Coverage of diarrhoea-associated Escherichia coli isolates from different origins with two types of phage cocktails

Eighty-nine T4-like phages from our phage collection were tested against four collections of childhood diarrhoea-associated Escherichia coli isolates representing different geographical origins (Mexico versus Bangladesh), serotypes (69 O, 27 H serotypes), pathotypes (ETEC, EPEC, EIEC, EAEC, VTEC, Shigella), epidemiological settings (community and hospitalized diarrhoea) and years of isolation. With a cocktail consisting of 3 to 14 T4-like phages, we achieved 54% to 69% coverage against predominantly EPEC isolates from Mexico, 30% to 53% against mostly ETEC isolates from a prospective survey in Bangladesh, 24% to 61% against a mixture of pathotypes isolated from hospitalized children in Bangladesh, and 60% coverage against Shigella isolates. In comparison a commercial Russian phage cocktail containing a complex mixture of many different genera of coliphages showed 19%, 33%, 50% and 90% coverage, respectively, against the four above-mentioned collections. Few O serotype-specific phages and no broad-host range phages were detected in our T4-like phage collection. Interference phenomena between the phage isolates were observed when constituting larger phage cocktails. Since the coverage of a given T4-like phage cocktail differed with geographical area and epidemiological setting, a phage composition adapted to a local situation is needed for phage therapy approaches against E. coli pathogens.


Introduction
The antibiotic crisis has rekindled the interest in phage therapy approaches that were developed in Eastern Europe (Brüssow, 2012). Phage therapy has theoretical advantages over antibiotic therapy. Phage is a selfamplifiable antibacterial agent that increases in titer only when it meets in vivo its target (Payne et al., 2000). Phage is quickly eliminated from the body in the absence of the pathogen as shown in pioneer work in mice (Merril et al., 1996) and later in humans (Bruttin and Brüssow, 2005). In addition, phages are mostly species-specific bacterial viruses. Exceptions are phages infecting Listeria innocua and Listeria monocytogenes or Salmonella and Escherichia coli (Santos et al., 2010). Consequently, and in notable contrast to antibiotics, no collateral damage on the commensal microbiota is to be expected. However, the full scale of ecological relationships between coexisting bacterial species in a given microbial community is quite complex and is not fully described or understood yet. Data from human healthy subjects do not suggest phageinduced gut microbiota changes induced by oral phage (Sarker et al., 2012;McCallin et al., 2013). The specificity of phages has also a downside: many phages are not only species-specific, but are also strain-specific. This property becomes a coverage problem when phage therapy targets a disease caused by a complex pathogen like diarrhoea-associated E. coli, which is represented by many different pathotypes and serotypes (Chibani-Chennoufi et al., 2004a,b). The constitution of complex phage cocktails then becomes also a commercial problem of producing a standardized preparation. For some bacterial pathogens, broad host-range phages have been described (Salmonella: Santos et al., 2010;Staphylococcus aureus: Kelly et al., 2011). However, no such broad host-range phages have so far been isolated for E. coli. Recent data suggest that coliphage phi92, isolated from an encapsulated K1 E. coli strain and belonging to the novel genus of rv5-like phages, might have a broad host range (Schwarzer et al., 2012) like the multivalent Salmonella phage PVP-SE1 infecting both Salmonella and E. coli (Santos et al., 2010). These phages have not yet been tested against large collections of pathogenic E. coli strains. Phage therapy against E. coli must therefore rely on phage cocktails. The structure of host-phage interaction matrices has recently become an active research area (Flores et al., 2011;Weitz et al., 2013). Different types of interaction were identified: specialist phages that may infect a unique host leading to a nearly diagonal matrix when columns indicate phage and rows host isolates; phage-host interactions that may describe block-like matrices, which exhibit high degrees of modularity; generalist phages and phage-host interaction that are random.
