Professional Documents
Culture Documents
Zambuzi 2016
Zambuzi 2016
Cytokine
journal homepage:
articleinfo
abstract
Article history:
Received 6 June 2016 Although much research has been done related to biomarker discovery for tuberculosis infection, a
Received in revised form 23 August 2016 set of biomarkers that can discriminate between active and latent TB diseases remains elusive. In
Accepted 26 August 2016 the current study we correlate clinical aspects of TB disease with changes in the immune response
as determined by biomarkers detected in plasma. Our study measured 18 molecules in human
plasma in 17 patients with active disease (APTB), 14 individuals with latent tuberculosis infection
Keywords: (LTBI) and 16 uninfected controls (CTRL). We found that active tuberculosis patients have increased
Pulmonary tuberculosis plasma levels of IL-6, IP-10, TNF-a, sCD163 and sCD14. Statistical analysis of these biomarkers
Immune response indicated that simultaneous measure- ment of sCD14 and IL-6 was able to diagnose active
Biosignature tuberculosis infection with 83% accuracy. We also demonstrated that TNF-a and sCD163 were
Diagnostic correlated with tuberculosis severity. We showed that the simultaneous detection of both plasma
Prognostic
sCD14 and IL-6 is a promising diagnostic approach to identify APTB, and further, measurement of
TNF-a and sCD163 can identify the most severe cases of tuberculosis.
© 2016 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.cyto.2016.08.030
1043-4666/© 2016 Elsevier Ltd. All rights reserved.
1
triggering receptor expressed on myeloid cells-1 (TREM-1) in
diameter of cavitation did not exceed 4 cm in patients
serum currently function as a diagnostic tool in critically ill
classified at this stage. Patients with advanced disease (stage
sepsis patients [12]. For TB diagnosis, the accepted approach
3), had lesions that were of greater extent than defined in
for biomar- ker discovery has been the use of gene expression
moderate disease. Table 1 presents the demographic and
profiles, as demonstrated by Berry et al. [13], in which
clinical characteristics of active pulmonary tuberculosis
peripheral blood leuko- cytes from TB patients were found to
patients (APTB), latent infected subjects (LTBI) and non-
have elevated expression of upstream and downstream
infected controls (CTRL).
elements of the type I and II interferon signaling pathway.
Another study has demonstrated that expres- sion of CD64,
LTF, and Rab33A constitute a biosignature, which allows 2.3. Immune biomarkers quantification
discrimination between TB patients and healthy donors [14].
Furthermore, serum levels of IL-6 may be a marker to deter- Blood was collected using Heparin blood collection tubes
mine the effectiveness of antibiotic treatment in patients with (BD Biosciences, San Diego, CA), and plasma separation was
active TB infection [15]. Other investigators have identified performed to quantify cytokine, chemokine and
immune mediators [reviewed by [16]], metabolic markers monocyte/macrophage biomarkers. To evaluate IL-1b, IL-4, IL-
[17] and a proteomics fingerprint [18] that form possible 5, IL-6, IL-10, IL-12p70,
biosignatures indicative of TB. If widely used these IFN-a2, TNF-a, IFN-ɣ, IP-10, RANTES, MCP-1, GM-CSF, IL-17, MIP-
biosignatures may aid in diagnosis and monitoring of 1a and MIP-1b, we used a customized multiplex kit (16-plex,
treatment effectiveness in pulmonary TB patients. EMD Millipore Corporation, Billerica, Massachusetts, USA) and
In the current study we correlated clinical aspects of fol- lowed the manufacturer’s instructions. Following
specific incubation with fluorescent magnetic microspheres coated with
stages of TB infection with changes in the immune response, capture antibodies specific for the above listed cytokines,
as determined by measurement of cytokines detected in samples were evaluated in a fluorescent bead-based instrument
plasma. We demonstrated that the plasma levels of cytokines (Luminex® MAGPIX® System; Luminex Corporation, Austin, Texas,
in patients with active pulmonary TB were elevated when USA). Luminex bead-based data were analyzed using Milliplex
compared to latently infected individuals or uninfected Analyst software v3.5 (Millipore; VigeneTech Inc., Boston,
controls. Specifically we found increased levels of TNF-a, IL-6, Massachusetts, USA) and a three-parameter logistic curve fit.
