QuantiFERON-TB Gold In-Tube conversion:
M.tb infection was assessed by measuring mycobacteria-specific IFN- production using the QFT test, performed prior to enrolment, at day 336 and at the end of study. All infants had a negative QFT response at enrolment. 16/43 had a positive QFT ≥ 0.35 IU/ml on day 336 but only two of these remained QFT positive at end of study; they had QFT values of 2.54 IU/ml and 0.82 IU/ml on day 336 and 5.05 IU/ml and 1.33 IU/ml at end of study, respectively. 22/43 infants who were QFT negative at day 336 had QFT values ≥ 0.35 IU/ml at the end of study. 5/43 infants converted to positive QFT at time points other than day 336 and end of study (Figure 2).
Ag85A-specific IgG response on Day -7 was associated with increased risk of M.tb infection
Baseline characteristics used to investigate possible associations between the studied immune parameters and M.tb infection are presented for M.tb-infected and uninfected infants (Supplementary Table 4). The number analysed was dependent on the variables used in the model and how many samples were missing for each of the included variables. The primary aim of this study was to investigate if the three immune correlates of risk of TB disease identified in our previous study in the same population were similarly associated with risk of M.tb infection [6]. We evaluated HLA-DR expression on CD4+ T-cells, frequencies of BCG-specific IFN- in PBMC and plasma Ag85A-specific IgG responses. HLA-DR expression on CD4+ T-cells (OR 1, 95% CI 0.95 – 1.06,p = 0.921, FDR = 0.921, adjusted only for the three immune correlates of TB disease, same for the remaining two parameters) and BCG-specific IFN- - producing cells (OR 0.99, 95% CI 0.97 – 1.01, p = 0.238, FDR = 0.357) were not associated with risk of M.tb infection, whereas Ag85A-specific IgG response on Day -7 was associated with increased risk of M.tb infection (OR 8.85, 95% CI 2.05 – 38.09, p = 0.003, FDR = 0.009).
Systematic upregulation of cytokines, chemokines and complement component factors in infants with subsequent M.tb infection:
As an exploratory analysis, we quantified a number of cytokines, chemokines, and complement factors in plasma samples collected from the study subjects in a multiplex assay. The statistical significance of these immune parameters was much higher than those measured by other assays including ELISpot, flow cytometry and MGIA (Figure 3A, Supplementary Table 3). Almost all of the immune parameters measured by the multiplex assay had an odds ratio >1, indicating that the increased values of these immune parameters were associated with an increased risk of M.tb infection (Figure 3A, Supplementary Table 3). Given that the immune parameters measured by the multiplex assay were closely correlated with each other (Figure 3B), it is unsurprising that almost all of them have a consistent odds ratio. In addition, the AUROC values of the immune parameters measured by the multiplex assay were also higher than those measured by other assays (Figure 3C, Supplementary Table 3). Therefore, it is likely that multiple testing correction with other immune parameters caused the loss of statistical power of the immune parameters measured by the multiplex assay. On the basis of these observations, we performed multiple testing correction within the immune parameters measured by the multiplex assay. The results showed that 26 out of the 45 available immune parameters measured by the multiplex assay had an FDR of no greater than 0.2, indicating that systematic upregulation of chemokines, cytokines and complement components in the plasma was associated with the increased risk of M.tb infection (Supplementary Table 5).
Approximately one third of the blood samples from infants who went on to be infected with M.tb were collected on Day 28, while all samples from M.tb-uninfected infants were collected on Day -7. To investigate the possible confounding effect of the sample collection time points on the univariate conditional logistic regression, we compared data from Day -7 samples only with combined Day -7 and Day 28 data from infants who went on to be infected with M.tb. We selected parameters that are different between Day 28 and Day -7 samples from infants who would be infected with M.tb (unadjusted p-values less than 0.1, Supplementary Table 6A). The odds ratios, calculated by univariate conditional logistic regression, of some of these parameters using Day -7 samples only changed qualitatively (namely from less than 1 to larger than 1 or vice versa) compared to those calculated using all samples (combined Day -7 and Day 28) (Supplementary Figure 2A and Supplementary Table 6B). Of the immune parameters which were significantly different between M.tb-infected and M.tb-uninfected infants (Supplementary Table 5), only Complement Factor I and Complement C5 differed when we looked at data from Day -7 only compared to Day 28 and Day -7 data combined (Day 28 vs Day -7, unadjusted p-value less than 0.1) (Supplementary Figure 2B and Supplementary Table 6B). Taken together, the results suggested that the significant change of 24/26 cytokines, chemokines and complements were not confounded by the time point of sample collection.
