Urinary chitinase 3-like protein 1 for early diagnosis of acute kidney injury: a prospective cohort study in adult critically ill patients

Background Acute kidney injury (AKI) occurs frequently and adversely affects patient and kidney outcomes, especially when its severity increases from stage 1 to stages 2 or 3. Early interventions may counteract such deterioration, but this requires early detection. Our aim was to evaluate whether the novel renal damage biomarker urinary chitinase 3-like protein 1 (UCHI3L1) can detect AKI stage ≥2 more early than serum creatinine and urine output, using the respective Kidney Disease | Improving Global Outcomes (KDIGO) criteria for definition and classification of AKI, and compare this to urinary neutrophil gelatinase-associated lipocalin (UNGAL). Methods This was a translational single-center, prospective cohort study at the 22-bed surgical and 14-bed medical intensive care units (ICU) of Ghent University Hospital. We enrolled 181 severely ill adult patients who did not yet have AKI stage ≥2 based on the KDIGO criteria at time of enrollment. The concentration of creatinine (serum, urine) and CHI3L1 (serum, urine) was measured at least daily, and urine output hourly, in the period from enrollment till ICU discharge with a maximum of 7 ICU-days. The concentration of UNGAL was measured at enrollment. The primary endpoint was the development of AKI stage ≥2 within 12 h after enrollment. Results After enrollment, 21 (12 %) patients developed AKI stage ≥2 within the next 7 days, with 6 (3 %) of them reaching this condition within the first 12 h. The enrollment concentration of UCHI3L1 predicted the occurrence of AKI stage ≥2 within the next 12 h with a good AUC-ROC of 0.792 (95 % CI: 0.726–0.849). This performance was similar to that of UNGAL (AUC-ROC of 0.748 (95 % CI: 0.678–0.810)). Also, the samples collected in the 24-h time frame preceding diagnosis of the 1st episode of AKI stage ≥2 had a 2.0 times higher (95 % CI: 1.3–3.1) estimated marginal mean of UCHI3L1 than controls. We further found that increasing UCHI3L1 concentrations were associated with increasing AKI severity. Conclusions In this pilot study we found that UCHI3L1 was a good biomarker for prediction of AKI stage ≥2 in adult ICU patients. Electronic supplementary material The online version of this article (doi:10.1186/s13054-016-1192-x) contains supplementary material, which is available to authorized users.


CO N C LU S IO N S R E GA RD IN G TH E V A L ID A T IO N O F TH E AN A LY T IC A L S TAB I LI T Y O F CH I3 L 1 AN D TH E P AR T I AL IN -H O U S E VA L ID A TI O N O F TH E
Figure S1 Area under the receiver-operating characteristics curve (AUC-ROC) with 95% confidence interval (CI) of (A) urinary chitinase 3-like protein 1 (UCHI3L1) and (B) urinary neutrophil gelatinaseassociated lipocalin (UNGAL) at enrollment for predicting acute kidney injury (AKI) stage ≥2 based on the Kidney Disease: Improving Global Outcomes (KDIGO) serum creatinine (SCr) or urine output (UO) criteria (AKI SCr/UO ) within 12-

Sample collection and handling
Blood (serum) Approximately 6 ml of blood was obtained via an indwelling arterial line at each studyspecific sampling moment. Blood was collected in clot activator collection tubes (Venosafe 6 ml, ref. VF-106SAS, Terumo Europe, Leuven, BE) for serum. After clotting at 4°C (in weekends: storage up to ±40-hours (h ; Table S3A) at 4°C based on our stability results (Table   S4D)), serum samples were centrifuged (Heraeus Labofuge 400 R, swinging bucket rotor with round bucket, Thermo Fisher Scientific, Waltham, MA) at 4°C and 1992 x g for 15-minutes (min). The supernatant was divided into 4 aliquots: 1 for chitinase 3-like protein 1 (CHI3L1), 1 for creatinine (Cr), and 2 as backup. These eppendorf tubes containing the supernatant were immediately stored at -80°C. No preservatives were added. Samples were thawed at room temperature immediately prior to analysis and vortexed before pipetting.

