Our goal here was to develop procedures to explore the dynamical underpinnings of success and failure in treating bacterial infections. To illustrate the application of these procedures, we used the E. coli K1 mouse thigh infection model of Smith and Huggins , and like them, applied treatments of phages and antibiotics. As part of this effort, we repeated their experiments to ascertain if their results were robust. Across a variety of treatments and conditions, mouse mortality rates were remarkably similar between our study and theirs, even though we used a different strain of mice, phages from different sources, and an independently isolated strain of E. coli 018:K1:H7. Thus, in the absence of treatment, inoculations of 108 bacteria were fatal in over 95% of mice, but immediate treatment with phage that were specific for the K1 capsule or with streptomycin essentially eliminated mortality from the infection. Immediate treatment with phages that were not specific for the capsule also reduced mortality, but mortality rates for treatment with these phage were greater than with the K1-specific phage. Both studies also observed that mortality increased if treatment was delayed by 8 hours, under most forms of treatment.
Factors affecting treatment success
Two observations from these combined studies seem especially interesting. (i) Phages vary in treatment success. (ii) Delaying treatment by 8 hours substantially increases the rate of mortality relative to that of immediate treatment (for many of the treatments applied here), despite the fact infected mice survive at least 16 hours longer. In attempting to better understand these observations, we developed two assays. One assay, the RCA (R esistance C ompetition A ssay) measures the efficacy of a treatment in killing or inhibiting the growth of bacteria based on changes in the frequency of treatment-resistant bacteria after treatment is applied. The other assay, the PRA (P hage R eplication A ssay), measures the ability of a phage to replicate on populations of a bacterium under controlled, in vitro conditions.
The RCA and PRA values reflect what is observed with other (microbiological) measures of the efficacy of treatment. Consider first the effect of delayed treatment, which reduced mouse survival in most treatments. Increased mortality with delayed treatment could have various causes: (i) By the time of treatment, the numbers of bacteria are already sufficiently high that mortality is caused by toxins and tissue damage ensuing from them, rather than from further growth of the bacteria. Under these conditions, reducing the number of bacteria by treatment would have little effect on the clinical outcome of the infection. (ii) By the time of treatment, the numbers of bacteria are so high that the mouse's defenses cannot prevent further growth of the bacterial population, even with treatment, and bacterial numbers simply increase to the point of mortality. This model is similar to that in (i), except that a treatment which controlled bacterial growth would prevent mortality. (iii) As a consequence of the host's response to the proliferating population of bacteria, the site of the infection becomes less accessible to the antibiotic or phages, and the proliferation of the bacterial population can no longer be controlled by these antibacterial treatments. (iv) The bacteria become physiologically refractory to the treatment.
The RCA specifically supports the latter two models. If by eight hours the bacteria were as accessible and responsive to the phage or antibiotic as they were initially, the RCA values would remain high. We instead observed a drop in RCA values. The data do not rule out additional processes consistent with models (i) or (ii), but the data clearly reveal a reduced bacterial susceptibility or reduced access of treatment to the bacteria when treatment is delayed.
More than a half a century ago, H. Eagle and colleagues observed a dramatic decline in the efficacy of antibiotics with the term of the infection in a mouse thigh model. In their now classical investigations of the within-host dynamics of antibiotic treatment, they followed the course of penicillin treatment of Treponema pallidum and Group A and Group B Streptococcus [26–28]. Eagle  postulated that the most likely reason for the declining efficacy of treatment with the term of the infection was due to (i) a decline in the rate of metabolism and replication of bacteria during the infection, and (ii) that slowly growing or non-growing bacteria are more refractory to penicillin than those that replicating at higher rates. More recent work by E. Tuomanen, A. Tomaz and their colleagues supported this interpretation, a phenomenon they called phenotypic tolerance [29–31]. They also provided evidence that nutrient limitation was the cause of the decline in the rate of bacterial growth . Not only do bacteria become increasingly refractory to the majority of antibiotics as their rate of growth declines, but they also adsorb to and replicate bacteriophage less efficiently [24, 33], as is evident from the PRA values in Fig. 2. Thus the decline in the efficacy of phage and antibiotic therapy with delayed treatment in our experiments is plausibly attributed to a decline in the rate of replication of E. coli K1 once inside the mouse.
The other intriguing observation from Smith and Huggins that we corroborated here is the consistent superiority of phages requiring the K1 antigen for infection, when measured as mouse survival. Our RCA values showed a significant difference between the phages in vivo, suggesting that the K1-specific phages replicate at a higher rate than the non-K1-specific phages inside the mouse. Yet although Smith and Huggins casually observed that their non-K1-specific phages were inferior to K1-specific phages at lysing cultures of bacteria grown in artificial media, suggesting intrinsic differences between the phages more generally, our PRA estimates indicated that there was no intrinsic difference between our two phages in artificial media. Instead PRA differences were consistent with mouse survival and RCA data only when the PRA was measured on cells grown in mouse serum. Thus the PRA enables one to begin unravelling the environmental bases of differences in phage growth in vivo.
