Plasma samples from 4 groups of adults were included in this study: ATB/HIV+ (n = 12), LTB/HIV+ (n = 22), ATB/HIV- (n = 21), and LTB/HIV- (n = 22) ( Table 1 ). All subjects were recruited from Cape Town, South Africa. LTB was defined as the absence of TB symptoms, no previous history of TB diagnosis or treatment and presence of IFN-γ based on an IFN-γ release assay (IGRA). ATB was defined as a positive culture for Mycobacterium tuberculosis (Mtb) growth or positive sputum smear microscopy. Blood from ATB individuals was obtained between 0 and 7 days of standard course anti-TB treatment following South African National Health Guidelines. None of the patients included in the study were on antiretroviral therapy (ART) at the time of enrollment. For each participant, whole blood was collected in sodium heparin Vacutainer tubes (BD Biosciences). Plasma samples were isolated by centrifugation under 500g after 5 minutes, within 4 hours of collection. Prior to the study, a written and informed consent was given by all study participants. Consents were approved by the Human Research Ethics Committee of the University of Cape Town and the Western Cape Department of Health, as well as the study institutional review board at Massachusetts General Hospital and Partners Healthcare.

Two hundred and nine Mtb antigens were used in this project. Purified LAM was received from BEI Resources and purified protein derivative (PPD) was obtained from the Statens Serum Institute. The remaining 207 proteins were recombinantly expressed Mtb antigens received from Dr. Tom Ottenhoff and Kees Franken ( Table S1 ), and prepared as described previously (33, 34). These antigens were selected based on their immunogenicity and their discriminatory potential for TB diagnosis (33, 35, 36).

Antigen-specific antibody subclass and isotype levels present in the plasma of individuals with TB were measured using a custom multiplexed Luminex assay, as previously described (37). This customized Luminex platform has been used extensively across diseases (38–41). Mtb antigens were coupled to magnetic carboxylated fluorescent Luminex beads (Luminex Corporation) by carbodiimide-NHS ester coupling, with one bead region per antigen. Before the coupling with the antigens, beads were activated in an activation buffer containing 100 mM monobasic sodium phosphate (pH 6.2), in addition to 50 mg/ml N-hydroxysulfosuccinimide (Sulfo-NHS; Pierce) resuspended in H₂0 and 1-ethyl-3-[3-dimethlyaminopropyl]carbodiimide-HCl (EDC; Pierce) resuspended in activation buffer. After a 30-minute incubation at room temperature (RT), beads were washed in coupling buffer (50 mM morpholineethanesulfonic acid (MES; pH 5.0)), then incubated with Mtb antigens for 2 hours at RT. Beads were then blocked during a 30 minute incubation at RT in phosphate buffered saline (PBS)-TBN (0.1% bovine serum albumin [BSA], 0.02% Tween 20, and 0.05% azide [pH 7.4]). Finally, beads coupled to proteins were washed in PBS-Tween (0.05% Tween 20) and stored in PBS with 0.05% sodium azide at 4°C. As LAM is a glycolipid, a modification by COOH-4-(4,6- dimethoxy[1,3,5]triazin-2-yl)-4-methyl-morpholinium (DMTMM) was required before the coupling to Luminex beads, as described previously (42). For this protocol, 2 μl of DMTMM (200 mg/ml; Sigma-Aldrich) were used for 25 μg of LAM. After 1 hour of incubation at RT, the excess of DMTMM was removed by using Sephadex G-25 PD-10 desalting columns (GE Healthcare). LAM was then added to the beads and incubated overnight at RT. The next day, LAM-coupled beads were washed in PBS then stored in PBS with 0.05% sodium azide at 4°C.

Antigen-coupled beads were added to plasma samples from TB individuals at 1:100 dilution in PBS and incubated at 4°C for 18 hours of shaking. Beads were then washed 3 times with PBS-Tween (0.05% Tween 20) and incubated with phycoerythrin (PE)-conjugated mouse anti-human IgG1, IgG3, IgA1 or IgM (Southern Biotech) at 1.3 μg/ml. After 1 hour of incubation at RT with shaking at 800 rpm, beads were washed 3 times with PBS-Tween (0.05% Tween 20) and resuspended in sheath fluid (Luminex Corporation). Each plasma sample was tested in duplicate, and PE median fluorescence intensity (MFI) levels were measured via the iQue screener plus (Intellicyt) analyzer.