For regulatory reasons, we worked with a cocktail containing only a single phage genus, reasoning that it might be easier to demonstrate safety for a single phage group than for a cocktail containing numerous different phage genera. We settled for the T4-like genus of Myoviridae (Bruttin and Brüssow, 2005;Sarker et al., 2012) because professional virulent phages that destroy the host genome during the infection process are considered by many researchers as best candidates for phage therapy (Brüssow, 2012). Early on, it was reported that T4-like phages recognize lipopolysaccharides (LPS) in the outer membrane of E. coli (O antigens) as primary phage receptors (Tamaki et al., 1971). However, LPS synthesizing genes represent a hotspot of genetic diversity in E. coli (Heinrichs et al., 1998). In fact, the specificity of LPS became the basis for the definition of more than 150 different O serotypes, although not all O serotypes are encountered in pathogenic E. coli (Robins-Browne, 1987). If LPS is the receptor for T4-like phages on E. coli, one would predict problems for achieving high coverage of a phage cocktail against diarrhoea-associated E. coli. Other researchers identified the outer membrane protein OmpC as receptor (Hashemolhosseini et al., 1994;Yu et al., 2000) for T4 phages. Since this gene is well conserved across different genera of Enterobacteriaceae when using the E. coli K-12 gene in BLAST searches, multivalent T4-like phages should be expected. In the present report, we explored the coverage of phage cocktails composed of T4-like phages against E. coli isolates from different epidemiological settings compared with a commercial Russian phage cocktail containing many different genera of coliphages (McCallin et al., 2013). We conclude that pathogen coverage is dependent on the epidemiological context requiring flexible and locally adapted phage cocktails.

Phage propagation and lysis tests
The test phages were propagated on the E. coli strain K-12 derivative K803, which lacks prophage lambda, obtained from E. Kutter, Evergreen College, Olympia, Wa, USA. Luria-Bertani (LB) broth was inoculated with 1% of an overnight bacterial culture and incubated at 37°C for 2 h to reach 1 × 10 8 cfu/ml. Phage lysate was added to the culture to get 1 to 5 × 10 8 pfu/ml and the infected culture was incubated at 37°C for 3 to 5 further hours with agitation (100 r.p.m.). After removal of bacterial cells and cellular debris by centrifugation for 15 min at 14 000 × g and passing the lysate through a 0.22 μm Millipore filter, the phage titers in the lysates were determined by the plaque assay as described previously (Bruttin and Brüssow, 2005). In small experimental series 100 μl of an appropriate phage dilution and 100 μl of an exponentially growing test pathogenic E. coli culture to achieve an moi of 3 were incubated at 37°C in 10 ml tubes containing LB broth, and the OD development was read at different time points against the uninfected test cell as control. For larger experimental series (100 phage isolates against about 100 E. coli isolates at three moi in duplicates resulting in 60 000 tests) we used the microtiter lysis test; 96-well plates with 10 μl of centrifugation-cleared and sterile-filtered (0.22 μm pore size) phage lysate and 180 μl of fresh LB broth were inoculated with 20 μl of the test E. coli target strains to obtain an moi of 0.1, 0.01 and 0.001 respectively. The plate was incubated at 37°C for 4 h under agitation (200 r.p.m.). Every hour the OD 595 was meas-ured in a Dynex reader (MRX model, Magellan Biosciences, Chantilly, VA). The E. coli strain without phage and non-inoculated LB medium were used as positive and negative controls respectively.