CMV infection is not associated with increased risk of M.tb infection:
Cellular and humoral immune responses to CMV were measured by IFN- ELISpot and IgG and IgM ELISA, respectively. We used the manufacturer’s recommendation to determine the cut-off for a positive CMV-specific IgM response. For the CMV-specific IgG response, considering that IgG antibodies can be vertically transmitted from mothers to babies [21], we introduced a more stringent cut-off criteria: only infants who were IgG positive (according to the manufacturer’s cut-off) on day 28 and maintained ≥ 90% of their day -7 IgG responses on day 28 were considered CMV-specific IgG positive (Figure 4A). Infants who had a CMV-specific IFN- response of >17 SFC/1 106 PBMC were considered positive, based on the cut-off used in our previous study on samples collected from the same population [11] (Figure 4B). All infants who had a positive CMV-specific IgM response and all with positive CMV-specific IFN- responses, except one infant who maintained 85% rather than 90% of their day -7 CMV-IgG response on day 28, had a positive CMV-specific IgG response, indicating that measuring IgG is more sensitive in detection of CMV infection (Figure 4C). Therefore, we defined the CMV infection status of infants as follows: infants were defined as CMV-uninfected if they had a negative CMV-specific IgG response; infants were defined as CMV-infected if they had either a positive IgG response, or IgM response or IFN- response; if the infant had negative IgM and IFN- responses with an unknown IgG response because of sample unavailability, they were considered ‘unknown’ (Figure 4C). Using this definition, M.tb infection incidence (based on QFT conversion) was 33.9% (21/62) in the CMV-infected group, compared to 16.7% (6/36) in the CMV-uninfected group (p = 0.0699, Conditional logistic regression (OR 2.607, 95% CI 0.9252-7.344), not corrected for multiple testing) (Figure 4D).
Although we did not detect a direct association between the 3 immune correlates of risk of TB disease and risk of M.tb infection, we detected increased frequencies of HLA-DR+ CD4+ T-cells and HLA-DR+ CD8+ T-cells and decreased levels of BCG-specific IFN- -producing cells in CMV-infected infants (Supplementary Figure 3 and Supplementary Table 7).
We evaluated the association between immune parameters measured by the multiplex assay and CMV infection and found that IP-10, TNF- and complement C2 were all significantly (FDR 0.2) upregulated in CMV-infected infants, compared with CMV-uninfected infants (Supplementary Figure 3; Supplementary Table 7). The plasma levels of these three proteins were also significantly upregulated in infants who became infected with M.tb (FDR 0.2, Supplementary Table 5).
Differential gene expression between infants with subsequent M.tb infection and uninfected infants (controls):
Gene expression analysis was performed in samples from 21 M.tb-infected who had QFT values ≥ 0.35 IU and 52 matched controls. Day 28 samples were used for 2 out of the 21 M.tb-infected because of lack of samples. When cell availability allowed, each infant was represented by a set of 2 samples: BCG-stimulated and unstimulated. A linear model was fitted to determine differential gene expression using the matched case-control set number as the strata variable and adjusted for the stimulus (BCG stimulated or unstimulated), given that BCG stimulation had a strong influence on gene expression. After differential gene expression analysis (Supplementary Figure 4A, Supplementary Table 8A), we performed gene set enrichment analysis using two methods and gene set annotation. The first method was the CERNO algorithm using gene sets adapted from Li et al, 2014 [18]. The second method was overrepresentation analysis using gene sets defined in Gene Ontology (GO).
The number of infants who became infected with M.tb was higher in the CMV-infected group, though this did not reach statistical significance. We therefore performed differential gene expression analysis in CMV-infected and CMV-uninfected infants separately (Supplementary Figure 4B-C, Supplementary Table 8B-C). We also identified transcripts differentially expressed between CMV-infected and CMV-uninfected infants (Supplementary Figure 4D, Supplementary Table 8D).
