Type I-F CRISPR-Cas resistance against virulent phage infection triggers abortive infection and provides population-level immunity

Type I CRISPR-Cas systems are the most abundant and widespread adaptive immune systems of bacteria and can greatly enhance bacterial survival in the face of temperate phage infection. However, it is less clear how these systems protect against virulent phages. Here we experimentally show that type I CRISPR immunity of Pectobacterium atrosepticum leads to rapid suppression of two unrelated virulent phages, ΦTE and ΦM1. However, unlike the case where bacteria are infected with temperate phages, this is the result of an abortive infection-like phenotype, where infected cells do not survive the infection but instead become metabolically inactive and lose their membrane integrity. Our findings challenge the view of CRISPR-Cas as a system that protects the individual cell and supports growing evidence of an Abi-like function for some types of CRISPR-Cas systems.

To respond to the pressure of phage infection, bacteria have evolved various lines of defence 1-1 3 . The adaptive arm of these defences is provided by CRISPR-Cas, which provides immunity 2 through CRISPR RNA guided cleavage of phage genomes 4,5 . CRISPR-Cas systems are 3 incredibly diverse, and are currently classified into two major classes (1 and 2), six types (I to 4 VI) and >30 subtypes 6,7 (for recent reviews, see 4,5,8 ). Crucially, recent studies revealed that at 5 least some CRISPR-Cas variants, belonging to types VI and III, induce cell dormancy through 6 collateral RNA cleavage following target recognition 9-13 . Furthermore, it is possible that type 7 V systems induce cell death through ssDNA cleavage 14 . In contrast, the most abundant type I 8 CRISPR systems, which make up around 60% of all CRISPR systems 15 , as well as the 9 somewhat less common type II systems, immunity are thought to increase survival of infected 10 individuals 16,17 . However, so far experimental studies on type I systems have almost 11 exclusively focused on interactions between bacteria and filamentous phages or obligate killing 12 mutants of temperate phages, and it is therefore less clear how bacteria with CRISPR immunity 13 resist virulent phages. Here we explored this question using Pectobacterium atrosepticum, 14 which carries a type I-F system, and two unrelated virulent phages as a model system. We 15 found that CRISPR-Cas immunity reduced the number of cells that released phages and of 16 those that produced progeny, the burst size was decreased. Infected cells did not survive phage 17 infection, yet they reduced phage amplification, which provided protection at the population 18 level. The observed CRISPR-Cas immunity phenotype to virulent phage infection has key 19 implications for the way natural selection operates on these genes 18 and is analogous to that 20 observed for kin-selected altruistic defences such as abortive infection systems, which also 21 provide population-level benefits at high individual cost. 22