E. coli strains from Mexico
Several hundred E. coli test strains were collected between 1985 and 2003 from children who experienced an episode of diarrhoea in two regions of Mexico (Mexico City and the State of Morelos). Eighty-nine strains representing the diversity of O and H serotypes encountered were selected for phage susceptibility testing. The E. coli strains isolated from children in the State of Morelos belonged to a field study reported previously (Cravioto et al., 1990). In short, the isolates were characterized as E. coli strains by the Vitek automated bacterial identification system from BioMérieux. The strains were then serotyped by agglutination assays (Ørskov and Ørskov, 1984) using 96-well microtitre plates and rabbit serum (SERUNAM) obtained against 187 somatic antigens and 53 flagellar antigens for E. coli, as well as against 45 somatic antigens for Shigella species. Polymerase chain reaction (PCR) assays were used to determine the presence of virulence genes with primers specified in Supplementary Table 1. The presence of the following genes was investigated: for enteropathogenic E. coli (EPEC), intimin gene (eae) (Karch et al., 1993) and bundle forming pilus (bfp) gene (Gunzburg et al., 1995); for enterotoxigenic E. coli (ETEC), heat-labile enterotoxin LTh (Schultsz et al., 1994), heat-stable enterotoxins STh and STp genes (Schultsz et al., 1994); and for enterohemorrhagic E. coli (EHEC), the Shiga-like toxin Stx genes (Pollard et al., 1990). The bacterial strains were stored at −20°C on 'Protect Bacterial Preservers' beads (Technical Service Consultants, Lancashire, UK).

E. coli strains from Bangladesh
A prospective community-based study was conducted in an urban slum in Mirpur, Dhaka, Bangladesh, from April 2002 to October 2004. Pregnant mothers were screened and 300 newborn children were followed prospectively for diarrhoeal diseases (Qadri et al., 2007). E. coli colonies were isolated as described previously (Albert et al., 1995;Qadri et al., 2000). In case of diarrhoea, six lactose-fermenting E. coli colonies from MacConkey agar were inoculated on GM1-coated microtiter plates containing LB broth for 18 h. The supernatant was tested for ST using an inhibition ELISA procedure (Svennerholm et al., 1986) and for LT by using an anti-LT monoclonal antibody (Svennerholm and Wiklund, 1983). ETEC strains were cultured on CF antigen agar (McConnell et al., 1989) and tested for the expression of CFA/I, CS1 to CS8, CS12, CS14, CS17, CS19, and CS21 by monoclonal antibody-based dot blot assay (Qadri et al., 2000).

E. coli pathogens from a community survey of childhood diarrhoea in Mexico
Since LPS, the chemical determinant of the somatic O antigen, was identified as receptor for T4-like phages on E. coli cells by some authors (Tamaki et al., 1971;Thomassen et al., 2003), we first tested our T4-like phage collection against a collection of O serotype-defined, diarrhoea-associated E. coli strains. E. coli strains were isolated from children with diarrhoea living in two regions of Mexico (Cravioto et al., 1990). From this collection, we chose 89 E. coli strains for screening against our T4-like phage collection (Table 1). The test E. coli strains represented 69 different O serotypes. Six O serotypes were associated with two or three different H serotypes. Overall, 26 different H serotypes were observed, several H serotypes were associated with up to 4 O serotypes (the strains were selected for maximal serotype diversity). Fifty isolates showed the EPEC-specific eae1 gene. Only three of them also showed the bfp gene characterizing them as classical EPEC strains, making the majority of the Mexican isolates atypical EPEC. Fourteen isolates showed enterotoxin genes -the most prevalent combination were strains with STh+LTh genes (n = 7). Only few strains showed STp+LTh gene combination (n = 2), only LTh (n = 2), only STp (n = 2) and STp+STh+LTh genes (n = 1). Compared to the 50 EPEC isolates, only 14 isolates were thus diagnosed as ETEC strains. Shiga-like toxin was found in a single isolate.