Urine
Approximately 5-10 ml of urine was obtained via an indwelling bladder catheter at each study-specific sampling moment. Urine was collected directly from the catheter (never from the collection bag) via the needle-free port-system in a standard (non-coated) transport container that can also be used as centrifuge tube for sediment recovery (Urine Monovette 10 ml, ref. 10.252, Sarstedt, Nümbrecht, DE). Urine was immediately (in weekends: storage up to ±40-h (Table S3A) at 4°C based on our stability results (Table S4D)) centrifuged (Heraeus Labofuge 400 R, swinging bucket rotor with round bucket, Thermo Fisher Scientific, Waltham, MA) at 4°C and 1029 x g for 10-min. The supernatant was divided into 5 aliquots: 1 for CHI3L1, 1 for neutrophil gelatinase-associated lipocalin (NGAL), 1 for Cr, and 2 as backup. These eppendorf tubes containing the supernatant were immediately stored at -80°C.
No preservatives were added. Samples were thawed at room temperature immediately prior to analysis and vortexed before pipetting. Eppendorf tubes that still contained visible sediment were very shortly (<15-seconds) centrifuged (Heraeus Biofuge Fresco, Thermo Fisher Scientific, Waltham, MA) at 4°C and 7697 x g before pipetting.

Biomarker measurements
Creatinine Creatinine concentrations were measured in the 24-h laboratory of Ghent University Hospital with a kinetic rate-blanked Jaffé assay (commercial reagents, Roche Diagnostics, Basel, CH) on a Cobas c502.

CHI3L1
The concentration of CHI3L1 was determined in-house by J. De Loor, K. Demeyere, and K.
Van Nuffel with a sandwich enzyme-linked immunosorbent assay (ELISA; Human Chitinase 3-like 1 Quantikine ELISA Kit, ref. DC3L10, R&D Systems, Minneapolis, MN).The standard procedure that was followed when measuring CHI3L1 by ELISA is as follows. All samples and reagents were brought to room temperature. Samples and standards requiring dilution were accordingly prepared using calibrator diluents (1/200 or 1/500 dilution for serum; 1/5 or 1/10 dilution for urine; Table S3B). To each well pre-coated with a rat monoclonal antibody against recombinant human CHI3L1 we added 100 µl of assay diluents followed by 50 µl of the appropriate sample or standard. This mixture was allowed to react for 2-h at room All samples were analyzed within 13-mo after collection (median: 7-mo; IQR: 3-10 mo).
When stored without preservatives (personal communication with Johansen, JS) at -80°C, serum CHI3L1 (SCHI3L1) is stable for 8-y [3]. Additionally, we showed that urinary CHI3L1 (UCHI3L1) may even be measured after a second thawing step within at least 30-mo after the first freezing (Table S4E).

NGAL
The concentration of urinary NGAL (UNGAL) was measured in the clinical chemistry laboratory of Ghent University Hospital with a particle-enhanced turbidimetric immunoassay (PETIA; NGAL Test, ref. ST001-3CA, BioPorto, Hellerup, DK) on a Modular P. The standard procedure that was followed when measuring UNGAL by PETIA is as follows. All samples and reagents were brought to room temperature. After calibrating and running the controls 150 µl of sample was provided in a specific sample cup, as described by the manufacturer. To read over the measuring principle of this PETIA, we refer to other literature [4].
All NGAL analyses were performed in batch in November 2014. Storage at -80°C for 2-y without preservatives has been shown not to affect UNGAL [5].

UO calculation
For urine output (UO) calculations we accepted a margin of 10% under the 1-h block.
Therefore, all urine volume measurements over a period less than 0.9-h (54-min) were first counted up with the following measurement. Then, we redistributed all blocks ≥1.8and <2.7h into 2 blocks, and all blocks ≥2.7-and ≤3-h into 3 blocks, generating blocks of ≥0.9and <1.8-h. Blocks >3-h were considered as unreliable for redistribution and UO calculation.