Fcγ-receptors (FcγRs) were purchased from Duke Human Vaccine Institute to study the relative binding levels of Mtb-specific antibodies to individual FcγRs (43). Avi-tagged FcγR2AR, FcγR2B, FcγR3AV and FcγR3B were biotinylated with a BirA biotin-protein ligase (BirA500; Avidity) and the excess of biotin was removed with Zeba spin desalting columns (7K MWCO; Thermo Fisher Scientific).

Diluted plasma samples (1:100) were added to antigen-coupled beads to form immune complexes as described above, and after 18 hours of incubation at 4°C, beads were washed 3 times with PBS-Tween (0.05% Tween 20). At the end of the incubation, streptavidin-R-phycoerythrin (ProZyme) was added to each biotinylated FcγRs in a 4:1 molar ratio during an incubation time of 20 minutes at RT. Fluorescent labeled FcγRs (1 μg/ml in 0.1% BSA-PBS) were then added to immune complexes and incubated for 1 hour at RT. Finally, beads were washed 3 times with PBS-Tween (0.05% Tween 20), then resuspended in sheath fluid (Luminex Corporation) and the median PE intensity was measured via the iQue screener plus (Intellicyt) system. Samples were tested in duplicate.

Data analysis was performed using R version 4.0.2 (2020-06-22). Comparisons between active and latent individuals were performed using a Mann-Whitney U-test test followed by a Benjamini-Hochberg (BH) correction for multiple comparisons.

A multivariate approach, combining a least absolute shrinkage and selection operator (LASSO) for feature selection and classification using partial least square discriminant analysis (PLS-DA) with the LASSO-selected features was used to define feature that discriminated active and latent antibody profiles. Prior to analysis, mean fluorescence (MFI) values were first log-transformed and all data were normalized using z-scoring. Models were built using the R package “ropls” version 1.20.0 (44) and “glmnet” version 4.0.2. Model accuracy was assessed using five-fold cross-validation. For each test fold, LASSO-based feature selection was repeated 100 times, features selected at least 90 times out of 100 were identified as selected features. A PLS-DA classifier was applied to the training set using the selected features, and a prediction accuracy was recorded. The model was also validated via permutation testing, where the performance of the selected model was evaluated against randomly shuffled active-latent labels. Selected features were ordered according to their Variable Importance in Projection (VIP) score, and the first two latent variables (LVs) of the PLS-DA model were used to visualize the samples.

To explore humoral responses in individuals with LTB and ATB, we began by profiling isotype and subclass levels across 8 common Mtb antigens (LAM, PPD, ESAT6/CFP10, Ag85A and 85B, groES and PSTS3) in ATB and LTB individuals that were HIV positive or negative. As previously described, HIV negative ATB individuals exhibited higher levels of IgG1, IgG3 and IgA1 responses to particular antigens, including LAM, PPD, and groES ( Figures 1A, B ). Conversely, an opposite profile was observed in HIV positive individuals, marked by significantly elevated humoral immune responses, and particularly IgM responses to Ag85A and PSTS3 among HIV positive LTB ( Figures 1A, C ). However, these features did not result in complete discrimination of ATB and LTB patients across HIV serostatus ( Figures 1D, E ) (cross-validation accuracy of 62.8% for HIV-; 67.6% for HIV+), suggesting that common antigen-specific antibody profiles provide only a modest level of discrimination across TB infection and disease states.

