Mycobacterium tuberculosis exploits the PPM1A signaling pathway to block host macrophage apoptosis

The ability to suppress host macrophage apoptosis is essential for M. tuberculosis (Mtb) to replicate intracellularly while protecting it from antibiotic treatment. We recently described that Mtb infection upregulated expression of the host phosphatase PPM1A, which impairs the antibacterial response of macrophages. Here we establish PPM1A as a checkpoint target used by Mtb to suppress macrophage apoptosis. Overproduction of PPM1A suppressed apoptosis of Mtb-infected macrophages by a mechanism that involves inactivation of the c-Jun N-terminal kinase (JNK). Targeted depletion of PPM1A by shRNA or inhibition of PPM1A activity by sanguinarine restored JNK activation, resulting in increased apoptosis of Mtb-infected macrophages. We also demonstrate that activation of JNK by subtoxic concentrations of anisomycin induced selective apoptotic killing of Mtb-infected human macrophages, which was completely blocked in the presence of a specific JNK inhibitor. Finally, selective killing of Mtb-infected macrophages and subsequent bacterial release enabled rifampicin to effectively kill Mtb at concentrations that were insufficient to act against intracellular Mtb, providing proof of principle for the efficacy of a “release and kill” strategy. Taken together, these findings suggest that drug-induced selective apoptosis of Mtb-infected macrophages is achievable.

the upstream signals and molecular regulators that control these processes in the context of Mtb infection remain largely unknown.
We here demonstrate that the Protein Phosphatase, Mg 2+ /Mn 2+ -dependent 1A (PPM1A), which we recently identified as a key regulator of the innate antibacterial and antiviral response in macrophages 21 , is targeted by Mtb to prevent host macrophage apoptosis.
Host serine/threonine phosphatases are known to play important roles for regulation of cellular apoptosis 22,23 , and this has been extensively described as potential drug targets in the cancer field 24,25 . However, phosphatases have not received much attention in the context of infectious diseases or more specifically pathogen-mediated host cell apoptosis. Kinome analysis provided us with a basic understanding of the protein-protein interaction network governed by PPM1A and allowed us to identify pharmacologically addressable targets to bring proof of principle that therapeutic restoration of the ability of macrophages to undergo apoptosis in response to Mtb infection can be achieved. Beyond this, we demonstrate that selective killing of Mtb-infected macrophages increased the efficacy of the first-line anti-tuberculosis drug rifampicin, which has been reported to be less effective against intracellular Mtb 26,27 due to inefficient penetration into cells 28,29 . As such, a "release and kill" strategy to deprive the replicative niche of Mtb by inducing Mtb-infected macrophage cell death would be a means to more efficiently expose the bacteria to already existing Mtb drugs, thereby shortening the currently long treatment times.

Results
PPM1A inhibits the macrophage apoptosis pathways. Apoptosis regulation is a critical component of the antibacterial response that has clear implications on pathogen clearance, stimulation of cell mediated immunity, and ultimately disease progression 2,3,5,6 . Kinome analysis of persistently Mtb-infected THP-1 macrophages that had guided our previous research on the effect of PPM1A expression on the innate antibacterial response of macrophages 21 , also suggested a possible link of Mtb-induced upregulation of PPM1A with the downregulation of a number of proteins with functions in the apoptosis pathways 21 . Indeed, there is precedence that phosphatases play important roles in the regulation of cell death 22,23,30 .
Cellular apoptosis can be induced by intrinsic (cell stress events leading to collapse of mitochondria membrane potential) and extrinsic (death ligand mediated) signals, leading to the eventual cleavage and activation of caspase 3, the executioner caspase 31 . As THP-1 cells are the most commonly used human monocyte/macrophage model, including frequent use for Mtb infection experiments 32,33 , we addressed the question whether upregulation of PPM1A, as observed in Mtb infection 21 , could prevent macrophage apoptosis through either of these apoptosis pathways using genetically manipulated THP-1 cells. To induce the intrinsic apoptotic death pathway, we used etoposide, a topoisomerase II inhibitor 34 , and ionomycin, a calcium ionophore 35 . A single addition of etoposide at 0.3 μ M induced 30% apoptosis in THP-1 cells after 48 h, but only 13% in THP-1 cells overexpressing PPM1A (THP-PPM1A), as measured by Annexin V assays (Fig. 1A). Etoposide-induced apoptosis levels increased as a function of time. After 96 h, etoposide induced apoptosis in > 60% of the THP-1 cells, but THP-PPM1A cells remained protected (< 25% apoptotic cells) (Fig. 1B). A single ionomycin treatment at 10 μ M for 24 h induced 20% apoptosis in THP-1 cells, but only 11% in THP-PPM1A cells (Fig. 1A). As Annexin V staining is mostly suitable for the detection of early apoptotic events, we also stained for active caspase-3 as an alternative indicator for apoptosis using the Fluorochrome Inhibitor of Caspases (FLICA) method 36 . When apoptosis levels were measured by FLICA caspase-3 assays, cells stimulated with 1 μ M etoposide for 24 h showed that 66% of THP-1 cells stained positive for active caspase-3, while only 26% of THP-PPM1A cells stained positive for active caspase-3 ( Fig. 1C), confirming the previous results.
