Introduction
Septic shock leads to significant morbidity and mortality in critically ill adult and pediatric patients [
1,
2]. Meanwhile, acute kidney injury (AKI) is also known to be independently associated with mortality and morbidity in critically ill patients. The treatment of sepsis costs the US population over $15 billion/year for adults and over $2 billion/year for children, while the costs for AKI approach $10 billion/year for adults alone [
3,
4]. Sepsis is the most common precipitant for AKI in both populations, and the development of kidney injury in the context of sepsis is a poor prognostic sign. Together the two disease processes carry up to 75% mortality [
5‐
9].
Effective therapies for septic-shock-associated acute kidney injury (SSAKI) are lacking. Detection schemes for SSAKI have been and still are dependent on serum creatinine, a flawed real-time marker of AKI [
10,
11]. Diagnoses of SSAKI based upon changes in creatinine, therefore, are considerably varied and create heterogeneity between studies investigating AKI therapy. Biomarker research seeking to identify more robust markers of AKI has yielded promising results. Neutrophil gelatinase-associated lipocalin (NGAL), IL-18, and cystatin C have all demonstrated encouraging efficacy for predicting ischemic or nephrotoxic AKI and its severity [
12‐
14]. Sepsis intrinsically induces increased expression of some AKI biomarkers (for example, NGAL), and studies of NGAL levels in patients with SSAKI often demonstrate high sensitivity with modest specificity [
9,
14‐
17].
Importantly, the pathophysiology of SSAKI may be unique from that of ischemic or nephrotoxic AKI [
18]. Therapies aiming to restore renal perfusion in ischemic AKI [
19‐
21] have not been demonstrated to be particularly effective and may be even less effective in SSAKI, a process that may not be secondary to impaired glomerular preload. Persistent SSAKI may fall into the class of fluid-unresponsive AKI [
22]. Renal replacement therapy has been used as therapy for AKI and data exist demonstrating that initiation prior to accumulation of excessive fluid overload may improve outcomes [
23,
24]. The aggregate data, however, show that patients with SSAKI have consistently increased mortality, even with early renal replacement therapy initiation [
25,
26]. The modest efficacy of biomarkers at identifying SSAKI also underscores the notion that the pathophysiology of SSAKI is unique from other etiologies of AKI. There is a need to identify novel candidate biomarkers of SSAKI, which would expedite early treatment aimed at preventing the effects of this highly morbid complication of sepsis.
We have generated an extensive genome-wide expression database from children with septic shock by way of microarray technology and have now leveraged this database to identify candidate biomarkers for SSAKI [
27‐
32]. Herein we report the identification of 21 unique gene probes upregulated in patients with SSAKI, within the first 24 hours of admission to the pediatric intensive care unit (PICU), and their ability to robustly predict SSAKI. Two readily measurable gene products from this list, matrix metalloproteinase-8 (MMP-8) and neutrophil elastase-2, show high sensitivity for SSAKI in a cohort of patients with septic shock.
Materials and methods
Patients and data collection
The study protocol was approved by the Institutional Review Boards of each participating institution (
n = 11). Children ≤10 years of age admitted to the PICU and meeting pediatric-specific criteria for septic shock were eligible for enrollment [
33]. Controls, used to normalize the microarray data across the patients with septic shock and to conduct the three-group analysis of variance in the first derivation analysis, were recruited from the ambulatory departments of participating institutions using published inclusion and exclusion criteria [
28‐
32]. These controls were required to reliably compare data across different batches of samples. All patients and controls in the derivation cohort were previously reported in microarray-based studies addressing hypotheses entirely different from that of the current report [
28‐
32]. All microarray data have been deposited in the NCBI Gene Expression Omnibus (GEO:GSE26440 and GEO:GSE26378). The patients in the validation cohort have not been previously reported in any manner.
After informed consent from legal guardians, blood samples were obtained within 24 hours of initial presentation to the PICU with septic shock. Clinical and laboratory data were collected daily while in the PICU, and were stored using a web-based database. Mortality was tracked for 28 days after enrollment.
SSAKI was defined as a >200% rise of serum creatinine relative to the median normal value for age [
34,
35], persistent to day 7 of PICU admission. For example, if a patient had a >200% rise of serum creatinine upon admission to the PICU but returned to <200% by day 7 (without renal replacement therapy), then that patient was classified as not having SSAKI. Although stringent, this definition allowed us to identify those patients with persistent fluid-unresponsive AKI. No patient in this study had known pre-existing chronic kidney disease. Patients who died before day 7 with a creatinine rise >200% from baseline were included.