CRISPR-Cas reduces phage infectious centres and burst size 25
To investigate the outcomes of phage infection in the presence of CRISPR-Cas immunity, we 26 examined the response to phage infection by P. atrosepticum, which contains a type I-F 27 system 19,20 . We used two different phages, ɸTE and ɸM1, members of the Myoviridae and 28 Podoviridae, respectively. Phage infectivity was assessed using strains with one or three 29 spacers in the chromosomal CRISPR arrays, with or without phage-targeting spacers. CRISPR-30 Cas provided protection against ɸTE and ɸM1 infection, reducing the efficiency of plating 31 (EOP) by at least 10-fold, with additional spacers increasing resistance to 10 5 -fold (Fig. 1A, 32 Table S1). To determine what stage of phage reproduction was impeded, we investigated the 1 effects of CRISPR-Cas on defined aspects of infection. CRISPR-Cas caused a decrease in the 2 efficiency of centre of infection (ECOI) formation (Fig. 1B), meaning that for ɸTE, only 4 or 3 1% of infected cells released at least one infectious phage (for the 1× and 3×anti-ɸTE strains, 4 respectively). Following ɸM1 infection, only 23 or 6% of cells released phages (for 1× and 5 3×anti-ɸM1 respectively). Next, one-step growth curves were performed to observe phage 6 growth on the resistant hosts ( Fig. S1 and Table S1). The average phage burst size was 7 determined for each host and the number was significantly reduced by CRISPR-Cas (Fig. 1C). 8 For ɸTE, both the 1× or 3×anti-ɸTE strains almost completely suppressed the burst and for 9 ɸM1 it was reduced by >90% on the 3×anti-ɸM1 strain. As expected, adsorption was 10 unaffected by CRISPR immunity (Fig. S1 and Table S1). Therefore, the P. atrosepticum type 11 I-F CRISPR-Cas immunity reduced both the number of cells releasing phages and the average 12 number of phages released per cell. 13 We previously characterised an Abi system in P. atrosepticum, ToxIN, which functions 14 as a toxin-antitoxin system 21,22 . ToxIN provides protection against both ɸTE and ɸM1 phages, 15 acting as an abortive infection mechanism, so we included ToxIN to compare the phenotypes 16 provided by CRISPR-Cas and Abi immunity genes 21-23 . The ToxIN Abi system provided strong 17 phage protection, reducing the EOP by 10 6 and 10 5 -fold against ɸTE and ɸM1, respectively 18 ( Fig. 1A). For both phages, only 1% of phage infected cells harbouring ToxIN released any 19 new viral progeny (Fig. 1B) and the average burst size was undetectable (Fig. 1C). As expected 20 for a post-adsorption phage resistance mechanism, ToxIN had no effect on adsorption (Table  21   S1). The outcomes of ToxIN and CRISPR-Cas-mediated immunity on the different aspects of 22 infection were therefore qualitatively similar with respect to phage adsorption and 1 amplification. The type I-F CRISPR-Cas system does not enable survival of infected cells 28 Next, we assessed cell survival of bacteria with CRISPR-Cas immunity upon infection with 29 the virulent phages. Surprisingly, CRISPR-Cas immunity provided no enhancement in cell 30 survival measured in viable count assays compared with the phage sensitive WT or the ToxIN 31 Abi system ( Fig. 2A), regardless of the multiplicities of infection (MOI) that were used (Fig.  32 S2A). To further investigate cell survival, we assessed membrane integrity and cellular 33 metabolic activity of phage infected cells ( Fig. 2B and C, Fig. S2B and C). Phage infection led 34 to significant reductions in both membrane integrity and cellular metabolism even in the 35 presence of CRISPR-Cas or ToxIN immunity. As a control, surface mutants (i.e. bacteria 36 carrying mutations in the phage receptor genes on the bacterial cell surface) were isolated that 37 were resistant to either phage. As expected for adsorption inhibition, surface resistance against 38 either phage resulted in cells retaining membrane integrity and metabolic activity upon phage 39 challenge, but not when challenged with a phage that uses a different receptor (Fig. S3). 1 Therefore, infected P. atrosepticum cells bearing type I-F CRISPR-Cas immunity limit phage 2 propagation at the expense of the individual -akin to altruistic abortive infection. 3 4

Figure 2. The type I-F CRISPR-Cas system does not enable survival of infected cells. A 5
Cell survival was assessed for the WT, 1× and 3×anti-ɸ strains, and ToxIN, using both ɸTE 6 and ɸM1 (infected at an MOI of 2). B The percentage of cells with intact membranes was 7 determined using LIVE/DEAD ™ staining and C the percentage of metabolically active cells 8 was assessed using the resazurin dye. For B and C cells were infected at an MOI of 2.5. Solid 9 outline bars represent mock infected samples, dashed outline bars represent phage infected 10 samples. Statistical significance was calculated using one-way ANOVA using Dunnett's 11 multiple comparison test, comparing strains with targeting spacers to the control with no-12 targeting spacers. No significance was detected, unless indicated (* p ≤ 0.05). 13

