Antibody Subclass and Glycosylation Shift Following Effective TB Treatment

With an estimated 25% of the global population infected with Mycobacterium tuberculosis (Mtb), tuberculosis (TB) remains a leading cause of death by infectious diseases. Humoral immunity following TB treatment is largely uncharacterized, and antibody profiling could provide insights into disease resolution. Here we focused on the distinctive TB-specific serum antibody features in active TB disease (ATB) and compared them with latent TB infection (LTBI) or treated ATB (txATB). As expected, di-galactosylated glycan structures (lacking sialic acid) found on IgG-Fc differentiated LTBI from ATB, but also discriminated txATB from ATB. Moreover, TB-specific IgG4 emerged as a novel antibody feature that correlated with active disease, elevated in ATB, but significantly diminished after therapy. These findings highlight 2 novel TB-specific antibody changes that track with the resolution of TB and may provide key insights to guide TB therapy.


INTRODUCTION
Tuberculosis (TB) continues to be one of the leading causes of death by infectious disease globally, and while the development of new protective vaccines continues to be a critically important goal for the fight against TB disease (1,2), detecting active TB (ATB), the TB state with the greatest likelihood of spreading Mtb, for immediate treatment could profoundly prevent TB spread (3)(4)(5). Current immune-based diagnostics, including the tuberculin-skin-test (TST) or the interferongamma release assay (IGRA), can detect individuals with TB but cannot distinguish individuals with ATB from latent TB infection (LTBI), which accounts for~95% of world cases, therefore limiting the ability to identify disease that requires immediate treatment (6,7). Furthermore, current immune diagnostics cannot distinguish those who have successfully completed therapy from those with ATB and actively replicating Mtb.
Given the heterogenous manifestation of disease in individuals exposed to Mtb, it is not surprising that immune responses to Mtb are also heterogenous. And in humans, features of the immune response such as numbers of circulating NK cells (8), neutrophils (9), B cells (9,10) and T cells (9) have been observed to differ in TB-diseased individuals depending on their disease severity and clearance of replicating Mtb following treatment. Phenotypic differences in the T cell response to TB have also been shown to associate with disease severity, with higher frequencies of proliferating and TH1-cytokine producing CD4 T cells observed in ATB compared to LTBI (11)(12)(13)(14); these T cells diminish from circulation following therapy (15,16). In addition, inflammatory signatures that include type I interferon, captured through whole-blood RNA sequencing, demonstrate a strong association with ATB disease, which also diminishes with treatment (9,17).
Antibody-based measures are attractive alternatives to cellular measures of disease activity; these disease-specific immune responses are easily and directly captured from serum in an antigen specific manner. Measuring antibodies in sera is also technically simple and relatively rapid when compared to the cumbersome and variable measures of cellular immune responses. Importantly, a single antibody molecule measures both antigen-specificity within the variable domain (Fab) and inflammatory state of the disease within the constant domain (Fc) (18). Alterations in disease-specific IgG properties including antigen specific titers (19,20), isotype switching (21), and glycosylation (22)(23)(24)(25)(26) provide insights into disease relapse or severity across diseases ranging from cancer (27,28), autoimmunity (29,30), and infections (20,31). While changes in disease-specific titers do not always reflect changes in disease activity (32)(33)(34)(35), alterations in disease-specific Fc-properties provide critical qualitative insights into disease activity. Given that changes in IgG Fc-profiles also track with altered Fc-effector function, unique humoral markers of disease activity may also provide additional insights into the mechanism(s) of enhanced disease control and even elimination. Along these lines, recent studies of TB-specific antibodies highlight the disease discriminatory activity of IgG Fc-glycosylation features (36,37) that may point to unexpected mechanisms of anti-microbial control (36,(38)(39)(40)(41)(42).
Recent data indicate that B cells change not only in number but also phenotype and function during TB disease and after treatment (10). While decreased TB-specific IgG titers have been noted in several studies following TB treatment (43,44), it remains unclear whether antibody Fc-profiles also shift with treatment. Humoral profile shifts could provide insights into long-term immunity after successful Mtb clearance and point to markers of TB treatment success. Thus, in this study we aimed to determine how the humoral immune response to Mtb differed among the TB states: LTBI, ATB, and txATB. We profiled the TB-specific humoral immune responses in the serum of individuals previously profiled for B cell phenotype and function (10) using a systems serology approach (45). These measures included total serum antibody titers, antigen-specific antibody titers, and antibody-mediated functional responses in human cells. And in light of our recent study of IgG glycans in LTBI/ATB discriminatory model performance, which found IgG-Fc glycans discriminated better compared to whole IgG and Fab-glycans (37); we focused on IgG-Fc glycans in the humoral profiling of this cohort. We observed significant differences in IgG-Fc glycosylation across individuals with LTBI and ATB disease, consistent with earlier observations in IgG from independent cohorts of LTBI and ATB individuals (37). Additionally, we found enrichment of TB-specific IgG4 among ATB individuals. Strikingly, these same antibody features distinguished ATB from txATB, suggesting that IgG4, in addition to IgG-Fc glycosylation, may also be a marker of ongoing inflammation and the ATB disease activity.

