Yadunanda Kumar, Steven Tannenbaum, EngEong Ooi and Jagath Rajapakse have applied for a patent on some of the markers identified in the study. This does not alter the authors' adherence to all the PLOS policies on sharing data and materials.
Conceived and designed the experiments: SRT YK EEO. Performed the experiments: YK CL ZB. Analyzed the data: YK ZB JCR SRT EEO. Contributed reagents/materials/analysis tools: ZB JCR EEO. Wrote the paper: YK SRT EEO JCR.
Infections caused by dengue virus are a major cause of morbidity and mortality in tropical and subtropical regions of the world. Factors that control transition from mild forms of disease such as dengue fever (DF) to more life-threatening forms such as dengue hemorrhagic fever (DHF) are poorly understood. Consequently, there are no reliable methods currently available for early triage of DHF patients resulting in significant over-hospitalization.
We have systematically examined the proteome, cytokines and inflammatory markers in sera from 62 adult dengue patients (44 DF; 18 DHF) with primary DENV infection, at three different times of infection representing the early febrile, defervescence and convalescent stages. Using fluorescent bioplex assays, we measured 27 cytokines in these serum samples. Additionally, we used multiple mass spectrometry methods for iTRAQ-based comparative analysis of serum proteome as well as measurements of protein adducts- 3-nitrotyrosine and 3-chlorotyrosine as surrogate measures of free radical activity. Using multiple methods such as OPLS, MRMR and MSVM-RFE for multivariate feature selection and classification, we report molecular markers that allow prediction of primary DHF with sensitivity and specificity of >80%.
This report constitutes a comprehensive analysis of molecular signatures of dengue disease progression and will help unravel mechanisms of dengue disease progression. Our analysis resulted in the identification of markers that may be useful for early prediction of DHF during the febrile phase. The combination of highly sensitive analytical methods and novel statistical approaches described here forms a robust platform for biomarker discovery.
While the majority of patients who exhibit febrile dengue infection recover within a week, a small proportion of the patients progress to develop severe symptoms that can be life-threatening if not managed in a hospital setting. Because there is no method to accurately identify this subgroup of patients, many dengue patients are hospitalized unnecessarily, which causes significant burden to the healthcare system. In our study, we have systematically measured a large number of molecules including cytokines and serum proteins in blood samples from a dengue patient cohort using highly sensitive mass spectrometry-based methods. We have further developed novel statistical methods that allow us to identify small panels of measureable blood markers, which can distinguish dengue patients that develop milder, self-limiting form of the disease from those that progress to develop severe symptoms. Because these markers can be applied within 48–72 hours of onset of febrile symptoms, we expect them to be useful for early classification of severe dengue disease.
Infection with dengue virus (DENV) causes a spectrum of clinical manifestations ranging from mild dengue fever (DF) to the potentially lethal dengue hemorrhagic fever (DHF) and dengue shock syndrome (DSS)
The majority of dengue patients recover uneventfully after 5–7 days of acute illness. In a small proportion of patients, however, the initial febrile period is followed by a rapid onset of vascular leakage, thrombocytopenia and hemorrhage indicating DHF. The continual loss of intravascular volume from plasma leakage can very rapidly lead to hypotension and cardiovascular collapse which, if not carefully managed, can result in death. In the absence of an effective antiviral drug, the management of dengue patients is primarily supportive. Early recognition of patients with plasma leakage is thus critical for the initiation of appropriate fluid management to prevent onset of hypovolemic shock. However, because these symptoms become evident only in the critical phase of infection, it is currently not possible to distinguish DF and DHF accurately during the early stages of illness, when the disease is less well differentiated
The mechanisms that trigger transition from mild DF to more life threatening DHF are poorly understood, hampering early classification of dengue patients who will progress to DHF. This not only delays treatment but frequently results in the over-hospitalization of patients contributing significantly to the financial burden imposed by dengue