When we analysed combined data from CMV-infected and CMV-uninfected infants, gene sets associated with B-cells, immunoglobulins, complement activation, type I interferon response and anti-viral response were upregulated in infants who were subsequently infected with M.tb. These included genes such as IGKV4-1, TNFRSF17, IFIH1 and IFIT2 (Figure 5A, Supplementary Figure 5A, Supplementary Table 8A, 9-10).
In CMV-infected infants, gene sets associated with anti-viral response, type-I interferon response, B-cells, immunoglobulins, complement activation, neutrophil activation and cytokine production and secretion were upregulated in infants who were subsequently infected with M.tb compared with those who remained M.tb-uninfected, these included genes such as DDX58, IFIH1, IFI35, CD19, IGKV4-1, C5, C3AR1, FUCA2, NCF4 and IL-10 (Figure 5A, Supplementary Figure 5B, Supplementary Table 8B, 9-10).
In CMV-uninfected infants, gene sets related to complement activation, B-cells and immunoglobulins were upregulated in infants that were subsequently infected with M.tb, these included genes such as C1QBP, CR2, C4BPB, CD19 and IGKV4-1 (Figure 5A, Supplementary Figure 5C, Supplementary Table 8-10). Gene sets related to neutrophil activation, natural killer cell activation, complement activation were downregulated in infants who were subsequently infected with M.tb, these included genes such as C1QA, C1QB, C1QC, C3, C3AR1, C5AR1, C5AR2, FUCA2, KLRB1, KLRD1, KLRC3 and NKG7 (Figure 5A, Supplementary Figure 5D, Supplementary Table 8C, 9-10).
Many genes that were differentially expressed between M.tb-infected infants and M.tb-uninfected infants in CMV-infected infants had the opposite patterns of fold change between M.tb-infected and M.tb-uninfected infants in CMV-uninfected infants (Figure 5B). This led to a lower number of differentially expressed genes, when we analysed combined data from CMV-infected and CMV-uninfected infants, compared to when we analysed them separately by CMV infection status. Gene sets associated with M.tb infection were strikingly different among CMV-infected and CMV-uninfected infants (Figure 5A and Supplementary Figure 5B-5D).
To further validate our RNA-seq results, we analysed a dataset from the training set of M.tb-infected adolescents in the South African Adolescent Cohort Study (ACS), who either progressed to active TB or remained disease-free [7]. Because the gene expression abundance in the ACS was quantified at the level of splice junction, we used the sum of all splice junction counts of each gene as the read count of the gene and performed differential gene expression analysis using DESeq2. Considering the high prevalence of CMV infection in South Africa [22], we compared the results in the ACS and our CMV-infected infants. We found that gene sets related to complement activation, neutrophils, anti-viral response and type-I interferon response were also upregulated in TB disease progressors in the ACS, which included genes such as CR1, C1QA, C1QB, C1QC, C2, C5AR1, NCF4, IFIH1, DDX58. However, gene sets related to B-cells and immunoglobulins were downregulated in progressors in ACS, though most of the differentially expressed genes in these sets were not immunoglobulin genes, which differs from those observed in our study (Figure 5C and Supplementary Table 11).
To investigate the association between CMV infection and susceptibility to M.tb infection we looked at the overlap between gene sets that were enriched in the differential expression analysis between M.tb-infected and M.tb-uninfected infants and also between CMV-infected and CMV-uninfected infants. The results showed overlap of differential expression of genes that were associated with B-cells and immunoglobulins (Figure 5A and Supplementary Figure 5A, 5E-5F, Supplementary Table 8D, 9-10).
Ag85A-specific IgG response correlates negatively with BCG-specific IFN-g response:
We evaluated the possible associations between the BCG induced antibody response and T-cell response in our infection cohort and found a significant negative correlation between the ELISpot BCG response and Ag85A-specific IgG response on Day 28. This was the case both when the analysis included the ELISpot BCG response data from Day -7 samples only and also when Day -7 and Day 28 data were combined (Figure 6A, p<0.05). We then classified the infants into BCG-specific IFN-g responders (SFC ≥ 60) and non-responders, selecting a cut-off (SFC = 60) which can separate the distribution of the BCG-specific IFN-g response into a high-density region and a low-density region (Figure 6B). The Ag85A IgG response was significantly higher in IFN-g non-responders (Figure 6C).