Controlled Expression of Branch-forming Mannosyltransferase Is Critical for Mycobacterial Lipoarabinomannan Biosynthesis

Lipomannan (LM) and lipoarabinomannan (LAM) are phosphatidylinositol-anchored glycans present in the mycobacterial cell wall. In Mycobacterium smegmatis, the mannan core of LM/LAM constitutes a linear chain of 20–25 α1,6-mannoses elaborated by 8–9 α1,2-monomannose side branches. At least two α1,6-mannosyltransferases mediate the linear mannose chain elongation, and one branching α1,2-mannosyltransferase (encoded by MSMEG_4247) transfers monomannose branches. An MSMEG_4247 deletion mutant accumulates branchless LAM and interestingly fails to accumulate LM, suggesting an unexpected role of mannose branching for LM synthesis or maintenance. To understand the roles of MSMEG_4247-mediated branching more clearly, we analyzed the MSMEG_4247 deletion mutant in detail. Our study showed that the deletion mutant restored the synthesis of wild-type LM and LAM upon the expression of MSMEG_4247 at wild-type levels. In striking contrast, overexpression of MSMEG_4247 resulted in the accumulation of dwarfed LM/LAM, although monomannose branching was restored. The dwarfed LAM carried a mannan chain less than half the length of wild-type LAM and was elaborated by an arabinan that was about 4 times smaller. Induced overexpression of an elongating α1,6-mannosyltransferase competed with the overexpressed branching enzyme, alleviating the dwarfing effect of the branching enzyme. In wild-type cells, LM and LAM decreased in quantity in the stationary phase, and the expression levels of branching and elongating mannosyltransferases were reduced in concert, presumably to avoid producing abnormal LM/LAM. These data suggest that the coordinated expressions of branching and elongating mannosyltransferases are critical for mannan backbone elongation.

Mycobacterium tuberculosis is an etiologic agent of tuberculosis, an infectious disease that remains a global problem. Glycoconjugates from the mycobacterial cell wall are involved in pathogenesis and immune modulation. In particular, phosphatidylinositol mannosides (PIMs), 4 lipomannan (LM), and lipoarabinomannan (LAM) form a class of glycoconjugates found in all species of mycobacteria, including pathogenic M. tuberculosis and saprophytic and experimentally tractable Mycobacterium smegmatis, and are known to have potent immunomodulatory activities (1)(2)(3). PIMs are anchored to the plasma membrane by a phosphatidylinositol (PI) and carry two or six mannoses, which are directly linked to the 2-OH and 6-OH of the inositol residue (4, 5) (Fig. 1A). Monomannose attached to the 2-OH of inositol is further modified by a fatty acid, making triacylated PIMs the major lipid species. The 3-OH of inositol can be further modified by a fatty acid to become a tetra-acylated species (6). LM carries a much longer chain of mannoses. For example, in M. smegmatis, 20 -25 ␣1,6-mannose residues were thought to form a linear chain, which is elaborated by 8 -9 ␣1,2-mannose monomer branches (7). A more recent report estimated M. smegmatis LM to carry 21-34 mannose residues (8), highlighting even greater heterogeneity of this molecule. LAM is an arabinosylated LM, in which ϳ70 residues of D-arabinofuranose form arabinan(s) comprising ␣1,5 linear stretches linked by multiple ␣1,3 branch points (9,10). It is not known if a single molecule of LAM carries a single large arabinan moiety or multiple smaller arabinans.
PIMs, LM, and LAM all contain PI as a common lipid core moiety and are thought to be biosynthesized in a sequential order. PIMs are synthesized from a PI by sequential additions of mannoses, resulting in dimannosyl or hexamannosyl PIMs (6,(11)(12)(13). PimA and PimBЈ mediate the transfers of first two mannoses in a GDP-mannose-dependent manner, and an acyltransferase, encoded by MSMEG_2934 in M. smegmatis, adds one fatty acid onto a mannose residue to produce a dimannosyl species, AcPIM2 (14 -16). Then, an additional four mannose residues are transferred onto AcPIM2, producing AcPIM6.