In the last decade, numerous efforts have been made to identify serum markers that may predict severe dengue disease, with an emphasis on cytokines
The early dengue infection and outcome (EDEN) study in Singapore prospectively recruits and follows-up adult dengue patients in Singapore through early febrile, defervescence as well as convalescence stages
The EDEN study is a multi-center longitudinal study of adult febrile infections that was carried out at a number of clinics island-wide in Singapore. Enrollment of eligible individuals was based on written informed consents and the protocols were approved by the National Healthcare Group (DSRB B/05/013). The study protocols have been described earlier
A detailed hematological and virological analysis was performed and a subset of 15 clinical indicators was selected for our statistical analysis. These included white blood cell count (WBC), red blood cell count (RBC), blood hemoglobin (HGB), hematocrit (HCT), macrophage cell volume (MCV), MCH, MCHC, platelet count (PLT), lymphocyte percentage (LYMPH%), lymphocyte count (LYMPH), mixed cell distribution (MXD), neutrophil percentage (NEUT%), neutrophil count (NEUT), red blood cell distribution width-coefficient of variation (RDW-CV), and viral titers. Additionally, we used plasma samples from 50 asymptomatic healthy army recruits collected during their annual physical examination in Singapore as controls in our analyses. A comparison of cytokines in dengue patient sera with healthy plasma is shown in supplementary data (
Cytokine measurements were performed with 12.5 µl sera in duplicates using the Bioplex 27-plex human cytokine kit from BioRad as per manufacturer's instructions. The standard curves were optimized automatically by the software (Bioplex manager) and verified manually. The Bioplex manager software was used to calculate cytokine concentrations and only measurements that showed a coefficient of variability (CV) of <10% were included for further analysis. Levels of interferon-induced cytokine IP-10 in 30% of the dengue patient samples during visit-1 were above upper limit of detection. We repeated the analysis after diluting the serum 100 fold for this subset of samples. Six of the visit-1 samples in DF group and 3 in DHF group still had very high levels of IP-10 and for the statistical analyses, we included these as missing values since levels of other cytokines for these samples were within detectable range. Measurement of 9 acute phase proteins was performed using the Bioplex Pro Acute phase multiplex kit (BioRad laboratories) as per manufacturer's instructions. Two different dilutions of sera were used-1∶1000 for ferritin (FT), serum amyloid A2 (SAA), procalcitonin (PCT), tissue plasminogen activator (tPA), fibrinogen (FB) and 1∶100,000 for alpha-2-macroglobulin (A2M), haptoglobin (HPT), C-reactive protein (CRP) and serum amyloid P (SAP).
We performed an isobaric tagging for accurate quantitation (iTRAQ) method for multiplexed analysis of serum proteome in dengue patients. For this, we first pooled serum samples (5 ul each) from 10 DF and 10 DHF patients selected randomly from our study population. The same groups of patients were followed for pooling of the samples from each visit resulting in a total of 6 pools (3 from DF and 3 from DHF). From 10 microliters of each pooled sample, albumin and immunoglobulins were depleted using the Vivapure anti HSA/IgG kit (Sartorius-stedim, USA). Following concentration and desalting, the protein concentration was measured in each sample using the BCA method in a kit
The samples were analyzed on an Agilent 6520 Accurate-mass QTOF-LC/MS system equipped with a 1200 series HPLC-Chip/MS system. We separated peptides on a HPLC-Chip with 75 um×150 mm analytical column HPLC-Chip and a 160 nL enrichment column. Three injections (1 ul each with approximately 2 ug peptides) of each sample were separated using a 60 min gradient (5% at 0 min 10% 2 min, 50% 42 min 80% 42–50 min, 5% 50–60 min) with Water/0.1%formic acid as aqueous phase and 95%acetonitrile/5%water/0.1%formic acid as organic phase at flow rates of 300 nL/min. Peptides eluting from the LC were injected online into the accurate-mass QTOF and examined in positive ion mode with the following settings for MS mode: 4 spectra/sec, m/z 110–2300, MS/MS mode: 8 spectra/sec, m/z 60–1097 with drying gas flow 5 L/min, 325degC, collision energy slope 3, intercept 2.5, and a capillary voltage of 1950.