Statistical analysis
The primary analysis was based on comparison of the areas under the receiver-operating characteristics curves (AUC-ROC) of UCHI3L1 with those of UNGAL for predicting the defined endpoints, which was performed in MedCalc 15.2.1 (MedCalc Software, Oostende, BE). The method by DeLong et al. was selected for calculation of the standard error of both the AUC-ROC and the difference between 2 AUC-ROCs [6]. For the AUC-ROC, a binomial exact 95% confidence interval (CI) was calculated. This program also lists the Youden index [7] , defined as the maximum of [sensitivity plus specificity minus 1], with its associated criterion for each AUC-ROC. We also calculated Spearman's coefficients of rank correlation with this program.
In SPSS 22 (IBM, Armonk, NY) we performed: (a) Mixed model analysis with log 10 (UCHI3L1) as outcome variable, diagnosis of the 1 st episode of acute kidney injury (AKI) stage ≥2 based on the Kidney Disease: Improving Global Outcomes (KDIGO) serum Cr (SCr) or UO criteria (AKI SCr/UO ) within 24-h after sampling as predictor variable, and patient as random factor.
(b) Unpaired comparison of a variable between 2 independent samples. Categorical variables were analyzed with Fisher's exact or the chi-square test, and continuous variables with the nonparametric Mann-Whitney U test. The SPSS 'Descriptives' menu uses Method 6 from the article by Hyndman and Fan for calculation of the IQR [8]. Additionally, we calculated the 95% CI for a proportion using the Wilson procedure without a correction for continuity [9,10].
(c) Paired comparison of a continuous variable between 2 related samples (Tables S4D and E) using the Wilcoxon matched-pair signed-rank test.
(d) Paired comparison of a continuous variable between >2 related samples using the relatedsamples Friedman's two-way analysis of variance by ranks.
Box and whisker plots were generated in GraphPad Prism 5 (GraphPad Software, San Diego, CA), which also uses Method 6 for calculation of the 1 st and 3 rd quartile [8]. The method by Tukey was selected for drawing of the whiskers [11].
For all analyses, 2-sided P values <0.05 were considered significant.
We will now describe how the urinary biomarkers were introduced into the statistical models.

Review and adjustment of CHI3L1 concentrations before input in statistical programs
Remark: CHI3L1 concentrations are expressed in pg/ml in these raw data.
Step 1: Evaluation of the standard curve of the ELISA Only if the coefficient of determination (R 2 ) ≥0.995, there is a good fit of the 7 standard points in the 4PL model. Additionally, we evaluated whether the assured dynamic range of the standard curve was in fact as dynamic as guaranteed (Table S3C).
Step 2: Re-analysis of samples with a concentration (not adjusted for dilution) outside the validated dynamic range of the curve The details of this standard procedure are outlined in Table S3D.
Step 3: Adjustment of CHI3L1 concentrations of optimally diluted samples with OD sample <OD 62.5 P a g e | 9 Table S3E in detail outlines the standard procedure that was followed in this case. Note that, based on the acceptable recovery limits (70-130%), the back-calculated Concentration(Conc) 62.5 ranges from 43.8-81.3 pg/ml.

Adjustment of UNGAL concentrations before input in statistical programs
In the 'performance data and application note for Roche Modular P' (http://ngal.com/media/30866/the_ngal_test_roche_modular_p_ivd.pdf) can be read that the limit of quantification (LOQ) of the NGAL Test was determined to be 25 ng/ml on this analyzer model. As the limit of detection (LOD) was not tested on this model we were advised (personal communication with BioPorto) to use the LOQ that was estimated on the Roche Hitachi 917 (http://ngal.com/media/30857/the_ngal_test_roche_hitachi_917_ivd.pdf), which was 12 ng/ml. The measured UNGAL concentrations were adjusted based on these reported limits (Table S3F). P a g e | 10

