Chitinase-3-like 1 is a biomarker of acute kidney injury and mortality in paediatric severe malaria

The study was conducted between 2011 and 2013 at the Jinja Regional Referral Hospital in Jinja, Uganda. All children were treated with intravenous artesunate followed by oral artemisinin-based combination therapy. Inclusion criteria were age 1–10 years, P. falciparum by RDT (First Response Malaria Ag. HRP2/pLDH Combo Rapid Diagnostic Test, Premier Medical Corporation Limited, India), selected severe malaria criteria (decreased consciousness, repeated seizures, prostration, and/or respiratory distress), and plasma sample available for CHI3L1 testing. Exclusion criteria included: known chronic illness, severe malnutrition, known haemoglobinopathy, prior treatment with quinine in the emergency department, suspicion of acute bacterial meningitis. Study gas was delivered continuously by non-rebreather mask for up to 72 h, as previously described [13, 22].

Creatinine and BUN were assessed at the point of care using i-STAT CHEM8+ or Crea cartridges (Abbott Laboratories, Saint-Laurent, Québec). Creatinine measured by i-STAT is calibrated traceable to the isotype dilution mass spectrometry reference measurement and is free of interference from haemoglobin, bilirubin, and glucose [23]. Estimated glomerular filtration rate (eGFR) was calculated using the Bedside Schwartz equation, using a constant for children (k = 0.413) [24]. Presence of AKI was determined retrospectively using KDIGO guidelines [25]. Children were considered to have AKI if they had either a ˃26 μmol/L rise in creatinine within 48 h or > 1.5-fold increase in creatinine from estimated baseline. Baseline creatinine was estimated assuming a normal GFR of 120 mL/min/1.73 m² and using the Schwartz equation to back calculate creatinine based on the child’s height. The children with AKI were further classified by stage: stage 1 (risk; 1.5–1.9-fold increase in creatinine from nadir), stage 2 (injury; 2.0–2.9-fold increase), and stage 3 (failure; greater than 3.0-fold increase, single value greater than 354 µmol/L during hospitalization, or an eGFR of < 35 mL/min/1.73 m²) [25]. The incidence of AKI in this cohort has been previously reported [6].

K₂ EDTA anticoagulated plasma samples were collected daily from children during hospitalization (day 1–4), and at follow up (day 14) and stored at − 80 °C until testing. Cystatin C and CHI3L1 were measured by ELISA (DuoSet, R&D Systems, Burlington, Canada) with the investigator blinded to treatment group and outcome [6]. LDH activity was measured using a colorimetric assay according to manufacturer’s protocol (BioVision, Milpitas, CA, USA). The limit of detection for CHI3L1 was 4.5 ng/mL with inter- and intra-assay reproducibility of 5.9 and 6.3%, respectively. Markers of intravascular haemolysis (haem, haemopexin) [26], endothelial activation (Ang-2, sICAM-1) [27], and inflammation (CRP, sTREM-1, CXCL10/IP-10) were measured by ELISA as previously described [27, 28]. Biomarkers were selected for analysis based on an established association with disease severity and mortality in Ugandan children with severe malaria [27, 28].

Data were analysed using Stata/SE v14.0, GraphPadPrism v7.03, and R [29]. For access to the dataset see Additional file 1. Continuous data are presented as median (interquartile range, IQR) and analyzed using Wilcoxon rank sum test or non-parametric test for trend. Categorical data are presented as n (%) and analysed using Pearson’s Chi Square or Fisher’s exact test, as appropriate. To compare biomarker levels at admission, Spearman’s non-parametric correlation was used. To assess the relationship between CHI3L1 and mortality, generalized linear models were used opting first for a log-binomial model with robust standard errors. In the event of failed convergence Poisson regression was used with robust standard errors.

R [29] and lme4 [30] were used to perform a linear mixed effects (LME) analysis of in-hospital longitudinal course of log (CHI3L1) over time in patients without AKI and with different stages of AKI. Time, nitric oxide treatment arm, and AKI stage were entered as fixed effects with and without interaction terms. Intercepts and slopes were modeled for each subject as random effects. The intercept for treatment arm was constrained to zero at baseline as children were randomly allocated to treatment arm. Visual inspection of residual plots did not reveal deviations from homoscedasticity or normality. p values were obtained by likelihood ratio tests of the full model (including AKI stage) against the model without AKI stage.

