Participant description and baseline characteristics
A total of 100 asymptomatic individuals (49 TBEx and 51 healthy individuals) were enrolled. Figure 1 illustrates group categorisation based on IGRA results. Results from 4/100 (4%) participants were IGRA test indeterminate thus excluded from further analysis. Compared to healthy participants (n=48), TBEx participants (n=45) were older (Median[IQR] age = 36years [28-42years] versus 27years [25-32years]; p=0.0008, Mann-Whitney U test), yielded less total RNA (Median[IQR] 66.4ng/µl [44.2 – 89.4 ng/µl] versus 83ng/µl [64.8 – 107.5 ng/µl]; p=0.001, Mann-Whitney U test), and had lower levels of formal education (p=0.006, Chi-squared test).
Twenty-five (51%) TBEx individuals were QFT-Plus positive and classified as LTBI. Eleven (22%) of the healthy participants were IGRA positive and denoted HCIGRA+. Baseline characteristics of participants in each category are shown in Table 1. Fourteen of the enrolled TBEx individuals were HIV positive (5 LTBI and 8 TBExIGRA-) and their median CD4 count was 501cells/mm3 and 509cells/mm3 respectively (Table 1). Subsequently, after exclusion of HCIGRA+ and TBExIGRA-, compared to HC (n=37), LTBI individuals (n=24) were older (p=0.0002; Mann-Whitney U test), yielded less RNA (p=0.002; Mann-Whitney U test) and were less educated (p=0.006; Chi-squared test). All other variables didn’t differ between groups (Table 1).
Table 1: Patient baseline characteristics
Participant Demographics
|
Level
|
LTBI [n=24]
|
TBExIGRA- [n=21]
|
HC [n=37]
|
HCIGRA+ [n=11]
|
Sex; n [%]
|
Male
|
7 [34.7]
|
8 [38.1]
|
19 [52.9]
|
7 [63.6]
|
|
|
|
|
|
|
Age; years
|
Median [IQR]
|
36 [30 - 42]
|
36 [28 - 42]
|
26 [25 - 31]
|
32 [24 - 39]
|
|
|
|
|
|
|
Weight; kgs
|
Median [IQR]
|
62.5 [–53 -71]
|
57.3 [53 -64]
|
62 [57.2 – 68.8]
|
59.0 [55.6 - 69]
|
|
|
|
|
|
|
HIV Status; n [%]
|
Negative
|
19 [79.2]
|
13 [61.9]
|
37 [100]
|
11 [100]
|
|
Positive
|
5 [20.8]
|
8 [38.1]
|
0
|
0
|
CD4 Count; cells/mm3
|
Median [IQR]
|
501 [459 - 563]
|
509 [427 - 588]
|
NA
|
NA
|
|
|
|
|
|
|
Smoking; n[%]
|
No
|
23 [95.8]
|
21 [100]
|
32 [86.5]
|
11 [100]
|
|
Yes
|
1 [4.2]
|
0
|
5 [13.5]
|
0
|
|
|
|
|
|
|
Firewood; n[%]
|
No
|
22 [91.7]
|
20 [95.2]
|
35 [94.6]
|
11 [100]
|
|
Yes
|
2 [8.3]
|
1 [4.8]
|
2 [5.4]
|
0
|
|
|
|
|
|
|
Alcohol; n[%]
|
No
|
20 [83.3]
|
13 [61.9]
|
27 [73.0]
|
9 [81.8]
|
|
Yes
|
4 [16.7]
|
8 [38.1]
|
10 [27.0]
|
2 [18.2]
|
|
|
|
|
|
|
BCG Vaccine; n[%]
|
Yes
|
24 [100]
|
21 [100]
|
35[94.6]
|
11 [100]
|
|
|
|
|
|
|
Can read; n[%]
|
Yes
|
23 [95.8]
|
21 [100]
|
37 [100]
|
11 [100]
|
|
|
|
|
|
|
Work status; n[%]
|
Employed
|
13 [54.2]
|
12 [57.1]
|
19 [51.4]
|
7 [63.6]
|
|
|
|
|
|
|
Marital Status; n[%]
|
Married
|
18 [75]
|
18 [85.7]
|
25 [67.6]
|
7 [63.6]
|
|
|
|
|
|
|
Education; n[%]
|
Secondary
|
11 [45.8]
|
13 [61.9]
|
23 [62.2]
|
8 [72.7]
|
|
Tertiary
|
4 [16.7]
|
2 [9.5]
|
12 [32.4]
|
2 [18.2]
|
|
|
|
|
|
|
RNA Yield; ng/µl
|
Median [IQR]
|
59 [44 -84.6]
|
72 [52.4 - 112]
|
91.2 [72.4 - 117]
|
71 [59 - 83]
|
|
|
|
|
|
|
IGRA; n[%]
|
Positive
|
24 [100]
|
0
|
0
|
11 [100]
|
|
Negative
|
0
|
21 [100]
|
37 [100]
|
0
|
Comparative analysis of Interferon gamma release level and TM expression in LTBI compared to HC participants.
