Phage cocktail strategies for the suppression of a pathogen in a cross‐feeding coculture

Predicting which combinations (cocktails) of bacteria‐infecting viruses (bacteriophage) will suppress pathogenic bacterial growth is complicated, in part, due to the complex ecological interactions of host bacteria, but important for effective treatment. We found that when hosts cross‐feed (exchange nutrients) with nonhost bacterial species, cocktails that include a bacteriophage infecting the nonhost cross‐feeding partner can be more effective at suppressing pathogenic bacteria than targeting the pathogenic bacterium with multiple bacteriophage species. This is a novel cocktail formulation.


Introduction
Phage has been used to treat pathogenic bacteria in human health, agriculture and the food industry. Phage therapy and biocontrol often use multiple phages simultaneously in 'cocktails' to suppress pathogen growth (Pires et al., 2017;Culot et al., 2019;Gordillo Altamirano and Barr, 2019;Kakasis and Panitsa, 2019;Mahony et al., 2020). Attacking a bacterial population with multiple phages can reduce the rate at which phage resistance evolves (Filippov et al., 2011;Gu et al., 2019;Ramírez et al., 2020). However, we understand less about how treatment outcomes are affected by complex interactions among bacteria in a microbial community (Fazzino et al., 2020). One bacterial interaction of particular interest is cross-feeding, in which metabolites secreted by one bacterium are used as a nutrient source by another. This is a common interaction in natural systems (Schink, 2002;D'Souza et al., 2014;Mee et al., 2014;Zelezniak et al., 2015;Adamowicz et al., 2018). Understanding how complex ecological interactions involving pathogens affect phage treatment outcomes will be critical for designing effective therapies. Here, we explore how two important factorsthe potential for cross-resistance evolution and relative bacterial growth ratesinteract with targeting strategies to suppress growth of a focal pathogen cross-feeding in an engineered coculture.
Experiments using cocultures with well-defined interactions have helped elucidate a range of responses to phage infection which may be leveraged for phage therapy. For example, adding non-host bacteria that compete with phage hosts for nutrients limits phage resistance evolution, thereby magnifying the efficacy of the phage (Harcombe and Bull, 2005;Brockhurst et al., 2006;Wang et al., 2017;Yu et al., 2017;Testa et al., 2019). Microbes can also engage in cooperative mutualistic interactions, where bacteria depend on others to cross-feed nutrients (Schink, 2002;Mee et al., 2014;Zelezniak et al., 2015;D'Souza and Kost, 2016;Adamowicz et al., 2018). For example, Pseudomonas aeruginosa cross-feeds with anaerobic fermenters in the lungs of cystic fibrosis patients (Flynn et al., 2016(Flynn et al., , 2020, and cross-feeding is an important interaction between Bifidobacterium and butyrate-producing anaerobes in the human gut (Belenguer et al., 2006).Targeting one species in a cross-feeding mutualism can reduce the population of both mutualists, leading to the hypothesis that phage therapies could target a pathogen's mutualists (Fazzino et al., 2020). However, it is unknown how cocktails should be assembled to maximize pathogen suppression in a community.
If pathogenic bacteria cross-feed with other community members, then we can consider novel strategies of phage cocktail design that also target the nonpathogenic cross-feeding partner. Phage cocktails classically contain multiple phages that target a focal species to better limit the growth of a pathogen while also decreasing the rate of resistance evolution (Chan and Abedon, 2012;Wang et al., 2017;Betts et al., 2018). However, cross-resistance can evolve during treatment with classic pathogentargeting cocktails when a single mutation blocks infection to multiple phages (Cairns and Payne, 2008;Wei et al., 2011;Kortright et al., 2019). We hypothesize that whether a focal pathogen is engaged in an obligate mutualism, including a phage that targets the cross-feeding nonpathogen will increase suppression of the pathogen. It has been shown that off-target inhibition of crossfeeding partners can inhibit focal bacterial strains (Shou et al., 2007;Adamowicz et al., 2018). Combining phage that targets the pathogen, and its cross-feeding partner in a 'multispecies-targeting' cocktail would require the coculture to evolve two resistance mutationsone in each cross-feeding partnerto continue growing. Here, we hypothesize that this novel cocktail strategy that targets pathogens and cross-feeding nonpathogens will limit cross-resistance evolution and lengthen pathogen suppression.