IP-10 and monocyte acti- vation markers, sCD163 and sCD14. The concentration of soluble monocytes/macrophages plasma
Of these, we found that IL-6 and sCD14 were sufficient to biomarkers CD14 and CD163 were measured using a DuoSet
discriminate active TB disease from both latently infected and ELISA kit (R&D Systems, Minneapolis, Minn), according to the
uninfected individuals. Analyzed together, levels of these manufacturer’s instructions.
molecules may discriminate between latent and active
infection and thus provide an effective TB biosig- nature.
Additionally we demonstrate that sCD163 and TNF-a may be 2.4. Statistical analysis
used to indicate the severity of Mtb infection.
Conventional statistical analyses were performed using
GraphPad Prism 5.0 software (Graph-Pad Software, San
2. Materials and methods Diego, CA, USA). Differences between groups were first
evaluated to test their normality. Considering the
2.1. Ethical aspects nonparametric nature of all data sets, statistical analyses
between the groups were performed by the Mann-Whitney
This research has approval of the Ethics Committee at the test. The correlation between immune biomarkers and stages
Hospital das Clínicas de Ribeirão Preto and FMRP-USP of pulmonary lesions visualized by radiog- raphy were
(Protocol #6481/2013). All patients and controls provided a calculated using Spearman rank correlation. Values of P <
written informed consent form for participation in the study. 0.05 were considered significant.
Fig. 1. Quantification of cytokines and chemokines from plasma. Plasma was separated from whole blood drawn from pulmonary tuberculosis patients (APTB), latent
infected subjects (LTBI) and non-infected controls (CTRL), by centrifugation and chemokines/cytokines were measured using a customized kit (MILLIPLEX ® MAP Human
Cytokine/ Chemokine; Millipore). Each point represents an individual sample. The results are expressed in pg/mL and were analyzed by the non-parametric Mann-Whitney
test, with significance P < 0.01667 (corrected by multiple comparisons).
1
Additionally, the cytokine profile of each individual was
play an important role in the immune response against bacilli,
assessed to determine individual immune biosignatures. Initially,
in one case, acting to contain the spread of the
all data were used to calculate the global median value of
microorganism through the Th1 INF-c/TNF-a axis that promotes
each marker, which was used as a cut-off point to classify
macrophage acti- vation and increases the intracellular killing
each individ- ual as ‘‘low” or ‘‘high” producers of a given
of Mtb, while the Th2 associated cytokines IL-4, IL-5 and IL-
immune marker. Data were organized in gray-scale diagrams
13 have an inhibitory effect on macrophage mediated killing
to calculate the frequency of high producers for each clinical
of microbes [24]. In this current study, we evaluated levels of
group. Relevant data (P50%) were highlighted in both bold these inflammatory mediators in plasma at different stages of
and underline format. Radar charts was performed in TB infection (Fig. 1). Therefore, we found an increased level
Microsoft Excel (Microsoft Office 2013, Las Vegas, USA) to
of IL-6, IP-10 and TNF-a in APTB patients compared to both
characterize the overall frequency of individuals with high
other groups (Fig. 1C–F).