Using Mann Whitney U test, we assessed whether TM expression was different in LTBI participants compared to HC. Sixty-one participants (24 LTBI and 37 HC) were analysed. Mean [SD] interferon gamma (IFN-g) level of the LTBI individuals was 4.2±3.6 IU/ml and 4.03 ±3.5 IU/ml for QFT-Plus tubes TB1 and TB2 respectively. Correspondingly, low levels of TM ZNF296 (310.5copies/µl vs 535copies/µl), KLF2 (18625copies/µl vs 35650copies/µl), ASUN (364.7copies/µl vs 902.5copies/µl), NEMF (3577.5copies/µl vs 7780copies/µl), PTPRC (16350copies/µl vs 30100copies/µl), DUSP3 (58.7copies/µl vs 110.5copies/µl), GBP6 (155.5copies/µl vs 332copies/µl), DHX29 (407.8copies/µl vs 719copies/µl), GBP5 (878copies/µl vs 1195.5copies/µl), ARG1 (1952.5 copies/µl vs 3265copies/µl), and C1QB (56copies/µl vs 149.5copies/µl) were observed in LTBI compared to HC respectively using Kruskal-Wallis test and Dunn's Test for multiple comparisons. Figure 2 illustrates significant group comparisons.
Robust modelling confirmed suppressed expression marker levels were significantly associated with LTBI
Robust logistic models were obtained by first evaluating collinearity, possible transformations of the covariates and the presence of influential observations on a first set of logistic regression models where the effect of the host gene markers on the probability of LTBI was corrected by age and yield. Influential point analysis detected two observations with strong influence (statistic >10) on the regression parameters limiting the performance of the fitted models (supplementary analysis 1). These 2 observations were assessed (see discussion) and removed. Model selection among multiple subsets of the covariates was carried out using the AIC and likelihood ratio tests. Our results showed that models including the suppressed expression of one of the host gene markers ZNF296 (p=0.008), KLF2 (p=0.012), or DUSP3 (p=0.037), significantly discerned LTBI from HC individuals, after controlling for age and yield (Table 2 and supplementary analysis 2). The regression equations for host gene marker ZNF296 showed that, holding constant age and RNA yield, a decrease of 91 copies/µl (0.1log10 decrease) in expression level of an individual from the median of 444 copies/µl was associated with 48% increase in the odds of having LTBI. The same decrease in the log expression marker KLF2 (6458 copies/µl) and DUSP3 (19 copies/µl) from the median log expression level of 31400 copies/µl and 92 copies/µl respectively, was associated with a 46% and 35% increase in the odds of LTBI. Reduced expression levels of host gene markers ASUN, PTPRC, and GBP5 also showed a trend in association with LTBI although this was not statistically significant possibly due to the small sample size (supplementary analysis 2).
Table 2: Showing regression estimates for the significant expression markers.
Variable (log10)
|
Estimate
|
Std. Error
|
z value
|
Pr(>|z|)
|
ZNF296
|
-6.558
|
2.502
|
-2.621
|
0.009 **
|
KLF2
|
-6.244
|
2.477
|
-2.521
|
0.012 *
|
DUSP3
|
-4.308
|
2.075
|
-2.076
|
0.038 *
|
Models with ZNF296 and KLF2 best fit the data and are reproducible for LTBI classification
The Likelihood ratio test and delta AIC showed that inclusion of the expression marker of either ZNF296 or KLF2 to the model with age and yield alone significantly improved the model fit with a p-value of 0.0009 and 0.0036 respectively and a noticeable reduction in the AIC (Table 3) thus TM preceded all the rest and were further evaluated for robustness. Model check analysis confirmed that these TM met the linearity assumptions (Supplementary Analysis 3). Hosmer-Lemeshow (HL) goodness of fit test for the best models including ZNF296 and KLF2 showed no evidence of poor fit for all risk categories (supplementary Analysis 4). Lastly, bootstrapping analysis revealed that the bootstrap standard errors of the estimates of ZNF296 and KLF2 were similar to those obtained in the standard logistic regression model and the coefficient estimates were within the 95% CI of the bootstrap estimates (Supplementary Analysis 5). These results are highly suggestive that the role of suppressed expression levels of the host gene markers ZNF296 and KLF2 in LTBI detection could be confirmed in a larger population.
