The Dynamics of Bacterial Death and Replication during Extended Macrophage Infection
The establishment of a productive infection in the macrophage is not dependent merely on the short-term response following phagocytosis but the subsequent stages of adaptation that lead to a net increase in intracellular Mtb. To appreciate the adaptations required by Mtb to enter this replicative state one has to employ a temporal approach that accurately correlates bacterial numbers, including the assessment of death versus growth, with changing transcriptional profiles. We employed three different methods to probe the growth states of Mtb during a 14-day intracellular infection model.
First, resting primary M
isolated from C57BL/6 mice were infected with Mtb
at a low multiplicity of infection (MOI) of 1
1. At two-day intervals, infected monolayers were lysed and dilutions plated to determine viable colony forming units (CFU). Supernatants were also diluted and plated (prior to lysis of M
) to monitor extracellular bacilli that could contribute to reinfection. An initial 0.5 log decrease in cell-associated viable CFU at day 2 was followed by a period of minimal change (days 2–6) and then a more steady increase in CFU (days 6–14) (). This profile, consistent with our previous observations 
, suggested an early killing phase followed by delayed adaptation of surviving organisms to survive and replicate within M
phagosomes. At every time point, extracellular Mtb
totaled less than 5% of the intracellular burden (data not shown) suggesting that re-infection is minimal.
Life and death dynamics during long-term intracellular survival of Mtb.
The intramacrophage growth of Mtb
was also monitored by parallel transmission electron microscopy (TEM) analysis of fixed infected M
samples (). Enumeration of both morphologically normal and damaged Mtb
in 100 randomly chosen macrophages revealed a rapid increase in the number of detectable bacteria per cell from 1.5 Mtb
/cell at 2 hr post-infection (p.i.) to 10.3 Mtb
/cell by day 4 (~1.7-fold increase/day). From day 4 to day 14, the average bacterial burden increased at a notably slower rate (~0.3-fold increase/day) (, Figure S1A
). Interestingly, this early period of rapid Mtb
replication following M
invasion coincided with the marked decrease in viable CFU (). This indicated that early in the infection bacterial replication is rapid, but is countered by effective bacterial killing by the M
To further validate this conclusion, we calculated Mtb
growth and death rates during long-term infection of resting M
based on the loss of an unstable replication clock plasmid 
. The proportion of intracellular Mtb
retaining the clock plasmid pBP10 was determined by plating bacteria on media with or without 25 µg/ml kanamycin. The profile of pBP10 plasmid loss (, red) indicated an initial period of rapid replication (day 0–day 2) followed by an extended phase of much slower cell division (day 2–day 9). The cumulative bacterial burden (CBB) – the total number of Mtb
live, dead, or degraded within M
during the infection – was calculated based on the mathematical model developed by Gill et al.
. The predicted CBB (, black) at day 2 p.i., closely mirrored the quantitative TEM results enumerating all detectable bacteria regardless of viability, at ~12-fold higher than the number of viable CFU determined by plating. As shown in , this indicates substantial replication coinciding with even greater death rates during the early phase of infection. However, following the adoption of a slower growth rate from day 2 of infection, Mtb
exhibited a steady increase in viable CFU.
These data indicate that, following invasion of M
encounters a bottleneck during which the rate of bacterial killing outpaces its relatively rapid rate of replication. A similar early phase of pronounced bacterial killing was noted in vivo
by Gill et al.
during the first 14 days in an Mtb
-mouse model of infection 
. Following day 2, a period of apparent adaptation ensues extending to approximately day 6, during which both the rate of bacterial replication and the rate of killing decrease. The subsequent overall increase in viable CFU reflects further enhanced survival and the successful establishment of a productive infection. The shift in the balance between replication and death over time () suggests that the successful adaptation of some Mtb
cells to avoid killing by M
-derived effectors is at least as important as mechanisms for sustained replication within phagosomes. It is tempting to speculate that the slower growth rate of Mtb
at later time points may contribute to the shift in growth
death balance, perhaps by rendering Mtb
more resistant to M
-derived pressures. Studies modeling the dynamics of Mtb
-host interactions within human granulomas by Segovia-Juarez et al.
support this idea, showing that slow intracellular growth rates are correlative with Mtb survival 
. This novel insight into the life
death equilibrium of Mtb
during a sustained model of infection provides a physiological context for the interpretation of the global gene expression profiles discussed hereafter.
