Sensitivity of Interferon-γ release assays is not compromised in tuberculosis patients with diabetes

SUMMARY

Setting

The sensitivity of the IFN-γ release assays (IGRAs) to detect Mycobacterium tuberculosis infection or disease may be affected by immune derangement in diabetes. Since millions of type 2 diabetes patients are at risk for tuberculosis worldwide, it is important to determine if the sensitivity of IGRAs is compromised in this vulnerable population.

Objective

Determine if IGRA sensitivity is reduced in tuberculosis patients with diabetes.

Design

The sensitivity of IGRAs (QuantiFERON®-TB Gold and T-SPOT®.TB) was evaluated on specimens from newly-diagnosed adults with microbiologically-confirmed TB with and without diabetes. We also evaluated the association between QuantiFERON-TB Gold results and diabetes-associated conditions (dyslipidemia, obesity).

Results

QuantiFERON-TB Gold sensitivity was 70% among tuberculosis patients. Those with diabetes, chronic hyperglycemia or overweight/obesity were more than twice as likely to have positive test results multivariate models (p<0.05). Low HDL cholesterol or high triglycerides were not associated with assay results. In a separate group of tuberculosis patients (n=43) the T-SPOT.TB was 93% sensitive, with similar performance in patients with and without diabetes.

Conclusion

IGRA sensitivity is not compromised by diabetes in TB patients. Accordingly, IGRAs may also be suitable to diagnose TB infection in diabetes patients, which is required to assess TB risk.

INTRODUCTION

The impact of diabetes on the burden of tuberculosis (TB) is of growing concern as the global prevalence of type 2 diabetes mellitus (hereafter referred to as ‘diabetes’) is expected to double by 2030.^1–3 There are millions of diabetes patients at risk for TB: seven of the high-burden countries for TB are among the top ten countries for diabetes prevalence worldwide.⁴ Diabetes increases the risk of active TB by 3-fold and may account for up to one-third of the TB patients in regions where both diseases are frequent (Restrepo et al, submitted).⁵ These findings highlight the importance of detecting TB infection among diabetes patients in order to design strategies for TB prevention.

The tuberculin-skin test (TST) has been used for more than 100 years to assess immunological memory to latent M. tuberculosis infection or as an alternative support to TB diagnosis when microbiological confirmation is inconclusive or unavailable. However, its cross-reactivity with the BCG vaccine hinders its widespread use in TB-endemic countries where the vaccine is routinely applied. Such is the case in our study population on the Texas-Mexico border. The advent of commercially-available Interferon-γ (IFN-γ) release assays (IGRA) to detect M. tuberculosis infection or TB, circumvents this limitation by using two purified antigens (ESAT-6 and CFP-10) that do not cross-react with BCG.^6–8 Since IGRAs such as the QuantiFERON-Gold (QFT-G) and T-spot.TB provide a better correlation to M. tuberculosis exposure than TST, they may be better predictors of TB development.⁹ However, the sensitivity of IGRAs may be lower in diabetes patients. Studies in TB patients show that diabetes is associated with an altered IFN-γ response to M. tuberculosis antigens in vitro, with some suggesting reduced and others increased cytokine secretion.^10–12 Furthermore, a recent study indicated that TB patients with diabetes were over 4-fold more likely to yield indeterminate QFT results by univariate analysis, even though the effect was lost after adjusting for confounders.^13;14 Since diabetes patients are more likely to present metabolic alterations that can affect immunity^15;16, IGRA sensitivity may also be compromised by diabetes-associated conditions such as obesity, which reduces the risk of TB^17–19, or dyslipidemia.

In this study we evaluated whether the sensitivity of commercially available IGRAS was compromised by diabetes status in TB patients. The association between IGRA results and diabetes-defining (hyperglycemia) or –associated conditions (high body-mass index, dyslipidemia) was also evaluated. Finally, the strength of the IFN-γ response to the M. tuberculosis antigens in the assay was evaluated to determine whether deficient IFN-γ secretion underlies the susceptibility of diabetes patients to TB.