Increased CRISPR-Cas resistance does not enhance survival of infected individuals 14
One possible explanation why CRISPR-Cas did not promote survival following infection could 15 be due to an insufficient immune response, leading to incomplete phage clearance. Since the 16 phage-targeting spacers are in CRISPR arrays carrying 30 (CRISPR1) and 11 (CRISPR2) other 1 spacers, most effector complexes will be loaded with non-phage-targeting crRNAs. To explore 2 if an increased abundance of Cas complexes loaded with phage-targeting crRNAs would result 3 in survival of infected cells, phage targeting spacers were overexpressed from plasmids in the 4 presence or absence of Cas overexpression (Fig. 3). Increased phage-targeting crRNAs 5 significantly boosted phage resistance compared with chromosomal expression, and induction 6 of Cas expression further enhanced resistance, by up to ~10 4 -10 7 fold compared to the WT 7 (inducible mini CRISPR array with no anti-expI spacer) "Cas+ crRNA-" or pTargeted and 20 pCRISPR (anti-expI spacer) "Cas+ crRNA+". Solid outline bars represent CRISPR repressed 21 samples, dashed outline bars represent CRISPR induced samples. Statistical significance was 22 calculated using one-way ANOVA using Dunnett's multiple comparison test, comparing 23 strains with targeting spacers to the control with no-targeting spacers. The Cas overexpression 24 and Cas overexpression with 1×anti-ɸ plasmid strains, as well as the strains in D, were 25 compared using an unpaired T-test. No significance was detected, unless indicated (* p ≤ 0.05). 26 While these data show that CRISPR-immune bacteria do not survive virulent phage infection 1 even under artificially high CRISPR expression levels, it is unclear whether this is due to 2 programmed cell death induced by CRISPR (analogous to that observed for type VI systems 10 , 3 or due to the phage, which may express lethal genes prior to clearance of the infection. To 4 explore this question, we examined the outcome of targeting plasmid DNA for the cells with 5 CRISPR-Cas immunity ( Fig. 3C and D). The P. atrosepticum CRISPR-Cas system effectively 6 inhibits transformation and conjugation 24 , but those assays fail to assess the outcome for cells 7 eliciting effective CRISPR immunity (they are killed by the antibiotic). To directly test whether 8 plasmid targeting by the I-F system reduces cell survival in P. atrosepticum, we induced a 9 mini-CRISPR array with a spacer targeting a plasmid and assessed total cell counts and plasmid 10 loss. Plasmid targeting decreased cells bearing the plasmid by 10 6 -fold in 18 h but did not 11 decrease total cell numbers. Hence, these experiments show that the Abi phenotype is phage-12 dependent, since cells survived plasmid targeting. 13

CRISPR-Cas provides population-level protection at low phage doses 14
Even though an Abi-like phenotype is costly for the infected individual, they may be favoured 15 by natural selection because of their population-level benefits if these are predominantly 16 directed at clone mates (i.e. kin selection). To explore these kin-selected benefits, we compared 17 population growth of cells carrying CRISPR-Cas or Abi under increasing phage pressures 18 (increasing MOIs) (Fig. S4). Phage sensitive WT P. atrosepticum populations were susceptible 19 to phages at any MOI. The phage effects on population growth were stronger and faster with 20 increasing phage numbers, but even with an MOI of 0.0001, WT populations collapsed (Fig.  21 4A). As predicted for an Abi mechanism, cultures containing ToxIN grew with low phage 22 doses, but when phages equalled or exceeded bacteria (MOI of 1 or higher) population growth 23 was inhibited. Likewise, CRISPR-Cas immunity enabled population growth at low phage 24 doses, but at higher MOIs, the populations either collapsed when infected with ɸM1, or became 25 static when infected with ɸTE (Fig. 4A, Fig. S4). We predicted that CRISPR-Cas was 26 providing population-level protection by reducing the phage epidemic. To test this, the effect 27 of CRISPR-Cas on phage titres was determined (Fig. 4B). Both phages replicated extensively 28 on the phage-sensitive WT bacteria, reaching ~10 10 -10 11 pfu ml -1 irrespective of the initial 29 phage dosage (Fig. 4B). ToxIN reduced the population phage burden regardless of the initial 30 phage abundance. CRISPR-Cas immunity limited the phage epidemic when initial viral 31 abundance was low, but when initial phage numbers were higher, CRISPR was unable to 32 suppress the phage burden. In summary, immunity provided by the type I-F CRISPR-Cas 33 system enables population growth under low viral load by reducing virulent phage burden, but 1 both CRISPR-Cas and ToxIN fail to cope with high phage pressures. ɸ strains, and ToxIN were grown with phages, added at a range of doses (MOIs: 0 (buffer), 10 -5 4 , 10 -3 , 10 -2 , 10 -1 , 1, and 10). A The final bacterial growth levels (OD600) and B the final phage 6 titres were determined after 16 h. Statistical significance was calculated using one-way 7 ANOVA using Dunnett's multiple comparison test, comparing strains with targeting spacers 8 to the control with no-targeting spacers. No significance was detected, unless indicated (* p ≤ 9 0.05). 10 11