EM Analysis Reveals Considerable Heterogeneity in the Bacterial Population during Adaptation
Descriptions of host-pathogen interactions based on population-level data, such as microarrays, are often interpreted without appreciation of the heterogeneity within that population. With this in mind, we performed detailed TEM image analysis of Mtb
at 2 day intervals over 14 days simultaneously with survival assays () and microarrays (see below).
Consistent with previous observations 
, at 2 hours p.i. most single bacteria resided in a vacuole surrounded by a phagosomal membrane tightly apposed to the Mtb
cell wall (). While intracellular Mtb
proved to be quite effective at resisting fusion with lysosomes, identified following endocytic uptake of colloidal gold (), ~25% of single Mtb
did traffic to phagolysosomes (P-L) and colocalize with colloidal gold as early as 2 hr p.i. (). At 2 days p.i., image analysis of Mtb
revealed an increased frequency of morphologically-damaged Mtb
in large granular lysosomes (), consistent with the early phase of killing of replicating Mtb
indicated by the clock plasmid experiments. Whereas 99% of bacteria in the inoculum cultures were scored as intact, by day 2 only 75% of phagocytosed Mtb
appeared morphologically normal. We also noted heavily damaged Mtb
that appeared as hollowed out “ghosts”, but these were not scored as damaged. The discrepancy between the magnitude of killing, calculated based on clock plasmid data (), and the small proportion of visibly damaged bacteria as well as the constant ratio of normal to damaged bacilli (Figure S1A
) suggests a relatively efficient clearance of nonviable organisms. At later time points, the bacterial population continued to increase (, Figure S1B
) with up to ~150 Mtb
. In contrast to a recent report that a large subset of Mtb
H37Rv translocated into the cytosol of M
within 48 hrs 
, phagosomal membranes were definitively detected surrounding ~90% (88% at day 4 p.i., 86.5% at day 8 p.i.) of intracellular Mtb
CDC1551 throughout the duration of the infection (25 M
and >100 Mtb
examined per time point).
Electron microscopy analysis of long-term Mtb-macrophage interactions.
Mtb occupies heterogeneous intracellular niches.
There was a surprising degree of heterogeneity of compartments in which visibly intact Mtb
appeared to reside during long-term infections of M
, including “replicating” Mtb
in typical tight niches, small fused P-L, and putative double membrane-bound autophagosomes that colocalized with gold (, Figure S2
). The diversity of intra-host environments undoubtedly contributes to heterogeneity within the bacterial population.
Mtb Transcriptional Adaptation during Long-term Macrophage Infection
The transcriptional profile of Mtb
during the course of the 14 day infection should mirror the physiological states and transitions defined in the previous section. Therefore, overlaying the temporal transcriptional changes with these physiological states will generate a correlative linkage between the two datasets. We conducted microarray analysis of RNA isolated from intracellular Mtb
at 2 hr p.i. and at 2 day intervals over a 14 day period (GEO accession GSE35362). Fluorescent amplified RNA (aRNA) targets from each time-point were prepared from linearly amplified total Mtb
RNA and hybridized against a 2 hr “no M
” control as previously described 
. This enabled us to determine dynamic expression ratios over time relative to a single common denominator mRNA sample, in this case an uninfected control time-matched with the earliest time-point (2 hr). Semi-quantitative RT-PCR of select target genes (aprAB, hspX, bfrB, icl, groEL2, katG, whiB7
) was conducted to confirm temporal array profiles (
and data not shown). We identified 3626 genes with significant changes in expression during extended M
infection by combining both static statistical cutoffs (p<0.05 in at least one time-point) and EDGE (E
xtraction and analysis of D
xpression) methodology () 
. EDGE analysis identifies temporal changes in transcript levels that would not be deemed statistically significant at any single time-point by standard p-value measurements.
