Contrasting roles of the innate receptors TREM2 versus Mincle in the recognition and response of macrophages to mycolic acid-containing lipids in mycobacterial cell walls


 Mycobacterial cell-wall glycolipids elicit an anti-mycobacterial immune response via FcRγ-associated C-type lectin receptors, including Mincle, and caspase-recruitment domain family member 9 (CARD9). Additionally, mycobacteria harbor immuno-evasive cell-wall lipids associated with virulence and latency; however, their mechanism of action remains unclear. Here, we show that the DAP12-associated triggering receptor expressed on myeloid cells 2 (TREM2) recognizes mycobacterial cell-wall mycolic acid (MA)-containing lipids and suggest a mechanism by which mycobacteria control host immunity via TREM2. Macrophages responded to glycosylated MA-containing lipids in a Mincle/FcRγ/CARD9-dependent manner to produce inflammatory cytokines and recruit inducible nitric oxide synthase (iNOS)-positive mycobactericidal macrophages. Conversely, macrophages responded to non-glycosylated MAs in a TREM2/DAP12-dependent but CARD9-independent manner to recruit iNOS-negative mycobacterium-permissive macrophages. Furthermore, TREM2 deletion enhanced Mincle-induced macrophage activation in vitro and inflammation in vivo and accelerated the elimination of mycobacterial infection, suggesting that TREM2-DAP12 signaling counteracts Mincle-FcRγ-CARD9-mediated anti-mycobacterial immunity. Mycobacteria, therefore, harness TREM2 for immune evasion.


Introduction
Tuberculosis (TB) is a chronic infectious disease caused by Mycobacterium tuberculosis (Mtb) and remains a major cause of morbidity and mortality worldwide. Most infected individuals do not manifest clinical symptoms of TB (referred to as latent infection). In this phase, the bacteria are dormant with a non-replicating phenotype and resistant to reactive intermediates and antibiotics, consistently evading host immune recognition 1.
A distinctive trait of mycobacteria is their highly lipid-rich outer membrane, which is not only critical for their replication, acid-fast properties, and drug resistance, but also plays a key role in pathogenicity2,3 Some of these lipids trigger immunopathologic response, whereas others have immuno-evasive functions associated with virulence and latency4. Mycolic acids (MAs) are mycobacterium and related spp.speci c lipids with extremely long fatty acids (C 60 -C 90 ) and represent predominant component in mycobacterial cell-walls. MAs form the cell wall skeleton by covalently linking to the arabinogalactan (ABG) and peptidoglycan (PG) basement layer and also exist on the cell wall surface in either glycosylated or non-glycosylated forms5. Their composition is altered dynamically depending on mycobacterial life-cycle stage and external environmental conditions6,7, and this affects bacterial virulence and latency by controlling host immunity. For example, glycosylated MA-containing lipids, such as trehalose dimycolate (TDM; also known as cord factor) and glucose monomycolate (GMM), possess potent pro-in ammatory and adjuvant activity and are capable of inducing lung granulomas when injected into animals8,9. These glycolipids are predominantly synthesized in actively replicating mycobacteria, while their levels are markedly diminished in dormant mycobacteria6. Non-glycosylated MA lipids, such as free MA (fMA) and glycerol monomycolate (GroMM), are associated with mycobacterial persistence, immune suppression, and bio lm formation. fMA is increased in cell walls in In contrast to CLR-associated FcRγ, the other ITAM-bearing signaling adaptor, DAP12, might negatively regulate the anti-mycobacterial immune response, because DAP12 de ciency accelerates the clearance of mycobacteria and the granuloma formation in lungs upon Mtb or Mycobacterium bovis Bacille de Calmette et Guerin (BCG) infection35, 36. These observations implicate unknown DAP12-associated regulatory receptors that might possibly recognize immune-suppressive ligands in mycobacteria.
However, these theoretical receptors, as well as the precise mechanisms of immune suppression via DAP12, have not been described.
In the present study, we present evidence suggesting that the DAP12-associated receptor triggering receptor expressed on macrophage 2 (TREM2) recognizes mycobacterial MA-containing lipids that are distinct from those recognized by Mincle and counteracts the Mincle-FcRγ-CARD9-mediated antimycobacterial immune response. We show that the glycosylated MA-containing lipids induce mycobactericidal macrophages in a Mincle/FcRγ/CARD9-dependent manner. Conversely, the nonglycosylated MA-containing lipids induce mycobacteria-permissive macrophages in a TREM2/DAP12dependent but CARD9-independent manner. Moreover, loss of TREM2 markedly enhances Mincle-induced macrophage activation in vitro and in ammation in vivo and accelerated the elimination of mycobacterial infection in mice. These results suggest that mycobacteria evade host immunity via TREM2 to avoid mycobactericidal macrophage activation through the Mincle-CARD9 pathway. This nding describes a mechanism by which mycobacteria controls the host immune response through a host regulatory innate immune receptor, which may have implications for treatment of mycobacteriosis including TB.