Coverage of the Mexican E. coli pathogens with coliphages
The susceptibility of the individual 89 E. coli strains for the 98 T4-like phages from our collection (Sarker et al., 2012) was tested in a lysis test at three different multiplicities of infection (moi = 0.1, 0.01, 0.001). The lytic effect of the phage was rated according to the observed lysis: grades 1, 2 and 3 lysis represented growth inhibition only at the highest moi, at two and all three moi levels, respectively ( Fig. 1). Most T4-like phages had only a narrow host range on the 89 serotyped test strains, but no clear correlation with O serotypes was seen (data not shown). A few phage isolates lysed many E. coli strains independent of the O serotype (Supplementary Table 2). The T4-like phage with the broadest host range showed lytic activity on 44 bacterial strains (49% of the total). When combining three T4-like phages with the broadest individual host ranges, an arithmetical coverage of 87% should have been achieved. However, the physical combination of these three T4-like phages only resulted in the lysis of 54% of the E. coli strains suggesting interference between the phages (Table 1). When six T4-like phages were combined into a cocktail, we could increase the coverage to 69%. Further increasing the number to nine phages did not result in enlarged coverage (Table 1). A commercial Russian phage cocktail (Coli-Proteus phage cocktail from Microgen) only lysed 17 of the 89 test bacteria (19% coverage) ( Table 1).

Coverage of E. coli pathogens from a community diarrhoea survey in Bangladesh
The commercial Russian phage cocktail was presumably selected for good coverage on E. coli strains from Russia. Its low coverage on the Mexican E. coli pathogens raises the question whether phage cocktails display geographical specificities. We do not know the selection process of phages for the Microgen cocktail, but at the Eliava Institute in Tbilisi/Georgia where Soviet phage therapy took its origin, new phages against E. coli pathotypes not covered by the previous stock are in regular intervals added to the cocktail (Chanishvili, 2012).
To test the hypothesis of geographical diversity, we studied the coverage of the above-mentioned phage cocktails against 40 E. coli isolates from a community survey of childhood diarrhoea in Bangladesh (Qadri et al., 2007). In contrast to the Mexican collection, only 5 isolates represented EPEC strains while 35 strains were ETEC strains with different combinations of ST and LT enterotoxins and colonization factors as assessed by ELISA and monoclonal antibody-based dot blot assays, respectively (Table 2). Our nine-phage cocktail lysed only 30% of the pathogenic E. coli strains from Bangladesh compared with its 69% coverage on the Mexican strains. We then increased the number of constituting T4-like phages to 14 in the cocktail, based on the lytic properties of individual phage strains from our cocktail against ETEC strains from Bangladesh. We achieved 53% coverage, while the Microgen phage cocktail showed 33% coverage on this strain collection from Bangladesh compared with 19% coverage it had achieved on the Mexican collection. For a subset of the E. coli strains, the phage susceptibility of the strain was confirmed by the plaque assay ( Table 2).

Coverage of E. coli pathogens from diarrhoea patients hospitalized in Bangladesh
The  (Table 3). Two different batches of Microgen ColiProteus phage cocktails showed a nearly identical lysis pattern on these strains (data not shown) while two different batches from Microgen 'Sextaphage' cocktail showed an overlapping, but distinct lysis pattern with an overall lower coverage of 35% (Table 3). Since we were not satisfied with the 24% coverage by our nine-phage cocktail, we modified its composition to obtain a broader coverage. Modification was done on the basis of lytic host range determined for each individual phage strain against the E. coli pathogens isolated in 2008 from patients at icddr,b. A new cocktail consisting of 10 T4-like phages led to an increased coverage of 50% on these hospital-derived E. coli pathogens.
With the 14-phage T4 cocktail, 61% coverage was achieved (Table 3). Notably, the 10-phage cocktail lysed only 46% of the Mexican strains, less than the 68% lysed by the 9-phage cocktail (Table 1). Changing the phage composition of a cocktail in order to achieve a higher coverage in one geographical region might thus result in lesser coverage in another region. E. coli pathogen coverage by a phage cocktail is thus strongly context-dependent, suggesting that phages for therapy should be tested, if not isolated, using the most prevalent pathogenic strains in a given geographical area.