Study Subjects
Sample collection was approved by the ethical committee of INMI, approval number 72/2015; informed written consent was obtained before collection. ATB was confirmed via Mtb sputum culture and patients were enrolled within 7 days of starting the TB treatment (isoniazid, rifampicin, ethambutol and pyrazinamide for 2 months, followed by isoniazid, rifampicin for 4 additional months (46). TxATB subjects were patients who completed a 6-month course treatment for culture-positive pulmonary TB and were culture-negative at 2 and 6 months of therapy. LTBI subjects were mainly contacts recently exposed (within the previous 6 months) to smear-positive ATB patients with positive QuantiFERON TB Gold In tube (QFT-IT) (Quiagen, Germany) but without symptoms or radiological signs of ATB. Healthy donors were QFT-IT-and HIVindividuals not undergoing immunosuppressive drug treatments. Serum samples were collected in heparin tubes. An additional 10 healthy HIVdonors from the Greater Boston, Massachusetts area were recruited by Ragon Institute of MGH, MIT, and Harvard for serum assay controls. Blood samples were collected in ACD tubes from different donors in the Boston-area, for the isolation of Neutrophils for ADNP assays. NK cells used for assessing antibody function were derived from buffy coats of healthy HIV-donors collected by the MGH Blood Donation Bank. All study participants for additional healthy negatives and primary cell isolation gave written, informed consent, approved by the institutional review boards at Massachusetts General Brigham Hospital.

Antibody Dependent Cellular Phagocytosis (ADCP)
Biotinylated PPD was used to coat 1mm Neutravidin labeled yellow-green, fluorescent beads. Immune complexes were formed by combining PPD-beads, and combined with serum samples diluted 1:30, 1:100, 1:300, 1:1000 in PBS and incubated at 37C for 2hrs. Complexes were washed, 2x10 4 THP1 cells were added per well of 96-well plates and incubated for 1hr at 37C. Samples were then washed and fixed for analysis of bead uptake on an iQue Screener. Phagocytic scores were calculated as previously described (36) across sample dilution series and were used to calculate AUC.

Antibody Dependent Neutrophil Phagocytosis (ADNP)
Neutrophils were isolated from healthy donor blood collected in ACD tubes as previously described (36). Complexes were formed and incubated with isolated neutrophils as described for the ADCP above. After incubation, samples were stained with CD66b and fixed with 4%PFA. Bead phagocytosis was measured as described above, and the enrichment of neutrophils was confirmed with CD66b staining (BioLegend 305112). Phagocytic scores were determined as in ADCP across a serum dilution series ranging from 1:30 to 1:1000 and used to calculate Phagocytic Score AUC.