RNA extraction and microarray hybridization
Total RNA was isolated from whole blood using the PaxGene™ Blood RNA System (PreAnalytiX, Qiagen/Becton Dickson, Valencia, CA, USA). Microarray hybridization was performed as previously described using the Human Genome U133 Plus 2.0 GeneChip (Affymetrix, Santa Clara, CA, USA) [
27‐
32].
Serum protein biomarker measurements
Measurements of serum MMP-8 and elastase-2 protein levels were performed using a Luminex 200 multi-plex instrument (Luminex Corporation, Austin, TX, USA) and antibody-coated magnetic beads (Millipore, Billerica, MA, USA) following the manufacturer's specifications.
Data analysis
Analyses were performed using one patient sample per chip. Image files were captured using an Affymetrix GeneChip Scanner 3000. CEL files were subsequently preprocessed using robust multiple-array average normalization and GeneSpring GX 7.3 software (Agilent Technologies, Palo Alto, CA, USA). All signal intensity-based data were used after robust multiple-array average normalization, which specifically suppresses all but significant variation among lower intensity probe sets [
36]. All chips representing patient samples were then normalized to the respective median values of controls.
Differences in mRNA abundance between the study categories were measured using GeneSpring GX 7.3 software, as were the class prediction procedures. Receiver operating characteristic (ROC) curves and calculations of biomarker performance were performed using SPSS (IBM Corporation, Somers, NY, USA). Ordinal and continuous clinical variables not normally distributed were analyzed via analysis of variance on ranks. Dichotomous clinical variables were analyzed using a chi-square test (SigmaStat software; Systat Software, Inc., San Jose, CA, USA).
Discussion
We have demonstrated that microarray data can be leveraged to identify gene expression patterns common in SSAKI. Although mRNA levels do not necessarily correlate with protein expression, analysis of MMP-8 and elastase-2 protein levels in the serum compartment indicates that our gene expression data can identify novel serum protein biomarkers for SSAKI.
Although sepsis is the most important predictor of AKI in critically ill patients [
37,
38], few clinical studies are dedicated to identifying the early phenotype of patients with SSAKI. Interrogation of several large databases of critically ill adult patients has yielded variables that may be predictive of SSAKI but are not consistently reliable (Table
6). The conclusions of these studies simply state that elevated proinflammatory markers portend adverse outcomes in acute kidney disease and death in patients with septic shock [
39,
40].
Table 6
Variables predictive of SSAKI in selected major adult trials
| Pre-infusion log IL-6, APACHE II score | 127/547 (23.2) | Hazard ratio = 1.16 (95% CI = 1.07 to 1.26); P <0.001 |
| | | Hazard ratio = 1.23 (95% CI = 1.09 to 1.39); P <0.001 |
| Plasma IL-6 | 112/537 (20) | Not significant |
| Plasma TNFα | | Not significant |
| Soluble TNFα receptor type I | | Higher in SSAKI; P <0.0001 |
| Soluble TNFα receptor type II | | Higher in SSAKI; P <0.0001 |
| Plasma TNFα | 34/97 (35) | Higher in patients with SSAKI (P = 0.001) than those without sepsis |
| Plasma IL-6 | | Not significant |
Hoste and colleagues [ 55] | APACHE II score | 30/185 (16.2) | APACHE II score higher (P = 0.002) |
| pH <7.35 | | OR for developing ARF = 6.25 (95% CI = 1.92 to 20.4); P = 0.002 |
| Serum creatinine >1 mg/dl | | OR for developing ARF = 7.56 (95% CI = 2.16 to 26.5); P = 0.002 |
Martensson and colleagues [ 56] | Peak serum NGAL | 18/45 (40) | Not significant in SSAKI vs. septic shock without AKI |
| Peak urine NGAL | | Higher in SSAKI than septic shock without AKI (P <0.05) |
Shapiro and colleagues [ 43] | Emergency room plasma NGAL | 24/661 (3.6) | NGAL >150: sensitivity = 96% and specificity = 51% |
de Geus and colleagues [ 42] | Serum NGAL | 171/632 (27) | OR of developing ARF = 1.7 |
| Urine NGAL | | OR of developing ARF = 1.42 |
| Sepsis | | OR of developing ARF = 9.15 |
NGAL has received considerable attention as a potential biomarker for AKI. NGAL derivation and validation studies were primarily performed in ischemic or nephrotoxic AKI with great efficacy [
13], while investigations of NGAL biomarker utility in SSAKI have demonstrated variable results. In the general ICU population, serum NGAL levels have not been reliably demonstrated to be specific for SSAKI (Table
