T Cell Responses against Mycobacterial Lipids and Proteins Are Poorly Correlated in South African Adolescents

Mycobacterium tuberculosis is a pathogen of global importance that infects >1 billion people and causes >1 million deaths annually (1). Several lines of evidence in human studies and animal challenge models underscore the importance of T cells in controlling M. tuberculosis infection (2–5). T cells recognize both peptide and nonpeptide Ags produced by mycobacteria, so the potential catalog of Ags mediating protective immunity extends beyond the proteome of M. tuberculosis (6–8). Nonpeptide Ags are fundamentally different from peptide Ags in their chemical structure, subcellular location within the pathogen, and pathways by which they are processed and presented to T cells. Thus, one hypothesis is that the adaptive immune system evolved the ability to recognize nonpeptide Ags to diversify the T cell response to infection.

T cells recognize mycobacterial cell wall lipids bound to CD1 proteins, which are homologous to MHC class I but are functionally nonpolymorphic (9). The human CD1 locus codes for four proteins (CD1a, CD1b, CD1c, and CD1d) that are expressed at the cell surface and are capable of presenting lipid Ags to T cells. At least eight cell wall lipids have been identified as CD1 Ags for human T cells. Five of these lipids are presented by CD1b: mycolic acid (MA), glucose monomycolate (GMM), glycerol monomycolate (GroMM), diacylated sulfoglycolipids, and phosphatidyl myoinositol mannosides (PIMs) (6, 10–13). Information regarding the frequency and phenotype of CD1b-restricted T cells targeting mycobacterial glycolipids is limited. CD1b tetramers were used to estimate the frequency of GMM-specific T cells to be 0.01% in four subjects with positive purified protein derivative (PPD) skin tests (14). T cell clones that recognize GMM using a semi-invariant TCR express CD4 (15). However, other CD1b-reactive T cells with lower affinity for GMM express CD8 or are CD4⁻CD8⁻ (10, 15, 16). A small cross-sectional study revealed that proliferative T cell responses to GMM were present in both PPD-positive and PPD-negative study subjects (17). By contrast, lymphocytes from PPD-positive but not PPD-negative subjects produced IFN-γ in response to mycobacterial sulfoglycolipid and GroMM (11, 12). More recently, it was demonstrated that MA induced IFN-γ production from lymphocytes in patients with tuberculosis but not healthy donors and that these responses decreased while receiving antituberculous therapy (18). Collectively, these studies provide clear evidence that populations of CD1b-restricted T cells targeting mycobacterial glycolipids are expanded in vivo, but no study has systematically assessed the phenotype of these T cells ex vivo or examined how lipid-specific T cell responses compared with T cell responses to protein Ags with regard to magnitude or timing.

Ex vivo analysis of CD1b-restricted T cells has been hampered by the lack of specific surface markers and challenges inherent to identifying and cloning cells in vitro. Further, ex vivo analysis typically requires the generation of autologous dendritic cells (DCs) to facilitate Ag presentation, and this approach carries inherent difficulties. First, the Ag-presenting molecule on DCs is difficult to define because DCs simultaneously express CD1a, CD1b, CD1c, and CD1d. Second, trace amounts of contaminating peptides in the lipid preparations would be efficiently presented to MHC-restricted T cells, confounding interpretations of whether responses are due to lipids or peptides. Third, and most importantly, generating DCs is a time and reagent intensive process that renders the use of cryopreserved PBMCs virtually impossible. Thus, a major barrier in the field is the lack of an activation-based assay that would enable large human cohort studies of lipid-specific T cells.

To enable large scale ex vivo studies of CD1b-restricted T cells in human populations, we took advantage of an assay using K562 cells, which are a human myelogenous leukemic cell line that expresses very low levels of MHC class I and class II and does not express CD1. Thus, these cells do not efficiently elicit allogeneic T cell responses (19). When stably transfected with single isoforms of human CD1 proteins, these cells are capable of lipid Ag presentation to T cell clones and T cell lines derived after long-term and short-term in vitro culture, respectively (14, 19). We modified this assay to enable the detection of 64 cytokine-defined CD1b-restricted CD4⁺, CD8⁺, and CD4⁻CD8⁻ T cells and conducted a cross-sectional study of ex vivo T cell responses to five mycobacterial glycolipids in South African adolescents. We integrated this assay with a novel analytic platform, providing a comprehensive study of human T cells targeting pathogen-derived lipids. Our findings reveal a previously unappreciated functional diversity among circulating CD1-restricted T cells that are not redundant with T cell responses to proteins. These results support the premise that the CD1 Ag presentation system evolved to complement functions performed by the MHC locus, ultimately leading to a diverse T cell response to mycobacterial infection.