TREM2 recognizes mycobacteria
To screen novel ITAM-coupled receptors capable of recognizing mycobacteria, we analyzed the binding of 20 known ITAM-coupled CLR, TREM, and leukocyte mono-Ig-like receptor (LMIR; CD300) family receptors22,37-39 fused to Fc-antibody fragments to heat-killed mycobacterial strains, including the virulent strain Mtb H37Rv, the attenuated strain Mtb H37Ra, and the vaccine strain M. bovis BCG, using owcytometry. In addition to several CLRs previously reported to recognize mycobacteria, including Mincle, speci c ICAM-3 grabbing nonintegrin-related (SIGNR)1, SINGNR3, and DC-SIGN30,40, this screening identi ed TREM2 and LMIR5 as novel receptors capable of binding to mycobacteria (Fig. 1a). Especially, TREM2 demonstrated a strong binding capacity to all tested mycobacterial strains, which prompted us to further characterize this interaction.
To investigate whether recognition of mycobacteria by TREM2 activates intracellular ITAM signaling, we used nuclear factor of activated T cells (NFAT)-driven green uorescent protein (GFP)-reporter cells (2B4)30 ectopically expressing TREM2 and its signaling subunit DAP1241,42. As reported previously30, Mtb H37Rv, Mtb H37Ra, and M. bovis BCG stimulated NFAT-GFP signaling in reporter cells expressing Mincle and FcRγ ( Supplementary Fig. 1a). We found that all these mycobacterial strains clearly activated reporter cells expressing TREM2 and DAP12, but not in those expressing only DAP12, in a dose dependent manner (Fig. 1b). Whereas, none of these strains activate reporter cells expressing TREM1 plus DAP12 (Supplementary Fig. 1b). Importantly, the level of stimulation by TREM2 and Mincle differed between the strains ( Fig. 1b and supplementary Fig. 1a), suggesting that TREM2 and Mincle recognize different ligands commonly expressed by these mycobacterial strains.
TREM2 recognizes MAs TREM2 binding to mycobacteria implies that its ligand(s) exist on the cell-wall surface, where mycobacteria express a wealth of unique lipids that in uence the host immune responses 43. Given that TREM2 binds to various endogenous mammalian lipids44,45, we rst examined whether mycobacterial lipids contain TREM2 ligand(s). De-lipidation of Mtb H37Ra with chloroform and methanol (C:M) (Fig. 2a) markedly diminished the level of stimulation in TREM2-reporter cells (Fig. 2b), as well as in Minclereporter cells as reported30 (Fig. 2c). Accordingly, the lipid-containing C:M fraction ( Fig. 2a) but not the hydrophilic M:W fraction showed strong stimulating activity in TREM2 reporter cells (Fig. 2b) as with Mincle reporter cells (Fig. 2c), implicating that TREM2 ligands were present in the cell wall lipid fraction as with so Mincle ligands 30. We then next tested the TREM2-stimulating activity using known major constituents of mycobacterial cell wall, including the glycans PGN and AbG, the immunostimulatory glycolipids LAM and TDM (Fig. 2d), and fMA (Fig. 2d). We observed that AbG, PGN, and LAM did not activate the TREM2-reporter cells, even at higher concentrations ( Fig. 2e-g), whereas we observed substantial ligand activity with TDM and fMA (Fig. 2h). Importantly, loss of trehalose moiety from TDM (i.e. fMA) markedly increased the TREM2-stimulating activity (Fig. 2h), while it completely abolished the Mincle-stimulating activity as reported (Fig. 2i) 30. These results suggest that TREM2 recognizes the MA moiety of TDM, and that the sugar moiety interferes with this recognition.
We then investigated the structure of MA necessary for TREM2 recognition. Extremely long (C 60 -C 90 ) and branched alkyl chains are hallmarks of mycobacterial MAs. To assess the importance of the length and branching, we compared the stimulatory activity of mycobacterial MA with that of Rhodococcus equi (R. equi) MA, which has shorter alkyl chains (C 35 ), and of behenic acid (BA), which is a linear fatty acid with an alkyl chain (C 22 ) of similar length with the mero-chain of R. equi MA (Fig. 2j). R. equi MA exhibited a similar level of stimulatory activity in TREM2-reporter cells as mycobacterial MA, whereas the level of stimulation with BA was very low, even at higher concentrations of BA (Fig. 2j), suggesting that the branched structure rather than the alkyl-chain length was important for TREM2 recognition of MA. To further investigate this, we tested the ligand activity of palmitic acid (PA; C 16 ), 2-tetradecylhexadecanoic acid (THA) (C 14 /C 16 ), which bears a PA-based synthetic fatty-acid with similar branching structure as MA but lacking its 3-hydroxyl residue, and 3-hydroxybutyric acid (HBA), which has only the carboxyl and hydroxyl residue at the branching moiety of MA (Fig. 2k). We detected TREM2 stimulation with THA, albeit at a lower level than MA, but could not detect stimulation with PA and HBA (Fig. 2k), suggesting that TREM2 recognition requires a branched fatty acid structure with an alkyl chain.
The R47H mutation in TREM2, which is associated with a higher risk of several neurodegenerative diseases46-49, impairs the recognition of brain lipids45,50. We found that the TREM2 R47H mutation also attenuated MA recognition ( Supplementary Fig. 2), suggesting that this residue is important for MA recognition by TREM2.
TREM2 and Mincle preferentially recognize distinct MA-containing lipids based on their glycosylation Mycobacteria express a variety of MA-containing lipids on their cell wall51. Since both Mincle and TREM2 have the capacity to recognize MA-containing lipids, we next compared their binding activity to the glycosylated (TDM and GMM) or the non-glycosylated (GroMM and fMA) MA-containing lipids (Fig.  3a) using the receptor-Fc fusion proteins. We observed that Mincle-Fc not only showed strong binding to TDM as expected, but also binding to GMM, whereas it showed no detectable binding to fMA and weak binding to GroMM only at higher concentrations (Fig. 3b). By contrast, TREM2-Fc showed strong binding to GroMM and fMA but showed relatively much weaker binding to TDM and GMM (Fig. 3b). We then examined whether the different binding of MA-containing lipids to TREM2 and Mincle re ect their receptor-stimulatory activities using NFAT-GFP-reporter cells. Consistent with the binding data, TREM2reporter cells strongly responded to MA and GroMM but weakly to TDM and GMM, whereas Minclereporter cells responded strongly to TDM and GMM but weakly to GroMM only at higher concentrations and did not respond to fMA (Fig. 3c). Therefore, TREM2 and Mincle preferentially recognize nonglycosylated and glycosylated MA-containing lipids, respectively.
Glycosylated and non-glycosylated MAs elicit distinct macrophage activation through Mincle and TREM2 To investigate the relevance of TREM2 and Mincle recognition of MA-containing lipids in the activation of innate immune cells, we stimulated peritoneal macrophages from wild-type (WT), TREM2-de cient (Trem2-/-), or Mincle-de cient (Clec4e-/-) mice with either glycosylated or non-glycosylated MAcontaining lipids and examined their production of MCP-1, a pivotal monocyte chemoattractant implicated in TB pathology52-54, and TNF, an essential cytokine for granuloma formation and TB control55,56. Interestingly, we found that activation of WT macrophages by the glycosylated MAcontaining lipids TDM or GMM or by lipopolysaccharide (LPS) induced production of both MCP-1 and TNF, whereas activation of macrophages with the non-glycosylated MA-containing lipids GroMM or fMA induced production of MCP-1 but not TNF (Fig. 4a), as well as other pro-in ammatory cytokines, including IL-6 and IL-12p40 ( Supplementary Fig. 3a). Loss of Mincle almost completely abolished TNF, as previously30, as well as MCP-1 in response to TDM as expectedly30. We found that Mincle de ciency also abolished these productions induced by GMM (Fig. 4a). However, Mincle de ciency did not affect MCP-1 production in response to GroMM or fMA. By contrast, loss of TREM2 almost completely abolished the MCP-1 production in response to GroMM or MA, but not to TDM or GMM (Fig. 4a). These results demonstrated that glycosylated and non-glycosylated MA-containing lipids elicited distinct macrophage activation, which was dependent on Mincle and TREM2, respectively.
Triggering of ITAM-coupled receptors on myeloid cells activates the ITAM-Syk-CARD9 signaling pathway to induce cytokine production57. To investigate whether MCP-1 and TNF production induced by MA-containing lipids depends on this pathway, we rst con rmed the requirement for DAP12 and FcRγ, which are the ITAM-containing signaling subunits of TREM2 and Mincle, respectively. As expected, we observed that FcRγ-de cient (Fcer1g-/-) and DAP12-de cinet (Tyrobp-/-) macrophages almost phenocopied Mincle-de cient and TREM2-de cient macrophages, respectively, with similar patterns of defects in TNF and MCP-1 productions in response to glycosylated or non-glycosylated MA-containing lipids (Fig. 4b).
However, FcRγ de ciency slightly dampened MCP-1 induction by fMA or GroMM at higher concentrations, implicating that FcRγ might partly contribute to TREM2 signaling. We then examined the requirement for Syk and CARD9 for each of these responses. Treatment of WT macrophages with a Syk inhibitor (BAY-613606) abrogated both MCP-1 and TNF production in response to all MA-containing lipids (Fig. 4c), indicating essential role for Syk in the response to these lipids. Intriguingly, although TNF production induced by glycosylated MAs through Mincle was abolished in CARD9-de cient (Card9-/-) macrophages, as expected58, CARD9-de ciency did not affect MCP-1 production induced by both glycosylated and non-glycosylated MA-containing lipids (Fig. 4c). Collectively, these results demonstrated that mycobacterial MA-containing lipids induced TNF production through Mincle via the canonical FcRγ-Syk-CARD9 pathway, while they induced MCP-1 production via the ITAM-Syk pathway but independent of CARD9 signaling ( Supplementary Fig. 3b).