Glycan Analysis of IgG-Fc
IgG was isolated from serum samples by incubating 10mL of serum diluted 1:20 in PBS with 25mL protein G beads (Millipore, Catalog #LSKMAGG10); the serum and beads were mixed at 4C for 16hrs. Excess serum protein was washed, and IgG-bound beads resuspended in digestion buffer containing 1uL IdeZ (NEB Catalog #P0770S) and IgG was digested at 37C for 2hr to remove Fab. IgG-Fc still bound to magnetic protein G beads were pelleted and washed on a magnet to separate Fc from Fab. Glycans from IgG-Fc were cleaved, enriched, and labeled with APTS according to manufacturer specifications in the Glycan Assure Kit (ThermoFisher A28676). To immune-precipitate and analyze antigen-specific IgG-Fc glycosylation, streptavidincoated magnetic beads (NEB Cat# S1420S) were coated with biotinylated-PPD, as described above for ADCP. Antigen-coated beads were incubated with 300mL of serum at 4C for 16hrs. Excess protein was washed off the beads with PBS and Fc of the antigen-bound IgG was cleaved with IdeZ as described above. Supernatants were taken from this IdeZ reaction for glycan cleavage and staining according to the Glycan Assure protocol. Samples were run with a LIZ 600 DNA ladder in Hi-Di formamide (Thermo Fisher 4408399) on an ABI 3130XLl DNA sequencer. Data were analyzed using ThermoFisher Glycan Assure Analysis software; peaks were assigned based on migration of standards of known glycans and peak area was calculated. The measured peak areas per sample were totaled to report a relative frequency of each glycan structure identified.

Data Visualization and Analysis
Univariate data visualization and statistical analysis were performed using GraphPad Prism (Version 8.3.1). For multivariate analysis, MATLAB computing environment (version 2018b, Mathworks, Natick, MA) was used, supported by the Statistics and Optimization toolboxes, as well as the third party PLS toolbox (Eigenvector Research, Inc, Manson, WA). Spearman network visualizations were performed using Cytoscape (version 3.6.0).

Identification of TB Signatures With LASSO and OPLSDA
Computational analysis was used to build classification models that identify key features that most effectively resolved pairs of the LTBI, ATB and txATB states. These classification models were built using previously described methods (25,45) combining (i) Least Absolute Shrinkage and Selection Operator (LASSO) method (47), for feature selection, and (ii) classification using the LASSO-selected features. For LASSO selection, a previously described nested cross validation framework was used to validate the robustness of the classification model (48). Orthogonalized Partial Least Square Discriminant Analysis (OPLSDA) (49,50) was used to visualize LASSO-selected variables and assess their predictive ability for classifying TB states (Figures 2, 3, 4A, B). These input variables were centered and scaled to a standard deviation of 1. PLSDA models consisting of two LVs were constructed and then orthogonalized to condense the Y-block variance (group separations) into the first Latent Variable (LV1). LV1 captures the variance in features that are in the direction of the pairwise separation of the groups, while LV2 describes the variation orthogonal to this predictive component. To assess each model, 5-fold cross validation (CV) was performed on the data (100 random 5-fold cross validation). To assess model significance, permutation test was performed on the cross validated models by randomly shuffling the labels. The OPLSDA models performed significantly better than random with CV Wilcoxon p values of lower than 2E-3 across pairwise group comparisons. Variable Importance in Projection (VIP) scores were calculated (51) to rank the importance of each variable in the projection of the PLS model. To emphasize the direction of the contribution of each variable, negative and positive signs were added to VIP scores to indicate negative and positive Loadings of each variable on LV1.

Construction of the Correlation Network of the LASSO-Selected Features
Spearman correlation of the LASSO-selected Fc features to all original 78 TB-specific antibody features were calculated. Each node is a feature and the thickness of the edges between nodes is proportional to their correlation coefficients. The p-value depicting the significance of these correlations were corrected for multiple comparisons (Benjamini-Hochberg q-value < 0.05, testing the hypothesis of zero correlation). Only correlations with corrected p-values<0.05 were included.

Three-Way PLSDA Model
The LASSO-selected features from the three pair-wise group comparisons were pooled (total of 8 features) and a PLSDA model was developed to separate the three groups of LTBI, ATB and txATB. This model was not orthogonalized to better capture and visualize the pairwise group differences. The twodimensional loadings on LV1 and LV2 were overlaid on the scores plot.

