Evidence of IL-17, IP-10, and IL-10 involvement in multiple-organ dysfunction and IL-17 pathway in acute renal failure associated to Plasmodium falciparum malaria

The Institutional Human Ethics Committee of S. C. B. Medical College, Cuttack and Institute Pasteur, Lille, approved the study. Blood and urine were collected after obtaining written informed consent from participants or, for comatose patients, the accompanying person.

The study was conducted at SCB Medical College and Hospital, Cuttack, Odisha. More than 85 % cases in Odisha are due to P. falciparum [5]. Subjects (n = 222) from coastal district areas with an API of 6.6 [24] were enrolled from 2008 to 2010. Parasite presence was tested with RDK (SD Bio Standard Diagnostic, India) and thick and thin blood smears. Nested PCR was used to detect P. falciparum/P. vivax co-infection [25]. Only falciparum positive cases were included in the study. Clinical histories, detailed physical assessments, demographic profiles and informed consent from all patients were recorded at admission. Using the World Health Organization’s severity criteria [26], patients were categorized as following: (1) 37 patients with uncomplicated malaria (MM) had fever without complications and were positive for P. falciparum, (2) 53 patients with severe non cerebral malaria (SNCM) which had one of the several manifestations of severe malaria without cerebral involvement such as severe anemia (haemoglobin ≤ 5 g/dl), jaundice (serum bilirubin ≥ 3 mg/dl), acute renal failure (serum creatinine ≥ 3 mg/dl), acute respiratory distress (PaO₂/FIO₂ ≤ 200), shock (systolic BP ≤ 80 mmHg) and haemoglobinuria (dark red or black coloured urine positive for haemoglobin), (3) nine patients with multiple organ dysfunction (MOD) were patients diagnosed for the presence of two or more organ involvement like CNS, respiratory distress, renal failure and hepatic dysfunction (ALT/AST ≥3 times of normal, prolonged prothrombin and albuminaemia), (4) 83 patients with fever and altered sensorium, unarousable coma with Glasgow Coma Scale (GCS) of ≤10 were categorized as cerebral malaria (CM) after exclusion of other causes of encephalopathy such as encephalitis, meningitis and metabolic encephalopathy by biochemical investigations in the CSF. Among CM, 41 patients with another organ failure was further defined as patients with MOD (CM-MOD), (5) 21 healthy subjects from malaria-endemic areas (EC) were patient’s relatives and who had not suffered from a bout of malaria for at least 2 years, nor were they asymptomatic carriers. We also added two control groups, 10 severe sepsis (SEPT) and 9 viral encephalitis (ENC) admitted in the Department of Internal Medicine for treatment. All blood smears were checked for the presence of malarial parasites (Table 1).

Plasma concentrations of 26 cytokines: GM-CSF, Eotaxin, IL-1α, MCP-1, IL-10, MIP-1α, MIP-1β, IL-8, IL-6, TNF-α, IFN-α2, IL-7, G-CSF, IP-10, IL-3, IL-5, IL-15, TNF-β, IL-2, IL-13, IL-4, IL-12p70, IL-17, IFN-γ, IL-12p40, IL-1β were measured using Luminex multianalytic profiling (MILLIPLEX^® MAP human cytokine/chemokine—Premixed 26 Plex), according to manufacturer instructions (Millipore, USA). Cytokine detection (range 1–10,000 pg/mL) and quantification were performed using a Bio-Plex 200 system (Bio-Rad, USA). Samples were tested in duplicate. Data were analyzed using Bio-Plex Manager 5.0 software (Bio-Rad).

All analysis was conducted on log₁₀-transformed data. Normality of the data was checked using the Shapiro–Wilk test. Differentially expressed cytokines were identified using the Student’s t test; p < 0.05 was considered significant. Cytokine expression heatmaps, and hierarchical clustering of cytokines and of the samples within each category, were generated based on the Euclidean metric and using R/Bioconductor. Multidimensional scaling (MDS) representation was generated using SVD-MDS [27]. MDS methods aim to represent the similarities and differences among high dimensionality objects into a space of a lower dimensions, generally in two or three dimensions for visualization purposes [28]. The Spearman coefficient of correlation was used to determine the association between cytokines. Linear Discriminant Analysis (LDA) was performed on log-transformed data from the Gondia study and the present study using macros and IgorPro software (version 3.16; WaveMetrics). Samples with incomplete information were rejected. No correction for multiple tests was applied to the p values.

R software’s glm function was combined with a stepwise selection procedure based on the Akaike information criterion for logistic regression analysis [29]. Such selection introduces, at each step, the cytokine yielding the best Akaike information criterion and removes non-significant cytokines (p > 0.05) [30]. The true positive rate (sensitivity) and true negative rate (specificity) were estimated for the obtained model; area under the Receiver Operating Characteristic curve was used to assess quality.

We quantified levels of GM-CSF, Eotaxin, IL-1α, MCP-1, IL-10, MIP-1α, MIP-1β, IL-8, IL-6, TNF-α, IFN-α2, IL-7, G-CSF, IP-10, IL-3, IL-5, IL-15, TNF-β, IL-2, IL-13, IL-4, IL-12p70, IL-17, IFN-γ, IL-12p40, IL-1β, by multiplex assay and generated an expression heatmap.

Although control groups (EC, SEPT, ENC) clustered together in the MDS representation, malaria subgroups, particularly severe forms, could not be clearly distinguished based on the whole set of cytokines (Additional file 1: Figure S1). Only the MOD group could be distinguished from the control groups. This suggests that the malarial inflammatory process is different from that of encephalitis and septicemia.