In 2011, we requested again random ETEC and EPEC isolates from children hospitalized with diarrhoea at icddr,b to explore temporal variation of pathogen coverage by given phage cocktails. The 10-phage T4 cocktail showed 29% coverage in the tube lysis test. The T4-like phage cocktail showed phage plaque counts in excess A. The phage got a lysis score of three because it lyses at all three moi. B. The phage got a lysis score of two because it lyses only at the two higher moi. C. The phage got a lysis score of one because it lyses only at the highest moi. D. It showed a bacterium-phage pair where no lysis was observed at any of the three moi. of 10 4 pfu/ml in all cases where lysis was observed ( Table 4). The Microgen Coliproteus phage cocktail lysed only 6% of these test strains, but formed plaques on half of them, but plaque counts were low when the cells were not lysed. Apparently, plaque counts higher than 2 × 10 3 pfu/ml are needed to cause lysis of the cells (Table 4). Lysis was occasionally observed when no plaques were seen in the plaque assay. Whether this represents 'lysis from without' by phage tail-associated lytic enzymes from phages that cannot complete an infection cycle or 'abortive infection' is not clear, but according to the low moi which we used and a theoretical paper (Abedon, 2011), the second diagnosis is more appropriate. Independent of this question of clear terminology,   (VTEC), Enterotoxigenic E. coli (ETEC)the latter were serotyped for O antigen. Age: P indicates pediatric patient. Phage cocktails: N9-p cocktail consisting of 9 T4-like phages, N10-p cocktail consisting of 10 T4-like phages (for composition see Table 1); N14-p cocktail consisting of 14 T4-like phages (for composition see Table 2), Micro: ColiProteus phage cocktail; Sexta: Sextaphage cocktail from Microgen. L: lysis in 10 ml test tube assay. bacterial cells do not experience the same environment when they are grown in solid medium as compared to liquid that leads to a distinct expression of some surface proteins which may affect the infection process.
Since Shigella is taxonomically not even a subspecies of E. coli, we extended the testing to Shigella isolates from 10 different pediatric patients. The isolates represented two different Shigella 'species'; one was represented with four different serotypes. All strains showed the ipaH virulence genes, while none contained the Shiga toxin gene. The Microgen ColiProteus phage cocktail formed plaques on 9 Shigella isolates while 10 T4-like phage cocktail showed lysis and plaques on 6 Shigella isolates (Table 5).

Discussion
T4-like phages from our collection were with few exceptions all isolated on a non-pathogenic laboratory E. coli strain, more specifically K803, a derivative of the K-12 strain lacking the lambda prophage (Chibani-Chennoufi et al., 2004b). This restriction could be a significant selection bias which might explain the relatively poor coverage observed in the current study. One might object that phages should be isolated using prevalent pathogenic E. coli as indicator strains even if this renders the whole experimental work more difficult to handle (need of a safety laboratory, need of many propagating strains). If these phages are subsequently propagated on a nonpathogenic laboratory strain for commercial purposes, then their ability to lyse the most prevalent pathogenic strains should be confirmed. In fact, in the initial phase of our work, we isolated phages from stool samples of diarrhoea patients using in parallel an enteropathogenic E. coli strain, which was commonly isolated from childhood diarrhoea cases, and the laboratory K803 E. coli strain (Chibani-Chennoufi et al., 2004b). We obtained less phage isolates on the pathogenic strain than on the laboratory E. coli strain. Furthermore, the phage isolates made on the pathogenic E. coli strain were Siphoviridae with 50 kb genomes showing the morphology of Jerseyand beta-4 like phages and a narrower lytic pattern on O-serotyped pathogenic E. coli strains than phage isolates from the laboratory K803 strain, which yielded Myoviridae with a 170 kb genome, the morphology of T4-like phages and a broader host range on O-serotyped pathogenic E. coli strains (Chibani-Chennoufi et al., 2004b). We also obtained rather narrow host-range phage isolates when using a few other pathogenic E. coli as indicator strains (unpublished observations).