Spectrum Mill software (Agilent Technologies) was used for protein identification and quantitation of iTRAQ reporter ion intensities. A minimum peptide score of 8 and a protein score of 10 was used to generate protein lists by searching against the Swissprot database. These thresholds were determined by comparing results from searching the Swissprot database and a reversed random database to identify error rates. The peptide and protein score thresholds indicated above ensured a false discovery rate of <5% in protein identification. Only proteins identified with two or more peptides were selected for relative quantification. A global weighted threshold for fold change was determined by comparing the ratio of summed intensities of each reporter ion for all peptides (115/114- 1.05; 116/114- 0.96; 117/114- 1.09) and then the intensities of each peptide ratio were further corrected by this factor. This weighted threshold was essential to make sure the fold changes observed were not simply due to an overall bias towards one or more reporter ion. Finally, fold change for each protein was calculated as a ratio of summed intensities of reporter ions across different peptides per protein.
Nitrotyrosine (NT) and chlorotyrosine (CT) in human serum were measured by a liquid chromatography-triplequadrupole MS method. Briefly, 2 mg of serum protein was spiked with 4 pmol internal standards (IS) L-3-chloro-[13C9, 15N]-tyrosine and L-3-nitro-[13C9, 15N]-tyrosine, and digested in the sodium acetate solution 0.1 M (pH 7.4) with 0.4 mg pronase E (freshly treated by the size-exclusive micro bio-spin column). The mixture was incubated at 50°C for overnight (16 hrs.) and filtered by Vivospin500 3KMW centrifuge filter at 15,000 rpm to remove undigested protein. The amino acids were further purified by Agilent 1200 series HPLC system (Waldbronn, Germany) on an Xbridge TM Phenyl column (3.5 µm, 4.6×50 mm, Waters, Milford, MA). The fractions containing nitrotyrosine and chlorotyrosine, together with internal standards, were collected and dried by SpeedVac for subsequent LC/MS/MS analysis. Subsequent mass spectrometry analysis of target compounds involved separation on an Xbridge TM Phenyl column (3.5 µm, 1.0×100 mm, Waters, Milford, MA) online injection into an Agilent 6460 triple quadrupole mass spectrometer. Two microliters of each sample was injected and eluted by isocratic 25% methanol (0.1% formic acid) for 13 min at 15 µL/min. CT along with IS were analyzed by regular multiple reaction monitoring (MRM) as follows: 216/170 (CT) and 226/179 (CT, IS). NT along with IS were measured by modified MS3 based in-source fragmentation as follows: 181/117 (NT) and 190/125 (NT-IS) by elevating the potential to 135 V at the ion source. The limits of quantitation achieved were 8.1 and 7.3 nM for CT and NT, respectively.
Overall, our dataset (cytokines, serum proteins and protein adduct measurements) had a largely non-gaussian frequency distribution (D'Agostino and Pearson Omnibus normality test, Graphpad prism), and were unbalanced with unequal sample sizes between different groups necessitating non-parametric data normalization and hypothesis testing where indicated. K-means clustering was performed on time courses of measurements using the Unscrambler-X statistical software package (CAMO software, Oslo, Norway). In order to perform
We used multiple feature selection approaches to select subsets (that is, the predictive markers) that give the highest predictive accuracy. To rank the ability of clinical and cytokine measurements to classify different types of dengue populations (e.g. DF vs. DHF), three feature selection strategies were used: the orthogonal projection least squares (OPLS), the multiple-support vector machine-recursive feature elimination (MSVM-RFE), and the maximum relevance minimum redundancy (MRMR) criterion. The OPLS finds features that maximize the correlation between input features and class labels (X) which has been used extensively for biological classification, for example microarray data analysis