LTBI and ATB Plasma Have Distinct Profiles of Fc-Glycosylation and TB-Specific IgG Subclass
We previously described the biophysical and functional features of purified IgG found in LTBI and ATB from cohorts of individuals from South Africa and US/Mexico (36,37). However, whether these differences persist, or differ following successful antibiotic treatment remains unclear. Thus, here we aimed to define the impact of therapy on shaping the TB-specific humoral immune in a cohort of LTBI (n=21), ATB (n=20), and txATB (n=23) from Italy, with an additional group of healthy control individuals (n=17) from Italy and the USA ( Table 1). Similar levels of circulating, non-antigenspecific IgM, total IgG and the subclasses IgG1, IgG2, Ig3, and IgG4 were observed in the serum of healthy, LTBI, ATB, and txATB individuals (Supplemental Figure 1A). In contrast, total IgA titers were significantly elevated in ATB compared to LTBI and txATB (Supplemental Figure 1A), consistent with previous observations in TB (52,53). Additional differences in antigen-specific isotype (Supplemental Figures 1B-E), subclass (Supplemental Figure 2), total IgG-Fc glycosylation (Supplemental Figure 3), and Fcmediated functional antibody responses in monocytes, neutrophils, and NK cells (Supplemental Figure 4) were noted across groups. The differences observed across this collection of measures suggested distinct humoral profiles existed not only in LTBI and ATB, but also txATB.
Given the univariate differences across this cohort, we sought to identify the humoral features of our dataset, which could best discriminate subsets of TB diseased individuals by applying a conservative multivariate analytic method to define a minimal set of distinguishing humoral features. We first aimed to determine whether LTBI and ATB were fully resolvable using antibody features as had been previously observed (36), and further, whether a minimal set of antibody-features could be identified, which resembled previous antibody profile differences. A least absolute shrinkage and selection operator (LASSO) (47)  was first applied to reduce the number of features in our highly correlated dataset (Data Sheet 1), to avoid overfitting, and define a minimal set of these features that could discriminate the LTBI and ATB disease states. Using PLSDA for visualization, clear separation was observed between the LTBI and ATB antibody profiles ( Figure 1A). Moreover, as few as 4 of the 78-total Fcfeatures selected by LASSO could resolve LTBI and ATB individuals, providing 91% cross-validation (CV) accuracy ( Figure 1B); these included totaled di-galactosylated (G2); the di-galactosylated, fucosylated, and bisected (G2FB) structure; PPD-specific IgG4 levels; and singly-galactosylated, sialylated, fucosylated (G1S1F) structures. Univariate plots of the LASSO-selected features revealed statistically significant differences in all 4 features when comparing LTBI and ATB groups (Figures 1C-F). Specifically, the levels of totaled G2 ( Figure 1C) and the G2FB structure ( Figure 1E) on total IgG-Fc were enriched in LTBI. While PPDspecific IgG4 levels and the G1S1F structure were enriched in ATB IgG-Fc ( Figures 1D, F). The differential IgG-Fc galactosylation seen across LTBI and ATB of this cohort is consistent with our previous observations in cohorts of LTBI and ATB samples from South Africa and the US/Mexico (36). However, the subclass enrichment of IgG4 antibodies in ATB represents a novel observation in TB disease. The coincidence of IgG4 and G1S1F on total IgG-Fc glycans of ATB is reminiscent of previous studies of subclass specific Fc-glycosylation, which found an enrichment in IgG4 of healthy individuals (54) and an elevation of G1S1F on IgG4 in patients with IgG4-related disease (55).
Given that the LASSO/PLSDA model selects features that solely account for the greatest variance across the antibody profiles being compared, additional distinctive antibody features that are correlated with the LASSO-selected features are not highlighted in this analysis. To explore the additional humoral features of our dataset that distinguished LTBI and ATB, we next generated a Spearman correlation network of the LASSO-selected features to highlight the relationship of this minimal set of features with the remaining 74 features Univariate plots (C-F) show median and interquartile range of each LASSO-selected measure; statistically significant differences between LTBI and ATB groups calculated using Mann-Whitney test: ***p < 0.0005 and ****p < 0.0001. The dotted lines represent median of healthy controls individuals tested. (G) Correlation analysis depicts other features that are positively (red lines) or negatively (blue lines) correlated with these four key features selected with LASSO (highlighted in yellow). The color intensity and width of the edges between nodes are proportional to the significance of correlation coefficients after correcting for multiple comparisons (Benjamini-Hochberg q-value < 0.05, testing the hypothesis of zero correlation). Only correlations with corrected p-values < 0.05 were included. measured ( Figure 1G). Two networks emerged from this analysis: a large network linking the two LTBI enriched features (G2 and G2FB) linked via negative correlations to the ATB enriched IgG4 signature ( Figure 1G, right), and a smaller second network consisting of features correlated with the ATBassociated G1S1F feature ( Figure 1G, left). Importantly, IgG4 titers against multiple TB antigens, including intracellular TB proteins HspX and GroES, were co-correlated, pointing to a shift to IgG4 responses in ATB. Additionally, di-galactosylated features enriched in LTBI were linked to several galactosylated structures, reinforcing an overall elevated galactosylation profile in LTBI. Finally, total serum IgG and IgA levels, found to be elevated in ATB (Supplemental Figure 1A), were inversely correlated to the glycan features elevated in LTBI profile. These networks link qualitative/quantitative changes in the humoral profiles between LTBI and ATB states with the LASSO-selected features and highlight the unique enrichment of Mtb-specific IgG4 responses. Furthermore, depletion of IgG4 resulted in increased antibody effector function in neutrophils and NK cells (Supplemental Figure 5), pointing to IgG4 as a mechanistic player in dampening antibody function, as previously shown in HIV (56) and cancer (57). In summary, we find a critical recapitulation of glycan features of latency in an Italian cohort, highlighting the universal presence of this biomarker of ATB, that includes IgG4 levels, perhaps previously overlooked due to the purification methods used in our original study. These data reinforce a set of qualitative antibody Fc-features that discriminate LTBI from ATB.