Polyprenol-phosphate-mannose-dependent mannosyltransferases are thought to be involved in these later reactions (11,13), but only the fifth mannosyltransferase, PimE, has so far been identified (17). AcPIM4 has been thought to be the branching point from which LM/LAM biosynthesis diverges from the PIM biosynthetic pathway, and a lipoprotein, LpqW, is thought to be involved in controlling the metabolic flux at this branch point (18,19). This branch point is structurally viable because the terminal two mannoses of AcPIM6 are ␣1,2-linked, whereas the LM/LAM mannan chain is ␣1,6-linked. LM is believed to be the precursor of LAM, but the precursor-product relationship has not been experimentally proven.
A number of mannosyltransferases involved in the PIM/ LM/LAM biosynthetic pathway have now been identified, but it remains largely unknown how these enzymes coordinate and control the complex and overlapping pathways of PIM/LM/LAM synthesis. As part of our study of the pimE gene (17), we generated an MSMEG_4247 gene deletion mutant. Here we report our analyses on the ⌬MSMEG_4247 mutant, revealing that controlling expression levels of MSMEG_4247 is critical for the proper synthesis of LM and LAM.
Gene Cloning and Complementation-A fragment containing the MSMEG_4247 gene was PCR-amplified (primers 053/ 054; supplemental Table S1) from M. smegmatis genomic DNA and cloned into the MscI and EcoRI sites of the episomal expression vector, pHBJ334 (25), driven by a strong promoter of a mycobacterial heat shock protein hsp60 (Phsp60), resulting in pYAB143. To create an integrative vector, an XbaI/SpeI fragment of pYAB143, containing Phsp60, the MSMEG_4247 gene, and the streptomycin resistance cassette, was blunt end-ligated with an SpeI/Eam1105I fragment of pMV361 (26), containing the attachment site and integrase gene, resulting in pYAB243. Site-directed mutagenesis was performed as described previously (17), using pYAB143 or pYAB250 as a template and primers 320/321, resulting in pYAB254 or pYAB255. Hygromycinresistant versions of the expression vectors were constructed as follows: pYAB251 by replacing the streptomycin resistance cassette of pHBJ334, pYAB250 by replacing the streptomycin resistance cassette of pYAB143, and pYAB184 by replacing the kanamycin resistance cassette of pMV306 (26). For an expression vector driven by the endogenous promoter, the MSMEG_4247 gene and 200 bp of upstream flanking sequence was PCR-amplified (primers 312/313) and ligated into the ClaI/XbaI sites of pYAB184, resulting in pYAB247. For the acetamide-inducible system, the MSMEG_4241 gene was PCR-amplified (primers 329/ 330) and ligated into the BamHI/XbaI sites of pJAM2 (27), resulting in pYAB262. To make an acetamide-inducible MSMEG_4247 expression vector (pYAB246), the kanamycin resistance cassette of pJAM2 was first replaced by a streptomycin resistance cassette, resulting in pYAB040, and the HindIII/ Bpu1102I fragment of pYAB143 containing the MSMEG_4247 gene was blunt end-ligated into the BamHI site of pYAB040. These constructs were transfected into M. smegmatis wild-type strain mc 2 155 and mutant cells by electroporation.
Lipid Extraction and Analysis-The cell wall lipids were extracted from the cell pellets as described previously (17). Extracted lipids were analyzed by high performance thin layer chromatography (HPTLC) using chloroform, methanol, 13 M ammonia, 1 M ammonium acetate, water (180:140:9:9:23, v/v/ v/v/v) as a solvent. Orcinol or molybdenum blue staining was used to visualize PIMs or phospholipids, respectively.
LM and LAM Extraction and Analysis-After lipid extraction, the delipidated pellets were resuspended in five volumes of phenol saturated with Tris-EDTA buffer (pH 6.6) (Nacalai Tesque, Kyoto, Japan) and five volumes of water and extracted for 2 h at 55°C. The extract was further purified by dialysis or octyl-Sepharose column chromatography (GE Healthcare) and electroelution (model 422 Electro-Eluter, Bio-Rad) from an SDS-polyacrylamide gel where appropriate. LM/LAM were separated by SDS-PAGE (10 -20% gradient gel) and visualized using the ProQ Emerald 488 carbohydrate staining kit (Molecular Probes) and a fluoroimage analyzer (FLA-5000, Fujifilm).