It is not unusual to select and propagate phages for phage therapy purpose against pathogenic bacteria on a nonpathogenic host strain. For example, a multivalent Salmonella phage was selected and amplified on a nonpathogenic E. coli strain. After serial passage on the two alternative hosts, the researchers observed a smaller burst size and longer rise period for the phage propagated on the heterologous E. coli host than on the homologous Salmonella host, but the lower cost of producing a safer phage product justified in their opinion these growth disadvantages. Notably, no difference in host range was observed between the phage propagated on the pathogenic Salmonella or non-pathogenic E. coli host (Santos et al., 2010).
Propagation on a laboratory strain avoids biosafety production conditions and the risk of transfer of bacterial or prophage-encoded toxins (Canchaya et al., 2003) from the pathogenic host into the phage preparation. The fact that K-12 displays only a truncated LPS molecule with a shortened O side chain (Qimron et al., 2006) might not only have changed biological phenotypes like pathogenicity for infection models in worms (Browning et al., 2013). Use of K-12 might have selected against phages recognizing the O side chain-specific sugar residues as recep-   (Tamaki et al., 1971;Qimron et al., 2006), potentially explaining the infection of many different O serotypes with a single phage. Recognition of alternative T4 phage receptors like the membrane protein OmpC proposed by other researchers (Hashemolhosseini et al., 1994;Yu et al., 2000) is not very likely for the isolated phages since one would expect a much broader cross-reactivity on O-serotypes E. coli pathogens than was actually observed.
One might also object that determining the host range of the isolated phages on E. coli growing in LB broth as done for enterobacteria propagation in the laboratory has nothing in common with the conditions encountered in vivo. For example, in our experiments, the infected E. coli cells were shaken to assure oxygen access while the lower parts of the gut display anaerobic conditions. To address this problem, we tested whether the investigated T4-like phages could infect E. coli under anaerobic in vitro growth conditions. This was the case: our T4-like phages yielded in vitro similar titers under aerobic and anaerobic growth (Weiss et al., 2009). Of course, there are many other conditions that differentiate in vitro from in vivo growth (Brüssow, 2013). For example, in their natural environment bacteria grow mostly in biofilms while broth culture represents growth in suspension. Indeed, T4 phage showed lytic infections on biofilms (Doolittle et al., 1995), but nutrient limitation in biofilms delayed the lysis process (Doolittle et al., 1996). One might therefore wonder whether in vitro conditions better adapted to in vivo growth conditions would not be preferable to standard broth culture. Since the in vivo growth conditions of E. coli are still poorly characterized, it is unfortunately not yet clear what in vitro conditions best mimic the in vivo condition.
It is widely believed that phage therapy approaches need phage cocktails (Brüssow, 2012). If resistance development is an issue during phage therapy (Cairns et al., 2009), it is less likely to develop against a phage cocktail than against a single phage. However, combining phages into a cocktail can create interference problems making the actual lytic capacity of a phage cocktail less than the sum of its parts. This phenomenon was here observed with cocktails of T4-like phages. Mechanistically, super-infection of E. coli cells with two T4-like phages causes lysis inhibition (Abedon, 1992) preventing an OD decrease, which was the read-out of our test. We tried different combinations of individual T4-like phages hoping to identify specific phage strain combinations displaying antagonistic behavior in order to eliminate them from the cocktail. However, no such specific interference problems between individual phage strains were spotted, suggesting merely random effects when increasing the overall number of T4-like phages in a cocktail. Retrospectively, one might also ask whether limiting the phage cocktail to a single phage genus like T4-like phages for safety reasons was a good strategy. Safety trials showed no difference when using a T4-like phage cocktail (Sarker et al., 2012) or the Russian phage cocktail containing in addition to T4and T7-like phages a dozen further phage genera (McCallin et al., 2013). As the in vivo phage growth conditions in the human gut are still largely undefined, working with a complex phage cocktail as done by the Eastern European phage tradition might represent a hedge betting strategy for diversifying risk. However, when considering only the coverage of the target pathogen with in vitro lysis tests, no clear superiority of the Microgen phage cocktails over the T4-like phage cocktail was observed.