We found PPM1A to also be involved in the control of the extrinsic apoptosis pathway. TNFα and Fas Ligand (FasL) are two death ligands that induce apoptosis through the extrinsic pathway. A single addition of TNFα (100 ng/ml) or FasL (1 μ g/ml) for 48 h induced apoptosis in ~25% of THP-1 cells, but in less than 15% of THP-PPM1A cells (Fig. 1D). Again, in TNFα -treated THP-1 cells, we observed a time-dependent increase in the frequency of apoptotic events to > 60%, whereas THP-PPM1A cells were mostly apoptosis-resistant with the percentage of apoptotic cells increasing to only ~20% (Fig. 1E). Thus, PPM1A appears to play an important role in the control of macrophage apoptosis as its expression directly inhibited both intrinsic and extrinsic apoptotic pathways.
PPM1A expression inhibits apoptosis of Mtb-infected macrophages. We next examined whether PPM1A plays a role during Mtb infection to control macrophage apoptosis. As it has been previously shown that a higher multiplicity of infection (MOI) induces macrophage apoptosis 37 , we infected THP-PPM1A cells with Mtb at an MOI of 20. In this scenario, PPM1A overexpressing cells retained higher cell viability relative to parental THP-1 cells (57% vs. 16%), as assessed by flow cytometry on day 2 post infection ( Fig. 2A). These results show that increased PPM1A expression levels have a clear impact on the survival of THP-1 cells during Mtb infection.
To detail this finding, we next addressed whether the observed Mtb-induced cell death was indeed caused by apoptosis. THP-1 or THP-PPM1A cells were infected with Mtb at an MOI of 20 and cells were stained for Annexin V at 2, 5, and 7 days post-infection. Over this period of time, apoptosis levels, as determined by Annexin V stains, increased continuously until 40% of THP-1 cells exhibited an apoptotic phenotype on day 7 post-infection (Fig. 2B). In contrast, only ~ 25% of THP-PPM1A cells underwent apoptosis during the same period (Fig. 2B). FLICA caspase-3 staining showed a similar reduction in apoptosis of Mtb-infected THP-PPM1A cells in comparison to parental THP-1 cells at the same time points post infection (Fig. 2C). These data show that increased PPM1A levels suppress apoptosis of Mtb-infected macrophages. While PPM1A has a major impact on macrophage apoptosis, it does not completely abrogate the apoptotic response, which would be consistent with reports for the involvement of other host-derived or bacterial factors that act outside of the PPM1A signaling axis 14 Measuring apoptosis over time by single cell staining techniques (FLICA, Annexin V) does not provide information on accumulative effects, as cells that underwent apoptosis at earlier time points disintegrate and are not accounted for at later time points. Since Mtb infection can also induce macrophage necrosis 11,18 , which could result in false positive Annexin V signals or the loss of active caspase-3 signal, both arising from plasma membrane destruction and eventual cellular breakdown, we wanted to confirm our findings with another method for the detection of apoptosis. We thus utilized the Apo-ONE assay 39 , which measures caspase-3/7 activity in a homogenous format following lysis of cells without the need to remove culture media that may already contain released caspase-3/7. The results using this method confirmed that PPM1A overexpression indeed inhibited Mtb-induced THP-1 cell apoptosis, as indicated by 4-fold reduction in active caspase-3/7 in the samples from THP-PPM1A cells on day 2 post-infection (Fig. 2D). These data collectively show that elevated PPM1A levels, as observed during persistent Mtb infections 21 , inhibit macrophage apoptosis.
PPM1A controls macrophage apoptosis through JNK activation. Our results clearly demonstrate that PPM1A controls both the extrinsic and the intrinsic apoptotic pathways of macrophages. However, PPM1A has not been previously linked to macrophage apoptosis, which suggests that PPM1A must control downstream effectors that directly mediate the apoptotic response. Given the diversity of proteins that have been reported to be involved in regulation of apoptosis, we used kinome analysis to identify a possible link between PPM1A and proteins known to be involved in the control of apoptosis pathways. The utilized Kinexus antibody array provide information on changes in the expression or phosphorylation state of 309 protein kinases, 38 protein phosphatases, 37 stress response proteins, 24 transcription factors, and 109 proteins involved in other signaling pathways.