TREM2/DAP12 signal inhibits macrophage activation through Mincle/FcRγ
Interestingly, we observed that loss of TREM2 markedly enhanced MCP-1 and TNF production by peritoneal macrophages in response to TDM or GMM (Fig. 4a). This was also true in Tyrobp-/macrophages ( Fig. 4b), suggesting an inhibitory role of TREM2/DAP12 signaling in Mincle/FcRγ-induced macrophage activation. We did not observe an increase in cytokine response when the same preparation of Trem2-/macrophages were stimulated with TLR ligands, such as Pam 3 CSK 4 (for TLR2), LPS (for TLR4), Poly (I:C) (for TLR3), CpG-ODN (for TLR9), or with the Dectin-1 ligand zymozan ( Fig. 4a and supplementary Fig. 4a, b). This selective enhancement of Mincle activation in the absence of TREM2 was more prominent in BMDMs, where TDM-induced TNF was detected at very low levels in WT BMDMs. Nevertheless, Trem2-/or Tyrobp-/-BMDMs showed substantial TNF production in response to TDM ( Supplementary Fig. 4c), whereas this enhancement was not observed following Pam 3 CSK 4 or LPS stimulation. Additionally, we observed an increased cytokine response following stimulation of Trem2-/-BMDMs with mycobacterial total lipids (C:M fraction in Fig. 2a) prepared from Mtb R37Ra or M. bovis BCG ( Supplementary Fig. 4d). This response was largely dependent on Mincle, as Clec4e-/-BMDMs produced markedly lower TNF in response to the total lipids compared to WT BMDMs ( Supplementary Fig. 4d). These data clearly indicated that TREM2/DAP12 signaling selectively inhibited macrophage activation induced by Mincle-FcRγ signaling.