Discussion 12
Here we show that the P. atrosepticum type I-F CRISPR-Cas system generates an immunity 13 phenotype similar to that Abi systems, in which infected cells limit phage propagation and as 14 a result, protect neighbouring cells from infection. CRISPR-Cas reduced phage infectivity, 15 resulting in fewer infectious centres with reduced phage burst sizes and infected cells did not 16 survive, became metabolically inactive and lost membrane integrity. However, population-17 level protection was achieved through CRISPR-Cas-mediated reduction in the phage epidemic. 18 Although the CRISPR-Cas response to phage infection has been investigated in other systems, 19 they are typically carried out with filamentous phages or virulent mutants of temperate phages, 20 and thought to enhance survival of the infected individual [25][26][27][28][29] . Cell survival was demonstrated 1 for S. thermophilus with a type II-A system 17 and in other studies survival may be inferred 2 since CRISPR adapted clones grow in the presence of phages 16, [30][31][32] . The observation that the 3 native P. atrosepticum system generates an Abi-phenotype upon infection with two virulent 4 phages helps to explain previous observations that infection by T7 or T5 virulent phages 5 targeted by the type I-E system of Escherichia coli slowed or inhibited bacterial growth 33 . 6 The 'suicidal' response of CRISPR-Cas to phage infection might occur through several 7 mechanisms 34 . These include: activation of toxic domains found in some Cas proteins, such as 8 Cas2 35 , off-target effects of the promiscuous RNA-targeting effector proteins from type III 11-13 9 and type VI 10 systems, and self-targeting due to increased spacer acquisition following 10 CRISPR-Cas activation 36 . None of these models explain the Abi phenotype of the type I-F 11 system in P. atrosepticum. For example, P. atrosepticum Cas2 does not have detectable 12 nuclease (i.e. toxic) activity 37 and although we have observed acquisition of self-targeting 13 spacers, this is at a low frequency that is unlikely to significantly impact cell survival 38 . Indeed, 14 PCR analysis of CRISPR array expansion following phage infection failed to detect spacer 15 acquisition (data not shown). Instead, our data suggest that post-infection immunity by 16 CRISPR-Cas response provides a window of time during which the virulent phage can express 17 toxic phage products 33 . Temperate and filamentous phages can transmit both horizontally and 18 vertically and therefore generally avoid immediate early expression of highly toxic genes as 19 this would be associated with severe fitness trade-offs when the phage enters the lysogenic 20 cycle. Early expressed virulent phage genes can lead to host DNA degradation, inhibition of 21 host RNA polymerase and other effects 39,40 . Although the exact mechanism of host cell 22 takeover for the two phages used in this study is unknown, ɸM1 encodes its own RNAP, 23 suggesting a rapid host-takeover, and we have also shown that a gene responsible for triggering 24 ToxIN immunity is toxic in P. atrosepticum 23 . Consistent with our phage-induced growth 25 inhibition hypothesis, the Abi phenotype was absent during type I-F plasmid targeting. 26 In contrast to our findings, type I-E and I-F CRISPR systems can provide resistance against 27 temperate and filamentous phages without apparent Abi phenotypes 16,41 . Chronic phage 28 infection (M13) or obligately lytic temperate phage mutants (e.g. DMS3vir) do not rapidly, or 29 strongly, manipulate bacterial physiology and therefore CRISPR immunity is sufficient to clear 30 infection and protect the cell. Nonetheless, P. aeruginosa deployment of type I-F CRISPR-Cas 31 causes a bacterial fitness cost, potentially due to decreased replication or repairing the damage 32 following phage infection 16 . Despite this, CRISPR-Cas was advantageous to P. aeruginosa 33 through the generation of diverse immunity against phages 42 . It is not clear how cells containing 1 a CRISPR system that functions through an Abi-like phenotype can acquire new spacers. 2 However, by analogy to work in a type II system, the type I-F system may acquire spacers 3 during infection by defective phages 28 , which might enable phage-resistance to arise when 4 bacteria are growing in a structured environment 18 . Indeed, Abi systems evolve in spatially 5 structured environments where clone mates benefit directly 18,43,44 and we predict that the P. 6 atrosepticum CRISPR-Cas system will be beneficial under such conditions. 7 The Abi phenotype of this type I-F system strengthens the view that CRISPR-Cas immunity 8 can sometimes coming at the expense of the individual, but providing a benefit for the 9 population 10,45,46 . The nature of the invading element, the relative efficiency of resistance and 10 type of CRISPR-Cas system are likely to influence whether CRISPR-Cas provides protection 11 to the infected individual and the population, or just to the population via Abi. We predict that 12 virulent phages are more likely to elicit Abi phenotypes, whereas temperate phages or other 13 mobile elements will be more likely be cleared and result in cell survival. These outcomes 14 likely fall on a spectrum determined by the invader vs host immune strength and will need to 15 be factored in to ecological and evolutionary analyses of CRISPR-Cas immunity. 16 17

