Only a fraction of bacterial genomes encode CRISPR-Cas systems, but the selective causes of this variation are unexplained. How naturally virulent bacteriophages (phages) select for CRISPR immunity has rarely been tested experimentally. Here, we show against a panel of genetically and functionally diverse virulent phages that CRISPR immunity was not universally beneficial to Pseudomonas aeruginosa, and its fitness effect varied strongly between phages in predictable ways. In addition to mechanisms known to alter the effectiveness of CRISPR immunity, such as encoding a matching spacer or a protective nuclear shell, we show that the fitness effect of CRISPR immunity negatively associated with the probability of evolving receptor-based resistance to the phage via spontaneous mutation. The supply of resistance mutations differed strongly between very closely related lipopolysaccharide-binding phages and was associated with variation at the C-terminus of the tail fibre protein altering residues involved in hydrogen bonding and the predicted binding site. Our results show that CRISPR immunity is more beneficial against virulent phages that are harder to evolve resistance to via receptor mutations, suggesting that virulent phage community composition and diversity will be important drivers of the prevalence of CRISPR immunity.

CRISPR immunity is not universally beneficial and varies by phage identity

CRISPR immunity significantly reduced the efficacy of both single and paired phage treatments (Fig 1A, S1 Table: CRISPR strain, F_1,16,842 = 57.03, p < 0.00001); efficacy describes the reduction in bacterial growth in the presence versus absence of each phage treatment. Significant factors driving variation in efficacy were identified using a linear mixed-effects model due to their robustness when handling skewed data (S2 Fig; [23]). Overall, phage pairs had higher efficacy than single phages (Fig 1A; S1 Table, number of phages, F_1,16,842 = 336.22, p < 0.00001), and those pairs containing phages both targeting LPS or targeting LPS and T4P achieved highest efficacy (Fig 1C, post-hoc Tukey test on full linear model: all pairwise receptor comparisons p < 0.0001 except LPS + T4P versus mixed). However, the efficacy of phage pairs was determined by the inclusion of more effective individual phages rather than synergy between phages, with few phage combinations outperforming their best constituent single phage (Fig 1B). Accordingly, phage identity was the strongest predictor of efficacy overall (S1 Table: phage identity, chi-squared = 5,895, p < 0.0001), but individual phages also had a significant interaction with CRISPR immunity (S1 Table: PA14P2, 14/1 and PA4P1), suggesting that the host fitness benefit of encoding CRISPR varied against different phages.

To quantify the host fitness benefit of encoding functional CRISPR immunity, we next calculated CRISPR effect scores per phage treatment (CRISPR effect = efficacy_{CRISPR KO} − efficacy_WT). Against individual phages, CRISPR effect scores ranged from positive to negative (Fig 2A). Phage identity, and interaction between phages, were the strongest predictor of CRISPR effect overall (S2 Table: phage identity, chi-squared = 427.1, p < 0.00001; interaction between phages, chi-squared = 930.8, p < 0.00001). Consistent with their known mechanistic interactions with CRISPR-Cas, the benchmarking phages PA14P2 and phiKZ had the expected positive versus neutral CRISPR effect scores, respectively (Fig 2A, S2 Table: PA14P2, F_1,8,393 = 186.21, p < 0.00001 and phiKZ, F_1,8,393 = 2.49, p = 0.115). Observed mean CRISPR effects against individual phages were generally predictive of the CRISPR effects against pairs of these phages (Fig 2B/2C; linear model: R² = 0.227, F_1,323 = 96.24, p < 0.0001). As such, CRISPR immunity was not universally beneficial against virulent phages or phage combinations but rather benefit depended upon phage identity.