We used the radial basis function (RBF) kernel-based support vector machines (SVM) to evaluate prediction accuracy of each selected subset. The original dataset was partitioned into training and testing datasets, randomly for 1000 times. The regularization and scaling parameters of the RBF kernel SVM were estimated using the leave-one-out cross-validation (LOO-CV) on the training datasets by choosing values selected from a grid formed by the values of regularization and scaling parameters. The hyper-parameters were determined by optimizing the weighted sum of errors in different classes. Here, the weights were used to handle the unbalanced sample sizes of different classes: the error in each class was weighted by the percentage number of samples of another class. The features were then selected by minimizing the same weighted error on the test dataset. This type of analysis was performed using OPLS+SVM, MRMR-MID, MRMR-MIQ, or MSVM-RFE after scaling and standardization as preprocessing of data samples. The performances of features selected by different methods were determined by the area under the curve (AUC) of receiver operating characteristics (ROC). An AUC value of >0.85 was used as a threshold for good predictive performance. Thereafter, one and two-sided two-sample
We selected 62 adult dengue patients from the EDEN cohort, of which 44 were diagnosed as DF, and 18 as DHF. The patients selected in DF and DHF groups had similar age and ethnic distribution (
Patient Groups | Serotype | Age | Race | Blood sampling Time |
DF (n = 44) | D1 (24, 54%) | 39±13.02 | Chinese (36, 81.8%) | 46±20 hours (visit-1) |
D2 (1, 2.3%) | Indian (2, 4.5%) | 73±31 hours (Vist-2) | ||
D3 (19, 43%) | Malay (2, 4.5%) | 16±5 days (visit-3) | ||
D4 (0) | Others (4, 9.0%) | |||
DHF (n = 18) | D1 (12, 66.6%) | 40±14.08 | Chinese (14, 77.7%) | 42±22 hours (visit-1) |
D2 (3, 16.6%) | Indian (2, 11.1%) | 77±32 hours (Vist-2) | ||
D3 (3, 16.6%) | Malay (1, 5.55%) | 20±12 days (visit-3) | ||
D4 (0) | Others (1, 5.55%) |
Average±SD time from fever to phlebotomy (visit-1), between visits 1 & 2 (visit-2) and between visits 2 & 3 (visit-3). DF- dengue fever; DHF dengue hemorrhagic fever; D1-dengue serotype-1; D2-dengue serotype-2; D3-dengue serotype-3, D4-dengue serotype-4.
We examined the key clinical indicators commonly used for the diagnosis of DHF. Blood platelet count dropped significantly from febrile phase to defervescence in both DF and DHF patient groups with DHF patients exhibiting significantly (p<0.05) lower platelet levels during defervescence (visit-2) than DF patients (
We measured the levels of 27 serum cytokines in our dengue patient cohort, using a multiplex assay. A majority of cytokines was maximally elevated in dengue patients during the early febrile phase (visit-1) of infection (
To identify temporal patterns in cytokine flux in patient sera, we performed K-means clustering to group cytokines in DF patients exhibiting similar patterns across the three stages of disease as detailed in the methods section. The cytokine IP-10 was the sole member of cluster-1 (
Overall, our results indicated that cytokines and chemokines associated with innate immune activity (e.g. IFN-γ, IP-10), Th2 cell response (IL-4, IL-10, and IL-13), inflammation (IL-1b, IL-6, and IL-8), chemotaxis of macrophages and neutrophils (Eotaxin, MIP-1b) are all maximally elevated in dengue patients during the early febrile phase. Cytokines IL-12, growth factors FGF and PDGF increased even at convalescence. TNF-α remained below detection levels in our analysis likely because production is transient and missed in our timeline of sample collection. Similarly, levels of IL-2, IL-15, GM-CSF and MIP-1a were below the detection limit in >85% of the samples and were excluded from further analysis.
Differences in temporal profile of a subset of cytokines between DF and DHF patients, identified in the clustering analysis outlined above, prompted us to examine these cytokines more closely across different time points of infection in DF and DHF groups. We observed that DHF patients had lower levels of IFN-γ during febrile phase, a time of peak interferon activity (
A subset of eight cytokines that exhibited different clustering between DF and DHF patient groups were examined individually. Mean values of IFN-ϒ (A), IP-10 (B), IL-4 (C), IL-1b (D), IL-17 (E), G-CSF (F), VEGF (G) and PDGF-BB (H) from patients (44 DF; 18 DHF) are plotted. Statistical confidence (p<0.05) was analyzed by ANOVA kruskall-wallis test (DF vs DHF: *p<0.05, **p<0.01, ***p<0.001, NS- not significant); dengue (DF or DHF) vs. healthy control: §§§p<0.001, §§p<0.01, §p<0.05, # not significant). Standard deviation from mean across populations is shown in the error bars (upper deviation only).