Metabolic Labeling of ⌬MSMEG_4247-Cells were washed and resuspended in a fresh aliquot of Middlebrook 7H9 broth at 0.2 g/ml pellet. Cells were radiolabeled with 50 Ci/ml D-[2-3 H]mannose (20 Ci/mmol; American Radiolabeled Chemicals) for 15 min, washed in 7H9 broth to remove unincorporated radioactive mannose, and chased in 7H9 broth containing 1 mM mannose. Extracted LM/LAM were separated by SDS-PAGE and visualized by fluorography.
Mannosidase Treatment-LM/LAM fractions, mixed with PIMs as internal controls, were treated with or without 0.4 milliunits of Aspergillus saitoi ␣1,2-mannosidase (Prozyme) for 24 h at 37°C. The treatment was repeated three times. LM/LAM were repurified by phenol and analyzed by SDS-PAGE as described above. Intensity profiles were obtained by the Multi Gauge software (Fujifilm). PIMs were purified by 1-butanol/water (2:1, v/v) partitioning and analyzed by HPTLC.
High Performance Anion Exchange Chromatography (HPAEC)-Acetolysis was performed following a published protocol (28). Monosaccharides were released by 2 M trifluoroacetic acid at 100°C for 2 h. Released carbohydrates were analyzed by a Nanospace high performance liquid chromatography system (Shiseido, Tokyo, Japan) equipped with Screbead II (2.0 ϫ 250 mm) (Shiseido) as an anion exchange column, 50 mM sodium acetate in 0.1 M NaOH as a mobile phase, and a pulsed amperometric detector.
Western Blotting-Rabbit antibodies were raised against a mixture of two peptides from MSMEG_4241 (MTPTETHK-PNPGLAEHVC and CRAPESAEATASRQ), MSMEG_4247 (RTHTGDAHETDEPLVPLC and MSKRQSPRGAGLAPC), or PimBЈ (CEHLPPGVDTDRFAPDPD and CGARLAELLSGRR-EARQA) and affinity-purified. Cell lysates were prepared by bead beating as described previously (17). Proteins were fractionated by SDS-PAGE (10 -20% gradient gel) under reducing conditions and blotted onto a polyvinylidene difluoride membrane (Millipore). The membrane was blocked with 5% skim milk and then incubated with a primary antibody (rabbit anti-MSMEG_4247, anti-MSMEG_4241, or anti-PimBЈ antibodies; 1 g/ml) for 1 h and a secondary antibody (anti-rabbit IgG antibody, horseradish peroxidase-conjugated; 1:2000 dilution; Amersham Biosciences) for 1 h. The bound probe was visualized by chemiluminescence (PerkinElmer Life Sciences). Images were captured by a luminescent image analyzer (LAS-4000, Fujifilm) and quantified by the Multi Gauge software.
Acetamide-induced Expression and Metabolic Labeling-Cells were grown in Middlebrook 7H9 medium, and 0.2% acetamide was added at midlog phase. After induction, the cells were washed in Sauton's medium (29), resuspended at 0.2 g/ml pellet, pulse-labeled with D-[2-3 H]mannose (24.7 Ci/mmol; PerkinElmer Life Sciences) for 15 min, and diluted 50-fold in Middlebrook 7H9 broth containing 1 mM mannose for chase. Extracted LM/LAM were separated by SDS-PAGE and visualized by fluorography.
Subcellular Fractionation-Cells were lysed by nitrogen cavitation, and the crude lysate was separated by a sucrose density gradient (25-56%) as described previously (30). Each fraction was tested for protein concentration, PIM biosynthetic activities (30), and Western blot detection of MSMEG_4247, MSMEG_4241, and PimBЈ.