Kinome analysis of THP-PPM1A cells in comparison to THP-1 cells revealed a total of 131 statistically relevant signals, which indicate up-or down-regulated protein expression or phosphorylation events. Of these, 78 (59%; p = 1.06e-35) proteins were associated with the regulation of apoptotic processes as indicated by Gene Ontology (GO) process analysis (Table 1). Using the proteins identified from this analysis as seed nodes, we generated a protein-protein interaction network (PIN) using Metacore and identified several nodes that could potentially function as downstream signaling effectors of PPM1A (Supplemental Fig. 1). This analysis identified several highly linked nodes in the JNK-AP1 pathway (c-Jun N-terminal kinase) (Supplemental Fig. 1). As JNK had been previously reported to be a substrate of PPM1A activity 40 and is known to play a central role in the control of the cellular extrinsic and intrinsic apoptosis pathways 41 , it made for a particularly interesting candidate that could act as a mediator of PPM1A signaling into the apoptotic pathway.
Using a multiplex bead assay with specific antibodies to total or phosphorylated JNK (Thr183/Tyr185), our data show similar baseline protein levels of JNK and no significant activation (phosphorylation) in either resting THP-1 or THP-PPM1A cells (Fig. 3A). Treatment of THP-1 cells with anisomycin, a potent JNK activator 42 , induced activation (phosphorylation) of JNK, whereas JNK activation was completely abrogated in THP-PPM1A cells (Fig. 3B). Importantly, following Mtb infection of THP-1 and THP-PPM1A cells, while total JNK protein levels remained similar, Mtb infection induced JNK activation (phosphorylation) in THP-1 cells, but not in THP-PPM1A cells (Fig. 3C). As PPM1A has also been reported to inactivate p38 MAPK 40 , we examined total and phosphorylated p38 levels in resting and Mtb-infected THP-1 macrophages. We found that p38 levels were unchanged at baseline and after Mtb-infection in THP-1 and THP-PPM1A macrophages (Supplemental Fig. 2), suggesting that p38 does not play a role in mediating cell death in our system. These data are consistent with the known role of JNK phosphorylation in pro-apoptotic signaling 41 , and thus link PPM1A to apoptosis control at the level of JNK phosphorylation.

Depletion of PPM1A promotes selective apoptosis of Mtb-infected macrophages. Since Mtb
infection induces elevated levels of PPM1A expression, which in turn inhibits macrophage apoptosis in response to pathogen invasion, inhibition of PPM1A signaling may be an attractive therapeutic strategy to selectively kill Mtb-infected macrophages by inducing infected macrophages to undergo apoptosis. To address this possibility, we generated PPM1A knockdown cells. Transduction of THP-1 cells with a PPM1A-specific shRNA vector  produced a cell population (THP-Δ PPM1A cells) in which PPM1A protein levels were reduced to ~10% of the protein levels in parental THP-1 cells (Fig. 4A). If PPM1A inhibitors, as hypothesized, would trigger selective apoptosis of Mtb-infected macrophages, we would expect Mtb infection to trigger increased apoptosis in THP-Δ PPM1A cells compared to parental THP-1 cells. To observe the effect of PPM1A depletion on apoptosis of Mtb-infected macrophages, we used an MOI of 5, a physiologically representative scenario where Mtb infection would naturally suppress host macrophage apoptosis. THP-1 and THP-Δ PPM1A cells were infected with Mtb and relative levels of apoptosis were measured by Annexin V, FLICA caspase-3 staining, and Apo-ONE assay. In line with our predictions, using three different methods, we indeed observed a 67% (Annexin V), 52% (FLICA) or 40% (Apo-ONE) increase in Mtb-infection induced apoptosis in THP-Δ PPM1A cells, relative to the parental THP-1 cells (Fig. 4B,C).
Our data would suggest that the underlying cause for this increased ability of Mtb-infected THP-Δ PPM1A cells to undergo apoptosis would be the restoration of the JNK pathway to respond to the incoming infection. Consistent with our data that PPM1A overexpression would abrogate anisomycin induced JNK activation (Fig. 3B), THP-Δ PPM1A cells stimulated with 1 μ M anisomycin showed a 2-fold increase in JNK phosphorylation compared to THP-1 cells (Fig. 4D), adding further evidence that PPM1A expression levels effectively control JNK activation in macrophages. Importantly, different from THP-PPM1A cells, where Mtb infection did not induce JNK activation (Fig. 3C), Mtb infection induced potent JNK activation in THP-Δ PPM1A cells (Fig. 4E).