Triggering of TREM2 induces permissive macrophages
Given that NO plays a pivotal role in controlling mycobacterial infections59,60, we investigated the action of TREM2 and Mincle on NO production by macrophages. We observed that TDM stimulation strongly induced NO production by bone marrow-derived macrophages (BMDMs), as reported previously30, while fMA did not substantially induce NO production (Fig. 5a). This was consistent with the observed increase in Nos2 expression, which encodes iNOS, after stimulation with TDM but not fMA ( Supplementary Fig.  5a). Additionally, the TDM-induced NO production by macrophages was inhibited following the addition of a TNF blocking antibody (Fig. 5b). Inversely, addition of recombinant TNF to the fMA-stimulated culture induced NO production by macrophages (Fig. 5b), indicating that NO production was dependent on TNF, which is consistent with previous ndings61-63.
We then characterize the in ammation triggered via Mincle and TREM2 in vivo. Because TDM induces lung granulomas in mice via the Mincle/FcRγ pathway30, we examined whether fMA had a similar effect.
Intravenous injection of a TDM (oil-in-water) emulsion into mice induced massive granulomatous lesions in the lungs, which was completely abolished in Clec4e-/mice (Fig. 5c), as reported previously30. Additionally, this effect was also absent in Card9-/mice, indicating that TDM-induced granuloma formation was dependent on Mincle/CARD9 signaling. By contrast, we did not observe lung granulomatous lesions in mice injected with an fMA emulsion ( Fig. 5c), suggesting that fMA cannot activate CARD9 signaling. To further explore the innate immune response induced by TDM or MA in vivo, we intraperitoneally injected the TDM or fMA emulsion into mice and measured the level of MCP-1 and TNF, and Nos2 mRNA levels, as well as the number of recruited in ammatory cells, induced in the peritoneal cavities. We detected MCP-1 in peritoneal lavages following respective fMA and TDM administrations at similar levels, with peak MCP-1 levels observed at 4-and 24-h post-injection, respectively (Fig. 5d). However, TNF production and Nos2 mRNA expression were induced only by TDM but not fMA (Fig. 5d), which was consistent with results from the in vitro-stimulated macrophages (Fig.  4). Moreover, we observed that TDM recruited substantial numbers of macrophages [CD11b+Ly6G-F4/80low small peritoneal macrophages (SPMs)], Supplementary Fig. 5b) and neutrophils (CD11b+Ly6G+ F4/80-, Supplementary Fig. 4b) to the cavity, whereas fMA recruited comparable numbers of macrophages as TDM but fewer neutrophils only immediately after administration (Fig. 5e). Importantly, MCP-1 level (Fig. 5f) and the number of recruited cells (Fig. 5g) induced by fMA in Trem2-/mice were almost the same as those induced by the control vehicle administration, indicating that the fMA-induced in ammation was dependent on TREM2.
To investigate the phenotype of TDM-or fMA-induced macrophages, we examined the expression of the in ammatory M1-macrophage markers iNOS and CD38 64 in the recruited macrophages (Fig. 5h). Intraperitoneal administration of TDM and fMA recruited comparable numbers of macrophages to the cavity (Fig. 5i); however, although TDM-induced macrophages exhibited a CD38highiNOS+ phenotype ( Fig. 5j), those induced by fMA had reduced CD38 expression and were negative for iNOS (Fig. 5k). This phenotype resembles a previously reported phenotype associated with permissive macrophages that are recruited by a subset of pathogenic mycobacteria expressing the virulence lipids PDIM and PGL in an MCP-1 and CCR2-dependent manner and provide a niche for mycobacterial propagation21,65.
Collectively, these results suggested that triggering of the Mincle-CARD9 pathway elicited lung granuloma formation and recruited M1-type mycobactericidal macrophages producing TNF and NO, whereas TREM2 activation recruited mycobacterium-permissive macrophages lacking TNF and NO production.