METHODS

Participant enrollment and characterization

TB suspects who were 20 years or older were prospectively identified at pulmonary clinics in the US-Mexico border: Hidalgo and Cameron County Health Departments in south Texas (referred to as ‘US-Texas’), and Secretaria de Salud de Tamaulipas in Matamoros, Mexico (referred to as ‘Mexico-Tamaulipas’). TB diagnosis was based on isolation of M. tuberculosis (culture routinely conducted in Texas, but not in Mexico) or in its absence, a positive smear for acid-fast bacilli. Patients who had received anti-TB treatment for more than seven days were excluded to avoid possible confounding by modulation of the immune response.¹¹ Sociodemographics and known risk factors associated with TB and diabetes were documented at enrollment as described previously.²⁰ Self-reported HIV status was complemented by blood testing in 95% of the participants analyzed for QFT-G results (n=161/169) and in the 43 participants analyzed for T-Spot.TB results. Body-mass index (BMI; weight in kg divided by height in meters squared) was calculated for classification as underweight (BMI<18.5), normal (BMI ≥18.5 and <25), and overweight/obese (BMI ≥25). Lipids [total cholesterol, low-density lipoprotein (LDL) cholesterol, high-density lipoprotein (HDL) cholesterol, HDL/LDL and total cholesterol/HDL ratios, triglyceride levels] were measured in heparinized plasma. Cut-off values were based on American Heart Association guidelines (online Supplement).¹¹ EDTA-anti-coagulated blood was used to assess hyperglycemia (fasting glucose ≥126 mg/dl or random glucose ≥200 mg/dl) and chronic hyperglycemia (HbA_1c≥6.5%) as described previously.¹³ Diabetes was defined according to the American Diabetes Association (ADA) 2010 revised guidelines: HbA_1c≥6.5% or hyperglycemia or self-reported diabetes.²¹ IGRAs (Cellestis, Melbourne, Australia for QFT-G and Oxford Immunotec, Oxfordshire, UK for T-SPOT.TB) were conducted following the manufacturer’s recommendations (online supplement).

Ethics Approval Statement

Participants were enrolled following guidelines from the Institutional Review Boards from the participating institutions of both countries, including signature of an informed consent.

Data Analysis

Data was compiled in SPSS (v.16, SPSS Inc, Chicago) and exported to SAS version 9.1 (Cary, North Carolina) for analysis. Descriptive statistics are provided. Comparisons between groups were done by chi-square tests (expected count > 5 per cell) or Fisher’s exact (expected count ≤ 5 per cell) for categorical variables and Kruskall-Wallis tests for continuous variables. Assessment of the International Units (IU) of IFN-γ secreted in response to each M. tuberculosis antigen or mitogen in the QFT-G kit was estimated by subtracting the background cytokine levels in the NIL from each stimulant. Multivariate models were built to determine whether diabetes, HbA_1c, BMI, or BCG vaccination remained associated with QFT-G results (logistic regression) or IFN-γ levels (linear regression) after adjusting for potential confounders and effect modifiers.^13;22 The independent variables of interest were chosen based on significant associations by univariate analysis with the QFT-G results (BCG, diabetes, chronic hyperglycemia, BMI). Other variables that could possibly affect the QFT-G response were also evaluated in the models. These included the host sociodemographics and risks for TB other than diabetes listed in Tables 1 and 2, and interactions between these variables and the outcome of interest, when the p-value <0.10 (data not shown). Backward removal of independent variables or interactions with the highest p-values was conducted until the only variables left in the final models were those with p-values <0.05.

Table 1.

Comparsion between the two groups of patients tested for either QFT-G or T-SPOT.TB

	QFT-G	T-SPOT.TB
	(n=169)	(n=43)	P value
	n(%)*	n(%)*
T-SPOT.TB/QFT-G results			0.004
Positive	119 (70.4)	40 (93)
Negative	50 (29.6)	3 (7)
Sociodemographics
Country of enrollment			0.03
Mexico-Tamaulipas	130 (76.9)	26 (60.5)
US-Texas	39 (23.1)	17 (39.5)
BCG vaccination			0.09
Yes	114 (74.5)	35 (87.5)
No	39 (25.5)	5 (12.5)
White Hispanic	168 (99.4)	43 (100)	1.00
Age			0.50
20–35	62 (36.7)	16 (37.2)
36–59	83 (49.1)	18 (41.9)
60+	24 (14.2)	9 (20.9)
Gender			0.77
Male	114 (67.5)	28 (65.1)
Female	55 (32.5)	15 (34.9)
TB characteristics at enrollment
Smear result			0.44
Positive	154 (91.7)	36 (87.8)
Negative	14 (8.3)	5 (12.2)
Culture performed			0.06
Yes	84 (50.3)	29 (67.4)
No	83 (49.7)	14 (32.6)
Days of anti-TB treatment			<0.001
None	42 (24.9)	28 (65.1)
Between 1 and 7	127 (75.1)	15 (34.9)
Diabetes and body-mass index
Diabetes			0.14
Yes	73 (43.2)	24 (55.8)
No	96 (56.8)	19 (44.2)
HbA1c ≥ 6.5%			0.10
Yes	56 (33.1)	20 (46.5)
No	113 (66.9)	23 (53.5)
Hyperglycemia			0.36
Yes	47 (27.8)	15 (34.9)
No	122 (72.2)	28 (65.1)
Body mass index			0.15
Underweight	33 (19.5)	3 (7.1)
Normal weight	94 (55.6)	26 (61.9)
Overweight or obese	42 (24.9)	13 (31)
Risks for TB other than diabetes
Alcohol abuse			0.20
Yes	32 (18.9)	12 (27.9)
No	137 (81.1)	31 (72.1)
Drug abuse			0.002
Yes	39 (23.1)	1 (2.4)
No	130 (76.9)	41 (97.6)
History of incarceration			0.49
Yes	10 (5.9)	4 (9.3)
No	159 (94.1)	39 (90.7)
HIV positive			1.00
Yes	5 (3)	1 (2.3)
No	164 (97)	42 (97.7)