To conclusively demonstrate that JNK functions in the PPM1A signaling pathway in the context of Mtb infection, we examined whether increased apoptosis associated with the absence of PPM1A (Fig. 4B,C) would actually depend on JNK activation. To this end, we treated Mtb-infected THP-Δ PPM1A macrophages with the specific JNK inhibitor SP600125, which has 300-fold selectivity over p38 and no effect on p38 phosphorylation at the concentration used in our experiement 43 . Consistent with our data showing increased JNK phosphorylation in Mtb-infected THP-Δ PPM1A macrophages, inhibition of JNK activity by SP600125 abrogated the ability of THP-Δ PPM1A macrophages to undergo apoptosis in response to Mtb infection (Fig. 4F), thereby providing further evidence for the involvement of JNK in the PPM1A-mediated block of macrophage apoptosis.
These data demonstrate that modulation of PPM1A levels controls apoptosis of Mtb-infected macrophages and suggest that PPM1A could be an attractive drug target to eliminate persistently Mtb-infected macrophages.

Inhibition of PPM1A activity increases JNK activation during Mtb infection. Given that genetic
manipulations of PPM1A expression showed that it controlled JNK activation as a means to inhibit apoptosis of Mtb-infected macrophages, we next sought to test whether PPM1A phosphatase activity was required to inactivate JNK signaling. This would also show whether pharmacological perturbations targeting PPM1A could restore proper JNK activation during Mtb infection.
Consistent with a previous report that showed the plant product sanguinarine as a potent and selective inhibitor of PPM1A activity 44 , we demonstrated that sanguinarine can block PPM1A phosphatase activity (Fig. 5A). Using sanguinarine as a chemical probe, we treated primary human monocyte derived macrophages (hMDMs) for 24 h, and then induced JNK activation with anisomycin. As we would predict, inhibition of PPM1A activity by sanguinarine increased anisomycin induced phosphorylation of JNK by ~2-fold (Fig. 5B). Importantly, sanguinarine treated hMDMs infected with Mtb showed ~45% increase in JNK activation, while levels of total JNK remained similar (Fig. 5C,D). These data show that PPM1A activity is essential for inactivation of JNK during Mtb infection and suggests that targeting the PPM1A-JNK signaling axis could restore the proper apoptotic response in Mtb-infected macrophages.

Anisomycin induces selective cell death of Mtb-infected primary human macrophages. While
we had shown that PPM1A in THP-1 cells and primary human macrophages were similarly upregulated in response to Mtb-infection, resulting in a severe impairment of the innate immune response 21 , we would have to verify that this similarity would extend to the ability of PPM1A to control apoptosis of primary macrophages in particular as a function of the identified PPM1A-JNK signaling axis. Differentiated primary macrophages are highly adherent, which requires aggressive manipulations to de-adhere the cells for downstream single cell analysis. Also, these methods are not conducive to study the kinetic effects of drugs on cell viability. To enable direct kinetic measurements of the effect of chemical research probes on the viability of the Mtb-infected primary macrophages, we used Real Time Cell Analysis (RTCA), a method that is frequently used to kinetically determine viability of adherent cells [45][46][47] . The RTCA system measures changes in impedance between electrodes at the bottom of E-well plates, which is then translated into a Cell Index (CI) measurement, a dimensionless value. An increase in the CI reflects an increase in macrophage adherence (differentiation), whereas a decrease in the CI reflects the loss of macrophage viability as they detach from the bottom of the well (Supplemental Fig. 3). Suspension cells trigger no measurable signal in this system.
To address the question whether interference with the PPM1A-JNK axis would provide a means to selectively kill Mtb-infected macrophages, we infected hMDM with Mtb, and 24 h post infection, treated the macrophages with or without anisomycin, and then used changes in the CI to kinetically quantify compound effects. Briefly, monocytes were seeded in 96 E-well plates and monitored for GM-CSF induced differentiation. As indicated by the increase in the cell index, primary human monocytes attached to the E-well plate over the first ~140 h (6 days) in response to GM-CSF and differentiated into hMDMs. The hMDMs were left untreated, were exposed to 3 μ m inert latex bead particles (phagocytosis control) or infected with Mtb. Wells for each condition were then mock treated or treated with anisomycin (75 nM) and macrophage viability (adherence) was followed kinetically by RTCA for the next 24 h to record changes in the CI.