In this study, we test the efficacy of phage cocktail treatment strategies to suppress a model pathogen obligately cross-feeding in a synthetic coculture. We performed wet-lab experiments with a synthetic engineered obligate cross-feeding coculture of an Escherichia coli methionine auxotroph that provides carbon to a methionine-secreting Salmonella enterica (Harcombe, 2010).
Here, E. coli is the model pathogen to be suppressed. We introduced all pairwise combinations of E. coli-specific T7 and/or P1vir lytic phage, and the S. enterica-specific P22vir lytic phage (Fig. 1A). We then compared 'pathogen-targeting cocktails' with 'multispecies-targeting cocktails'. We hypothesized and observed that targeting both cross-feeding partners was more effective at suppressing E. coli than targeting only E. coli with cocktails in our wet-lab experiments. We combined wet-lab experiments and mathematical modelling to uncover two reasons for this. First, as anticipated, we found evidence that E. coli evolved cross-resistance in the pathogen-targeting cocktail treatment. Second, the multispecies-targeting cocktail inhibited the slowest-growing crossfeeding partner, S. enterica, which limited how fast the coculture recovered from phage treatments. In fact, treatment with a single phage infecting S. enterica was as effective as targeting both cross-feeding partners in wetlab experiments and simulations. Ultimately, our study highlights a novel strategy for designing phage cocktails that suppress cross-feeding pathogens.

Results
We wondered whether the pathogen, E. coli, would be suppressed for longer by a phage cocktail combining two E. coli-targeting phage ('pathogen-targeting cocktail') or combining an E. coli-targeting phage with a S. enterica-targeting phage ('multispecies-targeting cocktail'). We grew control cocultures without phage ('phage-free'), and treatment cocultures with combinations of T7 and P1vir as E. coli-targeting phage and P22vir as a S. entericatargeting phage. The growth of each strain was tracked with unique fluorescent proteins (see Experimental Procedures for details). We predicted that the multispeciestargeting cocktail would provide the longest E. coli suppression because cross-resistance would not be possible and two mutations in separate bacterial species would need to evolve for the coculture to grow.
We quantified duration of E. coli suppression across phage treatments. To do this, we measured the amount of time required for a fluorescent protein in E. coli to reach 95% maximum intensity, which we refer to as time to maximum density. We calculated the relative suppression duration caused by each phage treatment as a foldchange relative to the phage-free cocultures, which required 34.4 h to reach maximum density (Table S1). As anticipated, all phage treatments increased E. coli suppression duration (Fig. 1B, Table S1 for absolute values). Notably, both multispecies-targeting cocktails delayed E. coli growth longer than the pathogen-targeting cocktail (P < 0.02 for T7 + P22vir and P1 + P22vir), but were not significantly different from each other (P = 0.43). Yet, the single phage treatment with the S. enterica-targeting phage P22vir suppressed E. coli equally as long as any of the cocktail treatments (P > 0.17 for any cocktail) (Fig. 1B, Table S1).
Two possible, but not mutually exclusive, reasons that targeting S. enterica suppressed E. coli growth longest are (i) that E. coli evolved cross-resistance to both phages in the pathogen-targeting cocktail, reducing its efficacy and/or (ii) that S. enterica sets coculture growth rate and that targeting it maximizes suppression of both species including E. coli.
We hypothesized that cross-resistance may be one reason why the pathogen-targeting cocktail was less effective than the multispecies-targeting cocktails. Multiple studies have reported that phage cocktails suppress focal bacteria less than expected given single phage treatments, suggesting that evolution of cross-resistance may be common (Cairns and Payne, 2008;Wei et al., 2011;Kortright et al., 2019). To determine whether cross-resistance evolved, we measured phage resistance of E. coli isolates from each coculture with crossstreak assays. As expected, E. coli isolates from phagefree controls were sensitive to both E. coli phage (Table 1). Additionally, E. coli clones treated with a single phage were resistant to that phage, but remained sensitive to phage with which they had not been treated. Half of the E. coli clones from pathogen-targeting cocktail treatments evolved resistance to both E. coli phage, suggesting that cross-resistance may have evolved in some replicate cocultures. We also observed that all E. coli isolates from the pathogen-targeting cocktail treatments evolved mucoid phenotypes, which has previously been shown to cause cross-resistance by a single mutation in various genes involved in lipopolysaccharide production (Radke and Siegel, 1971;Skurray et al., 1974;Mizoguchi et al., 2003;Scanlan and Buckling, 2012).