As recently detailed, 6363 adolescents were enrolled into a study that aimed to determine the incidence and prevalence of tuberculosis infection and disease (20). Twelve- to eighteen-year-old adolescents were enrolled at 11 high schools in the Worcester region of the Western Cape of South Africa. Subjects were screened for the presence of latent tuberculosis by a tuberculin skin test and QuantiFERON-TB GOLD In-Tube (Cellestis) at study entry. Subjects were all healthy without signs or symptoms of active tuberculosis. After matching for age and gender and ensuring sufficient PBMC for the planned studies, we selected 19 M. tuberculosis–infected and 22 M. tuberculosis–uninfected subjects for intensive immunologic study (Table I). PBMCs from 34 (83%) of 41 subjects underwent both lipid and peptide T cell profiling, whereas the remaining 7 were used for one or the other assay. Data from one M. tuberculosis–uninfected subject were dropped prior to analysis because of documented technical problems with the data.

The study was approved by the Institutional Review Board of the University of Washington and the University of Cape Town. Written informed consent was obtained from the parents of the adolescents, and assent was obtained from the adolescents for the parent study.

MA, GMM, GroMM, and PIMs were purified from mycobacterial cell wall extracts as previously described (6, 10, 12, 21). Diacylated sulfotrehalose (SL37) was synthesized as an analog of a purified diacylated sulfoglycolipid Ag (22). We generated peptide pools of 15 aa overlapping by 12 for 5 immunodominant proteins secreted by M. tuberculosis (Ag85A, Ag85B, CFP-10, ESAT-6, and TB10.4). For the analysis of CMV-specific T cell responses, cryopreserved PBMCs from healthy donors known to be high responders to CMV were stimulated with a peptide pool targeting pp65 consisting of 138 peptides 15 aa in length overlapping by 11 peptides.

K562 cells stably transfected with empty vector (K562-EV) or CD1b (K562-CD1b) (19) were maintained in RPMI 1640 (Life Technologies) supplemented with 10% FCS and G418 (Sigma-Aldrich) at a concentration of 200 μg/ml and periodically assessed for CD1b expression by flow cytometry.

On day 0, cryopreserved PBMCs were thawed, counted, and assessed for viability assessed by Guava easyCyte (EMD Millipore). PBMCs were rested in RPMI 1640 plus 10% FCS overnight and reassessed again by Guava. After resting, if there were not enough cells to test five Ags or if viability was <70%, the sample was discarded.

Lipids were evaporated to dryness from chloroform-based solvents under a sterile nitrogen stream and then sonicated into media. This lipid suspension was added to 50,000 K562-EV or K562-CD1b cells at final concentrations that were previously determined to be greater than those yielding maximal T cell responses: 5 μg/ml for MA and GMM and 10 μg/ml for GroMM, PIM, and SL37 (6, 10–12, 21). K562 cells were incubated for 18 h at 37°C, 5% CO₂ to facilitate lipid loading. The following day, rested PBMCs were split and added to the K562 cells without washing at a final density of 1 million/well. Thus, each subject’s PBMCs were tested against 10 conditions. The cell mixture was allowed to incubate for 6 h in the presence of anti-CD28/Ly29d Abs (BD Biosciences), brefeldin A at a final concentration of 10 μg/ml (Sigma-Aldrich), and GolgiStop containing Monensin (BD Biosciences), after which EDTA, at a final concentration of 2 mmol, was added to disaggregate cells. Plates were stored at 4°C until the following day when they were stained and acquired by FACS in duplicate. A total of 35 study subjects were analyzed in 9 batches in which the number of samples from M. tuberculosis–infected and M. tuberculosis–uninfected subjects were matched.