Complementation of MSMEG_4247-deficient Mutant with MSMEG_4247 Does Not Restore the Synthesis of Normal LM
and LAM-We deleted the MSMEG_4247 gene by homologous recombination and confirmed the gene deletion by Southern blotting (supplemental Fig. S1, A and B). We extracted LM and LAM from the deletion mutant and analyzed them by SDS-PAGE. We found even greater accumulation of LAM in the ⌬MSMEG_4247 mutant, but the mutant LAM appeared slightly smaller than that from wild type. More striking, the mutant lacked LM completely ( Fig. 1B; wild type (lanes 1-3) and ⌬MSMEG_4247 (lanes 4 -6)), as reported recently (see Fig. 2 of Ref. 22). The complete disappearance of LM was unexpected because the lack of the ␣1,2 side chain should still leave the ␣1,6 backbone intact. We considered the possibility that LM was synthesized but could not accumulate. To test this possibility, we radiolabeled the ⌬MSMEG_4247 mutant metabolically in a pulse-chase experiment using [ 3 H]mannose. The mutant cells produced a weakly radiolabeled species (Fig. 1C, asterisk), which had an electrophoretic mobility similar to that of wild-type LM. The radiolabeled LM in the wild-type cells remained during the 5-h chase period, whereas the mutant LM completely disappeared within 5 h, suggesting that LM could only accumulate transiently in the mutant. We next examined whether these phenotypes were rescued by the introduction of the MSMEG_4247 expression vector. We used an episomal expression vector, in which a strong hsp60 promoter (Phsp60) regulates the expression of MSMEG_4247 (26). Strikingly, the rescued mutant (⌬MSMEG_4247ϩ Phsp60 MSMEG_ 4247) produced LM and LAM, but they were both smaller than those produced by the wild type (Fig. 1B,  lanes 7 and 8).
Expression Level of MSMEG_ 4247 Is Critical for Controlling the Sizes of LM and LAM Produced-We considered the possibility that LM/ LAM synthesis was prematurely terminated because of the high expression level of MSMEG_4247. We found that ⌬MSMEG_4247ϩ Phsp60 MSMEG_4247 expressed MSMEG_4247 at Ͼ700 times the levels of the wild-type ( Fig. 2A, compare lanes 1 and lanes 3 and 4; note that the protein loading is 5 times less in lanes 3 and 4). To achieve a more controlled expression of MSMEG_4247, the ⌬MSMEG_ 4247 mutant was transfected with a vector carrying MSMEG_4247 with its own upstream region instead of an artificial promoter. This vector lacks the mycobacterial origin of replication and instead carries the attachment site and the integrase gene from the mycobacteriophage L5, allowing for the site-specific integration of a single copy of the MSMEG_4247 gene (26). Western blotting confirmed that the expression levels of MSMEG_4247 in the complemented strains (⌬MSMEG_ 4247ϩ P4247 MSMEG_4247) were much closer to those of wild-type (4 -5-fold overexpression; Fig 3 and 4). These results suggest that the sizes of LM and LAM are normalized if the ⌬MSMEG_4247 mutant is complemented with a wild-type level of MSMEG_4247 expression.
MSMEG_4247 is an ␣1,2-mannosyltransferase involved in the transfer of the monomannose side chain. It was therefore perplexing that its overexpression resulted in smaller LM and LAM. To investigate whether the smaller sized LM and LAM produced in MSMEG_4247-overexpressing strains were modified by ␣1,2-mannose side chains, we removed the ␣1,2-mannose branches of LM by ␣1,2-specific mannosidase treatment and analyzed the size changes by SDS-PAGE (Fig.  2C). A shift of LM was detectable upon digestion of LM from both the wild-type and overexpressing strains. To confirm further, we acetolyzed the LM/LAM mixture from select strains and quantified the frequencies of ␣1,2-mannose branching. Acetolysis cleaves ␣1,6-mannose linkages without digesting ␣1,2-mannose linkages, producing ␣1,2-mannobiose and mannose. The ratio of ␣1,2-mannobiose to mannose indicates the ratio of mannoses with ␣1,2-mannose branches to those without. As shown in Fig. 2D, whereas the ⌬MSMEG_4247 mutant completely lacked the ␣1,2-mannose branch, LM/LAM from all complemented strains showed frequencies of ␣1,2-mannose branches comparable with those of the wild-type LM/LAM (i.e. ϳ45-50% of backbone mannoses were branched).