^*Due to missing demographic information (BCG vaccine, n=153 for QFT-G and n=40 for T-spot.TB; Smear, n=168 for QFT, n=41 for T-spot.TB; drug abuse, n=41 for T-spot.TB), totals will not sum to the total sample; QFT-G, QuantiFERON®-TB Gold; T-SPOT, T.SPOT®.TB

Table 2.

Characteristics of TB patients associated with QFT-G results

	QFT-G positive	QFT-G negative	P value	ORadj (95%CI)	Variable(s) adjusted for in final model
	(n=119)	(n=50)
Sociodemographics	n(row %)*	n(row %)*
Country of enrollment			0.16
Mexico-Tamaulipas	88 (67.7)	42 (32.3)
US-Texas	31 (79.5)	8 (20.5)
BCG vaccination			0.06		Age, COE
Yes	75 (65.8)	39 (34.2)		0.4 (0.2, 1.2)
No	32 (82.1)	7 (17.9)		1
White Hispanic	118 (70.2)	50 (29.8)	0.51
Age			0.58
20–35	42 (67.7)	20 (32.3)
36–59	58 (69.9)	25 (30.1)
60+	19 (79.2)	5 (20.8)
Gender			0.24
Male	77 (67.5)	37 (32.5)
Female	42 (76.4)	13 (23.6)
TB characteristics at enrollment
Smear result			0.56
Positive	107 (69.5)	47 (30.5)
Negative	11 (78.6)	3 (21.4)
Culture performed			0.049		COE
Yes	65 (77.4)	19 (22.6)		1.8 (0.8, 3.8)
No	54 (63.5)	31 (36.5)		1
Days of anti-TB treatment			0.18
None	33 (78.6)	9 (21.4)
Between 1 and 7	86 (67.7)	41(32.3)
Diabetes and body-mass index
Diabetes			0.01		Age, BMI
Yes	59 (80.8)	14 (19.2)		2.4 (1.1, 5.0)
No	60 (62.5)	36 (37.5)		1
HbA1c ≥ 6.5%			0.007		Age, BMI
Yes	47 (83.9)	9 (16.1)		2.7 (1.2, 6.2)†
No	72 (63.7)	41 (36.3)		1
Hyperglycemia			0.003		BMI, COE
Yes	41 (87.2)	6 (12.8)		3.7 (1.4, 9.7)
No	78 (63.9)	44 (36.1)		1
Body mass index			0.049
Underweight	19 (57.6)	14 (42.4)		0.5 (0.2, 1.1)
Normal weight	65 (69.1)	29 (30.9)		1	None
Overweight or obese	35 (83.3)	7 (16.7)		2.6 (1.0, 6.2)
Risks for TB other than diabetes
Alcohol abuse			0.28
Yes	20 (62.5)	12 (37.5)
No	99 (72.3)	38 (27.7)
Drug abuse			0.83
Yes	28 (71.8)	11 (28.2)
No	91 (70)	39 (30)
History of incarceration			1.00
Yes	7 (70)	3 (30)
No	112 (70.4)	47 (29.6)
HIV positive			1.00
Yes	4 (80)	1 (20)
No	115 (70.1)	49 (29.9)

^*Due to missing demographic information (BCG n=153; Smear n=168), totals for these variables will not sum to the total sample; Variables initially controlled for in multivariate models prior to backward elimination included all the variable under sociodemographics, risks for TB other than diabetes, and body-mass index; COE, country of enrollment; QTF-G, QuantiFERON®-TB Gold

RESULTS

QFT-G studies

Among the 362 TB suspects identified between January 2006-May 2009, 176 were classified with TB and had a QFT-G test performed (Fig S1). Seven (4%) had indeterminate QFT-G results and were excluded from further analysis. The characteristics of the remaining 169 TB patients with valid QFT-G results are shown in Table 1. Over three-fourths were enrolled in Mexico. Accordingly, a similar proportion were BCG vaccinated (vaccination is not administered in Texas). Most TB patients had positive smears for acid-fast bacilli and culture was available for 50%. The most common medical risk for TB was diabetes (n=73, 43%), and among these, 56 (77%) had chronic hyperglycemia. Nearly one-fourth of all TB patients were overweight or obese. The most frequent social risks for TB were alcohol or drug abuse. Three percent of the TB patients had HIV co-morbidity.