The observed variation in CRISPR effect between phage treatments could not be explained by known mechanistic interactions with CRISPR-Cas. There was no relationship between CRISPR effect and spacer sequence matches, other than for benchmarking phage PA14P2, nor could we detect any homologs of characterised anti-CRISPR genes in the phage genomes. Furthermore, no acquisition of CRISPR spacers was observed. However, further inspection of individual replicates from the single phage treatments revealed a negative association between CRISPR effect scores and the frequency of bacterial escape from phage infection. Escape was measured as the proportion of replicate populations with <50% reduction in bacterial growth in the presence versus absence of phage (S2 Fig). Moreover, both for individual phages and phage pairs, whilst lower CRISPR effect scores resulted in higher frequencies of escape regardless of CRISPR immunity, higher CRISPR effect scores were only associated with high frequencies of escape in the presence of CRISPR immunity (Fig 2D; chi-squared test on interaction between CRISPR effect and CRISPR strain in generalised additive model (GAM) p < 0.00001); whilst the highest CRISPR effect scores were associated with the benchmarking phage PA14P2, this interaction between CRISPR effect and CRISPR strain was maintained in the absence of PA14P2 (S3 Fig; chi-squared test on GAM interaction p = 0.0228). Together these patterns suggest that the benefit of CRISPR immunity declines as the overall probability of bacterial escape increases, but that against phages for which there is a low escape probability, CRISPR immunity enabled increased survival albeit rarely.

Resistance mutation supply is associated with differences in tail fibre structure

In the absence of a functional CRISPR-Cas system, it is presumed that PA14ΔCRISPR relies upon gain of a surface receptor resistance mutation to escape phage infection. Therefore, we hypothesised that CRISPR immunity becomes more beneficial as the supply of such surface receptor resistance mutations decreases, for example, if fewer viable mutational targets in the LPS biosynthesis pathway alter phage-specific binding sites. To test this, we next quantified resistance mutation supply in the PA14ΔCRISPR background against a subset of 8 closely related PB1-like phages that consisted of 2 groups of phage which varied in CRISPR effect (Kruskal–Wallis: chi-squared = 9.32, p = 0.0023) and escape frequency (Fig 3A; Kruskal–Wallis: chi-squared = 11.36, p-value = 0.0008). Consistent with our hypothesis, both the probability of resistance (Fig 3B; R² = 0.525, F_1,6 = 8.73, p = 0.025) and absolute resistance mutation frequencies (Fig 3C; R² = 0.681, F_1,94 = 203.7, p < 0.00001) positively correlated with escape frequency in the PA14ΔCRISPR background.

The high nucleotide identity among the PB1-like phages then allowed us to identify possible genetic causes of the variation in resistance mutation supply. Variation in phage tail structures, such as tail fibres, is likely to indicate altered LPS-binding activity and influence the identify of phage-specific surface receptor resistance mutations. We identified a highly variable tail protein whose sequence divergence from a chosen representative (PA1P3’s gp49) varied by up to ~20% sequence identity across the 18 PB1-like phages (Fig 3D), which explained differences in escape frequency; lower tail protein sequence identity versus PA1P3 gp49 was associated with higher probability of escape in both bacterial strains (Fig 3D). Protein coding sequence alignments for all PB-1-like phages revealed 3 distinct variant groups (Figs 3E and S3): Group 1 resembled PA1P3; Group 2 had amino acid changes between positions 927–978 at the C-terminus of the protein; Group 3 had these changes plus further C-terminus amino acid changes between positions 809–870. Group 1 phages were associated with a low resistance mutation supply, but functional CRISPR immunity allowed rare bacterial escape from phage infection. By contrast, Group 2 and Group 3 phages were associated with higher resistance mutation supply (Fig 3B/3C: phages PA8P2, 14/1 and PA7P2), and functional CRISPR immunity had no effect on bacterial escape from phage infection (Fig 3A).