We aimed then to identify groups of cytokines having similar expression patterns. Hierarchical clustering identified cytokine groups having similar expression profiles across subphenotypes and in patients having similar expression profiles within each subphenotype (Fig. 1). Some cytokines presented large variations in their expression levels. Two clusters of cytokines were identified. The first consisted of cytokines abundantly present in patient plasma—GM-CSF, Eotaxin, IL-1α, MCP-1, IL-10, MIP-1β, MIP-1α, IL-8, IL-6, TNF-α, IFN-α2, IL-7, G-CSF, and IP-10 (Additional file 2: Figure S2); the second, characterized by lower cytokine levels, comprised IL-3, IL-5, IL-15, TNF-β, IL-2, IL-13, IL-4, IL-12p40, IL-17, IFN-γ, IL-12p70, and IL1-β (Additional file 3: Figure S3). All malarial subgroups, but not controls (EC, SEPT, and ENC), exhibited increased levels of TNF-β, IL-12p70, and IL-4. However, IL-15 and IL-12p40 levels were higher in SEPT and ENC patients. All malarial subgroups showed large variation in cytokine expression and had subpopulations with very low cytokine levels. Medians of cytokine levels tended to increase with disease severity.

We next aimed to identify the cytokines differentially expressed between the different biological conditions. First, we compared all patients with EC and each malaria subgroup with EC using the Student’s t test. Additionally, we compared the CM and CM-MOD groups with a group comprising SEPT and ENC—non-parasitic diseases with brain/multi-organ involvement—and EC. Of 26 cytokines/chemokines, 19 increased significantly during malaria. When cytokine profiles of patients in individual categories were compared to the cytokine profiles of controls, distinct patterns emerged. Many more cytokines were upregulated in MM and SNCM patients than in MOD, CM, and CM-MOD patients (Fig. 2). IL-10 clearly rose with severity (~fold increase of 29, 44, 93, and 61 in MM, SNCM, MOD, and CM/CM-MOD, respectively; Fig. 3).

Of note, there was no correlation between age, sex, and disease phenotype or cytokine levels in all the groups studied. In agreement with studies previously reported in Odisha, a higher percentage of males were observed among all malaria groups [31].

We investigated whether the cytokine clusters that discriminated MM from CM in Gondia [23] could differentiate these subgroups in Odisha. We selected cytokines common to both studies, i.e., IFN-γ, IL-1β, IL-2, IL-4, IL-5, IL-6, IL-10, IL-12p40, and TNF-α, and performed an LDA. Resulting Gondia LDA factors discriminated infection (Gondia LDA factor 1) and clinical manifestations (Gondia LDA factor 2; Fig. 4a). We projected the Odisha data onto Gondia LDA factor 1 and calculated a new LDA factor for the Odisha dataset (Fig. 4b). Although these factors distinguished malaria patients from controls (Fig. 4b), the cytokines that distinguished malarial subgroups in Odisha were very different from those in Gondia (Fig. 4c). Furthermore, the clear separation of clinical subgroups observed in Gondia (factor 2) was not seen in Odisha. However, LDA of the entire dataset distinguished the control groups from malaria subgroups and MOD from CM-MOD (Fig. 4d).

Using the Student’s t test, we characterized cytokine patterns associated with each severe malaria subgroup and patterns differentiating MM from severe malaria (SNCM, MOD, CM, CM-MOD). The MOD and CM-MOD groups were statistically indistinguishable (Additional file 4: Figure S4). Only IL-1α levels were significantly different between the CM-MOD and CM groups, with a fold-change of 1.43 (p = 0.009). By contrast, IP-10 and GM-CSF distinguished CM from the MOD group, with fold-changes of −1.46 (p = 0.006) and −1.66 (p = 0.03), respectively. We found the cytokine set associated with each severe malaria subgroup using logistic regression. This allowed determination of the discriminant power of some cytokines, accounting for multivariate dependence between predictors. Once a cytokine was identified as discriminant, we characterized its effect through an odds-ratio (OR). When the severe non-cerebral (SNCM + MOD) group was compared to the cerebral malaria (CM + CM-MOD) group (Additional file 5: Table S1), only MIP-1α discriminated the two (OR = 0.71; p = 0.0003. Within the SNCM + MOD group, only IL-17 (OR = 0.63; p = 0.04) and IP-10 (OR = 0.08; p = 0.03) differentiated SNCM from MOD. Levels of both cytokines were higher in MOD than SNCM. Most patients (SNCM, MOD, and CM-MOD) with ARF exhibited high IL-17 (Fig. 5a) or IP-10 (Fig. 5b) levels. In addition, in SNCM patients with ARF, IP-10 concentrations were negatively correlated to IL-17 (R = −0.59; p = 0.03) (Fig. 5c). Interestingly, levels of IL-17 and IL-12p40 were also correlated in ARF patients from the MOD group (Fig. 5d).

Cytokine patterns in severe malaria subgroups were then compared to identify cytokines contributing to a given clinical subphenotype (Additional file 5: Table S1). Levels of GM-CSF and IL-17 were significantly higher in the MOD than in the CM group, allowing for their differentiation (OR = 0.2; p = 0.03 and OR = 0.5; p = 0.02 for GM-CSF and IL-17, respectively). Besides IL-17, high levels of MIP-1α also differentiated CM-MOD from MOD. This approach was further used to define cytokines that discriminated all clinical malaria subgroups. The results are represented in Fig. 6. This cytokine/chemokine network could be helpful in stratifying severe malaria manifestations into clinical sub phenotypes (Additional file 6: Table S2).