TREM2 de ciency exacerbates Mincle-induced in ammation
To investigate the relevance of TREM2 inhibition of Mincle-induced macrophage activation in vivo, we assessed the impact of TREM2 de ciency on tissue in ammation induced by TDM administration. Intraperitoneal injection of TDM emulsion into Trem2-/mice resulted in signi cantly higher levels of TNF and MCP-1 production and increased Nos2 mRNA expression (Fig. 6a), as well as higher numbers of macrophages and neutrophils to the peritoneal cavities (Fig. 6b), as compared to those observed in WT mice. Intravenous injection of the TDM emulsion into mice induces lung swelling (increased lung weight index: LWI) and thymic atrophy (decreased thymic weight index: TWI) dependent on the Mincle-FcRγ pathway30. We observed that Trem2-/mice displayed signi cantly higher LWI and lower TWI than WT mice followingr TDM injection (Fig. 6c). Histopathological analysis of the lungs revealed that the granulomatous lesions were markedly more prominent in Trem2-/mice (Fig. 6d). In addition, speci c pathological features in the form of prominent vasculitis and edema, indicating accelerated in ammation, were observed in the lungs of Trem2-/-, but not WT mice ( Supplementary Fig. 6). Consistent with these results, the levels of TNF and MCP-1 production and Nos2 mRNA expression in the in amed lungs were signi cantly higher in Trem2-/mice than in WT mice (Fig. 6e). These results suggested that TREM2 suppressed Mincle-induced in ammation in vivo.

TREM2 de ciency accelerates the clearance of mycobacterial infection
Since our data suggested a suppressive role for TREM2 in the microbicidal innate immune response via Mincle, we investigated the impact of TREM2 de ciency on the clearance of mycobacterial infection. First, we infected WT and Trem2-/-BMDMs with M. bovis BCG in vitro and assessed NO production and mycobacterial killing in the macrophages. NO production in Trem2-/-BMDMs was signi cantly higher than in WT BMDMs (Fig. 7a). Accordingly, the number of intracellular BCG colony-forming units (CFUs) was signi cantly lower in Trem2-/-BMDMs than in WT BMDMs (Fig. 7b). Next, to investigate the relevance of this observation in vivo, we intraperitoneally infected WT and Trem2-/mice with M. bovis BCG and measured bacterial clearance. The bacterial burden (CFUs) in the peritoneal cavities at day 1 and 3 after infection were signi cantly lower in Trem2-/mice than in WT mice (Fig. 7c). Consistently, early MCP-1 production (Fig. 7d) and the number of in ltrated macrophages (Fig. 7e) after infection were signi cantly higher in Trem2-/mice than in WT mice. Moreover, Nos2 expression in in ltrated cells tended to be higher (Fig. 7f, left), whereas expression of the M2 marker arginase 1 (Arg1) was signi cantly lower in Trem2-/mice (Fig. 7f, right) relative to that in WT mice, suggesting that lack of TREM2 signaling favored M1 polarization. These results suggested that TREM2 contributed to immune evasion by mycobacteria by inhibiting the activation of microbicidal M1 macrophages.