Additionally, we wondered whether the observed resistances to multiple phages were caused by two independent mutations or a single mutation conferring crossresistance. We predicted how common resistance to both E. coli phage would be in populations unexposed to phage (i.e. standing variation of resistance). If resistance to the two E. coli phage required different mutations, then the frequency of resistance to both phages is the likelihood that each resistance mutation was acquired individually (f cross-resistance = f mut1 × f mut2 ). To quantify standing variation of resistance, we compared the number of resistant colonies on a phage-covered plate with the number of colonies on a phage-free plate for each ancestral bacteria (Luria and Delbrück, 1943). The frequency of resistance to both E. coli phage in the ancestral E. coli population was~100-fold larger than predicted if resistance to both phages required two independent mutations (Fig. 2 -T7 + P1vir and red asterisk). These data suggest that the evolution of crossresistance may be one reason the pathogen-targeting cocktail was less efficacious than the multispecies-targeting cocktail.
Cross-resistance cannot explain another result: that a single phage targeting S. enterica is just as efficacious at increasing suppression of E. coli as any of the best cocktail treatments. A hypothesis which could explain this result is that S. enterica's ability to recover from A. Schematic representation of the wet-lab engineered cross-feeding bacterial system with phage strains. E. coli = methionine auxotroph with cyan fluorescent protein, S. enterica = methionine secreter with yellow fluorescent protein. B. Relative E. coli suppression lengths of single and cocktail phage treatments standardized to the no phage control. General phage treatment indicated under the bar on the X-axis where #P X = # phages included that infect species X. Suppression length was calculated using 95% maximum cyan fluorescent protein measurement. Permutation statistical tests determined significance. P > 0.1 (NS), P < 0.05 (*). Exact P-values are in the text. Bars represent means AE SE (n = 4-5).
a. A representative isolate per treatment replicate was crossstreaked against the indicated phage. b. One repeat each of a T7-only treated community and a T7 + P1vir-treated community had no detectable E. coli at the end of growth and were omitted for phenotyping. phage infection in the multispecies-targeting cocktail treatment set the coculture growth response. We tested two possible ways that S. enterica could act as a response-setting species: (i) that resistance to S. enterica phage was the least likely to evolve (i.e. smallest standing variation of phage resistance) or (ii) that low S. enterica density caused by phage-mediated lysis delayed E. coli growth more than low E. coli density delayed S. enterica growth (i.e. S. enterica is the ratelimiting member of the coculture).
To determine whether resistance to S. enterica phage was less likely to evolve than resistance to E. coli phage, we measured the standing variation in phagefree cultures of ancestral bacterial populations, or the frequency of resistance without exposure to phage. If the ability to evolve resistance was the determining factor of cocktail efficacy, then resistance to the S. enterica phage P22vir should have the lowest frequency of standing variation for resistance. Resistance to P22vir was more common than resistance to T7 (P = 0.050, Fig. 2) and less common than resistance to P1vir (P = 0.047, Fig. 2), suggesting that resistance to T7, not P22vir, was hardest to evolve. If the ease of evolving resistance was the sole factor determining cocktail efficacy, then our results suggest that the longest suppression of E. coli should be caused by any treatment containing T7 phage. Yet, we observed that treatments including S. enterica phage P22vir suppressed E. coli growth the longest.