Subgroup analyses
We evaluated the biomarkers' diagnostic performances for predicting the primary endpoint as well as the 24-h AKI SCr secondary endpoint in subgroups of patients. The selected variables used for grouping were: age, baseline estimated GFR (eGFR) calculated with the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) formula, reason for ICU admission, patient's location prior to ICU admission, Sepsis-related Organ Failure Assessment (SOFA) score at d1, and presence of suspected bacterial infection, either leading to arterial hypotension or organ dysfunction, or leading to shock (Table S4A), at d1.
Additionally, we used the same grouping variables in both patients who did not develop AKI SCr/UO within 7-d after enrollment and those who did not develop AKI SCr within 7-d after enrollment. The distribution of UCHI3L1 and UNGAL at enrollment in selected subgroups of 7-d no-AKI SCr/UO patients was plotted against the distribution in all 12-h no-AKI SCr/UO patients and in all those maximally reaching AKI SCr/UO stages 1, 2, or 3 within 12-h after enrollment.
Likewise, the distribution of these biomarkers at enrollment in selected subgroups of 7-d no-AKI SCr patients was plotted against the distribution in all 24-h no-AKI SCr patients and in all those maximally reaching AKI SCr stages 1, 2, or 3 within 24-h after enrollment.

Additional AUC-ROC analyses
Additional endpoints of the study were: AKI SCr/UO stage ≥1 within 12-h and 24-h after enrollment; AKI SCr stage ≥1 within 12-h and 24-h after enrollment. As these endpoints include AKI stage 1, we additionally excluded the patients with AKI stage 1 at enrollment from the analysis cohort (n=181). The number of patients in the resulting sub-cohorts was 158 (AKI SCr/UO ) and 160 (AKI SCr ).

Validation of the analytical stability of CHI3L1
Short-term stability of CHI3L1 in serum and urine before centrifugation Høgdall et al. found no change in the CHI3L1 concentration when serum samples were left on the clot at 4°C for 24-h before centrifugation, however, after 72-h SCHI3L1 was significantly increased [12]. We further tested the stability of CHI3L1 in serum (n=2) and also in urine (n=4) of intensive care unit (ICU) patients when stored at 4°C for 6-h (urine), 24-h, and 48-h before centrifugation by comparing these concentrations with those of the immediately centrifuged samples and calculating the mean coefficient of variation (CV).
Combined long-term and freeze-thaw stability of CHI3L1 in urine As mentioned above, SCHI3L1 is stable for 8-y when stored at -80°C without preservatives.
Høgdall et al. additionally showed that repetitive freezing and thawing of serum samples up to 8 times does not influence the concentration of SCHI3L1 [12].
For our study the most relevant stability feature concerning UCHI3L1 was whether UCHI3L1 measured in an aliquot that was thawed for the first time stayed stable after refreezing followed by thawing for the second time. Indeed, this was the protocol when UCHI3L1 fell outside the range of the standard curve at the first measurement. To evaluate this combined stability we compared those concentrations measured after the second thawing with those measured after the first thawing and calculated the mean CV. Total freezing times (=time between first freezing and second thawing), specifically for this stability evaluation, ranged from 6-30 mo (n=2 for 6-mo, 12-mo, 18-mo, 24-mo, and 30-mo).

Within-run precision or intra-assay variability
Of the 101 ELISA runs performed for CHI3L1 measurement (with ELISA kits from 6 different lots), 31 runs had replicate analyses of at least 1 sample or standard. These replicate samples were divided into 3 (serum) or 2 (urine) groups, i.e. low, intermediate (serum), and high, covering the analytical range of the standard curve (0.06 -4.00 ng/ml). Note that for urine there was no intermediate group as there were no replicates available of urine samples with a UCHI3L1 concentration >1.0 and ≤2.5 ng/ml. Replicate analyses of the 0.25, 0.50 and 1.00 ng/ml standards were consistently performed as these constitute the middle 3 points of the standard curve. The intra-assay between-lot CV was calculated as the weighted mean of the mean intra-assay within-lot CVs.

Between-run precision or inter-assay variability
Upon inquiry it appeared that the inter-assay CVs reported by the manufacturer were generated using 40 different assays that were divided between 4 technicians. Each of them performed one ELISA per day. The 40 assays consisted of 2 different matched sets of reagents, just like 2 different kit lots (Table S4B), so lot-to-lot variation was taken into account. Therefore, this part of the validation process was not repeated in-house. Three samples (type not specified) with a known CHI3L1 concentration (±0.50, ±1.00, and ±2.00 ng/ml) were analyzed. The reported mean inter-assay between-lot CV was 5.3% for the ±0.50 ng/ml sample, 5.8% for the ±1.00 ng/ml sample, and 6.9% for the ±2.00 ng/ml sample. The mean of these 3 CVs is 6.0%.