If we accept the hypothesis that in vitro lysis is a necessary, although not sufficient condition for in vivo lysis, coverage is critical for efficient phage therapy. To illustrate it numerically: with 50% coverage, two E. coli patients need to be treated to expect an effect in one of them. If only one out of two to three hospitalized pediatric diarrhoea patients suffers from an E. coli diarrhoea as is the case in Bangladesh (Albert et al., 1999), four to six patients need to be treated for one to profit from phage treatment. This calculation neglects the possibility that many patients are infected with similar strains which is actually the case during a given epidemic (Qadri et al., 2005), while this might not be the same strains between epidemics (Chowdhury et al., 2011). This consideration put aside, relatively large numbers of children with diarrhoea need to be enrolled into a controlled clinical trial to observe a treatment effect raising even an ethical conundrum using a therapeutical preparation which only cures a limited proportion of the patients. This problem could be solved when targeting Shigella infections. Dysentery symptoms are so characteristic that it can be easily diagnosed clinically and its coverage by phage was much higher than for E. coli diarrhoea. The combination of both factors makes it likely that at least every second dysentery patient could profit from phage therapy. In line with other reports (Chang and Kim, 2011), Shigella was lysed by coliphages, but this should not be surprising since Shigella qualifies not even as a subspecies of E. coli (Sims and Kim, 2011). In support to this conclusion identifying Shigella infections as a suitable target for phage therapy, the only clinical trial published by Soviet phage therapy researcher reported successful prevention of Shigella dysentery (Sulakvelidze et al., 2001). Diarrhoeaassociated E. coli pathogens in contrast are a difficult target for phage therapy since the pathogen coverage by phages varies with geographical region, epidemiological setting and time as demonstrated in the present study. It is not clear whether our observations can be generalized. Due to frequent inundation by flood water, Bangladesh displays an exceptionally high dynamic of diarrhoea pathogens (Harris et al., 2008) making it not a typical, but a worst-case scenario.
Evolutionary biologists have asked whether in vitro phage-host systems are adequate models for the in situ situation in the real ecological niche (Forde et al., 2008). For clinical efficacy, phages must display the appropriate in vivo stability to survive the gastro-intestinal passage, a good in situ bioavailability at their site of action and a good replication potential on E. coli demonstrating different physiological states in different gut segments. None of these phenotypes can be easily tested in vitro. Some pathogenic E. coli strains multiply in the small intestine (e.g. EPEC), others replicate in the large intestine (e.g. EHEC, Shigella). Data on intestinal phage-host interaction in laboratory mice were equivocal. In our laboratory, a T4-like phage was less efficient than a T7-like phage in replicating on slowly growing commensal E. coli in the gut (Weiss et al., 2009). In another laboratory, a T4-like phage was a better replicator than a T7-like phage on a pathogenic E. coli strain asymptomatically colonizing the gut of mice (Maura et al., 2012a,b). In both models, oral T7-like phages replicated to high titers in the gut without, however, having a sustained effect on the fecal E. coli titer (Weiss et al., 2009;Maura et al., 2012a). When investigating phage-host interaction in different segments of the gut, marked differences were observed (Weiss et al., 2009;Maura et al., 2012b). Our understanding of coliphage E. coli interaction in the gut, their natural ecological niche, is apparently so limited (Brüssow, 2013) that testing a phage cocktail for in vitro coverage can only be seen as a step towards a successful phage therapy approach. In the absence of an E. coli diarrhoea model in small laboratory animals, we study currently the faecal excretion pattern in children hospitalized with E. coli diarrhoea who were treated with two different phage cocktails to get data on the in vivo replication of oral phage cocktails.