Alternatively, physiological differences between S. enterica and E. coli could cause a different coculture response to phage. Other studies have illustrated asymmetrical responses of cross-feeding systems to perturbations caused by differences in growth rates or production rates of cross-fed nutrients (Shou et al., 2007;La Sarre et al., 2017;McCully et al., 2017). Here, we examined the influence of physiology on coculture rebound after a population size reduction by manipulating starting coculture frequencies without phage. This manipulation isolated the impact of phage-mediated population size reduction from the impact of phage replication and/or resistance evolution. We started cocultures with either E. coli or S. enterica at 0.01% instead of 50% of the coculture population and tracked E. coli growth as before. We found that reducing the density of S. enterica lengthened the time to E. coli maximum density more than reducing the starting density of E. coli (P < 0.01, Fig. 3).
Given that cross-resistance and phage-target identity both influenced the cocktail efficacies, we wondered whether one was more important to consider when predicting outcomes of cocktail treatment. We used a mathematical model to investigate this question. To start, we modified a resource-explicit model of the coculture that simulated the abundance of bacteria, phage and resources through time ( Fig. 4A; Fazzino et al., 2020). We used model parameters informed by literature values and used wet-lab experiments to measure maximum growth rates (Table S2, Fig. 4B), and confirmed that the model accurately simulated the growth dynamics of the phage-free coculture ( Fig. 4C; Fazzino et al., 2020). Phage-resistant bacteria were seeded in at low frequencies to approximate standing variation for phage resistance. We included two different phage resistance mechanisms and then manipulated phage target identity. Phage-resistant mutants were either cross-resistant (resistant to two phage via one mutation) or dual-resistant (resistant to two phage via two independent mutations). The only difference between modelling resistance mechanisms was the doubly resistant mutants' starting frequencies. For cross-resistance, we seeded in mutants resistant to both phages at a frequency equal to the sum of single resistant mutant frequencies because resistance to either phage confers resistance to the other. For dual-resistance, doubly resistant mutants were seeded in at a frequency equal to the product of single resistant mutant frequencies to approximate the likelihood that two independent mutations evolved by chance (Table S3). With this model, we simulated treatment with single phage and cocktail treatments. As expected, cross-resistance to two E. coli phage decreased the time to maximum E. coli density in pathogen-targeting cocktail treatments and the multispecies-targeting cocktails suppressed E. coli better than the pathogen-targeting cocktail when cross-resistance evolved (Fig. 4Ddark bars). Furthermore, the relative efficacy of the cocktails depended on the resistance type. When we simulated cross-resistance, the multispecies-targeting cocktail was most effective (Fig. 4Ddark bars); however, when we simulated dual-resistance the pathogen-targeting cocktail was most effective (Fig. 4Dlight bars). These simulations suggest that for our experimental coculture, the evolved resistance type determines which cocktail treatment is most effective. Note though, for both simulated resistance types, the single P22vir phage treatments targeting S. enterica suppressed communities equally as well as the multispecies-targeting cocktail, suggesting that phage-target identity also contributes to treatment efficacy.
We wanted to know whether resistance type was always the determining factor of cocktail efficacy. Others have shown that differences in relative growth rates of cross-feeders change the recovery time from abiotic perturbations (Hom and Murray, 2014;La Sarre et al., 2017;Hammarlund et al., 2018). Therefore, we asked how changing relative growth rates of the cross-feeders altered phage treatment outcomes when simulating both dual-and cross-resistance mechanisms. In our experimental coculture, E. coli grows~1.3× as fast as S. enterica ( Fig. 4B and E arrow). If we made S. enterica grow faster than E. coli (left side of Fig. 4E graphs), then the most effective cocktail was the pathogen-targeting cocktail (2E) or the multispecies-targeting cocktail (1E + 1S), depending on the resistance type (Fig. 4E). Interestingly, the more similar the relative growth rates are, the smaller the differences in efficacy of cocktail treatment strategies. This suggests that targeting the slowest-growing cross-feeding partner is important for effective suppression, but predicting which cocktail suppresses the pathogen the longest depends on the evolved resistance type and relative growth rates of cross-feeders.