Calculation of the LOD and LOQ
The minimal detectable dose or LOD was determined by adding 2.6 [13]  Linearity check for urine P a g e | 13 The linearity of the assay was assessed by the manufacturer using samples from apparently healthy volunteers. Because our study population consisted of critically ill patients we rechecked the linearity for the specimen type urine. More specifically, we wanted to investigate 'how far undiluted' we could go as urinary components linked with severe illness may possibly interfere with the CHI3L1 measurement by ELISA. Therefore, undiluted urine was not tested and designated as unsuitable. The reference for our serial dilution experiment (1/2 -1/4 -1/8 -1/16) was the 1/2 diluted sample. The relationship between the measured (not adjusted for dilution) and the expected (1/2 as reference) analyte concentration was investigated by linear regression analysis (GraphPad Prism 5, GraphPad Software, San Diego, CA).

R E S U L T S Biomarkers' diagnostic performances in subgroups
Diagnostic performance at enrollment for prediction of AKI SCr/UO stage ≥2 within the next 12h, could be calculated in 9 of the 12 subgroups (Additional File 2: Supplemental Figure S1).
Likewise, diagnostic performance at enrollment for prediction of AKI SCr stage ≥2 within the next 24-h, could be calculated in 8 of the 12 subgroups (Additional File 3: Supplemental Figure S2). As for UCHI3L1, this biomarker showed decreased diagnostic performancedefined as an AUC-ROC <the lowest border of the 95% CI in the analysis cohort -for predicting the primary endpoint in patients either with a medical reason for ICU admission, or referred from either an emergency room, or operating room, or other hospital at ICU admission, or with a SOFA score <12 at d1 [15]. Its performance for predicting the 24-h AKI SCr secondary endpoint was decreased in patients either ≥65-y old, or with a SOFA score ≥12 at d1. As for UNGAL, this biomarker showed decreased diagnostic performance for predicting the primary endpoint in patients either ≥65-y old, or with a medical reason for ICU admission, or referred from the floor at ICU admission. In the latter 2 subgroups UNGAL could not predict the primary endpoint. Its performance for predicting the 24-h AKI SCr secondary endpoint was decreased in patients ≥65-y old.
We found that patients referred from the floor at ICU admission who did not develop AKI SCr/UO within 7-d after enrollment, had higher urinary biomarker concentrations at enrollment than all patients who did not develop AKI SCr/UO within 12-h after enrollment (P=0.002 for UCHI3L1 and P=0.001 for UNGAL; Additional File 4: Supplemental Figure   S3). Similarly, 7-d no-AKI SCr patients who were referred from the floor at ICU admission showed higher enrollment concentrations of both urinary biomarkers than all 24-h no-AKI SCr patients (P=0.001 for both biomarkers; Additional File 5: Supplemental Figure S4).

Additional diagnostic performances P a g e | 15
The AUC-ROCs for predicting AKI SCr/UO stage ≥ 1 within 12-h and 24-h in patients with no AKI SCr/UO at enrollment were markedly decreased for both UCHI3L1 and UNGAL (Table   S4C). Likewise, the AUC-ROCs for predicting AKI SCr stage ≥ 1 within 12-h and 24-h in patients with no AKI SCr at enrollment were markedly decreased as well (Table S4C). This can be explained by less renal stress or damage in patients with AKI stage 1.

Stability of CHI3L1 in serum and urine
Storage at 4°C up to 48-h before centrifugation had no effect on the CHI3L1 concentration in both serum and urine (P ≥0.05), with a mean CV ranging from 3.8 to 3.9% for serum and from 5.0 to 8.5% for urine (Table S4D).
Refreezing at -80°C followed by a second thawing step had no effect on the CHI3L1 concentration in urine (P ≥0.05), even when the time between initial freezing and second thawing was 30-mo, with a mean CV ranging from 0.7 to 18.8% (Table S4E).