For peptide stimulations, PBMCs were thawed and rested as above. The following day, PBMC were distributed at a final density of 1 million cells/well, and peptide pools dissolved in DMSO were added in single wells at a final concentration of 1 μg/ml per peptide. A negative control containing only DMSO was also plated. Cells were allowed to incubate with peptide for 6 h in the presence of anti-CD28/Ly29d, brefeldin A, and GolgiStop as above (BD Biosciences). A total of 40 study subjects were analyzed in 5 batches in which the number of M. tuberculosis–infected and M. tuberculosis–uninfected samples were matched.

The T cell clone LDN5 was maintained in culture through a combination of rapid expansion and Ag stimulation, as previously described (10). Polyclonal T cells were purified from total PBMCs obtained from a healthy donor by removing cells expressing CD14, CD16, CD19, CD36, CD56, CD123, and Glycophorin A using the Pan T Cell Isolation Kit II (Miltenyi Biotec).

We have previously published an optimized and validated 12-color intracellular cytokine staining assay to quantify Ag-specific T cells after vaccination (23, 24). This panel was modified to include anti–IL-17a and anti–IL-22 instead of anti–MIP-1β and anti-CD107a (Supplemental Table I). Briefly, cells were first stained with Avid Live/Dead (Life Technologies) viability dye and anti-CD14. After washing, the cells were permeabilized with FACS Perm II (BD Biosciences) and stained for the remaining markers (CD3, CD4, CD8, CD154, IFN-γ, TNF-α, IL-2, IL-4, IL-17a, and IL-22). Fully stained cells were washed and resuspended in 1% paraformaldehyde. Data were acquired on an LSR II (BD Biosciences) equipped with a high-throughput sampler and configured with violet (405 nm, 50 mW), blue (488 nm, 20 mW), green (561 nm, 50 mW), and red (641 nm, 150 mW) lasers. Raw data were manually gated in FlowJo (Tree Star). The raw FCS files and manual gates were imported into the R environment using the OpenCyto framework for data visualization and downstream analysis (25). For some of the assays with lipid Ags, we noted tandem dye breakdown within allophycocyanin-Cy7 CD4, leading to spurious signal within the allophycocyanin IL-4 gate, so this cytokine was not included in downstream analysis. Even though this artifact was not observed with peptide Ags, we removed IL-4 signal from peptide data to maintain consistency.

Cell counts were analyzed using COMbinatorial Polyfunctionality Analysis of Antigen-Specific T cell Subsets (COMPASS) (26). COMPASS is a novel computational approach developed for high-dimensional flow cytometry data analysis that can be used to detect changes across all observable functional T cell subsets, without the need to limit our analysis to very specific subsets based on expected biological significance. In COMPASS, responses are quantified using posterior probabilities that summarize for each subject and subset the evidence that the corresponding response is Ag specific by comparing the proportion of cytokine-positive cells in the Ag sample to the corresponding proportion in the control sample. Notably, COMPASS only reports the probability of detecting T cell responses with a particular functional profile rather than the frequency, which can be separately calculated. For a given subject, COMPASS can also be used to compute a functionality score that summarize the entire functionality profile into a single number. In this study, we have applied COMPASS to each of the 5 lipids in 3 (CD4⁺, CD8⁺, and CD4⁻CD8⁻) T cell subsets and 5 peptide pools in the CD4⁺ T cell subset, leading to 20 analyses. Each one of the analyses was unbiased and considered all of the 64 possible cytokine functions (defined as Boolean combination). Magnitudes of T cell responses were calculated independent of COMPASS as the maximum of zero or the proportion of gated events in the stimulated condition minus the proportion of gated events in the unstimulated condition.