We modelled the tail fibre protein structures for a representative phage from each group: PA1P3 for Group 1, 14/1 for Group 2 and PA12P2 for Group 3. An existing protein structure from a related phage suggests that this tail protein forms a trimer with a beads-on-a-string structure (20). All variable residues mapped to the protein surface, with those common to 14/1 and PA12P2 primarily mapping to the C-terminal tip of the tail fibre (Fig 3F–3H), which is involved in binding the LPS in Pbunavirus phages [24]. These amino acid changes had only very limited effects on charge (only a single charge change event was observed between 14/1 and PA12P2). In contrast, there were large changes to donor/acceptor hydrogen-bonding groups among the phages: 14/1 had 6 fewer donor/acceptor groups than PA1P3 or a total of 18 fewer sites per trimeric tail fibre molecule, whereas PA12P2 had 9 fewer or 27 fewer per trimeric tail fibre molecule (Fig 3H). This suggests changes in the C-terminus of the phage tail fibre may alter the binding target within LPS and hence the genes within the LPS biosynthesis pathway where mutations provide resistance against these phages.

Both 14/1 and PA12P2 are known to bind to the LPS O-antigen, with mutations in the corresponding polymerase, wzy, providing resistance to both of these phages [18,25]. To identify the binding receptor of PA1P3, we sequenced 5 independent spontaneous resistant mutants. Four out of five PA1P3-resistant mutants had single missense variants within the L-rhamnose biosynthesis pathway targeting either rmlA or rmlB, and the final mutant had 2 missense variants targeting wapR (rhamnosyltransferase) and algC (phosphomannomutase; core LPS biosynthesis) (S3 Table). These distinct mutational targets support our hypothesis that PA1P3 binds to a different region of the LPS than 14/1 or PA12P2, most likely a rhamnose residue within the LPS core. To further test this hypothesis, we performed a longer-term evolution experiment wherein replicate populations of PA14ΔCRISPR were propagated by daily serial transfer for 7 days with phages from Group 1, PA1P3 or PELP20, or from Group 2, 14/1 or PA8P2. Resistance evolved against each of the phages (S5 Fig) and for one random resistant colony per replicate population, we obtained a genome sequence. Ten out of twelve evolved clones had gained mutations in their LPS operons (S6 Fig; S4 Table). Consistent with phage groups binding to different LPS regions, these mutations were exclusively in genes encoding the LPS core for clones evolved against Group 1 phages or the O-antigen for clones evolved against Group 2 phages. Specifically, clones resistant to PELP20 gained a mutation in either rmlA, encoding a glucose 1-phosphatase thimydyltransferase, or rmlB, encoding a dTDP-glucose 4.6-dehydratase, and clones resistant to PA1P3 gained a mutation in rmlA or EIP97_RS27925, encoding a putative glycosyltransferase within the same LPS core operon. By contrast, clones resistant to 14/1 or PA8P2 had gained mutations in EIP97_R509780, encoding an O-antigen polymerase. Taken together, our findings suggest that altered hydrogen-bonding at the C-terminus of the phage tail fibre resulted in phage binding to different target sites within the LPS and thus distinct sets of LPS biosynthesis genes wherein mutations provide resistance, resulting in variation in resistance mutation supply.

Bacteria and phage strains

To understand the impact of CRISPR-Cas, we utilised a well-characterised model system [7] consisting of Pseudomonas aeruginosa strain UCBP-PA14 and the isogenic mutant csy3::LacZ (herein PA14ΔCRISPR or CRISPR KO), in which the type IF CRISPR-Cas system is not functional. A collection of 26 phage strains were used in this study, chosen due to their ability to infect PA14 and for their genetic and functional diversity (i.e., adsorption receptor usage; Table 1). Phage infection of PA14 was confirmed as ability to produce defined plaques on PA14ΔCRISPR lawn and/or reduce bacterial density of PA14ΔCRISPR in liquid culture (King’s B media) during a 24 h growth curve whereby absorbance was measured (OD 600 nm) every 20 min for 24 h at 37 °C (S7 Fig). Phage adsorption receptors were identified by loss of plaquing ability on bacterial receptor mutants: transposon insertion mutants PAO1::wzy (lipopolysaccharide O-antigen biosynthesis) and PAO1::pilB (type IV pilus motor protein) were provided by the Washington 2-allelle library [32] and plaquing on these hosts was compared to wild-type PAO1 which was susceptible to all 26 phages.