Despite the restoration of ␣1,2-mannose branching, LM and LAM from the ⌬MSMEG_4247ϩ Phsp60 MSMEG_4247 strain appeared small on SDS-PAGE. To determine which component of LAM was responsible for the SDS-PAGE mobility shift, we purified LAM from an SDS-polyacrylamide gel by electroelution and analyzed its sugar composition. As shown in Fig. 2E, the ⌬MSMEG_4247 mutant did not show significant changes in the sugar compositions of LAM despite the lack of ␣1,2mannose branches, suggesting that ␣1,6-mannose backbone is longer than that of wild-type LAM, and arabinosylation is occurring normally. The numbers of arabinose and mannose residues relative to inositol were somewhat smaller than previously reported (ϳ70 arabinose and 21-34 mannose residues; see Introduction). This might be due to the contribution of inositol from inositol phosphate capping, a minor modification of LAM arabinan in M. smegmatis and other fast growing mycobacteria (31,32). Strikingly, LAM from the ⌬MSMEG_ 4247ϩ Phsp60 MSMEG_4247 strain showed greatly reduced numbers of arabinose and mannose. The mannan core carried less than half the number of mannose residues found in wildtype LAM. Because the frequency of the side-chain mannose is not significantly different between the dwarfed and wild-type LAM, these data suggest that the length of mannan backbone is less than half the length of wild-type LAM. Furthermore, arabinan was found to be 4 times smaller. This gross dwarfing is apparently the cause of faster migration on the SDS-polyacrylamide gel, although the protein molecular weight standards do not precisely reflect the actual molecular weight shifts of LM/LAM (see also Ref. 33 for another example). Taken together, our data suggested that overexpression of MSMEG_ 4247 restored the branches of ␣1,2-monomannose but prematurely terminated the elongation of the ␣1,6-mannose backbone and dwarfed the arabinan size.
Next we investigated whether the effect of MSMEG_4247 overexpression is specific to the LM/LAM pathway. Because LM/LAM biosynthetic pathways are thought to diverge from PIM biosynthesis at the AcPIM4 intermediate (see Fig. 1A), we examined the lipid fractions to see if there are any changes in the PIM or phospholipid profiles. We found no significant differences in either PIMs or phospholipids (supplemental Fig.  S2), suggesting that the deletion or overexpression of MSMEG_4247 primarily affects LM/LAM biosynthesis.
Dwarfing LM/LAM Requires and Correlates with MSMEG_ 4247 Enzyme Activity-If the expression levels of MSMEG_ 4247 affect the elongation of the ␣1,6-mannose backbone, overexpression of MSMEG_4247 can affect the profile of LM and LAM in wild-type cells as well. To test this prediction, we introduced integrative or episomal MSMEG_4247 overexpression vectors and examined the effect of expression in the wildtype cells. Indeed, overexpression of MSMEG_4247 resulted in smaller LM and LAM in wild-type cells (Fig. 3A). A higher expression level was achieved using an episomal vector compared with an integrative vector. The hyperexpression in cells transfected with the episomal vector led to the production of LM and LAM, which appeared even smaller than the LM and LAM from cells transfected with the integrative vector (Fig. 3A, compare lane 2 with lanes 3 and 4). To examine whether the production of smaller LM and LAM was dependent on the enzyme activity of MSMEG_4247, we created an MSMEG_ 4247 expression vector, in which the gene carried a site-directed mutation at a conserved aspartic acid residue. The aspartic acid residue 45 was conserved among mycobacterial species and aligned with the aspartic acid residue of PimE that is essential for enzymatic activity (supplemental Fig. S3A) (17). The expression of D45A-mutated MSMEG_4247 could not alter the LM/LAM profile of the ⌬MSMEG_4247 mutant, suggesting the loss of enzyme activity (supplemental Fig. S3, B and C). When introduced in the wild-type cells, the D45A mutant of MSMEG_4247 did not show a shift in the migration patterns of wild-type LM and LAM (Fig. 3A, lanes 5 and 6). The lack of effect was not due to the lack of expression of MSMEG_4247 D45A mutant because it was expressed at a level comparable with that of wild-type MSMEG_4247 (Fig. 3B, compare lane 2 with lanes 5 and 6). These data further support our notion that the enzyme activity of MSMEG_4247 needs to be properly controlled to produce normal sized LM and LAM.