Treatment of ATB Correlates With Reduced TB-Specific IgG4 Titers and Inflammatory Glycan Signatures
Treatment of ATB and the resolution of replicating Mtb has been linked to the resolution of inflammatory cytokines (9), shift in T cell phenotypes (15) and NK cell abundance (8) in the blood. Along these lines, previous cellular profiling of this cohort of individuals indicated that B cells were less proliferative and produced fewer antibodies in ATB, while B cells functioned normally following treatment (10). Given these previously observed differences following TB treatment, we next aimed to determine whether a minimal set of humoral features could distinguish txATB from ATB and mirror the recovered B cell responses found in txATB individuals.
LASSO was applied to select minimal features that distinguished ATB and txATB, and PLSDA visualization of the selected features provided nearly perfect distinction between the disease states, with 92% cross-validation accuracy (Figure 2A). The LASSO-identified features included: total IgG-G2, -G2FB, -G1S1F glycans, and PPD-specific IgG4 ( Figure 2B). These antibody features point to the resolution of inflammatory Fcglycosylation with treatment, with elevated G2 ( Figure 2C) and G2FB ( Figure 2D) structures on txATB IgG-Fc compared to ATB. In contrast, the frequency of IgG-G1S1F structures was significantly higher in ATB compared to txATB ( Figure 2E). Finally, PPD-specific IgG4 titers were lower in txATB compared to ATB ( Figure 2F), with txATB levels of PPD-specific IgG4 equivalent to those found in healthy controls (Supplemental Figure 2D). Collectively, these data suggest that inflammatory glycans and PPD-specific IgG4 titers, markers of ATB, are diminished in txATB.
Again, to further probe the dataset for additional shifts in the humoral response related to the LASSO-selected features, we generated a Spearman correlation network between the LASSOselected features and the remaining features. A single correlated network emerged (Figure 2G), including a dense cluster of glycans and TB-specific IgG4 features that all diminished in the setting of treatment. In addition to the network of IgG4 titers that inversely correlated with glycan structures enriched in txATB, IgG2 and IgG3 titers specific for Ag85A/B and HspX features arose in the network and inversely correlated to G2 features. These relationships suggest that in addition to IgG4 titers, individuals with ATB utilize additional IgG subclasses, during persistent infection and exposure to Mtb antigens, that likely resolve following treatment. Finally, IgM-specific responses to PPD and ESAT6/CFP10 emerged in this analysis and were linked to G2 levels on IgG-Fc in the txATB group. As the first class of antibody produced in primary antigen exposure (58), the elevation of TB-specific IgM titers in txATB point to a development of novel naïve humoral responses following resolution of replicating Mtb (Supplemental Figure 1B). Together, these data highlight a significant shift in antibody isotypes following treatment, with an overall reduction of inflammatory IgG-Fc glycans and concomitant contraction of the ATB-specific IgG4 immunity across TB-specific antigens.