Dynamic Mtb transcriptome during long-term macrophage infection.
Transcriptional Regulatory Network Response to Phagosome Cues
In addition to analyzing the transcriptional profiles according to temporal dynamics and known functional themes we also conducted a systems-level analysis to characterize the behavior of the Mtb
Transcriptional Regulatory Network (TRN) underlying pathogen survival during the 14-day infection. This network-based approach incorporates extensive a priori
information on Mtb
gene regulation and network topology, combined with expression data, to assess network responses in surviving bacteria elicited by the M
We started by expanding a large-scale Mtb
TRN containing gene regulatory interactions extracted from both experimental and computational datasets 
. The previous TRN comprised 738 genes (18% of the genome) and 937 regulatory links obtained from three sources: (i
) literature mining; (ii
RegList database 
) inference from orthology with Escherichia coli
. To obtain the expanded network used in this study, we collected gene regulatory data from the following additional sources. (iv
) We added orthology-based interactions inferred from the closely related Corynebacterium glutamicum
available in the MycoRegNet database, which considerably expanded the TRN by adding 425 new interactions (Figure S9A
). The extensive overlap with regulatory links supported by experimental data validated these interactions identified from orthologous Transcription Factor (TF)-Target Gene (TG) pairs in the two organisms (p
, Fisher's exact test, Figure S9A
) Further, we expanded the TRN by adding 114 protein-DNA interactions discovered by a new bacterial one-hybrid reporter system termed TB1Hybrid 
, being that 31 interactions were exclusively identified by this method. (vi
) Finally, we performed operon-based network expansion, propagating a TF's regulatory effect to all members of the operon containing a given TG. The final expanded Mtb
TRN contained 1133 genes (28% of the genome) and 1801 regulatory links, more than a half of which were experimentally determined (Figure S9B
, C); the complete list of interactions is available in the Table S12
The global TRN provides a static summary of all possible regulatory interactions that mycobacteria may use when facing a broad spectrum of environments, ranging from normal to stressful conditions inside the M
. However, previous work has suggested that only parts of the network are utilized in specific conditions 
. Such parts (subnetworks) function as network modules regulated by a hierarchy of transcription factors in an environment-dependent fashion. To understand how the surviving subset of Mtb
bacilli specifically utilizes the TRN modules during prolonged intracellular infection, we analyzed the temporal response of the TRN by overlaying the 14-day M
infection time-course array data on the extended Mtb
We improved the earlier method called NetReSFun (Network Response to Step Functions), which identifies responsive TF-regulated subnetworks from time course microarray data 
. NetReSFun computes the Cov-score
(Methods) to quantify the expression change within the module (subnetwork) that we define as the total genes directly regulated by a given TF. A significant Cov-score
indicates subnetwork response, when the expression levels of the subnetwork's gene members are either down- or upregulated during consecutive time points (t
). Alternatively, simultaneous change of a TF's direct target genes may also be a surrogate of the TF's activity. Under this assumption, NetReSFun may recognize TFs that are “turned on” through posttranslational modifications such as phosphorylation and metabolite binding, but may or may not show increased expression levels themselves.
The temporal map of network responses () depicts specific TF-regulated subnetworks responsive during the time course, at a significance level of 0.05. The color scale indicates whether the overall trend of expression change within the subnetwork was positive (red) or negative (blue) at a given time interval. Intermediate colors denote subnetworks involving both up- and downregulated genes. The accompanying heatmap in indicates the source of regulatory links within the subnetwork (darker colors corresponding to higher fractions of links based on experimental evidence).
Temporal network response during macrophage infection.