While the CI remained stable for all control conditions (no anisomycin), and for pseudo-infected MDMs that had phagocytosed the inert latex beads, we observed a selective and rapid decrease in CI in anisomycin treated Mtb-infected hMDMs (Fig. 6A). The selective killing of Mtb-infected macrophages was indicated by the decrease in CI from 11 to 4, while the CI of untreated Mtb-infected macrophages remained above 10 (Fig. 6A). The RTCA experiment demonstrates that Mtb-infected primary macrophages were efficiently and selectively killed by the use of anisomycin as a JNK agonist. To unequivocally show that anisomycin induced selective killing of Mtb-infected macrophages is not mediated through a p38 dependent mechanism and that JNK activation is absolutely required in this process, we used a specific JNK inhibitor, SP600125 43 to probe the system. Using RTCA as described above, we show that the presence of SP600125 completely blocked the ability of anisomycin to induce selective apoptotic killing of Mtb-infected human macrophages (Fig. 6B). This result demonstrates that anisomycin induced selective cell death can be reversed by directly inhibiting JNK.
The RTCA data were confirmed by fluorescence microscopy analysis in an independent experiment. On day 6 post GM-CSF differentiation, macrophage differentiation was indicated by a strongly adherent, flattened, pancake-like morphology indicative of M1 polarization 48 , which was not altered by the addition of anisomycin   (Fig. 6C). Similarly, no influence of anisomycin on the viability or phenotype of hMDMs loaded with 3 μ m inert latex bead particles was observed (Fig. 6D). However, anisomycin treatment of Mtb-infected macrophages resulted in extensive cell death compared to Mtb-infected macrophages in the absence of anisomycin (Fig. 6E). While untreated Mtb-infected macrophages maintained the flattened, pancake-like morphology, anisomycin treated Mtb-infected macrophages exhibited features of cell shrinkage, detachment, and formation of apoptotic bodies with some extracellular bacteria, typical features of apoptotic cells (Fig. 6E).
Taken together, these data suggest that selective killing of Mtb-infected macrophages can be achieved by pharmacological perturbation and that targeting the PPM1A-JNK signaling network is a promising therapeutic strategy to induce selective elimination of persistently Mtb-infected macrophages.

Selective killing of Mtb-infected macrophages enhances antimicrobial chemotherapy to kill
Mtb. Apoptosis of infected macrophages stimulates the antibacterial response 5 and plays a significant role in promoting adaptive immunity 3 . Alternatively, given reports that several key TB drugs such as rifampicin 26,27 , streptomycin, and fluoroquinolones have decreased efficacy against intracellular Mtb likely due to penetration properties 29 , induction of apoptosis may also improve the ability of existing antimicrobial chemotherapy against Mtb.
To test this hypothesis, Mtb-infected THP-1 macrophages were treated with anisomycin to induce apoptosis-mediated bacterial release, and subsequently treated with rifampicin at a concentration below the minimal inhibitory concentration (MIC) necessary to kill intracellular Mtb.
Consistent with our data using primary human macrophages, we show that anisomycin treatment induces selective killing of Mtb-infected THP-1 macrophages (Fig. 7A). Subsequently, to measure Mtb viability, we used a well-characterized luciferase reporter system in Mtb 49,50 , which has been shown to correlate well with the standard colony unit formation plating method 51 . Our experiments show that while anisomycin or rifampicin at the used concentration had little to no effect on the killing of intracellular Mtb, combination treatment decreased the survival of Mtb by ~40% as measured by luciferase assays (Fig. 7B). This experiment suggests that selective induction of Mtb-infected macrophage apoptosis improves standard antimicrobial chemotherapy and should provide an attractive treatment strategy that has the potential to shorten the currently lengthy treatment period. To explore the possible effect of such a "release and kill" strategy on TB treatment kinetics, we used a simple mathematical model that assumes a biphasic decline of the bacterial load in patients following onset of chemotherapy (e.g. rifampicin treatment) (Supplemental Fig. 4). Based on a series of data published by other groups [52][53][54] , we assumed that during phase I (initial kill), therapy would eliminate extracellular Mtb and intracellular Mtb in macrophages located in tissues where sufficiently high drug concentrations can be easily achieved. During this phase, a reduction of the bacterial burden by ~70% would be achieved within a month of standard treatment. Phase II (slow kill) would be characterized by a much slower decline of the residual bacterial burden, which under standard treatment would be completely eliminated after ~6 months (24 weeks) of treatment  Fig. 4; black curve). This slow decline would be representative of the effect of drug treatment against intracellular Mtb, possibly in a persistent and therefore more therapy resistant form and in tissue locations that are poorly penetrated by drugs. The model provided us with the ability to predict various theoretical treatment scenarios, in which Mtb-release drugs that trigger selective apoptosis of Mtb-infected macrophages would be added as adjunctive therapy to the ongoing standard Mtb treatment regimen.