Higher TB-Specific Titers Distinguish txATB From LTBI
The overlapping features that distinguished both LTBI/ATB and ATB/txATB raised the question of whether humoral immunity in txATB and LTBI were largely similar or if humoral immunity could also distinguish these two states. Using LASSO/PLSDA on the humoral profiles, LTBI and txATB could be resolved with 81% cross-validation accuracy ( Figure 3A). The LASSO selected features were largely enriched in the txATB compared to the LTBI individuals ( Figure 3B); these included Ag85A/B-specific IgG and IgM titers as well as HspX-specific IgG1 titers ( Figures 3C, E, F). And while the LASSO-selected TB titers did not reach univariate significance, the amount of antibodymediated phagocytosis (ADCP) was significantly increased in txATB compared to LTBI ( Figure 3D). These LASSO-selected features indicate higher antibody levels and function amongst txATB individuals. Moreover, network analysis underscored the prevalence of higher IgG1 titers against multiple TB-specificities in txATB individuals and pointed to the persistence of TBspecific antibodies in this recently treated population ( Figure 3G). The elevated ADCP activity in txATB was linked to elevated neutrophil phagocytosis (ADNP), highlighting persistent antibody-mediated phagocytic activity following antibiotic treatment. Thus, txATB was distinguishable from LTBI in our multivariate analysis by higher antibody titers and enhanced opsonophagocytic function at the conclusion of treatment.

Antibody Titer and Glycosylation Distinguish txATB From ATB and LTBI
Given the ability of our pairwise models to distinguish TB states, we next aimed to resolve all three TB states simultaneously using the features previously selected by LASSO. Strikingly, this model discriminated between all three states with 80% classification accuracy ( Figure 4A). ATB was most distinct from LTBI using these features and showed some overlap with txATB. However, interdigitation was observed across LTBI and txATB individuals, pointing to an overlap of humoral profiles in these disease states. LASSO-selected features were superimposed on the PLSDA plot, in the quadrant in which it was enriched ( Figure 4A). PPDspecific IgG4 and IgG-Fc glycan, G1S1F, were uniquely enriched amongst the cluster of ATB individuals within the PLSDA. A markedly higher level of Ag85-specific IgG was also associated with the ATB cluster. Conversely, enhanced levels of PPD-specific phagocytosis, elevated HspX-IgG, and Ag85-IgM were observed among txATB. Finally, enhanced di-galactosylation was observed on IgG-Fc from LTBI highlighting the importance of titers, function, and glycosylation, many of which were significantly elevated in LTBI and txATB compared to ATB at a univariate level (Figures 4B-I).