Impact of CRISPR-Cas immunity on phage treatment efficacy

To investigate the impact of CRISPR immunity on the efficacy of virulent phages, we challenged the isogenic PA14 strains (with or without a functional CRISPR-Cas system) against 351 different phage treatments. Phage treatments included all single phages (N = 26) plus all possible pairwise combinations (N = 325). Phages were amplified in host PA14ΔCRISPR, then filter-purified (0.22 µm filter) and standardised to ~10⁹PFUml⁻¹ (plaque-forming units). Phage treatments were prepared in 96-well plate layouts, with pairwise combinations made by mixing equal volumes of standardised phage preparations.

Bacteria were cultured in synthetic cystic fibrosis medium (SCFM; 11), without the addition of mucin, with individual colonies selected to found 24 individual replicate populations per strain. After overnight growth (37 °C, 180 rpm), bacterial cultures were sub-cultured into 384-well plates (1 whole plate per replicate strain) to a final density of ~10³ cells per well in fresh SCFM. Phage treatments were added to each plate to give a final multiplicity of infection ~0.1 (i.e., a ratio of 10 bacteria per phage); each 384-well plate included all phage treatments (N = 351) plus negative controls (i.e., no-phage and/or no-bacteria wells). Assay plates were incubated at 37 °C, static, with 80% humidity, in sets of 10,384-well plates (replicate batches) for 18 h before reading optical density at 600nm to determine bacterial growth.

Phage treatment efficacy was calculated as bacterial growth (OD₆₀₀ at T = 18 h; S8 Fig) in the presence of each phage treatment relative to phage-free growth (Equation 1); average values for phage-free growth were taken within each plate to control for plate and replicate effects. All statistical analysis was performed in R (v4.6.1; [33]). Phage efficacy was tested using a linear mixed effects model (S1 Table; using lmer function from R package lme4 [34]) testing how efficacy varies by CRISPR strain (i.e., WT versus KO), number of phages (i.e., single or pairs), phage receptor group (LPS-only, T4P-only, other, or mixed), phage identity (as presence/absence matrix for all phages), and interactions between factors. Each effect was tested in a stepwise model, using forward selection and carrying over residuals from each model, to provide in-model test output to explore impact within levels (e.g., individual phages) and compared at model level (chi-squared test) to give overall significance. Replicate batch was included as a random effect. Additional pairwise comparisons within specific factors (e.g., phage receptor) were conducted using a post-hoc Tukey test on the linear model output.

For each bacteria-phage replicate population:

(1)

To quantify the impact of CRISPR immunity, the CRISPR Effect (Equation 2) was calculated as the difference between phage efficacy in the absence minus the presence of CRISPR; calculations were performed within replicate, values shown indicate mean ± S.E.M across 24 replicate populations. Significant drivers of the CRISPR effect were assessed using a linear mixed effects model (S2 Table; using lmer function from R package lme4 [34]) testing how CRISPR effect varies by number of phages (i.e., single or pairs), phage receptor group (LPS-only, T4P-only, other, or mixed), phage identity (as presence/absence matrix for all phages), and interactions between factors. Each effect was tested in a stepwise model, using forward selection and carrying over residuals from each model, to provide in-model test output to explore impact within levels (e.g., individual phages) and compared at model level (chi-squared test) to give overall significance. Replicate batch was included as a random effect. The predictive relationship between CRISPR effect of single phage versus phage pairs was tested using a linear regression.

