In a host gene expression survey comparing zebrafish larvae infected with wildtype Mm (WT) or RD1-deleted Mm (ΔRD1) (5, 6), we identified matrix metalloproteinase 9 (mmp9) and tissue inhibitor of metalloproteinase 2b (timp2b) as being RD1-induced during granuloma formation at 5 days post infection (5dpi) (Fig. 1A and 1B; tables S1–4; fig. S1A and S1B). To control for ΔRD1’s attenuated infection at 5 dpi (5), we confirmed RD1-dependent gene induction using higher ΔRD1 inoculations that produced similar bacterial burdens at 5 dpi with the expected paucity of ΔRD1 granulomas (5, 6) (Fig. 1A and 1C; fig. S1C). At 1 dpi, only mmp9 was induced, suggesting that timp2b induction at 5 dpi was a compensatory response to increased mmp9 (Fig. 1D). Mmp9 is a gelatinase and gelatin zymography confirmed that RD1-dependent mmp9 mRNA expression resulted in increased Mmp9 gelatinase activity (Fig. 1E). In contrast, mRNA expression and activity of another gelatinase Mmp2 were not altered by infection (fig. S1A and Fig. 1E).

MMP9 is implicated in the pathogenesis of several inflammatory conditions (7, 8) and is highly expressed in human tuberculosis as well as in the mouse model of tuberculosis (9–12)(table S1). In mice, MMP9 activity correlates to increased macrophage migration and granuloma formation; however it is reported to be a host resistance factor, perhaps because its expression is associated with variable effects on infection in different genetic backgrounds (10). In humans, MMP9 clearly mediates susceptibility as its increased activity is correlated with worse outcomes (9). To test mmp9’s role in promoting granuloma formation and virulence, we knocked down its expression transiently with three modified antisense oligonucleotides (morpholinos) (1, 13) (fig. S2). The morpholinos, singly or in combination, reduced gelatinase activity reliably up to 4 dpi with activity returning to control levels by 5 dpi (fig. S2). WT infection of morpholino-injected embryos (morphants) resulted in attenuated infection sharing several features of ΔRD1 infection of control embryos. First, morphants displayed reduced bacteria and granulomas, as well as increased host survival (Fig. 2A-C). Second, kinetic analyses of granuloma formation in the morphants confirmed a specific granuloma-forming deficit (Fig. 2D-F). We found a dynamic link between Mmp9 activity, granuloma formation and bacterial expansion: bacterial burdens and granuloma formation differed only up to 4 dpi, returning to control levels by 5dpi contemporaneous with restoration of Mmp9 activity (Fig. 2A-E; fig. S2). Finally, while the RD1 locus promotes macrophage recruitment to nascent granulomas, it is not required for initial phagocyte migration to infecting bacteria when they are still extracellular (5, 6). Similarly, mmp9 morphants displayed normal macrophage migration to extracellular bacteria when injected into the hindbrain ventricle (fig. S3).

RD1 probably contributes to granuloma expansion through pleiotropic effects including inducing apoptosis of infected macrophages and recruitment of new uninfected macrophages (5, 6, 14, 15). In contrast, Mmp9 was not required for RD1-induced cell death; morphant and control granulomas in WT infection contained similar numbers of TUNEL-positive cells, whereas control granulomas in ΔRD1 infection exhibited the expected reduction (13) (Fig. 2G). Thus RD1-induced apoptosis is Mmp9-independent and cannot mediate bacterial expansion in the absence of new macrophage recruitment, and Mmp9-mediated acceleration of macrophage recruitment to granulomas is an independent mediator of pathogenesis.

Multiple cell types express MMP9 in many inflammatory conditions (7). In the context of tuberculosis, it is induced in cultured Mtb-infected monocytes (9, 16–18) and in epithelial cells (19). In advanced human tuberculosis, induced expression is reported in some monocytes and multinucleated giant cells abutting necrotic centers of lymph node granulomas (17, 20) and in epithelial cells proximal to lung granulomas (19). To understand how the RD1-Mmp9 axis mediates granuloma formation we assessed localization of mmp9 expression during this process. Fluorescence whole mount in situ hybridization (FISH) (1) revealed RD1-dependent mmp9 induction in 5 dpi embryos in cells associated with granulomas as well as in distal single cells (Fig. 3A). Multiplex FISH combining the mmp9 and macrophage-specific fms probes, or mmp9 and neutrophil-specific mpo probes (1), showed that the distal single cells consisted largely of neutrophils with a minor macrophage contribution (fig. S4 and S5). However, mmp9 expression by neutrophils and macrophages was unlikely to be relevant for granuloma formation because their mmp9 expression induced by infection was RD1-independent, and most granulomas contained very few, if any, of these cells (Fig. 3B, fig. S4 and S5).