Discussion
In the present study, we identi ed the DAP12-associated TREM2 as a receptor for mycobacteria by the screening using the ITAM-coupled receptor-Fc fusion protein library. TREM2 is expressed on various myeloid cells, including macrophages and microglia65, and Trem2 de ciency is reportedly related to various neurodegenerative diseases. Homozygous loss-of-function mutations of Trem2 or dap12 cause Nasu-Hakola disease accompanied by demyelination and axonal loss66,67. Single nucleotide polymorphisms, including the TREM2 R47H variant, increase the risk of dementia, including frontotemporal lobar degeneration, Parkinson's disease, Alzheimer's disease, and amyotrophic lateral sclerosis46-49. These genetic associations highlight the physiological importance of TREM2 and its ligand recognition. Previous reports indicate that TREM2 binds to anionic ligands of bacteria68, lipooligosaccharide (LOS) of Neisseria gonorrhoeae69,70, several brain lipids associated with brillar amyloid β45, and apolipoproteins71. Although these ligands have been identi ed, ligand structures prerequisites for TREM2 recognition remain unknown. Here, we identi ed MA-containing lipids as TREM2 ligands and found that their branched alkyl chains are required for TREM2 recognition via a structureactivity relationship analysis. The removal of branched alkyl chains from LOS by O-deacetylation diminishes TREM2-LOS interaction70, which implicates the lipid moiety of LOS in the interaction with TREM2. Additionally, TREM2 ligands in the brain include phospholipids or sphingolipids carrying branched alkyl chains45; therefore, our ndings agree with previous results and might offer insight into future searches for unknown TREM2 ligands.
We found that macrophage activation by glycosylated versus non-glycosylated MA lipids resulted in different patterns of response with different receptor requirements. While glycosylated MAs were recognized by Mincle, and induced the production of MCP-1, TNF, and NO, non-glycosylated MAs were recognized by TREM2 and induced only MCP-1 production. A previous report showed that human Mincle but not mouse Mincle could recognizes GroMM and induces TNF production in primary human macrophages72, suggesting that the range of structures of MA ligands recognized by Mincle might differ between humans and mice. Nevertheless, Hattori et al showed in this report that the binding a nity of human Mincle to GroMM was ~100-fold lower than that to TDM, indicating the importance of the sugar moiety for the recognition by Mincle, regardless of species. Importantly, our data showed that mouse Mincle was also capable of binding to GroMM at higher concentrations, although it was not essential for macrophage response to GroMM. Future studies should clarify the contributions of TREM2 and Mincle to recognition of MA-containing lipids in humans.
MCP-1 plays a role in the early stages of Mtb infection53,54. Additionally, higher levels of MCP-1 production are associated with greater severity of tuberculosis in human patients73. The polymorphism of at position -2518 the MCP-1 promoter is associated with TB susceptibility74,75, and the odds of developing TB is 2.3-to 5.5-fold higher in patients with MCP-1 genotypes AG and GG relative to those with the AA genotype. Patients carrying the AG or GG genotype harbor extremely high concentrations of MCP-1, which inhibit the expression of IL-1275, suggesting that elevated MCP-1 production and lower in ammatory cytokine levels promote TB development. This is a similar phenomenon to that observed following TREM2 recognition of non-glycosylated MAs in the present study. Mycobacterial cell-wall components, PDIM and PGL, enhance infectivity by selective induction of MCP-1 for the recruitment of the permissive macrophages76. These ndings indicate that exclusive production of MCP-1 via TREM2 is presumably bene cial to mycobacteria.
The recruitment of permissive macrophages and inhibition of Mincle-FcRγ-CARD9 signaling are key functions of TREM2. Here, we found that TREM2/DAP12 signaling speci cally inhibited Mincle-FcRγ-CARD9 signaling but not TLR signaling. However, previous reports show that TREM2 de ciency enhances cytokine production induced by TLRs77,78. Although the reason for this discrepancy is unknown, upregulated signaling via Mincle might be responsible for the enhanced TLR response. Because Mincle recognizes damaged cell-derived endogenous ligands79,80, Mincle expression induced by TLR stimulation30 might subsequently amplify the TLR response81 depending on cell condition. Alternatively, TREM2 binding to TDM (Figure 3) but not TLR ligands could generate signi cant inhibitory signals upon TDM stimulation; therefore, TREM2 de ciency might result in more obvious effects on the signaling via Mincle than that via TLRs.
The composition of MA-containing lipids dynamically changes in response to the external environment.
TDM is a major glycosylated MA in the mycobacterial cell wall in culture, but in the host under glucoserich condition, TDM synthesis is downregulated, and GMM is produced by mycolytransferases using host-derived glucose7. Because glycosylated MAs exhibit potent adjuvant activity82, the presence of these lipids preferentially activates the immune response through Mincle to eliminate mycobacteria. Lipid composition differs in latent mycobacterial infections. GroMM might be associated with latent mycobacteria, given that GroMM-reactive T cells are observed only in latent but not active TB cases15,72. Additionally, levels of glycosylated MAs decrease in Mtb in a non-replicating dormant-like state6. Therefore, we speculate that the recognition of latent mycobacterial cell walls by TREM2 might lead to the recruitment of permissive macrophages, thereby promoting chronic infection ( Supplementary Fig. 7).
Macrophages or microglia phagocytize bacteria or apoptotic cell debris through TREM2 69,83-85. Macrophages are roughly classi ed into M1 and M2 subtypes that exert opposite effects on the in ammatory response86. M1 macrophages are pro-in ammatory, whereas M2 macrophages are antiin ammatory and exhibit phagocytic activity to promote tissue repair and homeostasis. Unlike FcRγcoupled mycobacterial receptors including Mincle, TREM2-DAP12 signaling seemingly confers M2-like anti-in ammatory properties on macrophages. Indeed, IL-4, which skews macrophage polarization toward M2, induces TREM2 expression in macrophages78. Additionally, TREM2 de ciency impairs wound healing of colonic mucosal injures accompanied by diminished M2 differentiation and increased TNF and interferon (IFN)-γ production87. Moreover, TREM2 plays an important role in maintaining homeostasis in the brain by clearance of dead cells or amyloid-β through brain-lipid recognition83. Therefore, mycobacteria presumably harness the functions of TREM2 by stimulating TREM2 with MAcontaining lipids to gain niches suitable for their propagation.
Collectively, our data suggest that TREM2-DAP12 signal activation by non-glycosylated MAs, which are associated with virulent or latent mycobacterial infections, recruits permissive macrophages and suppress the mycobactericidal immune response induced by glycosylated MAs through Mincle-FcRγ-CARD9 signaling. Therefore, this study elucidated al mechanism by which mycobacteria control the host immune response by stimulating a host regulatory receptor with their speci c cell-wall lipids. Our ndings suggest that targeting the TREM2-DAP12 pathway might represent a novel therapeutic intervention to control mycobacterial diseases.

Methods
Reagents MA from M. bovis BCG was prepared as described below. TDM, LPS, ABG, PA and HBA were purchased from Sigma-Aldrich. Poly (I:C), CpG-ODN, Pam3CSK were purchased from InvivoGen. Oxidized (OX)zymosan was kindly provided by Prof. Naohito Ohno (Tokyo University of Pharmacy and Life Sciences).
MA from R. equi was kindly provided by Dr. Jun Miyazaki (Tsukuba University). LAM was purchased from Nakalai tesque. THA was purchased from Wako pure chemical industries. GMM and GroMM were prepared as described previously 7,16 and the nal preparations provided no extra spots on analytical TLC plates, and the identity of the lipids was con rmed by mass spectrometry.