Discussion
We studied the optimal way to distribute two phage among two obligate cross-feeders to best suppress one focal bacterial species. In laboratory experiments, we found that a multispecies-targeting cocktail suppressed the model pathogen, E. coli, longer than pathogen-targeting cocktails. The simplest explanation for this result is that pathogen-targeting cocktails are overcome by a single E. coli mutation which confers cross-resistance to both phages. Consistent with this, we found an evolved mucoid phenotype in pathogen-targeting cocktails which did confer cross-resistance. However, we also found that even a single S. enterica phage suppressed E. coli as well as multispecies-targeting cocktails, which cannot be explained by cross-resistance. We first hypothesized that E. coli evolved resistance to T7 more easily than S. enterica evolved resistance to P22vir. However, resistance to P22vir was more common in S. enterica populations than resistance to T7 was in E. coli populations. An alternative hypothesis was rooted in population ecology: if S. enterica was the rate-limiting member of the obligate cross-feeding coculture, then reducing its population would limit growth longer than a similar reduction to E. coli. Experiments without phage, but where initial densities were manipulated, support this hypothesisa low starting density of S. enterica causes longer suppression than a similar low starting density of E. coli. Subsequent modelling showed that the cause of this effect was likely the differences in growth rate: S. enterica grows more slowly than E. coli, and this slower growth interacted with the population decrease caused by phage to enhance E. coli suppression duration. Our results highlight a novel multispecies-targeting strategy for designing phage cocktails when pathogens obligately cross-feed with other bacteria that is affected by relative growth rates and evolved resistance type. Our most effective cocktail strategy, the multispeciestargeting cocktail, included a phage that infected a nonpathogen, S. enterica, that cross-fed with our model pathogen, E. coli. This cocktail strategy used the ecological principle that inhibiting one cross-feeding partner effectively inhibits growth of other cross-feeding partners. By leveraging the same ecological principle, our laboratory previously showed that growth of a cystic fibrosis pathogen, Pseudomonas aeruginosa, can be inhibited by targeting its cross-feeding anaerobic partners with antibiotics (Adamowicz et al., 2018). While we are not the first to consider using multispecies-targeting cocktails, others have used them with a different goalto target co-occurring pathogens (Carson et al., 2010;Lehman and Donlan, 2015;Oliveira et al., 2018;Milho et al., 2019;Zhao et al., 2019). Additionally, others have explored phage treatment of pathogens in competitive ecological contexts, but limited their analysis to single phage treatments that targeted the focal bacterial species only (Harcombe and Bull, 2005;Brockhurst et al., 2006;Wang et al., 2017;Yu et al., 2017;Testa et al., 2019). Our research extended these foundational studies  Table S2 for details. B. Parametrizing bacterial growth rates from wet-lab data. The left panel is representative OD600 growth curves of E. coli (E, blue) and S. enterica (S, yellow) monocultures overlaid with Baranyi growth fits (black lines). The right panel shows calculated growth rates for each species. Bars are means AE SE (n = 5). C. Comparison of E. coli-specific phage-free coculture growth curves from the model and wet-lab experiments. Y-axis of the model growth curve is the total simulated E. coli biomass and the y-axis of the wet-lab growth curve is measured with CFP fluorescence units. D. Relative suppression (time to maximum E. coli density relative to phage-free simulations) of either cross-resistance (Cross-R) or dual-resistance (Dual-R) simulations with experimentally determined growth rates. X-axis labels refer to simulated phage treatment where #P X = # phages included that infect species X. Simulating resistance mechanisms used different starting densities of phage-resistant subpopulations (see text and experimental procedures for details). E. Simulation of relative suppression while modulating relative bacterial growth rates under cross-resistance and dual-resistance mechanisms. Simulated phage treatments included multispecies-targeting cocktail (1P E + 1P Sgreen), pathogen-targeting cocktail of two E. coli phage(2P E blue) or partner-targeting cocktail of two S. enterica phage (2P Syellow). Arrows indicate the relative growth rates of the experimental coculture measured in panel B. The multispecies-targeting and pathogen-targeting cocktails (green and blue lines) have experimental equivalents.
ª 2020 The Authors. Microbial Biotechnology published by Society for Applied Microbiology andJohn Wiley &Sons Ltd, Microbial Biotechnology, 13, 1997-2007 by including both pathogen-targeting and multispeciestargeting cocktails, and by addressing the role of cooperative cross-feeding between a pathogen and another coculture member. We highlight an additional way to leverage microbial ecological interactions to control pathogens.