To determine if lipid-specific T cells present at low frequency could be detected with sufficient sensitivity, we tested the ability of both K562 cells and DCs to activate LDN5, a T cell clone specific for GMM, at limiting concentrations. Using IFN-γ ELISPOT, we found that DCs and K562-CD1b cells were able to detect activated LDN5 T cells at a precursor frequency as low as 1:10,000 (Fig. 1A, 1B). Notably, background responses seen in the presence of DCs were lower or absent when K562 cells were used as APCs (Fig. 1C and data not shown). As a negative control, we then mixed K562 cells with PBMCs from healthy individuals known to have high frequency responses to the immunodominant CMV protein pp65. Even at an E:T ratio of 1:5, there was no effect on responding CD4 or CD8 T cells (data not shown). Thus, the mere presence of K562 cells does not lead to nonspecific secondary stimulation of T cells. We modified this assay to enable the analysis of cryopreserved PBMC by multiparameter flow cytometry (Fig. 1D, Supplemental Fig. 1). We did not find systematic evidence of NK cell activation as measured by IFN-γ production by CD3⁻ cells, nor did we find evidence of secondary stimulation of T cells by mock-transfected K562 cells after 6 h (data not shown). These data validate a sensitive and specific method of using CD1-transfected K562 cells to facilitate the detection of lipid-specific T cells within a polyclonal mixture of T cells.

We used this assay to perform a cross-sectional analysis of T cell responses to five mycobacterial glycolipid Ags presented by CD1b (Fig. 1E) in South African adolescents with known M. tuberculosis infection status (Table I). Our goal was to assess the frequency and cytokine profile of T cells recognizing each lipid Ag, the effect of M. tuberculosis infection, and the correlation with T cell responses to mycobacterial proteins. We employed a computational framework we recently developed called COMPASS to thoroughly characterize the functional profiles of CD1b-restricted CD4⁺ T cells (26). We found a high probability of detecting CD1b-restricted responses among CD4 T cells in only 7 of 64 possible functional profiles (Fig. 2A). Some of these responses were unique to a single lipid Ag, and some were shared among many lipids. For example, CD4 T cells expressing CD40L and producing TNF-α and IL-2 were detected in response to MA and GMM, but T cells that additionally produced IFN-γ were only detected in response to GMM (Figs. 2B, 2C, 3A, 3B). Because CD40L expression is transiently induced upon activation of T cells by Ag, we can infer that these responses were due to lipid-Ag–specific stimulation, a finding further supported by their absence in the presence of PIM and GroMM Ags (27). These data are consistent with the cytokine-producing profiles of GMM and MA-specific T cell clones derived after long-term in vitro culture and so confirm that the phenotype exists in the ex vivo state (15, 16). Notably, the magnitude of T cell responses containing CD40L was not significantly different between M. tuberculosis–infected and M. tuberculosis–uninfected adolescents (Figs. 2D, 3C). We also detected CD4 T cells expressing CD40L but no cytokines in the presence of SL37, suggesting that T cells specific to this lipid might have functional properties other than the ones we chose to study (Fig. 2A). Finally, we noted a high probability of detecting CD4 T cells producing only IL-2 in most individuals and in the presence of most lipids (Fig. 2A, Supplemental Fig. 2A, 2B). These results suggested that T cells restricted by CD1b with unknown Ag specificity might be present at low frequency in all subjects. Indeed, follow-up experiments confirmed that CD4 T cells producing only IL-2 were detectable in the presence of CD1b without the addition of a foreign lipid Ag (Supplemental Fig. 2C). Taken together, these data reveal the range of functional profiles exhibited by CD1b-restricted CD4 T cells, some of which are specific to mycobacterial lipid Ags.

We also detected CD1b-restricted CD8 and CD4⁻CD8⁻ T cells, and the pattern was different from what we observed for the CD4 T cell subset. A high probability of detecting CD8 T cell responses were limited to only 5 of 64 possible functional profiles and consisted of permutations of IFN-γ, IL-2, and TNF-α (Supplemental Fig. 3A). We found CD4⁻CD8⁻ T cell responses limited to 6 of 64 possible functional profiles, none of which included >2 simultaneous functions (Supplemental Fig. 3B). As with the CD4 T cell analysis, COMPASS revealed a high probability of detecting IL-2 single-producing CD8 and CD4⁻CD8⁻ T cells in most individuals and in the presence of most lipids. The lack of a clear association between functional profiles and lipid Ags or M. tuberculosis infection suggested that these responses might be the result of T cells recognizing CD1b expressed on the surface of K562 cells irrespective of the bound Ag. Because CD40L is not a useful marker of Ag-specific activation of CD8 and CD4⁻CD8⁻ T cells, we were unable to separate Ag-specific from Ag-nonspecific T cell responses in this analysis. Nevertheless, these data reveal that CD8⁺ and CD4⁻CD8⁻ T cells that recognize CD1b-presented Ags are likely present in all subjects and could contribute to host defense by elaborating the antimycobacterial cytokines IFN-γ and TNF-α.