LM/LAM Biosynthesis Is Immediately Affected by Induced Expression of MSMEG_4247-Secondary mutations sometimes occur to compensate for an existing mutation in the genome, and secondary mutants with improved growth fitness can outgrow the original mutant. For example, deletion of the lpqW gene leads to spontaneous mutations in the pimE gene, resulting in an improved growth fitness of the mutant (18). We therefore wanted to exclude the possibility that the constitutive overexpression of MSMEG_4247 resulted in secondary mutations and that smaller LM/LAM was the consequence of these secondary mutations. To address this possibility, we created a vector, in which MSMEG_4247 expression was controlled by an acetamide-inducible promoter (27) and introduced this vector into wild-type cells. As shown in Fig. 4A, the acetamide-inducible promoter is leaky and expresses MSMEG_4247 at a level significantly higher than the wild-type level (compare lanes 1 and 3). Nevertheless, a much higher expression level was achieved after 4 h of acetamide induction (lane 6). At the 4-h time point, cells were metabolically pulse-labeled with [ 3 H]mannose and chased for 25 min in the presence of excess non-radioactive mannose (Fig. 4B). LM and LAM were relatively small in size immediately after pulse (lanes designated P). After the chase (lanes designated C), increased LM/LAM sizes were detected in control strains (compare lanes  5 and 6 or lanes 7 and 8). Compared with control strains, the sizes of LM and LAM after chase remained smaller in the cells carrying acetamide-inducible MSMEG_4247 even without acetamide induction due to leaky expression of MSMEG_4247 (compare lanes 2 and 6; white arrowheads (LAM) and white lines (LM) indicate the peaks of the intensity profile shown in supplemental Fig. S4). Upon induction of MSMEG_4247 expression, production of normal sized LAM became even less efficient (compare black arrowheads in lanes 2 and 4), dwarfed LAM became even smaller (compare white arrowheads in lanes 2 and 4), and dwarfed LM became greater in intensity and slightly smaller in size (compare white lines in lanes 2 and 4). These observations suggest less efficient elongation of LM and LAM, being similar to those seen under constitutive overexpression of MSMEG_4247 (see Fig. 3A). Thus, MSMEG_4247 can exert its effect on LM/LAM biosynthesis immediately, and the timing appears to be too fast to be explained by secondary mutations.
The Effect of Overexpressed MSMEG_4247 Can Be Alleviated by the Overexpression of MSMEG_4241-Our data suggested that overexpressed MSMEG_4247 dominates the elongating enzymes, preventing the efficient elongation of the linear ␣1,6mannose chain. We wanted to test if this effect of MSMEG_ 4247 overexpression can be competed by the overexpression of an elongating mannosyltransferase, MSMEG_4241. Therefore, we created another mutant in which an episomal vector for acetamide-inducible MSMEG_4241 expression was transfected into the MSMEG_4247 overexpressing strain (used in lane 3 of Fig. 3). MSMEG_4241 expression was maximized at 4 h after the addition of acetamide (Fig. 5A)   LAM also accumulated over time (Fig. 5B). When MSMEG_4241 overexpression was induced, we found little accumulation of dwarfed LM and identified a radiolabeled species migrating at around 35 kDa, which was comparable with wild-type LAM (Fig. 5B). The synthesis of normal size LM (ϳ20 kDa) was less prominent, which might be due to the relative overexpression of MSMEG_4241. Next, we examined if the biosynthetic changes induced by the overexpression of MSMEG_ 4241 can lead into changes in cell wall composition of LM/LAM. Without acetamide induction, dwarfed LM and dwarfed LAM persisted over the 12-h time course (Fig. 5C, left). Being consistent with the metabolic labeling experiments, a species comparable with wild-type LAM started to appear at around 35 kDa after the 4-h induction and accumulated over the 12-h induction (Fig. 5C,  right). In contrast to metabolic labeling, dwarfed LM persisted during the time course, suggesting a slow turnover rate of this species. We purified dwarfed LAM (from 0-h induction) and LAM-like species (from 12-h induction) by electroelution (Fig. 5D) and examined the sugar composition (Fig. 5E). The LAM-like species carried a mannan moiety with a size comparable with the wild-type LAM (see Fig. 2E for comparison). In contrast, arabinan size was only partially restored, suggesting that the rate of arabinan biosynthesis falls short of that of mannan biosynthesis when both elongating and branching mannosyltransferases are overexpressed in wild-type cells. These data together suggest that the dwarfing effect of overexpressed MSMEG_4247 can be partially overridden by MSMEG_ 4241 overexpression.