In Supplemental Fig. 4, several scenarios are visualized. In a scenario where release drugs are not added to the regimen during phase I or their addition would not increase phase I treatment efficacy, this model would predict that a weekly application schedule of Mtb-release drugs that would result in a 10%, 20%, or 40% reduction in the bacterial burden (based on what can be achieved in Fig. 7B) would lead to a reduction in treatment times from 24 weeks to 15, 12, or 9 weeks, respectively. If Mtb-release drugs were to improve phase I treatment efficacy by 10% or 20%, and subsequently reduce the Mtb burden by even only 10%/week during phase II, treatment times would be reduced to 13 or 9 weeks, respectively. While these are only possible predictions based on simple, but realistic assumptions, these calculations suggest that a release and kill strategy-based adjunctive therapy to the standard TB treatment regimen could lead to a major reduction in treatment time.

Discussion
Mtb is a prototypical intracellular bacterial pathogen that replicates efficiently in the phagosome of macrophages 55 . We recently demonstrated that to enable this process, Mtb alters the kinase and phosphatase signaling networks within the infected macrophage resulting in an environment conducive to the establishment of a persistent infection. We found that Mtb-induced upregulation of PPM1A controls several aspects of the macrophage innate antibacterial response 21 .
In the current study, we demonstrate that PPM1A is an apoptosis checkpoint that controls both the intrinsic and extrinsic apoptosis pathways (Fig. 1), a critical innate immune response with major implications for intracellular pathogens. Kinome analysis enabled the identification of the JNK/AP-1 signaling pathway as a downstream effector through which PPM1A exerts its control on the apoptotic pathway in macrophages (Supplemental Fig. 1 and Table 1), which was confirmed by a series of experiments using PPM1A overexpressing and knockdown cells (Figs 3 and 4). These experiments revealed that the ability to control PPM1A expression provided Mtb with a means to hijack macrophages by suppressing host cell apoptosis. Suppression of apoptosis, which is otherwise an innate defense mechanism against intracellular pathogens, provides Mtb with a replicative niche within which it can persist for long periods. Indeed, apoptosis of infected macrophages has been shown to facilitate intracellular bacterial killing, priming of cell mediated immunity, and limits unnecessary tissue inflammation 2,3,5-7 . Suppression of apoptosis is also essential for eventual Mtb dissemination, which is achieved by Mtb-induced host cell necrosis. PPM1A upregulation thus seems to be a key factor for Mtb to establish a persistent infection in macrophages, by hijacking the innate immune response of its host cells, and disabling the natural ability of macrophages to undergo apoptosis in response to pathogen infection.
Our finding that a host phosphatase (PPM1A) controls both the extrinsic and intrinsic apoptotic pathways, and more importantly inhibits apoptosis of Mtb-infected macrophages corroborates with research on other members of the PPM (monomeric, metal-dependent) family of serine/threonine phosphatases 56 , which have well described functions in the regulation of cell death 23,30,57 . In addition, it was recently demonstrated that the Mtb phosphatase PtpA could dephosphorylate host Glycogen Synthase Kinase-3 (GSK3α ), thereby inhibiting apoptosis of Mtb-infected macrophages 38 . Consistent with this, kinome analysis from persistently Mtb-infected THP-1 cells 21 and that of PPM1A overexpressing cells (Supplemental Fig. 1) both show dysregulation of GSK3. There Mtbinfected THP-1 macrophages (MOI 5) were treated with or without 100 nM anisomycin for 3 days. Then cells were treated with or without 0.1μ g/ml rifampicin for 1 day. The amount of viable Mtb in each condition was quantified using the luciferase activity assay. Using relative luminescence units, data were normalized to control conditions (without rifampicin) as 100% Mtb survival. Results in this figure represent the means ± S.D. of three independent experiments. **p < 0.01 relative to untreated macrophages.
is also evidence showing that Mtb PtpA exploits host ubiquitin to inactivate JNK and p38, thereby suppressing innate immunity 58 . Our finding that PPM1A controls the innate immune response 21 and apoptosis during Mtb infection may suggest the existence of redundant mechanisms (via host and bacterial phosphatases) used by Mtb to ensure survival of the host macrophage. Such functional redundancy would also explain why knockdown of PPM1A in our experiments induced apoptosis in a large proportion of, but not all Mtb-infected macrophages.
