S. maltophilia. Participants were excluded if they had a history of S. maltophilia infection in the 5 years preceding the collection date and were only included after the 5-year mark if strain typing confirmed the occurrence of a new strain. Consistent with a previous study by our group (Acosta et al., 2017), two samples in the year before infection (Pre), the sample at incident infection (when a new strain of S. maltophilia was first identified) (At) and a sample in the year after incident infection (Post) were assessed, where available. A control cohort was identified contingent on not having had S. maltophilia identified in their sputum, including a period of at least 2 years of observed culture-free status. Controls were matched 2:1 based on age (±2 years) and sex. Two sputum samples from each control were used if they were collected within 2 years to the date of each corresponding incident infection case. Any sputum sample that had been collected within 4 weeks of a change in antimicrobial therapy, or systemic antibiotics were not considered for the study on the basis of potential confounding effects (Dickson and Morris, 2017; Noci et al., 2018).

Incident infection was defined as presence of an S. maltophilia strain that had not previously been identified in each pwCF as determined by pulse field gel electrophoresis (PFGE). For purposes of this study infection and colonization are not differentiated, and the descriptor “infection” used throughout the manuscript relates to the recovery of S. maltophilia from CF respiratory secretions. Infections were classified as either transient or persistent. Persistence was defined when cases had at least half of their cultures (rounded up) positive for S. maltophilia in the year following the incident date and a minimum of 3 sputum samples in that year.

Total genomic DNA was isolated from sputum following protocols as previously described (Acosta et al., 2017) and described in the Supplemental File.

A modified protocol (see the Supplemental File) of Bartram et al. (2011) was used to amplify the V3-V4 region of the 16S rRNA gene with 3VF and V4Rmod2 (Integrated DNA Technologies, Coralville, Iowa, USA) primers using 96 unique barcoded primers for paired-end Illumina MiSeq sequencing (Illumina, Inc., San Diego, USA [RRID: SCR_016379]) (Caporaso et al., 2011; Whelan et al., 2014). Amplified DNA samples were sent to the McMaster Genome Facility (Hamilton, Ontario) for sequencing (Whelan et al., 2014; Lam et al., 2015). The raw paired-end FASTQ files generated from sequencing were then processed for microbial community analysis.

Sequence processing and downstream analysis were performed in R Studio (RRID: SCR_000432) (v. 1.4), R (v. 4.1.1). In brief, the barcoded primers and adaptor sequences were removed using Cutadapt (RRID: SCR_011841) (v. 1.2.1) (Martin, 2011), followed by filtering trimming, sample inference, alignment, and finally, taxonomy assignment using the Divisive Amplicon Denoising Algorithm 2 (DADA2 [RRID: SCR_023519]) (Callahan et al., 2016).

A comparison of the taxonomic composition through the natural history of infection was assessed at the level of the 15 most prevalent ASVs to achieve <95% of read coverage (Nearing et al., 2022). As the distribution of read counts per amplicon were observed to be non-normally distributed, we conducted a chi-square goodness of fit test. To assess community composition, α-diversity was calculated using the Shannon Diversity Index (SDI) and Observed Diversity Index (ODI). The SDI and ODI were considered because they have been previously established measures that effectively assessed CF sputum community evenness and richness (Jost, 2006).