levels of each marker for each study group, showing potential To determine if any of the cytokine levels correlated with
immune biomarker signature. each other, we analyzed this data set using the Spearman
Finally, to understand how these markers could stratify our rank correla- tion test using Cytoscape software (Fig. 2). We
groups, we performed a heatmaps using the heatmap.2 found significant interactions between cytokines and
function in the R (Project for Statistical Computing Version chemokines in plasma from non-infected controls (CTRL),
3.3.0) and gplots package. A decision tree was generated latent tuberculosis infection (LTBI) subjects and active
using machine-learning algorithms and built using the WEKA pulmonary tuberculosis (APTB) patients. These interactions were
software (Waikato Environ- ment for Knowledge Analysis, further classified according to the value of the correlation
version 3.6.11, University of Waikato, New Zealand). The coefficient (r) into strong, moderate, weak and negative
algorithm C4.5 was used to build the decision tree using the correlations. Overall, patients with active tuberculosis had an
parameter J48. This method analyzed all phenotypic ele- vated number of correlations amongst most of plasma
characteristics to select a minimum set of markers that could cytokines analyzed when compared to the other groups, and
efficiently separate study groups. these correla- tions existed between cytokines that are known
to play an impor- tant role during Mtb infection. For example,
3. Results APTB patients had increased number of correlations amongst
TNF-a, IFN-c, IP-10, IL-17 and IL-6 when compared with
3.1. Identifying the biosignature of patients infected with Mtb LTBI subjects. In contrast to our findings in APTB patients,
increased IL-10 levels were corre- lated with latent infection.
During the progression of infectious disease, it is the This correlation was reduced during active disease,
activation of the immune system and the resulting release of demonstrating that an inflammatory profile charac- terized the
cytokines and chemokines that causes both bacterial progression of Mtb infection to the active form. Other studies
clearance and disease- associated pathology. In TB infection, have reported a protective role of IL-10, and found that IL-10
cytokines and chemokines and IFN-c levels were elevated and expressed simultaneously
Fig. 2. Network patterns among plasma cytokines. Networks were organized to determine the interactions among cytokines detected in the plasma from pulmonary
tuberculosis patients (APTB), latent infected subjects (LTBI) and non-infected controls (CTRL). The significant correlations with p < 0.05 are represented by lines according
1
to the intensity of the association, classified as weak, moderate, strong and negative according to correlation coefficient value (r).
1
and were related to the disease resolution in clinically cured (corrected by multiple comparitions).
Fig. 3. Quantification of plasma levels of sCD163 and sCD14. The levels of soluble
CD163 (a) and sCD14 (b) in plasma from pulmonary tuberculosis patients (APTB),
latent infected subjects (LTBI) and non-infected controls (CTRL) were measured
by ELISA. Each point is representative of an individual. The results, expressed in
ng/mL, were analyzed by the non-parametric Mann Whitney test, p < 0.01667
1 Using all markers that were significantly different among
the groups in our analysis showed in Figs. 1 and 3 (i.e., IL- Mtb infection of macrophages initiates a cascade of
10, TNF-a, IL-6, sCD163 and sCD14) we calculated the Z cytokine production including IL-12p40, IFN-c, TNF-a, IL-1b, IL-
score for all individu- als included in our study groups and 6 [30,31].
generated a heatmap, as shown in Fig. 5. This analysis
indicated that sCD14 was consis- tently highly expressed in
APTB individuals compared to the other two groups. This
analysis also highlighted differences in IL-6, TNF- a and
sCD163. These data indicate that these mediators could dif-
ferentiate active TB infection from both latent and uninfected
individuals.
Finally, using these results, we constructed a decision
tree to demonstrate how these markers can segregate APTB
individuals from other groups. Using this tree, we divided all
individuals into two groups, those with active pulmonary
tuberculosis and those without disease. This latter group
was comprised of individuals with latent infection and the
no-infection controls. From the 47 individuals included in the
construction of the decision tree, we obtained an accuracy
median of 97.87%, classifying 46 out of 47 individuals, as
shown in Fig. 6. To test if our model could accurately
diagnose TB we performed an analysis of leave-one-out
cross vali- dation (LOOCV) [29]. This analysis is commonly
used to validate statistical tests performed on data that are
difficult to confirm because of the difficulty of obtaining
replicates. Applying this vali- dation approach to our data,
we obtained an accuracy median of 82.97%, demonstrating
that the decision tree we built and thus tested can be used
to segregate patients with active disease from those with
latent or no disease. These data demonstrate that the most
important markers used in determination of active disease
status were sCD14, followed by IL-6. Importantly, sCD14 by
itself could classify 15 out of 17 APTB patients, highlighting the
effective- ness of this marker. When we analyze the AUC
(area under the ROC curve), i.e., true positive versus false
positives, good results were obtained. The classifier obtained
true positive results in 89.2%, for the both classes, APTB and
Non-disease (data not shown).