Partial in-house validation of the CHI3L1 ELISA
For the human CHI3L1 standards 0.25, 0.50 and 1.00 ng/ml of the ELISA, the intra-assay between-lot CV was 2.9% for the 0.25 ng/ml standard, 4.4% for the 0.50 ng/ml standard, and 5.0% for the 1.00 ng/ml standard (Table S4F). Serum samples with a low or a high CHI3L1 concentration showed an excellent intra-assay between-lot CV of 2.6% for the low and 3.2% for the high concentrations (Table S4G). For urine samples with a low CHI3L1 concentration an intra-assay between-lot CV of 4.1% was calculated (Table S4H). A mean intra-assay within-lot CV of 6.7% was obtained for serum samples with an intermediate CHI3L1 concentration. Likewise, a mean intra-assay within-lot CV of 1.9% was obtained for urine samples with a high CHI3L1 concentration.
The LOD was determined as 0.02 ng/ml, the LOQ as 0.06 ng/ml. For all 3 serial dilution experiments, each being performed with another urine sample, the 95% CI of the slope of the linear regression equation included 1, indicating 100% recovery (Table S4I). The corresponding R 2 was systematically ≥0.99. P a g e | 17

C O N C L U S I O N S R E G A R D I N G T H E V A L I D A T I O N O F T H E A N A L Y T I C A L S T A B I L I T Y O F C H I 3 L 1 A N D T H E P A R T I A L I N -H O U S E V A L I D A T I O N O F T H E C H I 3 L 1 E L I S A
As an important first step, we showed that serum and urine may be stored for at least 48-h at 4°C before centrifugation without affecting the CHI3L1 stability, in analogy with the reported data for urinary neutrophil gelatinase-associated lipocalin [16]. Based on these results the study personnel could flexibly plan the handling of the samples, especially in weekends (Table S3A). When subsequently stored at -80°C, UCHI3L1 may even be measured after a second thawing step within at least 30-mo after the first freezing, in analogy with the reported data for SCHI3L1 [3,12]. This information was required for our laboratory agenda, and is also relevant for biobanking purposes. Additionally, we can guarantee that our reported CHI3L1 concentrations are accurate and reproducible.

L I S T O F A B B R E V I A T I O N S
4PL: 4-parameter logistic; AKI: acute kidney injury; AKI SCr : AKI that was diagnosed and   (e) Describe any sensitivity analyses NA b Give information separately for exposed and unexposed groups. Based on our stability results (Table S4D), the responsible study coordinator had to come once a weekend: on Sat or Sun. The dilution of a serum sample for the initial measurement of CHI3L1 by ELISA was chosen based on a patient's serum CRP, while the dilution of a urine sample for the initial measurement was chosen based on a patient's SCr.       a The concentration of a sample that was stored for X-h at 4°C before centrifugation is represented by X-h, while that of a sample that was centrifuged immediately after collection is represented by 0-h. b The P values are the significance levels between samples that were immediately centrifuged and those stored for 6-, 24-, or 48-h at 4°C before centrifugation. a The concentration of a sample measured after 1 st thawing of an aliquot is represented by 1, while that of a sample measured after 2 nd thawing of an aliquot is represented by 2. The samples in this stability study were initially analyzed (conc. 1) after a period ranging from 1-9 mo.

Participants
b Mean inter-assay CV reported by the manufacturer = 6.0% (Table S4B). c The P values are the significance levels between samples that were thawed for the 1 st time and those thawed for the 2 nd time.

Table S4F
Within-run precision or intra -assay variabilit y of the chitinase 3like protein 1 enz yme-linked immunosorbent assay for standard points.    Underlined results represent the mean, SD and CV of the OD value (out of dynamic range of the curve).
Letters in superscript indicate different lots within each group. In the 2 different groups, letter X does not necessarily represent the same lot.       Note: Additional File 2 is a TIFF file of Figure S1. The dotted vertical lines delineate the AUC-ROC with 95% CI in the analysis cohort. The Note: Additional File 3 is a TIFF file of Figure S2.  Note: Additional File 4 is a TIFF file of Figure S3.