Spatial and Temporal Control of MSMEG_4241 and MSMEG_4247 Expressions in Wild-type Cells-
The above results suggested that MSMEG_4247 and MSMEG_4241 function in close collaboration with each other during the synthesis of LM/LAM. Therefore, we predicted that these two enzymes are located in the same domain of the plasma membrane. We tested this possibility using subcellular fractions of wild-type M. smegmatis (Fig. 6A). M. smegmatis has two functionally distinct membrane fractions designated PMf and PM-CW (30). PMf fractions contain membrane vesicles free of cell wall components, whereas PM-CW fractions represent plasma membranes tightly associated with the cell wall. PMf and PM-CW membranes can be separated by a sucrose density gradient sedimentation and are responsible for distinct biochemical reactions, such as the early (up to AcPIM2) and late (up to AcPIM6) parts of the PIM biosynthetic pathway, respectively (30) (Fig. 6B). As expected from the enriched PIM2 biosynthetic activities in PMf fractions (Fig. 6B), PimBЈ was specifically localized to these fractions (Fig. 6C). In contrast, MSMEG_4247 and MSMEG_4241 were both enriched in PM-CW fractions, suggesting that LM/LAM biosynthesis takes place in PM-CW fractions. These data are consistent with MSMEG_4247 and MSMEG_4241 acting in proximity to coordinate the LM/LAM biosynthesis. We have also tested co-immunoprecipitation of these two proteins to show their physical interaction, but it was not successful (data not shown; also see "Discussion"). Another interesting feature of LM/LAM biosynthesis is that it appears to be down-regulated in the late stationary phase (17) (Fig. 6D). Because imbalanced expressions of MSMEG_4247 and MSMEG_4241 can result in the production of aberrant forms of LM/LAM, we thought it important for the cells to down-regulate the expression levels of both enzymes in a concerted manner. Indeed, the expression levels of these two proteins were reduced simultaneously in the stationary phase (Fig. 6E), in correlation with the reducing levels of LM/LAM. Taken together, MSMEG_4247 and MSMEG_4241 were expressed in a temporally and spatially controlled manner.
Rv2181, an Ortholog of MSMEG_4247, Functions Similarly in M. tuberculosis-To examine whether the phenotypic characteristics of the ⌬MSMEG_4247 mutant were also displayed by other mycobacterial species, we generated an M. tuberculosis mutant in which the Rv2181 gene, the MSMEG_4247 ortholog, was deleted. The mutant was successfully generated (supplemental Fig. S5, A and B), and the ⌬Rv2181 mutant showed a smaller sized LAM and disappearance of LM (Fig. 7, lane 2), similar to the M. smegmatis mutant. When the Rv2181 gene was introduced with its endogenous promoter region, LM synthesis was restored, and the size of LAM was also restored to that of wild-type LAM (lanes 3 and 4). When Rv2181 was introduced via an integrative vector with Phsp60, the size of LM appeared normal, but the size of LAM appeared slightly smaller ( lanes 5 and 6). Although the phenotypes are less prominent than those in M. smegmatis, these data highlight the conserved role of the ␣1,2-mannosyltransferase in producing properly sized LM and LAM.

DISCUSSION
Our analysis on MSMEG_4247, a branch-forming ␣1,2mannosyltransferase involved in LM/LAM synthesis, revealed unexpected effects of its expression levels on the sizes and  quantities of LM/LAM (Fig. 8). Overexpression of MSMEG_ 4247 resulted in the production of dwarfed LM/LAM. In contrast, the deletion of MSMEG_4247 resulted in the disappearance of LM and accumulation of branchless LAM. We suggest that overexpression of the branching mannosyltransferase results in premature chain termination, whereas in the absence of branching, transiently synthesized branchless LM is either converted to LAM or degraded. Below, we discuss possible control mechanisms of mannan chain length and why branchless LM cannot be stably maintained.
Most important, the following two pieces of evidence suggested that MSMEG_4247 affects the sizes of LM/LAM primarily by competing with the elongating ␣1,6-mannosyltransferase. First, we showed that 15-fold or more overexpression of MSMEG_4247 resulted in short ␣1,6-mannose backbone (see Figs. 2 and 3). Second, overexpression of elongating ␣1,6-mannosyltransferase (MSMEG_4241) alleviated the dwarfing effect of MSMEG_4247 overexpression (see Fig.  5). Dwarfed LM/LAM are unlikely to be caused by substrate depletion because MSMEG_4241 overexpression can at least partially override the dwarfing effect of MSMEG_4247 overexpression. It is tempting to speculate that MSMEG_4241 and MSMEG_4247 compete for the growing non-reducing terminus of mannan polymer, and MSMEG_4247-mediated ␣1,2mannosylation of the non-reducing end blocks further elongation. Future studies using a cell-free assay system and synthetic substrates will address these possibilities.