As the PPM1A-JNK signaling axis presented itself as an attractive target to restore the ability of Mtb-infected macrophages to undergo apoptosis, we also address the important question whether this signaling network could be targeted to selectively kill Mtb-infected macrophages. Such a therapeutic intervention would (i) disrupt the intracellular replication cycle of Mtb, (ii) boost the adaptive immune response to Mtb 8 , and (iii) increase the susceptibility of Mtb to existing drug treatments through induction of macrophage apoptosis given that common anti-tuberculosis drugs (e.g. rifampicin) have reduced efficacy against intracellular bacteria 26,27 .
Given that we had identified JNK as a key downstream signaling effector of PPM1A, it was logical to target JNK to deliver proof of concept that selective apoptosis induction of Mtb-infected macrophages can be achieved. Indeed, anisomycin, a reported JNK phosphorylation agonist could counteract the PPM1A upregulation enforced suppression of JNK phosphorylation following Mtb-infection, and restored the ability of macrophages to undergo apoptosis in response to Mtb infection (Fig. 6).
Drugs that could selectively induce apoptosis of Mtb-infected macrophages are highly desirable as this would restore a key innate immune response that is actively blocked by Mtb infection, which will have therapeutic benefits through the generation of an improved adaptive immune response and the ability to potentiate vaccines. Beyond this, drug-induced selective killing of Mtb-infected macrophages could likely increase the efficacy of existing anti-Mtb drugs as they gain better access to the bacteria. This approach would be particularly useful as adjunctive TB therapy since the commonly used anti-Mtb drugs rifampicin 26,27 , streptomycin, and fluoroquinolones having decreased efficacy against intracellular Mtb likely due to penetration properties 29 . Indeed, in a process we term "release and kill", our results show that anisomycin induced killing of Mtb-infected macrophages removes the protective niche of the bacteria and increases the efficacy of rifampicin to kill Mtb at concentrations that are normally inactive against intracellular Mtb.
To examine the possible impact of the addition of Mtb-release drugs to standard TB treatment regimens, a simple mathematic model that allowed for some basic predictions suggested that a shortening of the treatment course by 35-60% seemed rather realistic. As clinical anti-tuberculosis drug treatment, particularly in resource poor settings, is primarily hampered by long treatment periods lasting over 6 months 59 leading to patient non-adherence, any significant shortening of the treatment period would be highly desirable.
Generation of THP-ΔPPM1A cells. HEK 293T cells seeded at 50% confluency were transfected with lentiviral plasmids expressing shRNA targeting human PPM1A or scrambled control using FuGENE (Promega, Madison, WI). Culture supernatants were harvested after 48 h and 72 h, aliquoted, and stored at − 80 °C. The supernatants containing lentiviral particles were used to transduce THP-1 cells. Cells were selected by puromycin and analyzed by Western blot to verify knockdown of PPM1A protein levels.
Bacterial infection. M. tuberculosis mc 2 6206 growing in log-phase was quantified by optical density measurement at 600 nm using the conversion of 3 × 10 8 (Mtb) bacteria per ml for OD 1.0. The amount of bacteria required for various MOIs were washed and resuspended in RPMI 1640 cell culture media without antibiotics. For infection of THP-1 monocytes, the prepared bacteria were added to the THP-1 cells and resuspended together to ensure even exposure of the monocytes to bacteria. For infection of differentiated THP-1 macrophages and primary human MDMs, bacteria were added to the cells and incubated at 37 °C for 4 h. Extracellular, non-phagocytosed bacteria were removed by 3 washes and infection was continued 37 °C for the desired time. Apoptosis assays. Cells were stained with Annexin V conjugated to AlexaFluor 488 or APC (Life Technologies) according to manufacturer's protocol. FAM FLICA ™ (green) or FLICA ™ 660 (far red) caspase-3/7 staining kits were used to stain for active caspase-3/7 in cells, and were used according to manufacturer's protocol (ImmunoChemistry Technologies, Bloomington, MN). Apo-ONE homogenous caspase-3/7 detection kit was purchased from Promega, and used according to manufacturer's protocol.

Survival of
Flow cytometry. Cells following Annexin V or FLICA staining were analyzed by flow cytometric analysis (FCM). FCM analysis was performed on a Guava EasyCyte (Guava Technologies Inc., Billerica, MA). Data analysis was performed using Guava Express software (Guava Technologies Inc.) or FlowJo V10 software (Ashland, OR).