For each phage treatment (single and pairwise combinations):

(2)

Escape from phage killing was determined at a breakpoint of efficacy equal to 0.5, i.e., less than a 50% reduction in bacterial growth, based on visual inspection of the distribution of efficacy values (S2 Fig). The frequency of escape was calculated as the proportion of replicates per treatment which had efficacy lower than 0.5. The relationship between CRISPR effect and frequency of escape was modelled using a GAM (R package mcgv [35]), with CRISPR strain fitted both as a main effect and interaction with smoothed CRISPR effect; a chi-squared test was used to interrogate the interaction between CRISPR effect and CRISPR strain.

Colony PCR was used to determine if escape was a result of acquisition of new spacers within either of PA14’s CRISPR arrays (Table 2) using previously described methods [36]. No acquisition of CRISPR spacers was observed following screening of whole replicate populations and individual colonies selected from each single-phage treatment where escape was observed during the phage efficacy screen.

Availability of de novo resistance mutations

To calculate the availability of de novo resistance mutations, we performed a modified fluctuation test challenging PA14ΔCRISPR against a set of 8 pbunavirus phages which contained 2 distinct patterns of CRISPR effect response and against which varying frequency of escape was observed (statistical significance measured by Kruskal–Wallis test). As CRISPR is not functional in this strain, phage resistance indicates acquired mutations within cell surface receptors preventing phage adsorption. Individual colonies of PA14ΔCRISPR selected to found 12 independent replicates which were then sub-cultured into 8 populations to generate diversity and grown in M9 with 20 mM glucose for 18 h at 37 °C. Independent bacterial subpopulations were then diluted into phage cultures to give a multiplicity of infection of ~10 phage per bacteria, with 12 independent replicates per phage treatment, mixed and incubated at 37 °C for 30 min to permit phage adsorption. Mixtures and phage-free controls were then plated and incubated overnight at 37 °C before final counts were taken. The probability of resistance evolution was calculated as the proportion of replicates for which any colonies grew after phage challenge. Mutation frequency was calculated as the relative bacterial density (colony-forming units per ml) in the presence versus absence of phage. Correlation in probability of resistance evolution and mutation frequency with frequency of escape from phage killing in the PA14ΔCRISPR background was tested with a linear regression

Five independent spontaneous mutants, in PA14ΔCRISPR background, selected for resistance to phage PA1P3 were whole-genome sequenced to identify resistance mutations. DNA was extracted from clonal overnight cultures using Wizard Genomic DNA Purification Kit (Promega, Cat. No.: A1120) according to the manufacturer’s instructions, then sequenced by Plasmidsaurus Bacterial Genome Sequencing service using Oxford Nanopore long-read technology. Raw reads were aligned to UCBPP-PA14 reference genome (accession NZ_CP034244.1) with minimap2 [37] then filtered for mapping quality and indexed using samtools [38] before variants were called by longshot [39] and annotated using SNPeff [40]. Variants were filtered by quality (read depth >10× coverage; proportion of alternate reads >0.8), compared to variants identified in control sequence (phage-sensitive PA14 ancestor) to exclude variants which differ from reference genome or common sequencing errors, and verified visually using igv viewer [41].

Phage whole genome sequencing

For whole genome sequencing, phage DNA was extracted from high-titre phage stocks (~10⁹ PFUml⁻¹) using a phenol-chloroform method described previously [42]. Library preparation followed by paired end short-read illumina sequencing on MiSeq was performed by Centre for Genomic Research, University of Liverpool. Phage genomes were assembled using spades [43], then annotated using pharokka [44] using default parameters and location of the large terminase subunit to reorient phage genomes for comparison. A pangenome of the 18 Pbunavirus phages (Table 1) was constructed using roary [45]. Phage genomes were searched for the presence of anti-CRISPRs using AcaFinder [46], but no anti-CRISPR genes were identified except for the previously described [47] anti-CRISPR genes present in DMS3vir which are not active against PA14’s IF CRISPR system. The presence of CRISPR spacers in PA14 matching our phage collection were identified using CRISPR Target [48]: only spacers matching PA14P2 were identified, corresponding to perfect matches with 3 spacer sequences.