We observed decreased levels of Th2 cytokine IL-4, in DHF patients during the febrile stage, (
Quantitative proteomics by isobaric tagging of peptides allows multiplexing of biological samples thereby reducing variability while increasing accuracy of protein quantitation
A. Functional grouping of proteins identified in the proteomics analysis (see
Protein | Uniprot ID | Peptides | Functional Class |
Fold Change (DF/DHF) | ||
Febrile | Deferv. | Conval. | ||||
|
||||||
Serum Amyloid A2 |
P02735 | 3 | ACR, HDL | 3.4/11.4 | 1.9/4.7 | 1.0/1.4 |
Leucine-rich-alpha-2 glycoprotein |
P02750 | 8 | Unknown | 2.0/6.4 | 1.3/3.6 | 0.9/2.5 |
Hemoglobin subunit-alpha |
P69905 | 4 | RBC | 1.9/5.9 | 0.8/1.0 | 0.8/1.1 |
Actin, cytoplasmic |
P60709 | 2 | cytoskeleton | 2.1/4.2 | 2.2/7.3 | 1.2/2.6 |
Hemoglobin subunit delta | P02042 | 5 | RBC | 1.2/4.1 | 0.9/2.8 | 0.8/1.3 |
Insulin-like growth factor-binding protein 3 | P17936 | 7 | carrier | 1.2/4.0 | 1.2/3.4 | 1.0/4.2 |
Hemoglobin subunit-beta | P68871 | 5 | RBC | 1.3/4.0 | 0.8/2.0 | 0.7/1.0 |
Phosphatidyl-inositol glycan specific phospholipase-D | P80108 | 7 | GPI-anchor cleavage | 1.2/3.8 | 0.9/4.0 | 1.0/2.2 |
Plasma protease C1 inhibitor | P05155 | 6 | ACR, Comp | 1.4/3.8 | 1.7/5.0 | 1.0/2.6 |
Hemoglobin subunit zeta | P02008 | 11 | RBC | 1.0/3.5 | 1.1/1.1 | 0.9/0.7 |
Coagulation factor XII | P00748 | 4 | Coagulation | 1.3/3.4 | 1.3/4.2 | 1.3/3.7 |
EGF-containing fibulin-like ECM protein | Q12805 | 4 | EGF regulation | 1.0/2.9 | 1.5/2.8 | 0.9/3.0 |
Alpha-1B-glycoprotein | P04217 | 10 | Unknown | 1.3/2.8 | 1.2/3.3 | 1.1/2.8 |
Tumor protein 63 | Q9H3D4 | 3 | DNA-binding | 1.5/2.8 | 0.8/2.5 | 0.8/3.8 |
Alpha-2 antiplasmin | P08697 | 4 | ACR, SERPIN | 1.2/2.7 | 1.6/2.5 | 1.3/2.5 |
Alpha-1-antichymotrypsin | P01011 | 18 | ACR, SERPIN | 1.3/2.5 | 1.4/2.6 | 0.8/1.4 |
Hemopexin | P02790 | 18 | Heme transport | 1.0/2.5 | 0.9/1.8 | 1.0/1.7 |
Transthyretin | P02766 | 4 | Hormone binding | 0.8/2.4 | 0.7/0.6 | 1.3/3.2 |
Complement C4-A | P0C0L4 | 34 | ACR, Comp | 1.3/2.4 | 1.6/2.2 | 0.9/1.8 |
Complement component C9 | P02748 | 4 | ACR, Comp | 1.5/2.3 | 1.2/2.9 | 0.9/1.5 |
Inter-alpha-trypsin inhibitor H3 | Q06033 | 6 | Hyaluronan-bind | 0.4/2.3 | 1.1/2.8 | 1.3/2.3 |
Alpha-1-acid glycoprotein 1 | P02763 | 6 | ACR | 1.5/2.3 | 1.4/2.2 | 0.8/1.0 |
Haptoglobin |
P00738 | 26 | ACR | 4.1/2.2 | 2.0/2.9 | 2.0/1.5 |
Antithrombin-III | P01008 | 17 | ACR, SERPIN | 1.2/2.2 | 1.4/3.2 | 1.1/1.9 |
Inter-alpha-trypsin inhibitor H2 | P19823 | 19 | Hyaluronan-binding | 0.8/2.2 | 0.7/1.9 | 1.0/2.4 |
Apolipoprotein A–I | P02647 | 32 | HDL | 0.7/2.1 | 0.9/1.1 | 0.7/1.5 |
Serum Paraoxonase/arylesterase 1 | P27169 | 7 | HDL | 1.0/2.1 | 1.0/2.1 | 1.1/1.9 |
Apolipoprotein E | P02649 | 15 | LDL, HDL | 1.3/2.0 | 1.2/2.8 | 1.5/2.3 |
Corticosteroid-binding globulin | P08185 | 4 | SERPIN | 0.9/2.0 | 0.8/1.9 | 0.9/1.8 |
Alpha-1-acid glycoprotein 2 | P19652 | 5 | ACR | 1.4/2.0 | 1.4/2.7 | 1.4/1.3 |