Altered lipid metabolism is more frequent in diabetes patients¹⁴, and may affect the immune response to M. tuberculosis in the case of hypercholesterolemia.²³ To assess the frequency of these alterations in our study population we randomly selected the specimens from 84 participants, including 33 (39%) with diabetes. As expected, the most frequent alterations in diabetes (versus no diabetes) patients were low HDL cholesterol levels and high triglycerides (Table S1).

Table 2 summarizes the QFT-G results by host characteristics. The overall sensitivity of the assay was 70% (119/169) among our TB patients. Univariate analysis indicated that diabetes or its defining characteristics (hyperglycemia and chronic hyperglycemia) were associated with QFT-G results. BCG vaccination, positive culture and BMI were marginally associated with QFT-G results. None of the remaining host characteristics in Table 2 or lipid levels (Table S1) were significantly associated with the assay results. Multivariate models indicated that TB patients with diabetes, hyperglycemia, higher HbA_1c, and higher BMI were more than twice as likely to have a positive QFT-G result. In contrast, BCG vaccination or positive M. tuberculosis culture were no longer associated with QFT-G results by multivariate analysis (Table 2).

We next explored whether the levels of IFN-γ secreted in response to ESAT-6 and/or CFP-10 explained the higher sensitivity of TB patients with diabetes versus no diabetes. Furthermore, to identify whether alterations in the secretion of IFN-γ also extended to non-MTB stimulants, we compared the secretion of this cytokine in response to the kit’s mitogen. The final multivariate logistic regression models for each stimulant indicated that diabetes patients or the subset with high HbA_1c secreted significantly higher levels of IFN-γ in response to the M. tuberculosis antigens, but not to the mitogen. Higher BMI was associated with higher IFN-γ secretion to the mitogen (but not the mycobaterial antigens), even after controlling for diabetes (Table 3).

Table 3.

Multivariate models for IFN-γ secretion levels in response to QFT-G stimulants by diabetes, HbA_1c or body-mass index

			Diabetes		HbA1c level			Body-mass index
	Diabetes*	No diabetes*	Padj	Variable(s) controlled for in final model	Adjusted β coefficient	Padj	Variable(s) controlled for in final model	Adjusted β coefficient	Padj	Variable(s) controlled for in final model
ESAT-6	0.4 (2.5)	0.2 (0.7)	0.004	Age, BMI	0.27	0.004	BMI	0.08	0.22	Alcohol, COE, Diabetes
CFP-10	1.4 (3.4)	0.5 (1.4)	0.04	Age, BMI	0.11	0.01	BMI	0.05	0.08	Alcohol, COE, Diabetes
Mitogen	16.9 (17.8)	12.9 (12.8)	0.40	Age, BMI	0.01	0.50	BMI	0.02	0.047	Alcohol, COE, Diabetes

^*Values indicate median IFN-γ level in International Units of IFN-γ/ml (interquartile range); variables initially controlled for in diabetes and HbA_1c multivariate models prior to backward elimination included all the variables listed in Table 2 under sociodemographics, risks for TB other than diabetes, and body-mass index; the same variables are controlled for in the body-mass index multivarite model, excluding body-mass index and including diabetes; COE, country of enrollment; P_adj, P value adjusted for variables controlled for in the final model

T-SPOT.TB studies

The higher sensitivity of the QFT-G among TB patients with diabetes compared to no diabetes motivated us to determine whether a more recently-available commercial IGRA, the T-SPOT.TB, would have a comparable performance in these patients. For this, 43 TB patients were identified among 65 TB suspects enrolled between October 2008–2009 with and without diabetes Their characteristics were similar to those of patients enrolled for the QFT-G studies, except for a higher proportion of patients who had not begun TB treatment, a lower proportion of drug users, and the absence of HIV co-morbidity (Table 1). The T-SPOT.TB was positive in nearly all the patients (93%), and therefore, no difference in sensitivity was identified by diabetes status.