We identified two independent factors that contributed to increased efficacy of the multispecies-targeting cocktail compared to the pathogen-targeting cocktail. First, the evolution of cross-resistance limited efficacy of the pathogen-targeting cocktail. We avoided this complication by using a multispecies-targeting cocktail strategy in which the individual phage could not infect both E. coli and S. enterica. Others have suggested alternative methods to prevent the evolution of cross-resistance. For example, Yu and colleagues designed cocktails with 'guard' phage that inhibit the evolution of phage resistance because they were previously experimentally evolved to infect likely-to-evolve resistant cells (Yu et al., 2018). However, many researchers have described multiple rounds of phage-host coevolution suggesting that protection by guard phage may be temporary on an evolutionary time scale, although this has not been tested (Koskella and Brockhurst, 2014;Jariah and Hakim, 2019). Others have used molecular techniques to identify phage binding sites and subsequently design cocktails that use multiple binding sites to increase both the number of mutations required for resistance and the cost of resistance (Filippov et al., 2011). While this would protect against receptor-mediated evolution of resistance, it would not prevent general resistance mechanisms that inhibit phage access to the cell surface, such as the evolution of mucoidy, which we observed when treating the cocultures with the pathogen-targeting cocktail. Yet, others have described phage that degrades this mucoid barrier and facilitate infection by other phage (Kim et al., 2015). Here, we identified an additional method for preventing cross-resistance from reducing the efficacy of phage cocktails by including phages that infect cross-feeding partners of targeted bacteria in our cocktail formulation.
Second, we found that including a phage that targeted the slower-growing cross-feeding partner was key to effectively suppressing pathogen growth. We used mathematical simulations to determine that relative growth rates of the cross-feeding partners altered how effective including a phage targeting the slower-grower was (Fig. 4E). In fact, inhibiting the slower-grower, S. enterica, with a single phage was as effective as inhibiting with the multispecies-targeting cocktails in experiments (Fig. 1B) and in simulations (Fig. 4D). Our findings agree with other studies that suggest that changes in relative growth rates of community members (Banks and Bryers, 1991;Raskin et al., 1996), particularly cross-feeders (Turner et al., 1996;Hammarlund et al., 2018) can alter responses to perturbations. To expand on these foundational studies, we used mathematical simulations to explore how relative growth rates impact the magnitude of response to perturbations. We found that the more similar the relative growth rates of the cross-feeders were, the smaller the difference in efficacy of cocktail strategies. Conversely, the more different the relative growth rates were, the more benefit we observed in targeting the slower-growing cross-feeding partner. While our simulations suggest that a S. enterica-targeting cocktail would be most effective at suppressing E. coli if cross-resistance did not evolve (Fig. 4E), we were unable to test this because our efforts to find a second S. enterica phage that replicated in our coculture were unsuccessful (Fig. S1). Our results suggest that including at least one phage targeting the slower-grower in a cross-feeding coculture is an effective method to extend pathogen suppression. Furthermore, our results indicate that if the relative growth rates of a pathogen and its cross-feeding partner are unknown, adding a nonpathogen-targeting phage could be one way to maximize the odds of inhibiting the pathogen.
A complication in a clinical setting or agricultural application could be that absolute pathogen population size, or pathogen load, may be more critical to treatment outcomes than how long the growth of a pathogen can be suppressed. Here, in two of fifteen communities treated with the T7 E. coli-targeting phage either alone or in a cocktail we observed complete eradication of E. coli populations and lower final E. coli densities in cocultures in which E. coli was not eradicated (Fig. S3). This would indicate that directly targeting the pathogen would be the fastest way to immediately decrease pathogen load. But the rate of resistance evolution would determine how long population sizes are kept low. Our results indicate that including phage targeting cross-feeding partners is one way to limit the recovery of knocked-down pathogen populations. These approaches are not mutually exclusivephage cocktails could include multiple phages targeting the pathogen, and one or more phage targeting its cross-feeding partner.