Haptoglobin-related protein | P00739 | 3 | ACR,SERPIN | 1.2/2.0 | 1.5/1.5 | 0.9/1.0 |
Clusterin | P10909 | 12 | Chaperone | 1.0/2.0 | 1.0/2.4 | 1.2/2.2 |
Alpha-2-HS-glycoprotein | P02765 | 10 | ACR | 1.1/2.0 | 1.3/1.8 | 0.9/2.4 |
Alpha-1-antitrypsin |
P01009 | 20 | ACR, SERPIN | 1.5/1.9 | 2.2/3.3 | 1.1/1.2 |
|
||||||
Apoliprotein CI | P02654 | 2 | VLDL | 0.3/1.05 | 0.4/0.9 | 0.7/1.9 |
Apoliprotein CII | P02655 | 3 | VLDL | 0.4/1.3 | 0.6/1.6 | 1.1/2.6 |
Apoliprotein CIII | P02656 | 2 | VLDL | 0.5/0.8 | 0.6/0.8 | 1.1/1.2 |
Apoliprotein CIV | P55056 | 2 | VLDL | 0.17/− | 0.8/− | 0.9/− |
Platelet basic protein | P02775 | 3 | Platelet cytokine | 0.5/1.2 | 0.4/0.8 | 1.0/2.0 |
Functional class obtained from Uniprot database (
indicates proteins that changed >1.5 fold in both DF and DHF samples during the febrile or defervescence stage. Fold change – refers to ratio of mass ion (115 –visit-1, 116-visit2, 117-visit-3) intensities over the 114 mass ion intensity representing a reference control sample (see methods) Febrile stage corresponds to visit-1, Deferv. –visit-2, Conval.-visit-3. DF- dengue fever, DHF- dengue hemorrhagic fever.
A major caveat of the sample pooling approach described above is the averaging effect which may result in a gross underestimation of fold changes despite the high accuracy and sensitivity of the proteomic quantification. As an alternative, we used a commercially available multiplex fluorescent-bead based ELISA assay, which simultaneously measures levels of 9 well-known acute phase proteins including two serum proteins (serum amyloid A2 (SAA) and haptoglobin (HPT)) that were identified in our proteomics analysis (
We used a previously established mass spectrometry based method
Total serum CT (A) and NT (B) levels in dengue patients during febrile, defervescence and convalescence stages. Levels of CT and NT measured in 15 healthy samples was found to be below detection limit (not shown). Statistical confidence was analyzed by ANOVA kruskall wallis-test, DF vs DHF (**p<0.01, NS- not significant).
We adopted a multiple-feature selection strategy to identify subsets of features from among the 47 blood parameters described above that may have predictive value in the identification of DHF during the early febrile phase. By analyzing the various feature classes (i.e. cytokines, serum proteins, protein adducts, and clinical features) measured at the early febrile phase (visit-1), both independently, as well as together we evaluated the relative predictive power of these various molecules. First, we analyzed 23 cytokines and identified a subset of 7 cytokines which displayed sensitivities and specificities >75% (
A. Receiver operator characteristic curve (ROC) analysis of subset groups A (cytokines only), B (cytokines+clinical indicators) and C (cytokines+clinical indicators+protein adducts) (see table-3 for subsets) is shown. Area under curve (AUC), values of >0.85, indicated good performance (see methods) with high sensitivity and specificity.