Efficiency of centre of infection assays (ECOI) 9
Overnight cultures were OD-adjusted and 1 ml was used to inoculate a 25 ml culture in a 250 10 ml flask, for a starting OD600 of 0.1. Cells were grown until early stationary phase (OD600 of 11 ~0.3) before 10 9 total phages (~ 4 × 10 7 pfu ml -1 ) were added at a multiplicity of infection 12 (MOI) of ~0.1 and cultures were incubated with shaking for 20 min. Aliquots of 1 ml were 13 extracted, washed twice in 1×phosphate-buffered saline (PBS), diluted and plated in top LBA 14 with P. atrosepticum before the infected cells starting lysing. The pfu ml -1 was determined for 15 each strain and since each plaque was formed from the phages released from an individual cell, 16 the titre represents the number of infectious centres formed. The ECOI was calculated as (pfu 17 ml -1 (test strain) / pfu ml -1 (control strain, P. atrosepticum)). Spontaneous ɸ-resistant surface 18 mutants, PCF333 and PCF334, were included to control for unadsorbed phages. 19

One-step growth curves 20
Overnight cultures were OD-adjusted and 1 ml was used to inoculate a 25 ml culture in a 250 21 ml flask, for a starting OD600 of 0.1. Cells were grown until early exponential phase (OD600 of 22 0.25-0.35) and 10 9 total phages (~ 4 × 10 7 pfu ml -1 ) were added, for an MOI of ~0.1. Duplicate 23 samples were taken at various timepoints, until 70 min post infection. One sample was plated 24 immediately (non-treated sample, free phages and phage-infected cells), while the second was 25 added to phage buffer containing chloroform (treated sample, free phages and phage 26 accumulated inside infected cells), which lysed the cells, allowing the assessment of the total 27 number of mature phages at each time point. Samples were diluted in phage buffer and plated 28 in top LBA with P. atrosepticum. Phage adsorption over time was determined from the treated 29 samples using the equation ((pfu ml -1 (t=0) -pfu ml -1 (t=0 to 70) / pfu ml -1 (t=0)). The average 30 phage burst size was also calculated from the treated samples, as number of phages released 31 ((pfu ml -1 (t=70) -pfu ml -1 (t=30)) / the number of cells infected ((pfu ml -1 (t=0) -pfu ml -1 32 (t=30)). The latent period was determined from the treated samples as was defined as the time 1 before the phage burst starts. 2