Table 3: Likelihood Ratio test and AIC (Akaike Information Criterion) results
Models
|
Likelihood Ratio test
|
AIC
|
Delta AIC#
|
Deviance Resid.
|
Df Resid.
|
Dev
|
Pr(>Chi)
|
|
Model Age - Model X
|
Model Age+Yield - Model X
|
NULL
|
58
|
77.94
|
|
|
Age
|
21.72
|
57
|
56.22
|
3.162e-06 ***
|
60.22
|
|
|
Age+Yield
|
8.92
|
56
|
47.3
|
0.002825 **
|
53.30
|
6.92
|
|
Age+Yield+ZNF296
|
10.90
|
55
|
36.4
|
0.0009606 ***
|
44.40
|
15.82
|
8.90
|
Age+Yield+KLF2
|
8.44
|
55
|
38.86
|
0.003670 **
|
46.86
|
13.36
|
6.44
|
Age+Yield+DUSP3
|
6.11
|
55
|
41.19
|
0.013421 *
|
49.19
|
11.03
|
4.11
|
Age+Yield+ASUN
|
4.78
|
55
|
42.53
|
0.028842 *
|
50.53
|
9.69
|
2.78
|
Age+Yield+GBP5
|
3.65
|
55
|
43.65
|
0.056105 .
|
51.65
|
8.57
|
1.65
|
Age+Yield+PTPRC
|
3.11
|
55
|
44.19
|
0.077857 .
|
52.19
|
8.03
|
1.11
|
Age+Yield+NEMF
|
2.98
|
55
|
44.32
|
0.084195 .
|
52.32
|
7.90
|
0.98
|
Age+Yield+GBP6
|
1.71
|
55
|
45.60
|
0.191
|
53.60
|
6.63
|
-0.29
|
Age+Yield+C1QB
|
1.28
|
55
|
46.02
|
0.258
|
54.02
|
6.20
|
-0.72
|
Age+Yield+ARG1
|
0.60
|
55
|
46.70
|
0.438
|
54.70
|
5.52
|
-1.40
|
Age+Yield+GAS6
|
0.52
|
55
|
46.78
|
0.469
|
54.78
|
5.44
|
-1.48
|
Age+Yield+DHX29
|
0.08
|
55
|
47.22
|
0.778
|
55.22
|
5.00
|
-1.92
|
Age+Yield+CD64
|
0.01
|
55
|
47.30
|
0.931
|
55.30
|
4.93
|
-1.99
|
Age+Yield+BATF2
|
0.31
|
55
|
46.99
|
0.578
|
54.99
|
5.23
|
-1.69
|
*Significant p-value for likelihood ratio test
# A change in AIC of ≥5 is considered noticeable
Similar gene expression profiles were seen in TBExIGRA- individuals as in LTBI.
The TM expression levels among TBExIGRA- (n=21) and HCIGRA+ (n=11) participants were profiled. Mean [SD] IFN-g levels were 0.03±0.08 IU/ml and 0.04 ±0.08 IU/ml for TBExIGRA- and 3.2±3.7 IU/ml and 2.6 ±2.8 IU/ml for HCIGRA+ individuals for QFT-Plus tubes TB1 and TB2 respectively.
TBExIGRA- expression profiles were significantly suppressed for host gene markers ZNF296, KLF2, ASUN, NEMF, PTPRC, DUSP3, DHX29, GBP5 and GAS6 compared to expression among HC (Refer to Figure 2). No difference in expression profiles among TBExIGRA- and LTBI participants was observed. Similarly, expression levels in HCIGRA+ didn’t differ from expression profiles among either HC or expression among LTBI individuals although this expression profile was slightly suppressed as in LTBI (Figure 2).
Suppressed expression profiles among both HIV- Negative and Positive TB Exposed contacts.
To verify whether the TM expression pattern among TBEx household contacts was not a result of HIV co-infection, we compared expression profiles of TBEx HIV negative (32/45) and TBEx HIV positive (13/45) participants to HC (n=37) , irrespective of IGRA status. No significant difference in gene expression profiles was observed between TBEx with either HIV negative (TBExHIV-) or HIV positive (TBExHIV+) status for all gene markers (Kruskal-Wallis test and Dunn's Test with Bonferroni corrected p-values for multiple comparisons). Compared to HC, gene expression in TBExHIV- participants was significantly suppressed for host TM KLF2, ZNF296, ASUN, NEMF, PTPRC, DUSP3, GBP6, GBP5 and C1QB. Similarly, gene expression among TBExHIV+ participants was suppressed for host gene TM KLF2, ZNF296, ASUN, NEMF, PTPRC, DUSP3, GBP5, C1QB, DHX29, ARG1, and GAS6 (Figure 3).
Gene expression strongly positively correlated.
Finally, correlation between the evaluated expression markers and the covariates age, and yield was assessed. Strong positive linear correlation was observed between several variables. The best performing host transcriptional genes ZNF296 and KLF2 were positively correlated with each other and with ASUN, NEMF, PTPRC, and DHX29 (Figure 4). Moderate positive association was observed between expression markers DUSP3, GBP5, GBP6 and ARG1 with each other and with the previously mentioned strongly associated expression markers. CD64 was positively correlated with BATF2 and GBP5. A moderate relationship was seen between yield and some expression markers namely DHX29, ARG1, ASUN, KLF2, DUSP3, NEMF and GBP5. Age was weakly negatively correlated with most of the evaluated host gene markers. No association between age and yield was observed.