An alternative method for targeting multiple species in a community is with polyvalent phage treatment or phage with host ranges that encompass multiple species. Descriptions of polyvalent phage have increased over the past five years likely due to directed changes in phage isolation protocols (Hamdi et al., 2017;Duc et al., 2020;Li et al., 2020). In fact, Zhao and colleagues used a soil-carrot microcosm system to compare the efficacy of a cocktail that included phage targeting two different plant pathogens with a treatment of a single polyvalent phage that infected both pathogens (Zhao et al., 2019). They found that both treatments effectively limited the growth of both pathogens, but the polyvalent phage treatment disturbed the soil microbiome less than the ª 2020 The Authors. Microbial Biotechnology published by Society for Applied Microbiology andJohn Wiley &Sons Ltd, Microbial Biotechnology, 13, 1997-2007 multipathogen-targeting cocktail. Some challenges with using polyvalent phage might include differences in host preference based on receptor-phage binding strength. If binding strength were different enough, the polyvalent phage should function like a phage that targeted a single species. However, one benefit is that phage populations could grow faster because more hosts would be available, although no studies have directly tested this yet. We suggest that future research could test including polyvalent phage with different cocktail strategies.
In conclusion, we have illustrated a novel phage cocktail strategy for targeting cross-feeding pathogens. Our strategy limits cross-resistance evolution and maximizes pathogen suppression by targeting both the slower-growing partner and the pathogen. These and other results indicate that leveraging microbial community ecological interactions is a promising approach to help control pathogen growth in a variety of applications in human health, agriculture and food safety.

Experimental procedures
Bacterial and phage strains in the cooperative coculture system The bacterial strains used in this experiment have been previously described ( Fig. 1A; Harcombe, 2010). Briefly, the E. coli K12-derivative has a metB deletion and cyan fluorescent protein (CFP) in the attB lambda integration site. S. enterica is an LT2 strain with mutations in metA and metJ causing methionine secretion and yellow fluorescent protein (YFP) in the attB lambda integration site (Douglas et al., 2016(Douglas et al., , 2017. E. coli metabolizes lactose and excretes acetate which S. enterica consumes. S. enterica excretes methionine which is used by E. coli. Bacterial stocks were stored at −80°C in 20% glycerol. E. coli-specific phage T7 and P22vir were provided by Ian Molineaux (UT Austin) and S. enterica-specific P1vir by Ross Carlson (Montana State University). Phage stocks were grown on monocultures of ancestral E. coli or S. enterica in lactose or acetate minimal medium at 30°C. Cells were lysed with chloroform, centrifuged to pellet cell debris and stored at 4°C.
Measuring E. coli suppression in the cross-feeding coculture Bacterial growth at 30°C was tracked every 20 min with OD600 and fluorescence measurements using a shaking plate reader (Tecan Infinite ProM200). E. coli was measured with CFP (Ex: 430 nm; Em: 490 nm) and S. enterica with YFP (Ex: 500 nm; Em: 530 nm). We used four or five replicates of each treatment, as indicated. To wells in a 96-well plate, 10 5 cells each of mid-log phase E. coli and S. enterica were inoculated into 200 µl of lactose minimal media with 5 × 10 2 virions as indicated (MOI = 0.05 per phage). Cultures incubated for 5 days until stationary phase was reached. E. coli suppression length in hours was estimated by calculating the time to 95% maximum CFP measurement.

Profiling resistance to phage via cross-streak assays
To assay for acquired phage resistance, we used crossstreak assays with representative isolates from treatments. 30 µl of ancestral phage stock (10 8 to 10 9 PFU ml −1 ) was dripped down a minimal medium agar plate and left to dry. Overnight cultures of isolates grown in minimal medium were streaked perpendicular to the phage. Plates were incubated at 30°C until growth was visible. Isolates were determined to be resistant if streaks were uniform across the phage line and sensitive if bacterial growth was interrupted.

Resistance to phage due to standing variation in ancestral bacterial stocks
To determine frequency of phage resistance of ancestral bacteria, we quantified the number of cells that grew on phage-saturated agar plates. Ancestral E. coli and S. enterica monocultures were grown in lactose + methionine or acetate minimal media, respectively, for 3 days at 30°C. LB plates were saturated with 1 ml of ancestral phage stock (~1 × 10 9 PFU ml −1 ), dried and spotted with 5 μl of bacterial monocultures in 10-fold dilutions. Plates were incubated at 30°C until phage-resistant colonies were counted. We compared the number of colonies of plates with and without phage for each phage-host combination.