We also performed a parallel cross-sectional study of T cell responses to the immunodominant protein Ags ESAT-6, CFP-10, TB10.4, Ag85A, and Ag85B in the same cohort of South African adolescents. We recently reported the association between a five-function T cell subset and M. tuberculosis infection using these data (26). In this study, we reanalyzed this dataset with a focus on T cell responses to individual proteins, which were exclusively detected within CD4 T cell subset (Fig. 4A). We considered the possibility that the peptide pools would preferentially stimulate CD4 T cells instead of CD8 T cells based on their length, but we would not expect this to be the case based on published data with these Ags (28, 29). Further, we were readily able to detect CD8 T cell responses against CMV using peptide pools designed and tested in the same way (data not shown). The probability of detecting CD4 T cell responses to proteins was highest with a single highly polyfunctional T cell subset (IFN-γ⁺IL-2⁺TNF-α⁺CD40L⁺). The five-function T cell subset (IFN-γ⁺IL-2⁺TNF-α⁺CD40L⁺IL-17a⁺) was preferentially detected in the presence of ESAT-6 and TB10.4 (Fig. 4A, 4B). As expected, the magnitude of highly polyfunctional T cell responses as well as total IFN-γ production was higher among M. tuberculosis–infected subjects (Fig. 4C and data not shown). Thus, COMPASS analysis reveals a highly polyfunctional CD4 T cell response that was surprisingly uniform across protein Ags.

Finally, we sought to compare CD4 T cell responses among lipids and between lipids and peptides. To account for the functional diversity of the observed T cell responses, we used the functionality score generated by COMPASS, which summarizes the posterior probabilities of Ag-specific responses across all cell subsets for each subject (26). Functionality score analysis clearly demonstrates a hierarchy in which GMM is dominant among the glycolipid Ags tested (Fig. 5A). Next, we compared the average functionality score among lipids Ags with the average functionality score among peptide Ags and found them to be poorly correlated (Spearman ρ = −0.01; p = 0.95) (Fig. 5B). We considered the possibility that this low correlation might be due to the inclusion of CD1b-restricted T cell responses with unknown Ag specificity within the lipid data. However, when we limited the lipid-specific T cell responses to only those that included CD40L as a proxy for specificity to mycobacterial lipids, there was still a poor correlation (data not shown). This was also true when we considered total IFN-γ production independent of COMPASS (Fig. 5C). Thus, our data demonstrate immunodominance among mycobacterial lipid Ags and a remarkably low correlation between T cell responses to lipids and peptides.

We developed an assay that enables the study of lipid-specific T cells ex vivo and used this assay to perform a systematic analysis of T cell responses to mycobacterial glycolipids in a population in which tuberculosis is endemic. We incorporated a recently developed computational approach to describe the functional profiles of CD1b-restricted CD4⁺, CD8⁺, and CD4⁻CD8⁻ T cells and their association with M. tuberculosis infection. Our results highlight properties of lipid-specific CD4 T cell responses that have not been previously appreciated, such as the lack of association with M. tuberculosis infection, the phenomenon of immunodominance, and a poor correlation with T cell responses to mycobacterial proteins.