DISCUSSION
With our growing appreciation of antibodies in TB immunity (36-39, 59, 60), changes in antibody isotype, subclass, and glycosylation are all emerging as biomarkers specific of disease activity. While many studies have described distinctive antibody features across LTBI and ATB disease, less is known about the changes in humoral immunity following treatment. Using systems serology, we observed significant differences in TBspecific antibody profiles across LTBI, ATB, and txATB, which highlight antibody changes that correlate with the burden of infection. As previously observed (36,37), we found that LTBI and ATB are marked by distinct IgG-Fc glycosylation patterns, with an enrichment of glycans associated with inflammation in ATB. These observations are in line with an accumulation of inflammatory-glycans observed in antibodies of diseases including HIV (25) and autoimmune disorders (61)(62)(63). Importantly, we describe for the first time in txATB individuals an enrichment of G2 structures on IgG-Fc after the successful completion of treatment, pointing to IgG Fcglycosylation as a marker of reduced replicating Mtb. A similar shift in IgG-Fc digalactosylation has been reported in a longitudinal study of IgG-Fc glycans in patients infected with hepatitis B; as treatment progresses and detectable viral DNA decreases, IgG-Fc digalactosylation increases (26). Additionally, we measured PPD-specific IgG-Fc glycans in the serum of a subset of TB diseased individuals from this cohort and found that levels of digalactosylated structures were also lower in ATB compared to LTBI and txATB (Supplemental Figures 6C, I).
Unlike total IgG-Fc glycan measures, PPD-specific IgG-Fc was enriched for agalactosylated structures in ATB (Supplemental Figure 6G), a glycan structure found to be enriched on IgG in other inflammatory diseases (25). Thus, inflammatory Fcglycans mark disease activity and track with the presence of replicating Mtb in ATB individuals. Beyond antibody glycan changes, humoral comparisons within this cohort pointed to TB-specific IgG4 as a novel humoral marker of TB disease activity. While consistently low in titer, previous studies noted elevated LAM-specific (64) and PPD-specific IgG4 titers (65) in HIV-and HIV+ individuals with ATB, respectively. IgG4 emerges through class-switch late in disease due to its distance on the human IgG locus, and with low Fc-receptor and complement affinity is selected under high antigen-burden when antibody titers are high. In diseases with prolonged antigen exposure and inflammation including parasitic infections (66,67), chronic Staphylococcus aureus (S. aureus) infection (68), chronic infectious aortitis (69), melanoma (57), and even following repeated high-antigen dose immunization (56,70,71) elevated IgG4 levels have been described. And in the wake of resolving antigen burden, a longitudinal analysis of patients chronically infected with Brugia malayi, demonstrated a rapid loss of IgG4 titers, and a preservation of IgG1 responses following treatment (67). Similarly, we observed no change in TB-specific IgG1 titers and a loss of TB-specific IgG4 titers in txATB, pointing  to a similar trajectory of IgG4 as TB antigen is eliminated with therapy. And analogous to IgG2 and IgG3 titer declines observed following Brughia malayi treatment (67), although not statistically significant, we noted trends of decreased IgG2 and IgG3 titers in txATB compared to ATB in our co-correlate network analysis ( Figure 2G and Supplemental Figures 2B, C). These data suggest that elimination of the antigen results in a shift of antibody subclass selection, with a more dramatic loss of IgG4 antibodies, suggesting that across TB disease, IgG4 may mark a more transient population of antibody-secreting cells that require high-antigenic stimulation to persist, making TB-specific IgG4 an attractive disease-specific marker of treatment success. The balance of subclass, isotype, and glycosylation within an antibody immune complex can have significant functional consequences for Fc-mediated immune responses. IgG4 may arise to compensate for the pro-inflammatory activity of agalactosylated glycans that accumulate in active inflammatory diseases. In chronic parasitic infections, IgG4 levels are elevated in individuals with asymptomatic parasitic worm infection compared to symptomatic patients, and this IgG4 has been linked to immune-suppressed states (72) and blocking of antibody-mediated hypersensitive responses in basophils (66). IgG4 tend to exhibit enhanced antigen-affinity (73,74); thus, IgG4 may outcompete binding of functional