Strikingly, the map reveals that the dynamic utilization of the TRN occurs in a defined pattern that can be mapped back to distinctive phases of intracellular growth. In the first 2 days p.i. - which corresponds to the stress phase in the growth curve of – we observed a high number of responsive TF-subnetworks (20 out of 83), mostly exhibiting increased expression of involved genes. Among these were DosR, HspR, KstR, members of the WhiB family (WhiB3, WhiB4), two-component response regulators (Rv0260c, Rv0818, RegX3), and alternative sigma factors (SigE, SigK, SigM). The sharp induction of a large number of subnetworks is reminiscent of the general Environmental Stress Response (ESR) in yeast 
and in Bacillus subtilis
In contrast, after ~6 days inside M
, we observed a reciprocal scenario where the TRN reflects a significant downmodulation of target genes, many of which had been induced immediately after invasion. This pattern is especially evident for a number of stress-responsive subnetworks that shift into downregulation during the slow growth phase, including RegX3, HspR and DosR. This “repressive” transcription phase indicates that the surviving bacteria have either adapted to stress, or they reside in a less hostile niche. For example, the SigH-controlled subnetwork displays a strong negative response only late in the time course (~8 days and onwards). As a global regulator, SigH modulates the transcription of SigE and SigB, as well as its own promoter. Although SigH is not required for growth in M
, mutants lacking sigH
caused reduced immunopathology and lethality in mice 
. Alternatively, it is possible that these regulatory changes are associated with the surviving bacilli reprogramming their physiology to assume the slow growth phenotype observed from day 4 onwards.
Importantly, we observed responsive subnetworks throughout the entire time-course, which indicates their importance for establishing productive infection. The presence of sustained responders such as HspR and DosR can have two possible implications. First, the opposite trends in the early phase of the infection (primarily upregulation) and later phases of infection (primarily downregulation) may indicate that the stress to which these modules respond initially is ameliorated at later time points. Alternatively, sustained responders may be necessary both to counteract initial phagosomal stress during the early phase of infection as well as for driving the persistor phenotype encountered in later phases of infection. This is the case of DosR, which is crucial for maintaining redox balance and energy levels during transitions into and out of dormancy-like conditions that perturb aerobic respiration, electron transport, or menaquinone pools 
. HspR, which activates a subset of the heat-shock general stress response upon M
, is also necessary in the persistent phase since ΔhspR
strains exhibited attenuated growth in the chronic infection 
Finally, TRN analyses revealed novel sustained responders that might be critical for Mtb
adaptation within the intracellular compartment. For example, the Rv2034-controlled subnetwork (inferred from C. glutamicum
orthology) contains multiple fadE
homologs implicated in β-oxidation of fatty acids and redox homeostasis. Notably, Rv2034 was recently characterized as an activator of the phoP
virulence regulator in mycobacteria 
, which makes this regulator an interesting candidate for follow-up studies.
By overlaying the genome-scale temporal expression data onto the TRN, we revealed the activation of additional regulons during macrophage survival not readily apparent in our supervised analyses. Thus, by leveraging the behavior of multiple TG as a readout of TF activity, TRN analysis of time-course microarray data further enhances the ability to detect adaptive changes in Mtb gene expression during productive infection of macrophages.
The process of infection is extremely dynamic as both host and pathogen seek to respond to the stimuli that they sense at their interface. In the current study we applied multiple analytical tools to establish a link between the transcriptional responses and physiological states through which Mtb transitions on its way to the establishment of a productive infection in its host macrophage. This analysis revealed several unexpected findings. Firstly, the initial phase of infection is marked by rapid bacterial replication coupled with effective bacterial killing by the macrophage. This is a period of marked stress for Mtb, which is illustrated by the greatest transcriptional response with respect to both up-regulated and down-regulated genes. Subsequently, the rate of replication slows and the bacterial number appears constant or at equilibrium, during which period the expression of many genes returns closer to control levels, whilst the divergent level of expression of others is sustained. Finally, the bacterial numbers start to increase indicating that the rate of replication exceeds that of death. At this time there is a marked down-regulation of many of the genes linked to general stress supporting the contention that Mtb has entered into a productive phase of infection characterized by enhanced intracellular survival. This is an important functional framework on which to hang the transcriptional profiles to determine which responses and/or metabolic themes impact which phase of infection.
