In this study we have shown that IP infection does result in chronic TB infection in mouse lungs similar to that observed during low dose aerosol infection. This type of infection is not common as the natural route of TB infection is absorption of bacteria by lung tissues. The kinetics of accumulation of bacteria in the organs after IP administration is slower than that of chronic TB via aerogenic infection (stabilization of CFU numbers at an almost constant level after 60 days vs 20–35 days for aerogenic infection) [14, 22]. The advantages of this approach are low level cross contamination between animals and low cost of the experiments (no need a special apparatus for aerogenic infection). Dhillon et al. have reported on establishing of chronic TB infection in Balb/c mice with CFU levels of ca 104 CFU in lungs and in spleens after 4 weeks of IP infection but the kinetic curves were not published . It is interesting that the significant portion of cells recovered from organs in this model  were non-culturable on solid medium but capable of growth in liquid media . These observations bear some similarity to dormant MTB cells obtained in vitro  and are suggestive of the formation of a population of dormant (latent) bacteria in the IP model. It is interesting to note that a similar population of non-culturable bacteria could also be isolated when MTB was grown in macrophages  Evidently, the stable CFUs in organs at 60 days post infection are controlled in an immune dependent fashion as with low dose aerosol infection since immune suppression by AG or anti-TNFα antibodies resulted in significant increase of CFU levels in lungs after 20–30 days of drug administration. Similar results were also observed for chronic infection induced after aerosol challenge . AG is known as an effective inhibitor of nitric oxide synthase (NOS) and induces reactivation of bacillary growth in mice with chronic TB infection [26, 27]. As expected AG administration in our model resulted in increased bacillary growth and an increase in the lesion area in the lungs of infected mice (Figure 2). Nitric oxide has been shown to alter the regulation of an endothelial derived mediator of vascular tone, endothelin-1 (ET-1). It was demonstrated that increasing endogenous NO generation through activation of NOS will lead to a decrease in ET-1 . It is possible that AG administration can result in an increase in ET-1 possibly followed by vasoconstriction, pulmonary hypertension and hypoxia. These factors will result in augmentation of lesion area in lungs as found in this study (Figure 2). In the present study the increase in CFU is transient in contrast to published results [22, 26]) as AG was given to mice for a short period only (10 days). This approach makes it possible to observe the direct response of bacterial multiplication to inhibition of NOS without interference by the negative site effects caused by long-term administration of AG. A similar reactivation effect was observed after administration of anti-TNFα antibodies. It has been previously shown that anti-TNFα antibodies were able to induce reactivation of TB in chronically infected mice and produced significant histopathological deterioration in organs . This effect of inactivation of TNFα is probably due to different mechanisms including modulation of inflammatory state of the lungs, attenuation of the expression of NOS2 in the lungs and altered expression of other cytokines and chemokines [27, 29]. The increase in CFUs after anti-TNFα treatment in the present study was similar to data reported in previous experiments with chronically infected mice (ca. 1.5 log after administration) . Again, due to the short regimen of antibody administration, the increase in CFUs is transient.
Generally, the IP infection model reveals features similar to the more commonly used low dose aerogenic infection model, such as the establishment a steady-state level of CFUs in organs that does not change significantly over time. Given that clinical latency in humans is characterized by low bacillary loads, the IP route of infection may in fact model TB latency even better than aerogenic infection as the steady-state bacterial load in this study was lower than for aerogenic infection for the same strain in our studies (unpublished results and see also [14, 15, 30]).