Interestingly, overexpression of MSMEG_4247 resulted in the dwarfing of not only the mannan chain but the arabinan moiety of LAM (see Fig. 2E). Because both MSMEG_ 4241 and MSMEG_4247 are mannosyltransferases, it is unclear how the expression levels of these enzymes affect the arabinan size. One possibility is that multiple arabinans are attached to a single LAM molecule, and a short mannan chain limits the sites available for arabinosylation. In another experiment, we showed that overexpressed MSMEG_4241 competed with overexpressed MSMEG_4247 in terms of the LAM mannan size, but the restoration of the arabinan size was only partial (see Fig. 5). We speculate that relatively low (wild-type) expression levels of arabinosyltransferases compared with overexpressed mannosyltransferases have led to the reduction in arabinan size. Based on these observations, it is conceivable that the expression levels of arabinosyltransferases are also controlled in concert with the two mannosyltransferases to coordinate LAM synthesis in wild-type cells.
Identification and characterization of the arabinosyltransferase that initiates the arabinan synthesis will provide further mechanistic insight into LAM biosynthesis.   several possibilities for why branchless LM cannot accumulate. First, in the absence of branching-dependent chain termination, MSMEG_4241 may be incapable of terminating the mannan polymerization by itself. We then speculate that arabinosylation acts as an alternative termination signal, thus resulting in the production of LAM only. Second, contrary to the first possibility, MSMEG_4241 may be able to terminate the polymer elongation after reaching a certain chain length, producing branchless LM. However, branchless LM cannot accumulate stably, perhaps because it is either efficiently converted to LAM or degraded. These possibilities are not exclusive to each other, and in either case, there appears to be an alternative chain termination mechanism in addition to mannose branching to control the mannan chain length. We consistently observed more enriched accumulation of branchless LAM in the ⌬MSMEG_ 4247 mutant compared with the wild-type levels. Although the precursor-product relationship of LM and LAM is not established, this observation favors the possibility that branchless LM intermediate is more avidly converted to LAM rather than being degraded. Curiously, although our M. tuberculosis ⌬Rv2181 mutant showed a complete lack of LM, another recently reported ⌬Rv2181 mutant accumulated small size LM (23). Further studies are necessary to identify factors that have led these two ⌬Rv2181 mutants into different phenotypes.
At least a fraction of LAM synthesized in the plasma membrane appears to be transported to the surface layer of the cell wall (34), suggesting a dynamic spatial control of LAM biosynthesis. We demonstrated that both MSMEG_4247 and MSMEG_ 4241 are topologically confined in a membrane domain known as PM-CW (see Fig. 6, A-C), suggesting that LAM is synthesized in a spatially controlled manner, and these two enzymes may function in topological proximity. If MSMEG_4247 and MSMEG_4241 form a heterodimer, overexpression of kinetically inactive MSMEG_4247 may interfere with the formation of a functional dimer in the wild-type cells, leading to a phenotype similar to ⌬MSMEG_4247 mutant. However, such dominant negative effects were not observed (see Fig. 3A), making heterodimer formation less likely. We do not know if the levels of overexpression achieved in this study occur in physiological situations. Nevertheless, our results suggest that the expression levels of MSMEG_4247 and MSMEG_4241 need to be controlled in concert to avoid the production of aberrant LM/LAM. Being consistent with this suggestion, disappearance of LM/LAM in the stationary phase was closely correlated with concerted down-regulation of both enzymes (see Fig. 6, D and E). Interestingly, truncated structural variants of LAM have been identified in Mycobacterium leprae and clinical isolates of M. tuberculosis (33), implying that the polymer length of LM/LAM can vary among species/strains and may be controlled in response to environmental stimuli. Controlling the expression levels of MSMEG_4247 and MSMEG_4241 in M. smegmatis or their orthologs in other species may represent a key feature of the polymer length control of mycobacterial LM/LAM. Understanding the mechanisms of transcriptional and post-transcriptional controls of MSMEG_4247 and MSMEG_4241 expressions may provide further clues to understand the regulatory mechanisms of LM/LAM biosynthesis.