Despite the lack of differences across a limited number of common Mtb antigens ( Figure 1 ), Mtb can express up to 4000 distinct antigens (45, 46), many of which may serve as additional antibody targets. Thus, we next elected to deeply profile the humoral immune response across LAM, PPD as well as 207 Mtb antigens ( Table S1 ), transcriptionally enriched in in vitro or in vivo infection models, including lung specimens (36). Specifically, IgG1, IgG3, IgA1 and IgM levels were profiled across all antigens in ATB and LTB among HIV positive and HIV negative individuals ( Figure 2 ). The data highlighted generally expanded IgG1 and IgG3 responses across HIV negative and positive ATB compared to LTB, with some sporadic elevated IgG1/IgG3 responses in HIV-positive LTB ( Figure 2A ). IgA responses were significantly elevated in HIV negative ATB compared to LTB, despite expanded IgA immunity in HIV-positive LTB ( Figure 2A ). IgM responses were more diffuse across HIV negative groups, despite the presence of a few high IgM responses in ATB compared to HIV negative LTB. Conversely, the IgM response was distinct in the setting of HIV co-infection, marked by significant expansions of particular Mtb-specific IgM responses that were either enriched in ATB or were enriched in LTB ( Figure 2A ). Thus, using this expanded antigen set, striking differences were noted in the evolution of the humoral immune response across HIV-serostatus, pointing to the presence of particular antigen-specific humoral immune responses that may discriminate between TB disease states.

To define the minimal set of specific antigens that were differentially targeted across the groups, we began by quantifying the number of antigens that were differentially targeted by each subclass or isotype of antibodies ( Table S2 ). Specifically, in HIV negative individuals, 79 Mtb antigens were differentially targeted by IgG1, 5 antigens were differentially targeted by IgG3, 106 were differentially targeted by IgA1, and 5 antigens were targeted distinctly by IgM ( Figure 2B and Table S2 ). Conversely, in HIV positive individuals, fewer statistically significant differences were observed after multiple correction but included: 34 Mtb antigens that were targeted statistically differently by IgG1, 30 that were differentially targeted by IgG3, and 19 antigens were targeted uniquely by IgM across the two groups ( Figure 2B ). Interestingly, for most comparisons, IgG1, IgG3, and IgA1 levels were higher in ATB individuals, in accordance with published data and likely reflective of the enhanced inflammatory state and bacterial burden observed in ATB infection among HIV positive individuals (22, 23). However, Mtb-specific IgM responses were targeted differently in HIV positive and negative individuals, with higher Mtb-specific IgM responses in ATB in the HIV negative group, but higher Mtb-specific IgM responses in LTB in the HIV positive population, reflecting a shift in B cell response profiles in the setting of HIV. Yet, collectively, these data highlight the striking differences in Mtb-specific humoral immune responses in the setting of HIV infection, but still associated to the persistence of differential humoral immune responses across ATB and LTB.

Beyond overall changes in antibody subclass and isotype responses to Mtb, emerging data point to significant differences in the inflammatory state of Mtb-specific antibodies across disease states (26, 29). These changes are induced by alterations in antibody-Fc-glycosylation, aimed at deploying antibody effector functions required for enhanced clearance and control of the pathogen (47). Because these Fc-glycosylation changes result in altered binding to Fcγ-receptors (FcR), we next examined whether Mtb-specific FcR binding differences (FcγR2AR, FcγR2B, FcγR3AV, and FcγR3B) existed against LAM, PPD and 207 Mtb antigens ( Table S1 ) in HIV positive and negative individuals with ATB or LTB ( Figure 3 ).

Thirty-nine Mtb antigens were differentially recognized by FcγR2AR binding antibodies in HIV negative LTB populations compared to ATB ( Figure 3B ). Moreover, 54, 31 and 50 Mtb antigens were differentially recognized by FcγR2B, FcγR3AV and FcγR3B-binding antibodies, respectively, with higher binding observed in ATB ( Figures 3A, B and Table S2 ). Conversely, FcR binding antibodies were globally shifted in HIV positive individuals, marked by disproportionately higher levels of FcγR2AR, FcγR2B, FcγR3AV and FcγR3B binding antibodies in HIV+ ATB compared to LTB ( Figure 3 ). Specifically, 145, 183, 162 and 148-Mtb-antigens were targeted more robustly by FcγR2AR, FcγR2B, FcγR3AV and FcγR3B binding antibodies, respectively, in ATB compared to LTB ( Table S2 ). Thus, beyond alterations in overall antibody subclass/isotype binding profiles across TB disease status, alterations are observed with FcR binding, that are amplified in the setting of HIV co-infection, again pointing to the possibility for antibody-based discrimination across LTB and ATB across HIV status.