TB-specific antibodies in immune complexes, thereby diminishing antibody effector activity. And a study focusing on IgG4 biology, using monoclonal antibodies, found that IgG4 antibody could block phagocytic functions of antibodies (57). Moreover, a study comparing Yanomami people and Brazilians of European descent with ATB, found an association of TST anergy and elevated TB-specific IgG4 in the Yanomami people with ATB (75), leading the authors to speculate that the Yanomami developed immune responses to Mtb infection that is poorly protective against TB disease. Thus, the presence of IgG4 titers may not only be indicative of high antigen-burden in ATB but might dampen the antibody-mediated functions in TB.
Our findings point to a unique antibody profile in txATB that differs from ATB and LTBI. Both inflammation-associated  Figures 2-4, a three-way PLSDA analysis summarizes the comparison of LTBI, ATB, and txATB. The depicted biplot overlays the scores plots of individuals color-coded based on their TB status on the two-dimensional loading plots of the 8 input features (red diamonds). LV1 accounts for 27% of variance in X and 32% of variance in Y, whereas LV2 explains 14% X variance and 11% Y variance. To assess the model performance, 5-fold cross validation was performed resulting in 80% CV accuracy. Permutation testing results showed that this model outperformed 90% of models based on shuffled group labels (Wilcoxon p = 0.1). (B) AUC of PPD-specific IgG4 titers. (C) frequency of G1S1F glycan on IgG (D) frequency of G2FB glycan on IgG (E) The frequencies of totaled G2 structures without sialic acid on IgG-Fc. (F) AUC Ag85A/B-specific total IgG titers. (G) PPD-specific ADCP Phago Score. (H) AUC Ag85A/B-specific IgM titers (I) AUC HspX-specific IgG1 titers. Univariate plots (B-I) show median and interquartile range of the measured values and statistically significant differences between LTBI, ATB, and txATB groups were calculated using Kruskal-Wallis test with Dunn's multiple comparison test correction: *p < 0.05, ***p < 0.0005, and ****p < 0.0001. glycosylation and IgG4 titers found in ATB are diminished upon completion of treatment, providing an attractive set of humoral features to explore more broadly in longitudinal studies tracking treatment success. It should be noted that our co-correlate network analysis in the pairwise comparisons of LTBI/ATB ( Figure 1G), ATB/txATB ( Figure 2G), and LTBI/txATB ( Figure 3G) also highlighted both IgA and IgM related features that highly correlated with the features selected by LASSO/PLSDA models with additional discriminatory potential for the TB disease states studied here. We found significantly elevated total IgA, enriched in ATB, inversely correlated with digalactosylated IgG-Fc levels found in LTBI and txATB, suggesting expanded IgA titers also mark ATB state. Consistent with total IgA expansion in ATB, IgA2 titers specific for Ag85A/B were positively correlated were G1S1F on IgG-Fc ( Figure 1G). While extensive antigens were not used to characterize the IgA response in this cohort, this observation is consistent with several findings of elevated TB-antigen specific IgA titers in untreated TB (52,53,76). Interestingly, PPD-IgM and HspX-IgG1 titers were significantly higher in txATB compared to ATB of this cohort (Supplemental Figures 1 and  2) and is consistent with a previous observation of expanded TBspecific antibody titers developing following TB treatment, which tracked with the control of replicating Mtb (77). Further studies will be important to identify Mtb antigen specificities that expand during therapy, which could be used to track treatment responses.

DATA AVAILABILITY STATEMENT
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

ETHICS STATEMENT
The studies involving human participants were reviewed and approved by INMI Ethical Committee Approval Number 72/ 2015. The patients/participants provided their written informed consent to participate in this study.

AUTHOR CONTRIBUTIONS
GA and PG conceived and designed the experiments. PG, AC, and LL performed the experiments. Data was analyzed by PG and SD. Multivariate analysis and modeling were performed by SD. Clinical cohort was established and collected by DG, FP, and LP. The manuscript was written by PG, SD, and GA with contributions from SJ, TO, DG, SF, and DL. All authors contributed to the article and approved the submitted version.

FUNDING
This work was supported by the Ragon Institute (to GA), the SAMANA Kay MGH Research Scholar Program (to GA), NIH T32 AI007061 (to PG), and Italian Ministry of Health Ricerca corrente, Linea 4 (to DG).