DISCUSSION

We show that sensitivity of the QFT-G and the T-SPOT.TB was not reduced in TB patients with diabetes despite the immune derangement thought to underlie TB susceptibility in these patients. In fact, the sensitivity of QFT-G was significantly higher in TB patients with diabetes versus those without diabetes (81% versus 63%, respectively). Moreover, the sensitivity of T-SPOT.TB was 93%. These data suggests that IGRAs, and the T-SPOT.TB assay in particular, could support the diagnosis of TB in diabetes as recommended for any other TB suspect. That is, IGRAs should only be used to support TB diagnosis when smear is negative, culture data is pending or unavailable, and nucleic acid amplification tests are not accessible.^23–26 Furthermore, since IGRAs are also positive in individuals with latent TB infection (which is their intended application), their use in TB suspects should be limited to low-incidence settings.^27;28

Based on our findings in TB patients, we anticipate that IGRA sensitivity will not be compromised in diabetes patients with latent TB infection. TB patients are prone to develop M. tuberculosis-specific anergy, which is accompanied by compromised IGRA sensitivity.^12;17;29 In contrast, this immunocompromise is not observed in individuals with latent TB infection. Therefore, the sensitivity of IGRAs should be as good (for the T-SPOT.TB) or even better (for QFT-G) in diabetes patients with latent TB infection when compared to those with TB. Accordingly, IGRAs are likely to be a reliable tool to assess latent TB infection in diabetes patients, which is the first step to determine the need for TB prophylaxis.

Previous studies reporting on the sensitivity of IGRAs in TB patients with diabetes were limited by sample size, lack of diabetes confirmation or absence of chronic hyperglycemia evaluation.²² Our study overcomes these limitations. We found that laboratory-confirmed diabetes, its defining characteristics (hyperglycemia and high HbA_1c), and higher BMI which is more frequent in diabetes patients, were independent risk factors for higher QFT-G sensitivity. In contrast, the two most frequent alterations in lipid levels in our study population, high triglycerides or low HDL, were not associated with QFT-G results. Hypercholesterolemia, which has been shown to affect immunity to M. tuberculosis^30;31, was not frequent in our study population. Altogether, our findings suggest that chronic hyperglycemia, and perhaps obesity, may underlie the altered levels of IFN-γ secretion to M. tuberculosis antigens, with negligible impact by low HDL or high triglyceride levels. A secondary finding from this study is the borderline association between low BMI and reduced QFT-G sensitivity. This subset of patients may have advanced TB, where weight loss and up-regulation of T-regulatory lymphocytes are more frequent.¹¹

The higher sensitivity of the QFT-G in TB patients with diabetes was accompanied by higher secretion of IFN-γ in response to the QFT-G antigens. This is in agreement with our previous ex-vivo studies using purified protein derivative (PPD) from M. tuberculosis as antigen. Here, higher levels of pro-inflammatory cytokines were secreted by TB patients with diabetes (versus no diabetes) or with chronic hyperglycemia (versus euglycemia).³² Altogether, the QFT-G results in this study and our previous findings indicate that the higher susceptibility of diabetes patients to TB is not explained by compromised secretion of IFN-γ despite the importance of this cytokine for TB control.³³ A possible explanation for this apparently counterintuitive observation is that diabetes patients have higher bacterial burden, resulting in a more robust stimulation of IFN-γ. This would be consistent with findings from the TB-diabetes mouse model in which mice with diabetes have delayed adaptive immunity, leading to higher bacterial burden and higher IFN-γ levels over time.³⁴ This is also consistent with the higher bacterial burden (proportion of smear-positive) in diabetes patients at time of TB diagnosis.²⁸

We acknowledge potential study limitations. First, the significantly-higher sensitivity of the T-SPOT.TB versus the QFT-G in TB patients is consistent with the literature, but we cannot rule out that differences between the participants evaluated by both methods could have accounted for this disparity (Table 1). Second, we recognize the gold standard for TB diagnosis is M. tuberculosis culture, which was not always available for TB patients in Mexico. Nevertheless, we anticipate misclassification would be low: we have recently begun conducting routine sputum culture confirmation with less than 0.5% of the smear-positive having mycobacteria other than tuberculosis, instead of M. tuberculosis.

In summary, our findings suggest that immune dysregulation in diabetes does not compromise the sensitivity of IGRA assays in TB patients. As the number of diabetes patients exposed to M. tuberculosis increases, IGRAs may prove to be a valuable tool to help tailor TB prevention. In the meantime, much remains to be learned on the association between TB and diabetes. The mechanism by which diabetes patients are more susceptible to TB despite their capacity to secrete high levels of IFN-γ is intriguing and deserved further study.