In a previous study, the behavior of MTB strains that were deleted for either one or three rpf genes were investigated in vitro and in vivo and we found that these mutants did not show significant growth attenuation in vitro although the triple mutants were significantly and differentially attenuated in B6 mice . Similarly, double mutants used in this study grew as well as the wild type in Sauton's medium either in broth culture or on plates (data not shown). These strains have also been checked for their ability to resuscitate from a "non-culturable" state after prolonged incubation in stationary phase under anaerobic conditions . Strains deleted for one rpf gene only showed no defects in their ability to resuscitate however, the triple mutants were significantly defective for resuscitation in this model. This defect for the triple mutants could be restored by addition of culture filtrate (taken from log phase MTB cells) to the resuscitation medium [B. Kana et al., in preparation] indicating that the inability of bacteria to spontaneously resuscitate was not only due to poor viability of the mutant strains in this model. We tested KDD6 for its ability to recover from a non-culturable state in vitro and we found no defects (data not shown).
In this study we show that one double (KDD7) and both triple rpf deletion mutants reveal significant attenuation in a chronic in vivo model after IP infection. Moreover, deletion of three rpf genes resulted in complete arrest of cell multiplication after immune suppression in the case of KDT8. In our model KDD7 seems to be more attenuated than KDT8 (which was derived from KDD7) at 90 days (3.5 log CFUs for KDT8 vs 2.9 log for KDD7) (Figures 3B and 4B). However, this difference is reduced as the experiment progresses and could also be associated with some natural variation in mice sensitivity in different groups. Deletion of the rpfA, rpfC and rpfD genes (KDT9) resulted in greater attenuation of MTB in the IP model than deletion of the rpfA, rpfC and rpfB genes (KDT8) (Figs 4A,B), in contrast to previously published data that indicated KDT8 was more attenuated for growth in B6 mice than KDT9 . We speculate that this discrepancy could be related to the different models used in the published experiment (intravenous, acute infection, B6 mice) compared to the current study. It is possible that the Rpf proteins could be differentially important for MTB proliferation depending on the route of bacillary dissemination. We are currently attempting to genetically complement the multiple rpf deletion strains and further pair wise experiments with complemented derivatives should confirm unequivocally the role of Rpfs in pathogenesis. However, our data are reflective of the notion that Rpf proteins may have a significant physiological role under conditions of bacterial stress in vivo. Indeed, actively grown mycobacteria were not sensitive to Rpf  whilst aged cultures or "non-culturable" cells were sensitive to Rpf addition in form of exogenously administrated protein or the plasmid coding for Rpf synthesis [18, 31]. We also found that bacterial growth from a small inoculum in non-optimum medium was sensitive to Rpf administration . Chronic infection (low MTB initial dose) could be considered as less permissive for mycobacterial multiplication in comparison with the acute infection phase and therefore, the role of Rpf proteins in the development of chronic TB infection and reactivation become more prominent and it is clear that different Rpfs contribute differently to the observed effect. It seems that RpfC is important for bacterial dissemination in the IP model in double knockouts while RpfD plays a more significant role in triple mutant combinations. At the same time RpfB is important for AG induced reactivation for the double mutants which is in accordance with the observations of Tufariello et al. for a single rpfB mutant .
The observed behavior of the different rpf mutants in vivo elucidates a sophisticated interplay between different Rpf proteins during cell growth and cell recovery after dormancy. Indeed, being almost redundant in single mutants, Rpf proteins exhibited less redundancy in double and triple mutants. We speculate that the different Rpfs perform similar functions in the cell (e.g. cell wall processing ) but at different loci of the target or with different efficiencies. This is supported by the fact that all Rpf molecules from MTB show differences in their variable domain which is probably important for cell wall binding . In multiple mutants it is likely, that some of the remaining molecules could substitute for the deleted Rpf better than others and the probability of such successful substitution decreases in the order of single-double-triple mutants. Alternatively, expression of a particular rpf gene could influence (directly or indirectly) the expression of the others. Future studies should elucidate these possibilities. Despite the fact that the mechanisms of Rpf activity are not fully understood it is clear that these proteins represent interesting targets for the design of chemical compounds against development of chronic\ latent TB infection. If dormant MTB cells sustain some metabolic activity, such anti-Rpf compounds could not only stop reactivation disease but eventually kill latent organisms.