Cell survival assays 3
Cells were grown to OD600 ~0.3 and for each culture, 1 ml was transferred into two universals. 4 One culture was infected with phages at a MOI of ~2, while the other was mock infected, with 5 phage buffer. Cultures were shaken at 180 rpm for 20 min for phages to adsorb and then cells 6 were pelleted and resuspended in PBS to remove unadsorbed phages. Finally, cells were diluted 7 and 100 µl samples were plated prior to the phage burst (40 min). Cell survival was calculated 8 as (colony forming units (cfu) ml -1 (phage treated sample) / cfu ml -1 (mock treated sample). 9 To assess cell survival at a range of MOIs, 100 µl of each exponential phase culture was 10 aliquoted into eight wells of a 96-well flat-bottomed plate for the addition of 10 µl phages at 11 seven MOIs as well as a mock infection control (phage buffer). Cultures were shaken for 20 12 min for phages to adsorb, and to reduce the burden of secondary infection, a viricidal solution 13 called TEAF (per ml: 680 µl of 4.3 mM FeS04, 320 µl 7.5% (w v -1 ) green tea solution (filter-14 sterilised) 49 ) was then added, at a ratio of 75% (v v -1 ) to each culture. The cultures were then 15 diluted, more TEAF was added to each dilution and cells were plated as 5 µl spots. Survival 16 for the ɸTE-infected cells was higher than predicted from the MOIs used, suggesting that 17 despite high adsorption rates (Table S1), the phage was not able to infect well in these assays 18 with high phage doses. 19

LIVE/DEAD staining for membrane activity 20
Cell membrane integrity was assessed using the LIVE/DEAD ™ BacLight ™ bacterial viability 21 kit, consisting of two nucleic acid stains, syto-9 and propidium iodide (Life technologies ™ ). 22 Cultures were prepared for live/dead staining as described above for the cell survival assays 23 performed at a range of MOIs. Cells were infected for one hour, to allow for one complete 24 round of infection, before being stained, according to the manufacturers' instructions. Culture 25 fluorescence was measured using a Thermo Scientific ™ Varioskan ™ plate reader, with 26 excitation / emission wavelengths of 485 / 530 nm for styo-9 and 485 / 630 nm for propidium 27 iodide. Cultures of exponentially growing cells and cells killed with 70% isopropanol were 28 combined at different ratios to generate a standard curve, from which the percentage of cells 29 with intact membranes at each phage MOI could be determined. 30

Resazurin assays for cell activity 1
For assays assessing cell activity after one round of phage infection, cultures were prepared as 2 described above for the cell survival assays performed at a range of MOIs. Cells were infected 3 for one hour before resazurin solution was added at a final concentration of 0.005% (w v -1 ). 4 Cellular oxidoreductases reduce the blue indicator to resorufin, which is pink. Resorufin 5 fluorescence was measured 30 min after it was added using a Thermo Scientific ™ Varioskan ™ 6 plate reader with excitation / emission wavelengths of 510 / 535 nm. Cells for the standard 7 curve were prepared as described for the live/dead staining, from which the percentage of 8 metabolically active cells at each MOI was determined. Cell activity was assessed, following 9 the 16 h growth assays, in the same way. 10

Isolation of spontaneous phage-resistant surface mutant strains 11
ɸTE and ɸM1 were plated on P. atrosepticum and cells from colonies that formed in the centre 12 of plaques were streaked to single colonies. Since ɸTE is flagella-trophic 50 , clones isolated 13 from plates with ɸTE were patched onto tryptic swimming agar (10 g Bacto tryptone, 5 g NaCl, 14 3 g agar, per litre) to assess flagella-mediated swimming. A clone that did not swim (PCF333, 15 Table S2) was resistant to ɸTE, but sensitive to ɸM1, which does not use the flagella for 16 infection, suggesting that it was a surface mutant. A clone isolated from a ɸM1 plaque 17 (PCF334 , Table S2) was ɸM1-resistant, but sensitive to ɸTE. 18