Subset group | Type of features | Specific features |
Sensitivity %* | Specificity %* | AUC* |
A | 23 Cytokines |
IFN-ϒ, IL-1b, IL-17, IL-8, | 78(12) | 76(8) | 0.86(0.05) |
IL-9, Eotaxin, FGF-basic, | |||||
B | 23 Cytokines |
IFN-ϒ, IL-1b, IL-17, IL-8, | 80(11) | 86(7) | 0.92(0.03) |
19 Clinical features |
G-CSF, LYMPH, PLT, | ||||
Viraemia | |||||
C | 23 Cytokine |
IFN-ϒ, IL-1b, IL-8, IL-9, | 71(18) | 85(7) | 0.88(0.05) |
19 Clinical Features |
IL-1ra, G-CSF, Eotaxin, | ||||
2 Protein Adducts |
CT, LYMPH, Viraemia | ||||
D | 23 Cytokines |
IFN-ϒ, IL-17, FGF-basic, | 83(14) | 75(13) | 0.90(0.06) |
19 Clinical features |
RANTES, IP-10, CT, | ||||
2 protein adducts |
LYMPH, SAA, HPT | ||||
5 Serum proteins |
n = 62 (44 DF; 18 DHF);
n = 54 (44 DF; 10 DHF);
n = 34 (24 DF; 10 DHF);
All measurements are from febrile stage (visit-1); Mean (SD) values shown; AUC-area under curve from receiver operator characteristics analysis. DHF - dengue hemorrhagic fever. Please see methods section for details on feature selection.
Finally, we expanded the dataset to include all measured features (i.e. 23 cytokines, 5 serum proteins, 2 protein adducts and 15 clinical features). The number of patients in this analysis was much lower (n = 34) than the previous analysis (n = 62 and n = 54) due to further exclusion of samples where the data was incomplete due to missing values. The subset from this analysis included a variety of features including serum proteins (SAA and HPT), cytokines (IFN-ϒ, IL-17) and protein adducts (CT) that achieved a sensitivity and specificity of >75% and AUC of 0.90±0.06 (
We have performed a comprehensive molecular analysis of serum molecules in a cohort of adults with primary dengue infections with the objective of identifying predictive markers of DHF. Traditionally, biomarkers studies have relied mostly on case versus control studies (reviewed in
A detailed cytokine analysis indicated that DHF patients are characterized by an attenuated serum cytokine response especially during the early febrile phase. In DHF, low levels of IFN-γ during febrile phase correlated with reduced levels of IP-10, indicating that an inability to mount a timely anti-viral response may result in high viremia. In cell culture models, pretreatment with interferons inhibits dengue viral replication
In contrast our findings, a number of previous studies have reported elevated levels of IFN-γ
In an attempt to identify serum protein markers of DHF, several groups have reported proteomic analysis of dengue patient sera
Nitric oxide (NO) production by phagocytes is an important inflammatory response to pathogens and although increased levels of both inducible NO synthase (iNOS) and NO levels have been reported in dengue patients
The comprehensive database of 47 blood parameters from dengue patients described in this study provides a unique opportunity to statistically query this dataset to identify -1) most significant molecules and 2) their relative importance in distinguishing DHF from DF during the early febrile stage. In the final analysis, a subset of 9 features was identified that included 5 cytokines, chlorotyrosine, blood lymphocyte count, and two serum proteins. Overall, cytokines involved in attenuated antiviral response; up regulation of acute phase proteins, and elevated neutrophil activity; together appear to be early signatures of DHF resulting from primary infections. The precise role of other cytokines IL-17, FGF-basic, and RANTES that were included in the predictive subset, in DHF pathogenesis is currently unclear and does not rule out the involvement of other cytokines in regulation of immune mechanisms in DHF patients.
Previously, a variety of statistical methods including classification and regression tree (CART) analyses
In conclusion, this study describes a comprehensive and systematic molecular analysis of serum samples from a cohort of patients with primary dengue infection. The analytical approach and statistical workflow we have outlined forms a robust platform for both future discovery and validation of biomarkers for prediction of severe dengue disease.
(TIF)
(TIF)
(TIF)
(DOCX)
(DOCX)
(DOCX)
The authors would like to thank Sharone Green as well as the anonymous reviewers for their comments and suggestions on improving the manuscript.