Protein modelling

Comparison of the phage tail proteins homologous to PA1P3 gp49 across the collection of 18 Pbunavirus genomes was made using clustal omega [49] to align coding sequences, then visualised using Jalview [50]. Protein structures were modelled with the alphafold3 server [51] using default parameters. Structures were modelled as homo-trimers. Since all of the sequence variation is in the region of residues 800–985 only this region was modelled. Types of amino-acid change were identified by inspection.

Resistance evolution assay

Overnight cultures of PA14ΔCRISPR were diluted 1:100 into fresh KB medium and infected with one of four phages (14-1, PA8P2, PELP20, PA1P3) at an initial MOI of 0.01, with 3 biological replicates per phage. Infected cultures were incubated at 37 °C with shaking at 180 rpm and 1% diluted into fresh medium every 24 hours for 7 days. After 7 transfers, cultures were plated out for single colonies and 12 colonies from each treatment were picked to test for phage resistance.

Overnight cultures of putative resistant colonies were diluted 1:100 into a 96-well plate both with and without the phage against which they were evolved at an MOI of 100 with colonies of the ancestral strain included as control. Plates were incubated with shaking at 37 °C for 24 hours with OD600 measured every 10 min. The area under the curve (AUC) was calculated for each sample and the ratio of AUC (phage-treated)/AUC (phage-free) used to determine phage resistance with a score >0.8 indicated phage resistance. A single phage-resistant clone from each population was selected at random for sequencing with Plasmidsaurus whole-genome sequencing to screen for mutations.

Raw reads were mapped to the UCBPP-PA14 reference genome (accession NZ_CP034244.1) with minimap2 [37], filtered and indexed with samtools [38], variants called with Clair3 using model r1040_e82_400bps_sup_v430_bacteria_finetuned [52] and annotated with SNPeff [35]. Background mutations common to all samples were removed with bcftools [53] and variants of low quality were removed manually (including any mutations in csy3, known to be mutated in the reference genome by lacZ insertion).

S1 Table. Statistical table for phage efficacy models.

Linear mixed effects model output testing how efficacy varies by CRISPR strain (i.e., WT vs. KO), Number of phages (i.e., single or pairs), Phage Receptor Group (LPS-only, T4P-only, other, or mixed), Phage identity (as presence/absence matrix for all phages), and interactions between factors. Models were built up stepwise (carrying over residuals) to provide in-model test output to explore impact within levels (e.g., individual phages) and compared at model level (chi-squared test) to give overall significance. Replicate batch was included as a random effect in each model.

https://doi.org/10.1371/journal.pbio.3004094.s001

(TIFF)

S2 Table. Statistical table for CRISPR effect models.

Linear mixed effects model output testing how CRISPR effect varies by Number of phages (i.e., single or pairs), Phage Receptor Group (LPS-only, T4P-only, other, or mixed), Phage identity (as presence/absence matrix for all phages), and interactions between factors. Models were built up stepwise (carrying over residuals) to provide in-model test output to explore impact within levels (e.g., individual phages) and compared at model level (chi-squared test) to give overall significance. Replicate batch was included as a random effect in each model. Non-significant interactions were dropped from the model.

https://doi.org/10.1371/journal.pbio.3004094.s002

(TIFF)

S1 Fig. Phage genetic relatedness.

Neighbour-joining tree describes phage relatedness based on overall nucleotide identity (% similarity; see heatmap key). Coloured dots indicate phage adsorption receptors (see key). The data underlying this figure can be found in S6 Data.

https://doi.org/10.1371/journal.pbio.3004094.s005

(TIFF)

S2 Fig. Distribution of efficacy values to determine breakpoint for escape.

Dashed line indicates a breakpoint of efficacy = 0.5; populations with efficacy<0.5 were classified as ‘escape populations’, indicating less than 50% reduction in bacterial growth. The data underlying this figure can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3004094.s006

(TIFF)

S3 Fig. Relationship between CRISPR effect and Frequency of escape excluding phage treatments containing phage PA14P2.