Our examination of dynamic alternations of gene expression across the Mtb TRN also highlights stress responses and survival mechanisms deployed during distinct phases of host interaction. Among other things, our results indicate adaptive changes in lipid and energy metabolism akin to those observed in various dormancy models. Based on this, it is tempting to speculate that early infection of resting macrophages may serve to prepare Mtb for conditions encountered within granulomas after the onset of the adaptive immune response. Detailed understanding of the sensory and regulatory pathways required for Mtb virulence remains rudimentary at best. This point is illustrated by the large number of “genes of unknown function” which are actively regulated during intracellular survival.
In the discussion
, we have presented the observed transcriptomic changes as the result of all Mtb
cells sensing and responding in concert to phagosomal cues in the intracellular environment. Clearly, the microarray profiles in this study capture an average behavior over time of a population of intracellular bacilli that, as we have shown, exist in distinct vacuolar niches and presumably metabolic states. It could be argued that the changes in gene expression represent the minority of Mtb
that fail to block P-L fusion mounting dramatic stress responses. However, our data appear to be inconsistent with this view. For example, we know that Mtb
in P-L encounter Fe-deplete conditions whereas the average behavior of Fe-responsive signature genes reflects an Fe-replete environment (data not shown). Alternatively, the apparent up- and down-regulation may be due to the selective enrichment of pre-existing phenotypic variant cells with randomly upregulated stress response subnetworks, similar to the cell-to-cell heterogeneity reported by Aldridge et al.
. The rapid killing observed immediately after macrophage entry supports this scenario, suggesting that a subpopulation of bacilli capable of surviving the bottleneck may be present prior to infection, due to random phenotypic variation within the microbial population. However, the ability to block the induction of an acid regulon following M
invasion by chemical manipulation of phagosomal pH 
would suggest that Mtb
are altering gene expression in response to host-derived cues.
Thus, our data suggest that the dominant transcriptional profiles highlighted here represent the adaptive responses of the majority population. Analysis at the single-cell level will be required to explore the strategies employed by Mtb
to survive across diverse host environments experienced by each individual bacterium. To address this, we have begun to exploit fluorescent reporter strains responsive to specific environmental cues to gain a high-resolution view of Mtb
intracellular adaptation and cell biology. Interestingly, the expression of an acid-inducible locus required for normal intracellular survival, aprABC
, is expressed at distinct levels by individual bacilli within the same host cell 
. The impact of this type of heterogeneity on Mtb
-host interactions and pathogenesis has yet to be determined. Previous work has shown that stochastic variation (intrinsic to individual cells) can aid survival in stressful environmental conditions 
. However, a similar role of extrinsic, environmental variation is yet to be established.
The identification of mutants attenuated for intracellular survival is a popular and powerful tool for defining in vivo
survival mechanisms, however, there are some limitations to this methodology that can be addressed by transcriptional profiling. Array-based mutant screens such as Transposon Site Hybridization (TraSH) 
are in essence end-point assays that are not readily amenable to quantitative kinetic analyses. In addition, there are several well-characterized examples of genes known to be required for M
survival, such as pckA
, or phoPR
, that have not been identified by TraSH screens indicating that the method is not comprehensive 
. The TraSH methodology may preferentially identify mutants with severe survival defects, while being less effective at isolating mutants with less extreme phenotypes, such as ΔphoPR
, that resist killing but fail to grow within phagosomes 
. Finally, mutant screens are limited by their inability to query the in vivo
role of genes that are essential in vitro
or identify genes whose phenotype upon inactivation is masked by compensatory changes in gene expression. Future studies would benefit from the coordinated application of these two distinct but complementary approaches to identify genes contributing to the pathogenesis of Mtb
We feel that the significance of this current study is that it transforms transcriptional profiling from a purely descriptive analysis to the generation of a predictive discovery tool that can be used to identify genes, and therefore metabolic pathways and physiological states, that are required to support distinct phases in the intracellular life cycle of M. tuberculosis.