Given the significant differences in isotype, subclass, and FcR binding differences across LTB and ATB in both HIV positive and negative individuals, we next aimed to identify a minimal set of potential candidate antigens that could be used to discriminate between ATB and LTB. A computational analysis that combined LASSO feature down-selection and PLS-DA classification was conducted in the HIV negative ( Figures 4A–D ) and positive ( Figures 4E–H ) groups. Out of the 1680 antigen-specific antibody features included in the analysis, only 4 features were sufficient to separate HIV negative ATB and LTB ( Figures 4A, B ): RV2034-specific FcγR2AR binding, which was higher in LTB, as well as LAM-specific FcγR3A binding antibodies, RV1528-specific FcγR2B binding antibodies, and RV2435c IgG1 levels that were enriched in ATB ( Figure 4B ). Three of the LASSO selected features were significantly differentially targeted across the groups, including RV2034-specific FcγR2AR binding, LAM-FcγR3A binding and RV1528-specific FcγR2B binding antibodies ( Figure 4C ). Given that LASSO selects a minimal set of features that account for the greatest variance across the groups, the LASSO-selected features may represent groups of correlated antibody features that diverge across TB groups. Thus, to gain enhanced insights into the global changes across groups, a co-correlate analysis was built on the LASSO-selected features highlighting the presence of a large, expanded FcγR2B network in ATB, as well as two smaller networks of LAM-biomarkers and FcγR2AR binding biomarkers enriched in HIV negative ATB individuals ( Figure 4D ).

Similar to the HIV negative LTB/ATB individuals, as few as 3 of the 1680 analyzed features were sufficient to separate LTB and ATB individuals with HIV ( Figures 4E, F ). Specifically, RV3583 FcγR2AR binding was enriched in ATB individuals, and RV1508 and Ag85A IgM levels were enriched among LTB ( Figures 4E, F ). All features were significantly different across the groups ( Figure 4G ). Interestingly, co-correlates analysis revealed that the single ATB feature was tightly linked to nearly all IgG/FcR binding levels that were globally expanded in ATB ( Figure 4H ). Additionally, a smaller network of IgM features emerged, marking the unique expansion of IgM responses among HIV positive LTB.

Finally, we aimed to estimate the classification accuracy of the LASSO-selected antibody features across ATB and LTB. An area under the receiver operating characteristic (ROC) curve (AUC) was calculated for each feature ( Figures 5A, B ). Importantly, the combination of the selected antibody features gave an AUC close to 1 (AUC = 0.98 for HIV-; AUC = 0.92 for HIV+). However, even using individual antibody features, classification accuracies of 0.96, 0.8, 0.71 and 0.74 were observed with Rv2034-specific binding to FcγR2AR, LAM-specific binding to FcγR3AV, Rv1528-specific binding to FcγR2B and Rv2435C-specific IgG1 levels, respectively, among HIV negative individuals ( Figure 5A ). Similarly, in HIV positive individuals, individual feature AUCs reached 0.81, 0.86 and 0.84 for Rv1508-specific IgM levels, binding of Rv3583-specific antibodies to FcγR2AR and Ag85A-specific IgM levels, respectively ( Figure 5B ). Our results showed that 4 parameters were systematically enriched in ATB compared to LTB among HIV negative and positive populations, which are IgG1 levels against Rv2435.C, as well as Rv3583-, Rv1528- and LAM-binding to FcγRAR, FcγR2B and FcγR3AV, respectively ( Figure 5C ). Thus, a minimal set of largely novel Mtb-specific humoral biomarkers may provide a unique opportunity to help discriminate between LTB and ATB. Given the ease of generation of rapid point-of-care antibody-based diagnostics, these data point to a simple opportunity to develop rapid point-of-care tests for the diagnosis of Mtb disease state.