Construction of the plasmids expressing crRNAs 19
Spacers present in strains targeting ɸTE (PCF190) and ɸM1 (PCF254) were cloned into 20 pPF975. Overlapping primers containing the spacer sequences were annealed and ligated into 21 the BsaI site in the mini CRISPR array (repeat-repeat loci) as previously described 51 to form 22 the plasmids, pPF1421 and pPF1423 (Table S2). Oligonucleotide sequences are listed in Table  23 S3. All plasmids used in this study were confirmed by sequencing. 24

Construction of the cas overexpression strains 25
The chromosomal cas overexpression strain (PCF610) was made by conjugating the suicide 26 vector, pPF1814, into P. atrosepticum. The vector, pPF1814 was constructed as follows: 27 pSEVA511 was digested with NotI and ligated with the T5/lac promoter and multiple cloning 28 site (MCS) from pQE-80L-stuffer, which had been amplified with PF3494 and PF3495 and 29 digested with NotI. The lacI gene was amplified from pQE-80L-stuffer (PF2511, PF2512) and 30 ligated into the MCS at XmaI and SalI sites. Finally, the first 500 bp of cas1 was amplified 1 using PF357 and PF669 and ligated into EcoRI and XmaI sites in the MCS. 2 Plasmid targeting assay 3 The effect of plasmid targeting on cell survival was assessed using a two-plasmid setup. The 4 first plasmid was either a control vector (pControl, pPF445, Ap R ) with an inducible mini 5 CRISPR array with a single repeat or pCRISPR (pPF452, Ap R ) carrying a spacer targeting expI. 6 The second plasmid was pTargeted (pPF459, Km R ), which carried the targeted expI gene. 7 pTargeted was made by PCR-amplifying expI from P. atrosepticum with PF314 and PF317, 8 digesting the product with BamHI and PstI and ligating the product into the same sites in 9 pPF260 (Km R -pQE-80L derivative). pControl and pCRISPR were made previously 52 . P. 10 atrosepticum ΔexpI (PCF81) was co-transformed with pTargeted and pCRISPR, or pControl, 11 under CRISPR repressing conditions (0.2% glu) with both antibiotics (Km and Ap). These 12 strains were for 6 hours in LB, 0.2% glu, Ap + Km with shaking. Cells were pelleted by 13 centrifugation, washed and the culture was split into two samples, repressed (0.2% glu and Ap) 14 and induced CRISPR conditions (0.2% ara and Ap). Following growth for a further 18 h, cells 15 were plated onto Ap (for total cell counts) and Km (for targeted vector-containing cell counts). 16 Efficiency of plasmid maintenance was calculated from the Km counts as (cfu ml -1 (pCRISPR) 17 / cfu ml -1 (pControl)). Cell survival was calculated for each strain as (cfu ml -1 (induced) / cfu 18 ml -1 (repressed). The cell counts for the induced CRISPR conditions were higher because the 19 growth rate of P. atrosepticum was increased with supplemented arabinose. 20 Bacterial population growth assays 21 P. atrosepticum cultures were grown to an OD600 of 0.3 and 100 µl was transferred to each well 22 (of a 96-well plate). Phages were added in 10 µl at multiplicities of infection (MOIs) ranging 23 from 0.0001 to 10 and cultures were grown in a Thermo Scientific ™ Varioskan ™ plate reader 24 with shaking at 480 rpm. Cell density was monitored for 16 h, measuring OD600 every 12 min. 25 Following growth, final phage titres were determined by chloroform treating the bacterial 26 cultures and titrating the phages. The data were processed using GraphPad Prism to generate 27 restricted cubic spline curves (324 points were calculated). 28

Data availability statement 29
The data that support the findings of this study are available from the corresponding author 30 upon reasonable request. pQE-80L-stuffer. We thank members of the Fineran laboratory for useful discussions. 10 11