Principal Component Analysis (PCA) using CLR (Center-log ratio) transformation was used to determine the dissimilarity between samples as represented by the β-diversity in the microbiome (Gloor et al., 2017). To assess the dissimilarity across the natural history of infection, PCA was performed on Pre, At, and Post samples in the S. maltophilia cohort. PCA was also conducted to compare the pre-infection and uninfected control samples and those with persistent versus transient infection. Permutational multivariate analysis of variance (PERMANOVA) using the Aitchison distance with the adonis function of the vegan package (Okasanen et al., 2022) was performed to investigate the significance of the factors that contributed to any observed differences (Garcia-Nuñez et al., 2020). PCA was selected as it is a commonly used visualization tool that can summarize variance among the data (Paliy and Shankar, 2016). CLR transformation was conducted on account of the compositional nature of the data and served to capture the relationships between the features in the data. Thus we can compare the abundances of features relative to other features (Gloor et al., 2017). This transformation has been shown to be suitable for multivariate analyses such as PCA (Paliy and Shankar, 2016). The top 6 taxa influencing the PCA dissimilarity were plotted on the PCA plot. Taxa were filtered with a minimum prevalence of a proportion of 10% of all samples/reads to reduce the sensitivity to extremely rare taxa (Cao et al., 2020). Permutational analysis was conducted using distance matrices with 1,000 permutations on normalized data using the CLR method (Anderson, 2001). The R package, DESeq2 (RRID: SCR_015687) was used for differential abundance analysis (DAA). DESeq2 was selected since it uses a negative binomial distribution, which is closely related to the observed Poisson Lognormal distribution of our data (Nearing et al., 2022). DAA was used to identify differences in the relative abundance of taxa among the Pre samples compared with the uninfected controls and between the persistent infection samples and the transient infection samples. Taxa were excluded if they were not observed in more than one sample (prevalence ≥2).

Pulse-field gel electrophoresis (PFGE) was used to identify incident strains of S. maltophilia and identify episodes of infection following previously published protocols (Parkins et al., 2014) and modified from Aaron et al. (2010) – full details in Supplementary File. PFGE was conducted for the 6 participants that had multiple episodes of S. maltophilia infection collected.

Quantitative PCR (qPCR) using TaqMan Fast Advanced was used to confirm the presence and absolute abundance of S. maltophilia (Fraser et al., 2019) and total-bacterial 16S rRNA (Cho et al., 2021) following adapted protocols (full details in the Supplementary File).

Non-parametric Wilcoxon rank-sum tests or two-tailed Fisher exact probability tests were performed on the comparisons of participant demographics and clinical characteristics (Heirali et al., 2017, 2020; Breen et al., 2022). A non-parametric Kruskal-Wallis test followed by the Dunn’s Test with the Bonferroni method to correct for multiple comparisons was used to compare the α-diversity of the natural history of S. maltophilia infection among the Pre, At, and Post samples (Heirali et al., 2017). Wilcoxon rank-sum tests were used to assess the differences in α-diversity. Additionally, Wilcoxon rank sum tests were used to compare the differences in abundance, both relative and absolute abundance between the case and control samples, as well as between the persistent and transient infection samples (Caverly et al., 2019; Heirali et al., 2020). All analyses were performed in R Studio (v. 1.41717), R (v. 4.1.1).

Twenty-five pwCF with 33 incident S. maltophilia infections met the inclusion and exclusion criteria (14 females; 11 males, median age 29.0 years (interquartile range [IQR] 24.1–39.6), median ppFEV₁ 61.0 (IQR 48.0–82.0). In the cohort, 19 pwCF (76%) had only one episode of S. maltophilia infection, 4 pwCF (16%) had two episodes (median time between episodes 2.1 years [IQR 1.8–3.0]), and 2 pwCF (8%) had three episodes (median time between episodes 1.7 years [IQR 1.4–1.7]). From the cohort, 57 pre-infection samples were collected (median 5.1 months (IQR 2.7–6.9 months) from the date of incident S. maltophilia infection), 22 at-infection samples, and 31 post-infection samples (median 3.0 months [IQR 2.0–5.0 months]). In pwCF with first incident infection, median ppFEV₁, as surrogate for disease severity, did not differ between Pre (60.0 [IQR 46.0–78.0]), At [67.0 (IQR 50.0–85.5)], and Post (67.0 [IQR 49.0–74.3]) p = 0.658). Of the 33 incident infection events, 20 (60.6%) included all three time points; at least one pre-infection, at-infection, and post-infection, 9 (27.3%) included only pre-infection and post-infection, 2 (6.1%) included only at-infection and post-infection, and 2 (6.1%) included only pre-infection.