Plotted values for CRISPR effect show mean of 24 replicates; frequency of bacterial escape was measured as proportion of populations which had phage efficacy lower than 0.5 within each treatment. Matching phage treatments are joined by a grey line. Coloured lines show smoothed line of best fit for each bacterial strain. The data underlying this figure can be found in S2 Data (CRISPR effect) and S3 Data (Frequency of escape).

https://doi.org/10.1371/journal.pbio.3004094.s007

(TIFF)

S5 Fig. Growth reduction in phage–treated ancestral and phage-evolved PA14ΔCRISPR.

Growth-reduction ratio (AUC (phage-treated)/AUC (NTC)) for four phages (14-1, PA8P2, PELP20, PA1P3) tested against ancestral (blue) and phage-evolved (orange) bacterial populations. A ratio of 1 (red dashed line) indicates no reduction in growth relative to the untreated control; lower values indicate stronger phage-mediated suppression. Boxes show the interquartile range of individual colony-level ratios (small points); diamonds show the mean ratio for each independent biological replicate (lineage; n = 3 lineages per phage per treatment). Ancestral populations showed reduced growth relative to NTC for all four phages, while the growth reduction ratio remained above 0.8 for all but two phage-evolved clones of PELP20, indicating phage resistance. The data underlying this figure can be found in S7 Data.

https://doi.org/10.1371/journal.pbio.3004094.s009

(TIFF)

S6 Fig. Mutations detected in phage-evolved Pseudomonas aeruginosa lineages, by gene and predicted impact.

Heatmap of mutations identified (relative to the ancestral genome) in three independently evolved lineages (columns, numbered 1–3) for each of four phages used for evolution (14-1, PA8P2, PELP20, PA1P3). Rows show gene mutation presence, labelled by locus tag/gene name and annotated product. Cell colour indicates the predicted impact of the mutation in that gene (SnpEff classification): HIGH (dark red)—mutations predicted to severely disrupt protein function (e.g., frameshift, premature stop, splice-site loss); MODERATE (red-orange)—missense mutations altering the encoded amino acid; LOW (yellow-orange)—mutations with minimal predicted effect (e.g., synonymous changes); light grey—no mutation detected. Row labels colored by predicted impact on LPS—green: O-antigen biosynthesis, purple: core oligosaccharide biosynthesis, grey: non-LPS associated loci. The data underlying this figure can be found in S4 Table.

https://doi.org/10.1371/journal.pbio.3004094.s010

(TIFF)

S7 Fig. Phage screening to determine infectivity in PA14 strain.

Growth curves show bacterial growth (PA14 CRISPR KO) in the presence (dashed line, mean of 3 replicates, shaded area shows the S.E.M.) and absence (solid line) of each phage. The data underlying this figure can be found in S8 Data.

https://doi.org/10.1371/journal.pbio.3004094.s011

(TIFF)

S8 Fig. Distribution of raw absorbance data used to calculate efficacy.

Absorbance (OD at 600nm) after 18 h growth in synthetic cystic fibrosis medium against different phage treatments (rows: none, single, pairs) for PA14 either with (left panels) and without (right panels) functional CRISPR-Cas. The data underlying this figure can be found in S9 Data.

https://doi.org/10.1371/journal.pbio.3004094.s012

(TIFF)

S7 Data. Phage resistance data for evolved bacterial clones.

Spreadsheet containing raw growth curve data for evolved PA14ΔCRISPR in the presence and absence of phage challenge (‘resistance’ tab), and details on column headings (‘readme’ tab).

https://doi.org/10.1371/journal.pbio.3004094.s019

(XLSX)

S9 Data. Raw data underlying efficacy calculations.

Spreadsheet containing all absorbance data collected during the phage efficacy screening experiment (‘Absorbance_efficacyscreen’ tab) and details on column headings (‘readme’ tab).

https://doi.org/10.1371/journal.pbio.3004094.s021

(XLSX)