MHC-restricted T cells circulate extensively as they move from the site of infection to secondary lymphoid organs and back. Accordingly, our data are consistent with many prior studies showing increased numbers of T cell responses to protein Ags during M. tuberculosis infection. For CD1-reactive cells, it is unclear if infection causes durable memory responses. Several published studies have documented that T cell responses to mycobacterial lipid Ags are increased postinfection, supporting the existence of memory and systemic action (11, 12, 18, 30). In this study, we did not find an association between lipid-specific CD4 T cell responses and M. tuberculosis infection at the time points measured, and there are several potential explanations for these discordant results. First, our study was conducted in an endemic setting to, so it is possible that intense exposure to M. tuberculosis among uninfected subjects induces lipid-specific T cell immunity even if it does not prime peptide-specific responses. It is also possible that these responses are the result of exposure to mycobacteria ubiquitously present in the environment or previous vaccination with bacillus Calmette-Guérin. Second, T cells recognizing lipid Ags might preferentially home to sites of infection, reducing the chance of detecting them in the blood of infected subjects. This would be similar to MR1-restricted T cells, which are not known to have memory or a requirement for trafficking to secondary lymphoid organs and show diminished numbers in peripheral blood during acute infection (8). Third, we used a statistical modeling approach to quantify Ag-specific T cells present at very low frequency ex vivo. This is in contrast with published studies in which lymphocytes were incubated with DCs up to 6 d, which could amplify small differences that are not initially apparent (11, 12). Fourth, our experimental approach was designed to study CD1b-restricted T cell responses only, whereas the results of published studies are confounded by the expression of all human CD1 isoforms on DCs. Finally, we did not specifically assess the contribution of CD8⁺ and CD4⁻CD8⁻ T cell responses, whereas these studies examined IFN-γ production from bulk lymphocyte preparations.

We identified GMM as an immunodominant glycolipid Ag recognized by CD4 T cells. GMM-specific CD4 T cells were polyfunctional, simultaneously producing IFN-γ, TNF-α, and IL-2 upon stimulation with Ag, and present at frequencies between 0.001 and 0.01%. These data confirm and extend published data using GMM-loaded tetramers to estimate precursor frequency in peripheral blood, characterization of GMM-specific T cell clones, and experimental immunization of cattle with GMM (15, 16, 31, 32). We also showed induction of CD40L by GMM and MA-specific CD4 T cells, suggesting these T cells are capable of interacting with cognate B cells to facilitate the production of high-affinity Abs mycobacterial surface lipids. Polyfunctional T cells responses to M. tuberculosis are associated with the clinical stage of disease as well as protection after vaccination in animal models (29, 33–35). It is not known if lipid-specific polyfunctional T cell responses or Abs are important for protective immunity against M. tuberculosis, but these questions can be addressed experimentally in ongoing human trials of whole mycobacterial vaccines (36, 37).

To our knowledge, our study is the first to comprehensively assess T cell effector function and compare responses to five mycobacterial lipid Ags with responses to five immunodominant protein Ags in a cross-sectional study. Thus, we are able to quantify the relative contribution of T cells targeting structurally distinct Ags produced by mycobacteria. The poor correlation we observed support a model in which the functions of T cells targeting lipids and proteins are different with regard to timing or localization. This is consistent with comparative genomic studies, in which CD1 gene families have been identified in virtually every mammalian species, suggesting they perform an important and nonredundant function (38).

Our data have several implications for the development of T cell–based diagnostics and therapies for tuberculosis. The definition of M. tuberculosis infection is based on detecting T cell responses to protein Ags, but <10% of those deemed infected will eventually develop active tuberculosis. Thus, including other immune parameters could improve the prognostic utility of an immunodiagnostic. The poor correlation between CD4 T cell responses to lipids and proteins suggest that measuring T cell responses to lipid Ags would provide nonredundant information that could be studied for its prognostic utility in a prospective clinical trial. Second, our results predict that whole-cell vaccines that included both lipid and protein Ags might activate T cells with a broader set of functions than either class alone. Functional breadth of T cells could partly explain the observed clinical efficacy of whole-cell vaccines, such as bacillus Calmette-Guérin and M. vaccae (39, 40). Third, a formidable challenge to performing population-based studies of mycobacterial T cell immunity is the highly polymorphic nature of MHC class I and II molecules. By contrast, virtually all humans express the same CD1b molecules, so assays like the one we have developed can be used to measure T cell signatures independent of genetic background.

We thank Blessing Kadira, Rebecca J. Smith, Bryce A. Manso, Saheli Datta, and Krystle K. Quan for technical assistance as well as the SATVI Study Team.

The authors have no financial conflicts of interest.