Preparation of MA
Free MAs were isolated from heat-killed M. bovis BCG (Tokyo 172 strain). The BCG cells were suspended in 85% tetrahydrofuran (THF)/water solution under a nitrogen atmosphere, followed by re ux with stirring for 1 h. The cell suspension was ltrated under pressure and washed with 75% THF/water solution. The residue was resuspended in 75% THF/water solution under a nitrogen atmosphere, followed by re ux with stirring for 1h. The suspension was ltrated under pressure and washed with 75% THF/water solution three times and with methanol twice. Then, the bacterial cells were suspended in 50% 2propanol/water solution containing 10% potassium hydroxide, followed by re ux with stirring for 2 h to complete alkaline hydrolysis of MA ester. After the re uxing, the suspension was cooled down on ice, and acidi ed with 6 M hydrochloric acid. The reaction mixture was extracted twice with n-heptane, and the nheptane fraction was washed twice with water and then twice with 90% ethanol/water. Finally, the nheptane fraction was concentrated in vacuo to obtain puri ed MAs. The product was applied for TLC (nhexane:methyl tert-butyl ether: formic acid = 8/2/0.5, v/v/v) to con rm no extra no spots and for MADI-TOF mass spectrometry analyses to con rm the identity of MAs.
Screening for ITAM-coupled receptors binding to mycobacteria Heat-killed M. tuberculosis H37Rv, M. tuberculosis H37Ra (Difco) or M. bovis BCG were incubated with 20 µg/ml of ITAM receptor-Fc fusion proteins or control Fc fragment protein in RPMI1640 medium for 1 h on ice. After 2 times washing with the medium, the mycobacterial cells were incubated with FITC-conjugated anti-human IgG secondary antibody for 30 min on ice. After a wash, the uorescence intensity was analyzed by ow cytometer as above. The receptors that showed higher uorescence intensities than the control values were judged as positive for binding.

Plate-coated lipid binding assay
The lipids (MA, TDM, TDB, PA, THA, GMM, GroMM and HBA) were dissolved in chloroform at 1 mg/ml and then diluted with isopropanol to the working concentrations. Lipid binding assay was performed as described previously91. Brie y, 0.5 µg/well of fMA, GroMM, GMM, and TDM were coated on ELISA plates and then incubated with 0.24 to 1000 nM puri ed Mincle-Fc or TREM2-Fc proteins in TSM buffer (20 mM Tris-HCl, 150 mM NaCl, 1 mM CaCl 2 , 2 mM MgCl 2 , pH 7.0) for 2 h at room temperature. The plates were washed 4 times with 150 µl of TSM buffer , and then incubated with anti-human IgG HRP secondary antibodies for 1 h at room temperature. The binding was detected by colorimetric assay using TMB substrate (Sumitomo Bakelite) and the absorbance at 450 nm was measured by VERSAMax microplate reader (Molecular Devices).

In vitro macrophage stimulation
Peritoneal macrophages were prepared as described previously92. Brie y, 2 ml of 4 % thioglycollate (Difco) solution was intraperitoneally injected into mice. Five days after the injection, peritoneal cells were collected by washing the peritoneum cavity with 5 ml of RPMI 1640 medium containing 10% FCS and 2mercaptoethanol (RPMI-10). The collected cells were cultured overnight in RPMI-10 and then the adherent cells were used for assays. Bone marrow-derived macrophages (BMDMs) were prepared as described previously 27. Brie y, bone-marrow cells collected from WT, Trem2-/-, Tyrobp-/or Clex4e-/-mice were cultured in RPMI-10 in the presence of 25 ng of recombinant murine M-CSF (PeproTech) for 3 days, and then the adherent cells were collected as BMDMs. For in-vitro cell stimulation, the lipid solutions were added into the 96-well at bottom plates at 20 µl/well and then the solvent was completely evaporated in a hood before plating macrophages, as described previously30. 1 x 105 cells per well were stimulated with plate-coated lipids or TLR ligands in RPMI-10. The culture supernatants after 24-h culture were collected and the concentrations of TNF (eBioscience), IL-6 (eBioscience), IL-12p40 (Biolegend), IL-10 (Biolegend) and MCP-1 (Biolegend) were analyzed by ELISA kits according to manufacturers' instructions. For Syk inhibition, cells were incubated with 1 µM BAY-613606 (Calbiochem) for 30 min prior to the stimulation.