Assessing the effect of starting frequency of microbial partners on coculture growth We tested the time to maximum density of E. coli in the coculture when starting frequencies were altered in the absence of phage. We started the rare partner of cocultures at 0.03% while holding the common species at 10 5 cells/well in lactose minimal medium (n = 5). Community growth was as described above (see Methods: Measuring Experimental Cross-Feeding. . .).
In silico modelling of communities To represent our cross-feeding microbial community, we modified a series of resource-explicit ordinary differential equations to simulate an E. coli and S. enterica crossfeeding system in which one species grows on nutrients secreted by the other (Fazzino et al., 2020). We used Monod equations with multiplicative limitation of lactose and methionine essential nutrients for E. coli. The model mimics the metabolic network of the synthetic experimental coculture.
The major metaboliteslactose, acetate and methionineare tracked throughout simulations. Lactose is seeded in and is depleted as E. coli grows. Acetate is produced by E. coli growth and is depleted by S. enterica growth. Methionine is produced during S. enterica growth and is depleted during E. coli growth. Simulated cocultures grow until all lactose is consumed.
Each species has multiple genotypes to simulate resistance to different phage, with the amount seeded in representing mutation rarity. Resistant genotypes had founder population sizes at a maximum of 0.1% of the sensitive genotype to simulate rare resistance. E. coli had four genotypes; Es for sensitive to both phages, ErT7 for resistant to only T7, ErP1 for resistant to only P1vir and ErT7P1 for resistant to both phages. S. enterica had two genotypes; Ss for sensitive to P22vir and Sr for resistant to P22vir. Resistance was modelled as complete and without cost. The replication of each phage strain-T7, P1vir and P22virwas determined by adsorption rates and burst sizes. Each phage species can only kill sensitive genotypes of a single bacterial species. Model parameters are informed by literature values and were parameterized to approximate coculture growth dynamics without phage (Fig. 4, Table S2; Fazzino et al., 2020). Bacterial growth rates were measured from wet-lab monoculture experiments (E. coli grown in lactose + methionine and S. enterica grown in acetate) where OD600 was measured every 20 min. Growth curves were fit with a non-linear least-squares Baranyi function of the growth rate parameter, as described (Baranyi and Roberts, 1994). Figure 4 shows a schematic representation of the following equations and parameters.
E. coli (E) growth: dEs dt ¼ Es * μ_E * Lcts ðLcts þ k _ELctsÞ * Met ðMet þ k _EMetÞ S. enterica (S) growth: dSs dt ¼ Ss*μ_S* Ace ðAce þ k _SAceÞ Population sizes (E or S) are multiplied by their species-specific growth rates per hour (μ x ) and a Monod saturation function with a species and resource explicit constant (k) for each necessary resource. During phage infection, cell lysis is simulated. For example, when P1vir infects an E. coli that is only T7-resistant: Simulations were run in R with the DESOLVER package, using the LSODA solver (Soetaert et al., 2010). Fig. S1. Screening of S. enterica-specific phage activity in cooperative coculture. P22vir, SP6, and Felix-01 S. enterica-specific phages were inoculated into E. coli-S. enterica cocultures and grown at 30°C while shaking until stationary phase was reached (4-5 days, n = 1-2). Initial and final PFU ml −1 were measured by plating with ancestral S. enterica. Only P22vir increased in concentration over the growth period. Fig. S2. Coculture-level suppression lengths caused by phage treatments. Relative coculture suppression lengths of single and cocktail phage treatments standardized to the no phage control. Suppression length was calculated using 95% maximum OD600. Bars represent means AE SE (n = 4-5). Fig. S3. Boxplots of final E. coli densities after phage treatments. Including T7 phage in treatments lowered final E. coli population size. Cocultures were grown with single phage treatments and cocktails and bacterial populations sizes were counted by plating with selective plates. Statistical significance was tested with a Two-sample Mann-Whitney U. (n = 15). Table S1. Absolute and relative suppression lengths of phage treatments. Table S2. Parameters for resource-explicit ODE mathematical model. Table S3. Starting densities of cross-and dual-resistance modeling.