In this investigation, we used the mouse thigh infection model because of the many precedents for its use and because of its established repeatability [34–37]. The fact that bacteria undergo such a profound change in susceptibility or accessibility to treatment in only 8 hours raises the question of whether the mouse thigh infection model is representative of natural infections. This question is not necessarily answerable at present, but the results do highlight the fact that development of specific protocols and phages for treatment in any one experimental model may be inadequate for treatment of the same bacterium under field conditions.
The Resistance Competition Assay
The preceding discussion indicates that the resistance competition assay is consistent with other measures of phage performance in vivo and provides specific insights not easily obtained in other ways. The RCA has other virtues that make it useful for studying phage therapy and other forms of treatment. (1) It is versatile. In addition to being used to study the population dynamics of phage and antibiotic treatment in general, the RCA could be employed to design and evaluate the efficacy of different antibiotic treatment protocols. It can be applied to virtually any experimental model, such as enteric infections [38, 39] or urinary tract infections . (2) The RCA provides a more direct measure of the in vivo action of antibiotics (or other treatments) than estimates of the concentrations of these compounds in serum or in solid tissue. Moreover, the RCA protocol controls for the contribution of the host defenses as well as variety other factors that could influence the density of bacteria in a particular tissue. Only the action of the treatment per se can account for the difference in the frequency of resistant bacteria between treated and untreated treated hosts. (3) The RCA is more humane and offers greater statistical power (per animal) than "outcome" measures of the efficacy of treatment based on survival or other clinical indications. The fact that the RCA yields a continuous statistic, rather than a binary one, enables the use of standard statistical analyses in which meaningful comparisons can be made with as few as two replicates per group. (4) The RCA can be applied to populations of bacteria infecting specific tissues at specific times, even allowing multiple measures per animal. However, individual tissues subjected to the RCA must not have high levels of bacterial migration from other tissues over the course of treatment.
The RCA is a sensitive and specific measure of treatment efficacy, and as such, will not correlate perfectly with other measures of treatment. To wit, our RCA s were similar between delayed φLH treatment (0.3) and immediate φLW treatment (0.2), even though mouse survival rates were significantly different between them (11 of 12 mice versus 6 of 15). These discrepancies highlight the complexity of the infection process and the fact that different measures of infection dynamics capture different properties. Because our RCA s were based on samples from the infected thigh, the values apply specifically to that thigh and do not necessarily reflect the efficacy of treatment in other tissues that may influence host survival.
The Phage replication assay
This method has a potential utility beyond that demonstrated here. If phage therapy is to be developed for particular infections, it would be useful to have an in vitro procedure to screen phages for their potential efficacy in vivo. The diversity of phage is enormous. For example among 40 phage isolated from different samples or different plaques on lawns of E. coli K12 and E. coli B, at least 32 distinct phage were found  (as measured by host range and/or restriction pattern). A far greater number of phage could certainly be isolated with a broader array of lawn bacteria and by sampling different sources. The implication is that it should be possible to isolate a substantial number of lytic phage capable of killing most strains of enteric and other bacteria, hence offering a compelling reason to pursue phage therapy as a solution to antibiotic resistance. Screening many phages for their therapeutic potential in experimental animals would be a time- and animal-consuming task. The PRA employed here could facilitate that screening. Our results suggest that phage performance in serum could be a sufficient indicator of in vivo performance, but there is no reason that this assay could not be performed in vitro with other modifications, such as solid tissues. Moreover, as suggested in a recent study , in vivo culture may selectively improve the capacity of a phage to replicate on and kill bacteria in a mammalian host.
Prospects for phage therapy
This is a methodological study to develop and experimentally evaluate procedures to measure the efficacy of antibacterial treatment and to screen bacteriophage for their therapeutic potential. Our purpose in performing these experiments and publishing these results is not to advocate the use of phages for the treatment of systemic infections. Nevertheless, because of the current novelty of and aspirations for phage therapy as an alternative to antibiotics, it seems appropriate to acknowledge that our results and those of several other studies offer promise that phage therapy is highly repeatable, can be successful, and is thus worthy of further research for clinical practice. Moreover the use of phage for therapy and prophylaxis needn't be restricted to humans, as phage could obviously be used for these purposes in domestic animals.
There are, of course, a number of problems associated with the use of phage as an alternative to antibiotics. To us the most serious biological problem is a restricted host range. Not only would one have to know the species of bacteria responsible for an infection, it would be necessary to know which phages can infect that strain of bacteria. These requirements are certainly inconsistent with current empiric therapy that uses broad spectrum antibiotics, which dominates how antibiotics are employed in the community as well as in hospitals. Commonly, it is not clear whether a bacterial infection is responsible for the symptoms being treated with antibiotics, much less the species, strain, and resistance profile of the bacteria responsible. On the other hand, there are situations, like epidemics, where this knowledge would be available. And as procedures to identify the bacteria responsible for symptomatic infections get better and more rapid, it soon may be quite easy to get this information from individual patients. As a consequence of the ever increasing frequency of antibiotic resistant bacteria, the range of antibiotics to which individual bacteria are resistant, and the limited number of targets to which current (and soon to be anticipated) antibiotics are directed, there is a pressing need to develop alternative methods of treating and preventing bacterial infections. Phages certainly offer some of the most readily-available and promising alternatives.