Of the 33 S. maltophilia incident infection events, 18 (54.5%) were defined as persistent infections. Transient and persistent infections among first episode cases were similar across demographic and clinical characteristics including age (Wilcoxon; p = 0.851), sex (Wilcoxon; p = 0.089), lung function as measured by ppFEV₁ (Wilcoxon; p = 0.328) and ppFVC (Wilcoxon; p = 0.103), BMI (Wilcoxon; p = 0.336), proportion homozygous for F508del (Fisher’s exact test; p = 0.102), and CF-related comorbidities (p > 0.05) (Table 1). There were no differences in trimethoprim-sulfamethoxazole susceptibilities detected between transient and persistent samples (p = 1) (Supplementary Table S1).

Fifty-six S. maltophilia uninfected controls were identified (20 pwCF were matched with two controls and 5 pwCF were matched with one control). Where only 1 control was identified it related to the inability to match on basis of age/sex – generally due to outlier status.

Case and control samples measured at the first infection episode (n = 44) did not differ based on lung function as measured by either ppFEV₁ (Wilcoxon; p = 0.371) or ppFVC (Wilcoxon; p = 0.103), BMI (Wilcoxon; p = 0.221), or on whether they were homozygous for F508del (Fisher’s exact test; p = 1.00). Case and controls were similar across CF-related co-morbidities (p > 0.05) (Table 2).

Across the 166 sputum samples assessed, a total of 6,713,780 reads (69,935.21 reads/sample, IQR, 40,728 – 95,364) with a total of 1,261 Amplicon Sequence Variants (ASVs) were identified. To assess the quality of the 16S sequencing, a species accumulation curve (SAC) analysis was performed (Supplementary Figure S1). The comparison of taxonomic composition through the natural history of infection showed the relative abundance of the 15 most prevalent ASVs accounted for 95.7% of the reads (Figure 1). All of the ASVs corresponded to taxa that have been previously identified as CF airway genus-level constituents: Streptococcus (27.8%), Pseudomonas (15.8%), Staphylococcus (14.8%), Haemophilus (7.8%), Prevotella (6.1%), Neisseria (5.7%), Gemella (3.8%), Granulicatella (3.6%), Rothia (2.0%), Porphyromonas (1.6%), Fusobacterium (1.5%), Stenotrophomonas (1.4%), Veillonella (1.3%), Sneathia, (1.0%), Prevotellamassilia (0.6%)(Stressmann et al., 2012; Surette, 2014). The distribution of the read counts per amplicon across samples were found to follow a Poisson-Lognormal distribution (p < 0.0001).

To determine the natural history of community composition through S. maltophilia infection, we compared the α-diversity of the microbiome between sputum Pre (n = 57), At (n = 22), and Post (n = 31) incident infection. We did not find any significant differences among Pre, At, and Post samples as measured by either the SDI (Kruskal-Wallis, p = 0.63) or the ODI (Kruskal-Wallis, p = 0.42) (Figures 2A,B). Similar results were observed when limiting the analysis to just the first episode of infection and when only considering the first pre-infection sample (data not shown). Principal component analysis (PCA) was used to visualize any potential clustering patterns among the sputum samples corresponding to the infection stage (Figure 2C). Samples did not cluster by infection stage (PERMANOVA, p = 0.991). The strongest driver of microbial community structure was participant ID, the unique identifier given to each participant (PERMANOVA; F = 3.9, R² = 52.2%, p = 0.001). Considering the strong contribution of participant ID to community structure, for each subsequent analysis we additionally blocked the metadata by the participant and then used PERMANOVA to support each analysis which showed similar results. We assessed the difference in relative abundance of S. maltophilia at Pre, At, and Post and found significant differences between Pre and At (p = 6.6e-05) and At and Post (p = 0.0094) (Figure 2D). These differences were also observed when comparing the ratio of the absolute abundance of S. maltophilia to absolute abundance of 16S rRNA total bacterial burden between Pre and At (p = 0.00016) and between At and Post (p = 0.006) (Figure 2E). Similar results were observed when limiting the analysis to only the first episode of infection and when only considering the first pre-infection sample (data not shown).