Measurement of NO production
BMDMs were stimulated in RPMI-10 with 0.1 or 1.0 g per plate of plate-coated MA or TDM in the presence of 10 ng/ml of recombinant IFNγ for 24 h. For the assessment of TNF involvement in NO production, carrier-free recombinant mouse TNF (Biolegend), LEAF puri ed anti-mouse TNF-α antibody (MP6-XT22) (BioLegend), or LEAF puri ed Rat IgG 1 κ Isotype Ctrl antibody (BioLegend) were added to the culture. The culture supernatants were collected and mixed with Griess reagent (1% sulfanilamide, 0.1% N-(1-Naphthyl)ethylenediamine dihydrochloride, 2.5% phosphoric acid) at a 1:1 ratio, reacted for 10 min, and the absorbance at 550 nm was measured by a VERSAMax microplate reader. The nitrite concentration was calculated according to the standard curve.
Quantitative real time PCR Total RNA was isolated from cells using Sepasol-RNA I Super G RNA-isolation kit (Nacalai Tesque). After the removal of DNA contamination by DNase I (Nippon Gene), the total RNA was reverse-transcribed with ReverTra Ace qPCR RT Master Mix (TOYOBO) to synthesize cDNA. Quantitative real-time PCR (qRT-PCR) was performed using THUNDERBIRD SYBR qPCR Mix (TOYOBO) and StepOnePlus (Thermo sher scienti c). The speci c primer pairs were "TTGGGTCTTGTTCACTCCACGG" and "CCTCTTTCAGGTCACTTTGGTAGG" for Nos2 and "AACTTTGGCATTGTGGAAGG" and "ACACATTGGGGGTAGGAACA" for Gapdh as an internal control. The relative expression was calculated by ΔΔCt.
MA and TDM emulsions for i.p. administration Ten mg MA or 2 mg TDM was dissolved in 1 ml of Bayol F (SERVA Electrophoresis) at 64°C, then mixed with 1 ml of PBS using a handy homogenizer (RELIEF). One hundred µl of the emulsion was intraperitoneally injected into WT or Trem2-/mice. The emulsion without lipids was injected as a vehicle control. Peritoneal lavages were collected by washing the peritoneum cavity with RPMI-10. Concentration of cytokine in the lavage was measured by ELISA kit as described above.
TDM emulsion for i.v. administration TDM emulsion was prepared as described previously30. Brie y, 1 mg of TDM was dissolved in 180 µl of Bayol F at 64°C, then mix with 1.8 ml of 1.1% Tween 80/0.9% NaCl solution to make emulsion. One hundred µl of the emulsion including 50 µg of TDM was intravenously injected into WT and Trem2-/mice. At day 7, the thymuses, lungs, and body weights were measured and the thymus or lung weights were divided by body weights to calculate lung or thymus weight index, respectively. The left lobes were then xed in 4% paraformaldehyde and embedded in para n and stained with hematoxylin and eosin solution. The rest of lung tissues were homogenized by gentleMACS (Miltenyi Biotec) as described previously93 and the concentration of cytokines in homogenates were measured by ELISA kit as

M. bovis BCG infection
In-vitro infection of M. bovis BCG was performed as described previously94. Brie y, BMDMs (1 x 105) from WT of Trem2-/mice were and infected in 96-well plate with 1 x 106 (MOI 10:1) of live M. bovis BCG (Tokyo 172 strain, Japan BCG laboratory) for 4h. After the infection, uninfected bacteria were washed-out with RPMI-10 and the infected cells were further cultured for 24 h in RPMI-10. The supernatant was collected and he nitrite concentration was measured by Griess assay. The cells were disrupted by water and plated on Middlebrook 7H10 agar (Difco) and number of BCG colonies was counted after 3 week culture for CFU determination. In vivo infection was performed as described previously95. WT and Trem2-/mice were infected intraperitoneally with 5.0 x 106 CFU of live M. bovis BCG. The peritoneal lavages were collected at indicated time point by washing the peritoneal cavity with 1 ml of RPMI-10. The concentration of cytokines in the lavage was analyzed by ELISA kit as described above. The recruited peritoneal cells were analyzed by ow cytometer as described above. For CFU counting, the peritoneal cells were disrupted by water and plated on Middlebrook 7H10 agar and colonies were counted as described above.

Statistical analysis
All statistical analyses were carried out using Prism 5 software. Student's t-test was performed for the comparisons between TDM and fMA in Fig. 5I-K or between WT and Trem2-/mice in Fig. 7A, B. Oneway ANOVA or two-way ANOVA with Bonferroni post hoc test were performed for the comparison of       monocyte-derived macrophages (CD11b+Ly6G-F4/80low SPMs) was analyzed by ow cytometry. (f, g) WT or Trem2-/-mice were injected intraperitoneally with fMA or control (vehicle) emulsion, peritoneal lavages were collected, and the concentrations of MCP-1 (control, n = 4; fMA, n = 6; at 4 h) and TNF (n = 4; at 72 h) was measured by ELISA (f). Recruitment of peritoneal exudate cells (n = 4; at 72 h) was analyzed as described in (e, g). (h-k) WT mice were injected intraperitoneally with TDM, fMA, or control (vehicle) emulsion, and in ltrated macrophages at days 1, 2, and 3 were analyzed by ow cytometry to measure the expression of the M1 macrophage markers CD38 and iNOS (h). The numbers of total recruited monocyte-derived macrophages (CD11b+Ly6C+F4/80+) (i), the M1 macrophages (CD11b+Ly6C+F4/80+CD38highNOS+) (j), and permissive macrophages (CD11b+Ly6C+F4/80+CD38dullNOS-) (k) at day 3 are shown. Data in (d-k) are presented as the mean ± SEM and are representative of at least three independent experiments. *p < 0.05, **p < 0.01, ***p < 0.001.  shown (d). Cytokine concentration in lung homogenates was measured by ELISA, and Nos2 mRNA levels in the lungs were measured by qRT-PCR (e). Data in (a-c) and (e) are presented as mean ± SEM and are representative of at least two independent experiments. *p < 0.05, **p < 0.01, ***p < 0.001.