4. Discussion
104 F.A. Zambuzi et al. / Cytokine 88 (2016) 99–107
1
Fig. 4. Representation of a possible global signature of pulmonary tuberculosis. (A) This diagram was designed using the value of the global median of each soluble
mediator as the cutoff point to mark each individual as ‘‘low” or ‘‘high” regarding their expression levels of a given analyte. Those marked in black correspond to values
higher than the overall median of the respective analyte, and blanks are those with less than the median values. Below each table are shown the frequency (%) of
individuals with high levels of mediators while the relevant data (P50%) are in bold and underlined format. (B) Radar charts summarize the signature of the mediators
analyzed in pulmonary tuberculosis, relative to control and latent subjects, wherein each axis shows the proportion of subjects with high levels of a given mediator. The
relevant data (P50%) are indicated by arrows (?). APTB: pulmonary tuberculosis patients, LTBI: latent infected subjects, CTRL: non-infected controls.
Fig. 6. Decision tree to segregate APTB patients from individuals without active
infection. The values of IL-6, TNF- a, IP-10, sCD163 and sCD14 were used to
construct the data set. The decision tree was generated based on the clustering of
all individuals into two groups, those with active tuberculosis disease (APTB) and
without disease (CTRL and LTBI individuals). The numbers beside the name of the
groups correspond to the values of correct/incorrect register.
Fig. 7. Correlation between plasma levels of TNF- a and sCD163 and lung injury in APTB patients. Lung injury of APTB patients was classified in a double-blind test and then
divided into three grades. Images are representative of each grade – minimal (1), moderate (2) and advanced (3) disease. Levels of sCD163 (A), TNF-a (B), IL-6 (C),
sCD14 (D), IP-10 (E) and IL-5 (F) were correlated with the degree of lung injury in patients with active disease through Spearman’s rank correlation test. The values of
p and r2 are
specified in each chart.
during Mtb infection is a better predictor of active disease The authors declare that they have no commercial or other
than any one marker alone. association that might pose a conflict of interest on the
Our study also demonstrated that plasma levels of sCD163 manuscript
and TNF-a correlated with disease severity in APTB patients.
Other studies have demonstrated that increased sCD163 in
plasma corre- late with mortality and advanced TB disease.
Our study confirms these results and demonstrates that sCD163
is a critical component in the TB biosignature. Although TNF-a
is associated with immune protection in patients with Mtb
infection, there is also evidence that high levels of TNF-a are
related to immunopathological responses and contribute
significantly to lung damage [46]. Other studies have
demonstrated a positive association between TNF-a serum
levels and the severity of TB [47]. Here we suggest that TNF-
a, along with sCD163, can be used to stage the progression
of pulmonary TB.
In conclusion, we have demonstrated that quantification of a
set of well characterized biomarkers in plasma can be used in
a novel way to classify different stages of disease progression
in patients infected with Mtb. Specifically, we have shown
that the simultane- ous detection of plasma sCD14 and IL-6
can identify APTB, and measurement of sCD163 and TNF-a in
APTB patients can further classify the degree of disease
progression. Measurement of these molecules in a larger
cohort of patients is needed to confirm the accuracy and
utility of these markers as an accurate diagnostic tool for Mtb
infection.
Funding
Conflict of interest
Acknowledgements
References
[29]