Milliplex assays. Cells following various treatments or Mtb infections were lysed according to Milliplex kit protocol (Millipore, Billerica, MA). Total and phosphorylated JNK (T183/Y185) or p38 (T180/Y182) protein levels were measured using separate 2-plex Milliplex kits. Experiments were performed according to manufacturer's protocol and read-out was performed using a Bio-Plex 200 instrument. Data analysis was performed using Bio-Plex manager 6.0 software (Bio-Rad, Hercules, CA).
PPM1A activity assay. The enzymatic activity of PPM1A was measured using the pNPP phosphatase assay kit (BioAssay Systems, Hayward, CA). In brief, 10 μ g/ml recombinant PPM1A (Abnova, Walnut, CA) was prepared in reaction buffer (25 mM HEPES, 100 mM NaCl, 2 mM MnCl 2 , pH 7.2) in the absence or presence of 0-30 μ M sanguinarine for 15 min at room temperature. The resulting reaction was mixed with pNPP substrate mix according to manufacturer's protocol and incubated at 37 °C for 15 min. The reaction was then stopped using the kit provided stop solution and absorbance was measured in a plate reader at 405 nm.
Real-time cell analyzer assay (RTCA). Macrophage adhesion was measured in specialized 96-well plates (E-plate 96) with the xCELLigence Real-time Cell Analyzer (RTCA) SP apparatus (ACEA Biosciences, San Diego, CA). Data was quantified by measuring impedance changes between the sensing electrodes located in the well-bottom, which changes as a function of the adhesion of cells to the surface of the plate. A dimensionless value, the Cell Index (CI), is representative of these impedance changes. Using this system, macrophage adhesion and therefore viability was monitored in real-time. Plates were removed at various time points for the addition of anisomycin or Mtb for infection. After each manipulation, plates were placed back in the RTCA apparatus for kinetic monitoring. The CI at every time point represents the mean of three independent measurements. A schematic description of this system is shown in Supplemental Fig. 3.
Western blot. Cells were harvested by centrifugation, washed once with PBS, and lysed in RIPA buffer (Cell Signaling) according to the manufacturer's instructions. Protein concentration of the lysates was determined by the bicinchoninic acid (BCA) method according to the manufacturer's recommendations (Thermo Scientific). About 10 to 20 μ g of protein per sample was separated on 10% Mini-Protean TGX gels (Bio-Rad) and subsequently transferred to a polyvinylidene difluoride (PVDF) membrane using an iBlot gel transfer system (Life Technologies). Western blot analysis was performed according to standard protocols. Total PPM1A, α -tubulin, or GAPDH proteins were detected with specific antibodies (see antibody section). A horseradish peroxidase-conjugated goat anti-rabbit or goat anti-mouse polyclonal antibody (Santa Cruz) was used as the secondary antibody. The blot was developed using the Western Lightning Ultra chemiluminescent substrate from Perkin Elmer, Inc., and detected in an EpiChemi3 darkroom (UVP BioImaging Systems).
Kinex ™ Antibody Microarray-based analysis. Kinex ™ antibody array analysis was performed as before 21 . Briefly, 50 μ g of lysate from each sample were covalently labeled with a proprietary fluorescent dye according to the manufacturer's instructions (Kinexus, Canada). After blocking non-specific binding sites on the array, an incubation chamber was mounted onto the microarray to permit the loading of one control (THP-1) and one experiment (THP-PPM1A) sample side-by-side on the same chip. Following sample incubation, unbound proteins were washed away. KAM-850 chips are spotted in duplicates with over 850 antibodies, including 517 pan-specific antibodies and 337 phosphosite-specific antibodies. By this means, the microarrays provide information about the expression and phosphorylation levels of these target proteins.
Each array produces a pair of 16-bit images captured with a Perkin-Elmer ScanArray Reader laser array scanner (Waltham, MA). Signal quantification was performed with ImaGene 8.0 from BioDiscovery (El Segundo, CA) with predetermined settings for spot segmentation and background correction. The background-corrected raw intensity data are logarithmically transformed with base 2. Z-scores are calculated by subtracting the overall average intensity of all spots within a sample from the raw intensity for each spot, and dividing it by the standard deviations (SD) of all of the measured intensities within each sample 61 . Z-ratios are further calculated by taking the difference between the averages of the observed protein Z-scores and dividing by the SD of all of the differences for that particular comparison. Calculated Z-ratios have the advantage that they can be used in multiple comparisons without further reference to the individual conditional standard deviations by which they were derived.
Statistical analysis. Data are expressed as the mean ± the standard deviation of three independent experiments. Statistical analysis was performed using the Student t test. Values of p < 0.05 were considered to be significant.