Levels of CHI3L1 were available for 159 children at admission (Fig. 1). A description of the population is included in Table 1. The median age of children was 2.0 years and 56% were male. At admission, children with CHI3L1 levels in the highest quartile had a median of 5 severe malaria criteria compared to 4 in children with the lowest three quartiles (p < 0.0001) [31]. Children with elevated CHI3L1 also had higher lactate (p = 0.001), lower bicarbonate, (p < 0.0001) and higher Cystatin C (p = 0.001) and BUN levels (p = 0.025). Children with CHI3L1 levels in the highest quartile were more likely to have AKI (p = 0.034), and there was a strong relationship between CHI3L1 and severity of AKI with 5.9% of children with CHI3L1 levels in the lowest three quartiles having stage 3 AKI compared to 28.2% of children with CHI3L1 levels in the highest quartile having stage 3 AKI (p < 0.0001). The relationship between CHI3L1 levels and kidney function was further explored using linear regression to investigate the relationship between creatinine, Cystatin C. and BUN as dependent variables and log₁₀(CHI3L1) levels at admission as the independent variable (Table 2). A log₁₀ increase in CHI3L1 was significantly associated with creatinine, Cystatin C, and BUN following adjustment for age and sex. When adjusting for other measures of kidney function, log₁₀(CHI3L1) remained independently associated with an increase in Cystatin C following correction for creatinine (beta, 95% CI 0.17, 0.05–0.30, p = 0.008), but not BUN. The relationship between log₁₀(CHI3L1) and BUN was not significant when adjusting for either creatinine or Cystatin C (p > 0.05), and the relationship between log₁₀(CHI3L1) and creatinine was not significant when adjusting for Cystatin C or BUN (p > 0.05).

CHI3L1 levels were compared at admission with subsequent in-hospital mortality and all-cause 6 month mortality. CHI3L1 levels in the highest quartile were significantly associated with death (Table 1). Further, median CHI3L1 levels were higher among children who died in-hospital than those who survived (p = 0.023, Wilcoxon rank sum test) and children who died by 6 months follow up compared to those known to survive (p = 0.046). By multivariable analysis, a one unit increase in log₁₀(CHI3L1) was associated with a 4.10-fold increased risk of in-hospital death (95% CI 1.32–12.75, p = 0.015) following adjustment for treatment arm and AKI status using a log binomial model. Further, a one unit increase in log₁₀(CHI3L1) was associated with a 3.21-fold increased risk of death by 6 months (95% CI 1.47–6.98, p = 0.003) following adjustment for treatment arm and AKI status.

Survival data were not available at 6 months follow up for 16 children lost to follow up (LTFU) and 19 children > 5 years of age and not followed to 6 months (according to the study protocol). To evaluate whether the association between CHI3L1 and 6 month mortality was robust to various assumptions a sensitivity analysis was performed first assuming all children < 5 years of age LTFU died (Table 3, model 1). Then all children < 5 years of age LTFU were assumed to have survived (model 2). Assuming all children < 5 years of age LTFU died, log₁₀(CHI3L1) was associated with a 2.13-fold increased risk of death (1.18, 3.84, p = 0.012, model 1), and a 3.11-fold increased risk of death (1.38–6.99, p = 0.006, model 2) if all children < 5 years of age LTFU survived. This analysis was then expanded to all children in the cohort (including children > 5 years of age that were not followed). Assuming all children LTFU or > 5 years of age died, log₁₀(CHI3L1) was associated with a 1.99-fold increased risk of death by 6 months (1.22–3.25, p = 0.006, model 3). Assuming all children LTFU or > 5 years of age survived, log₁₀(CHI3L1) was associated with a 3.10-fold increased risk of death by 6 months (1.34–7.16, p = 0.008, model 4).

In order to explore potential pathophysiologic mechanisms linking elevated CHI3L1 and mortality in severe paediatric malaria, admission levels of CHI3L1 were compared with other host markers of immune and endothelial activation. CHI3L1 was correlated with pathways implicated in the pathobiology of severe malaria including markers of endothelial activation (angiopoietin-2, 0.43, p < 0.0001; sICAM-1, 0.43, p < 0.0001), markers of inflammation (CRP, 0.36, p < 0.0001; CXCL10/IP-10, 0.26, p = 0.0008; sTREM-1, 0.23, p = 0.005); and markers of haemolysis (LDH, 0.28, p = 0.0004; haemopexin, − 0.23, p = 0.004; haem, 0.30, p = 0.0001).

LME models were used to explore the relationship between the longitudinal time course of CHI3L1 in patients in the placebo and treatment arms of the iNO trial. Assuming that differences at baseline between trial arms were due to chance alone (random allocation), we observed that CHI3L1 was elevated at baseline and decreased over the first 3 days of hospitalization at a different rate in children receiving iNO compared to placebo. The baseline (day 1) CHI3L1 concentration (estimate, 95% CI 213 ng/mL, 176–259) decreased by 34% per day (95% CI 31–38) in the placebo group and 29% per day (95% CI 25–33) in the iNO group (p = 0.007) (Fig. 2).

AKI is associated with higher mortality in paediatric severe malaria [6]; therefore, the relationship between longitudinal time course of CHI3L1 in patients with and without AKI was investigated. In an analysis including all trial participants and adjusting for the effect of iNO, AKI was associated with higher CHI3L1 levels (1.02-, 1.30-, and 2.50-fold higher in AKI grade 1, 2 and 3, respectively, relative to no AKI, p < 0.0001, Fig. 3). To confirm the association between AKI and elevated CHI3L1 independent of iNO exposure, a subgroup analysis was performed using patients in the placebo arm of the trial: AKI grade 1, 2 and 3 was associated with CHI3L1 concentrations 0.94, 1.31, and 2.41-fold higher than patients without AKI, over the course of the first 3 days of hospitalization (p = 0.017).