Differential Interference Contrast (DIC) and confocal microscopy revealed that granuloma-associated mmp9 expression was localized to epithelial cells proximate to infected macrophages (21) (Fig. 3B, movie S1). Expression was restricted to specific epithelial cell types: epidermal cells adjacent to the granuloma expressed mmp9 whereas immediately overlying peridermal cells did not (21) (movie S1). Epidermal cell-specific expression was highlighted in granulomas forming in muscle where mmp9 was expressed not by the immediately surrounding myocytes but by their closest epidermal neighbors (Fig. 3C, movie S2). Every granuloma analyzed had proximate mmp9-expressing epithelial cells (n=35 granulomas in 9 fish), including the smallest identifiable macrophage aggregates (fig. S6 and movie S3). Thus Mmp9 induction is critical for granuloma formation from the very earliest stages and probably in later stages as well, given RD1’s continued influence on granuloma structure in chronic tuberculous infection (5, 22).

Bacteria residing in macrophages could induce epithelial cell mmp9 in two ways: (i) RD1 might induce macrophage signals such as secreted cytokines (23, 24) that in turn elicit mmp9 secretion by epithelial cells, or (ii) bacteria (25, 26) or bacterial products (27) released from macrophages might interact directly with epithelial cells. To distinguish between these mechanisms, we assessed mmp9 induction in pu.1 morphants that lack macrophages and in which infection results in extracellular mycobacterial growth (1). pu.1 morphants exhibited RD1-dependent mmp9 induction, suggesting that bacteria or their products interact directly with epithelial cells to induce mmp9 by a macrophage-independent mechanism (Fig. 4A).

The observation that uninfected epithelial cell mmp9 induction can occur distant from infection foci (Fig. 3C and movie S3) implicated an RD1-dependent secreted determinant rather than direct bacterial contact with epithelial cells. Indeed, injection of WT but not ΔRD1 bacterial supernatants rapidly induced mmp9 expression (Fig. 4B). The ESX-1 secretion system secretes five proteins that are all mutually co-dependent for secretion, so distinguishing their individual roles in virulence has been difficult (14, 15). We pursued ESAT-6 as the lead candidate for inducing mmp9 for two reasons: ESAT-6 mediates virulence independent of secretion (15) and its pore-forming activity (28, 29) could allow it direct access to epithelial cells. Injection of 4.8×10⁻¹⁷ moles of purified ESAT-6 was sufficient to induce mmp9 within 4 hours (Fig. 4C). In contrast, 5.0×10⁻¹⁷ moles CFP-10, thought to bind ESAT-6 and serve as its chaperone (15), failed to induce mmp9 significantly (Fig. 4C). Moreover, co-injection of CFP-10 and ESAT-6 did not augment the induction observed with ESAT-6 alone, confirming an ESAT-6-specific effect (Fig. 4C). Finally, similar to RD1-competent bacteria (Fig. 4A), ESAT-6 induced mmp9 in pu.1 morphants (Fig. 4D), consistent with a direct interaction with epithelial cells. We next asked if epithelial cell mmp9 induction was dependent on MyD88 and TNF signaling, as each can enhance mycobacterial induction of mmp9 in cultured cells under certain conditions (18, 19). ESAT-6 induced mmp9 in myd88 and tnf-receptor 1(tr1) morphants (Fig. 4D) suggesting a novel pathway for this epithelial cell-specific interaction. Moreover, TNF-independent induction of mmp9 is consistent with the finding that TNF does not mediate granuloma formation either in the presence or absence of bacterial RD1 (13).

Thus ESAT-6 functions in virulence by promoting granuloma formation via interaction with epithelial cells, previously regarded as bystanders in the pathogenesis of tuberculosis (fig. S7). The co-option of epithelial cells may offer mycobacteria a means of amplifying MMP9 secretion in the vicinity of a single infected macrophage to establish the granuloma niche. In addition, the differential induction of inflammatory programs in macrophages and epithelial cells may generate a hospitable growth niche in macrophages while harnessing epithelial cells to facilitate the chemotaxis of additional macrophages for niche expansion (6) (fig. S7). Our work provides a mechanistic explanation for the implication of MMP9 in human susceptibility to tuberculosis (9, 11, 12) and suggests targeted inhibition of its expression as a host-directed antituberculous therapy. Because, increased MMP9 is detrimental in both tuberculosis and a variety of noninfectious inflammatory conditions (7), interception of this pathway may have broad utility in treating a variety of inflammatory conditions in addition to tuberculosis.