To determine if the constituents of the CF sputum microbiome differ between pwCF that acquire S. maltophilia infection and uninfected controls, we compared the α-diversity of sputum in the year preceding incident infection. We observed a difference in the SDI (Wilcoxon, p = 0.04) (Figure 3A) but not ODI (Wilcoxon, p = 0.25) (Figure 3B). Beta-diversity ordination was calculated by using PCA to determine clustering patterns between pre-infection and uninfected controls (Figure 3C). We found that community structure clustered based on the cohort (PERMANOVA, p = 0.01). Similar results were observed when data were stratified by participant ID (PERMANOVA, p = 0.001). We identified the top 6 ASVs that contributed to this dissimilarity in clustering and conducted DAA with DESeq2 to identify the ASVs that differed (p-adjusted<0.05) between groups. We found that Haemophilus, Neisseria, Campylobacter, Gemella, and Staphylococcus contributed to the difference in clustering in the pre-infection samples whereas Pseudomonas was a contributor to the clustering difference in the control samples (Figure 3C). Neisseria (p = 0.01), Staphylococcus (p = 0.02), Lautropia (p = 0.02), and Stenotrophomons (p = 0.002) genera were significantly enriched, and Haemophilus (p = 0.02) were diminished in cases versus controls (Figure 3D). Despite sputum samples collected before infection identification (Pre) being culture negative for S. maltophilia, we found its relative abundance to be increased compared to uninfected controls (Figure 3E) (Wilcoxon, p = 0.0013). To support this observation, we confirmed the absolute abundance of S. maltophilia total 16S both at pre-infection and at-infection by qPCR (Figure 3F) (Wilcoxon, p = 4.1e-05). Similar results were obtained when assessing only the first episode from each case.

We found that of the 33 at-infection samples, 18 (54.5%) were derived from outcomes that were classified as persistent. We assessed if the constituents of the microbiome differed between those with persistent versus transient infection. We did not find any difference in the α-diversity as measured by either SDI (p = 0.62) or ODI (p = 0.31) (Figures 4A,B). To determine if sputum from those with transient and persistent infections clustered together or separately, PCA was compared (Figure 4C). We found a significant difference in the clustering based on the Aitchison distance of the transient and persistent At samples (PERMANOVA, F = 1.7, R² = 7.8%, p = 0.03). However, these results were not supported when the data were stratified by participant ID (see Supplementary File). The top 6 taxa that contributed to PCA clustering dissimilarity between infection outcome were determined. DESeq2 was used to establish the DAA of the ASVs that differed in relative abundance in samples collected at-infection based on eventual outcomes (Heirali et al., 2019; Cuthbertson et al., 2021). We found that Pseudomonas and Stenotrophomonas correlated with the difference in clustering in the persistent At samples whereas Fusobacterium, Neisseria, Prevotellamassilia, and Haemophilus were associated with the clustering difference in the transient group (Figure 4C). According to the DESeq2 data, only Stenotrophomonas (p = 0.01) was significantly enriched at incident infection of those with persistent infection and Burkholderia (p = 8.8e-14) was enriched in those with transient infection (Figure 4D). We assessed if the difference in the relative abundance of Stenotrophomonas at incident infection correlated with outcome. We found those who ultimately developed persistent infection had higher S. maltophilia relative abundance than those with transient infection (Figure 4E) (Wilcoxon, p = 0.0043). This relative abundance was confirmed via qPCR establishing the absolute ratio of S. maltophilia to total 16S total bacterial load (Figure 4F) (Wilcoxon, p = 0.005). Those who ultimately progressed to persistent infection had higher absolute (normalized) bioburden of S. maltophilia at incident infection than those who were merely transiently infected.