6). We previously published that while day 1 serum NGAL levels were elevated, with acceptable sensitivity, in children who develop SSAKI, the specificity was quite poor (39%) [
41]. A recent prospective cohort study of 632 adult ICU patients demonstrated that the likelihood ratio for AKI (RIFLE Failure) was 1.71 for NGAL, and that elevated NGAL alone conferred an ROC area under curve of 0.77 for RIFLE Failure but when included in a 'most efficient clinical model' NGAL improved the ROC area under curve for RIFLE Failure to 0.96 [
42]. OK In this study, specificity was 50% for a plasma NGAL cut-off value of 168 ng/ml. Emergency room serum NGAL levels >150 ng/ml were highly sensitive for AKI within 72 hours (96%), but specificity was only 51% [
43]. NGAL is known to be released by activated neutrophils and appears to be intrinsically elevated in sepsis, which may explain its limitation as a biomarker specific for SSAKI.
A variable that may complicate delineation of biomarkers for SSAKI is that the pathophysiology of SSAKI may be different from that of ischemic or nephrotoxic AKI. Although acute tubular necrosis has traditionally been considered the etiology of AKI in sepsis, no conclusive pathologic evidence has demonstrated that this is true [
18]. Animal models of SSAKI demonstrate increased, rather than decreased, renal blood flow and a state of hyperdynamic AKI [
44]. The mechanisms mediating disease progression are multifactorial, and AKI during sepsis may be a result of acute tubular
apoptosis rather than acute tubular
necrosis [
18]. Nonhemodynamic-dependent mechanisms of injury during sepsis, inflammatory and immunologic, may trigger this apoptotic response. This pathophysiology has probably decreased the utility of traditional tests based on the functionality of proximal tubular filtration [
45].
Our study demonstrates that microarray-derived gene expression data can provide a tool for discovery of candidate biomarkers for SSAKI. Outside the neonatal population, we are the first group to attempt to characterize biomarkers specific to SSAKI (and not all-cause AKI) in children. Our data represent the first 24 hours of presentation to medical care, a timeframe that can be considered a therapeutic window for acute intervention. It is important to note that our patients were restricted to having septic shock, an exclusion criterion not previously used in pediatric studies investigating AKI biomarkers. Additionally, our definition of SSAKI identifies patients with severe, persistent kidney injury up to day 7. These candidate biomarkers thus identify patients with resuscitation unresponsive AKI (that is, patients having high serum creatinine levels at admission who subsequently
did not normalize with standard resuscitation). The expression patterns of the 21 upregulated gene probes demonstrate excellent sensitivity and good specificity for SSAKI (Figure
1). The negative predictive value of nearly 100% carries obvious import to the bedside practitioner, but should be tempered by the fact that the prevalence of AKI in our cohort was relatively low (31/179, 17%). Individual performance calculations of two of these gene products, MMP-8 and elastase-2, demonstrate reasonable sensitivity in the derivation cohort (Table
3), but even higher sensitivity in the validation cohort (Table
5). Additionally, in the validation cohort, the specificity and negative predictive values of MMP-8 and elastase-2 increased further. Although MMP-8 and elastase-2 will require further prospective validation studies for confirmation, the results are a relevant demonstration that our gene expression methodology can successfully be leveraged against the tremendous heterogeneity in patients with SSAKI.
The biological links between MMP-8, elastase-2, and SSAKI are unclear at the present time. MMP-8 is a collagen-cleaving endopeptidase, secreted by neutrophils, involved in extracellular matrix remodeling. MMP-8 is implicated in the pathogenesis of numerous rheumatologic diseases and has been demonstrated as a biomarker for disease processes ranging from odontogenic inflammation, to left ventricular remodeling after myocardial infarction, to preterm labor [
46‐
48]. We have recently identified that MMP-8 may be important in the pathogenesis of septic shock [
49], and Hu and colleagues demonstrated that the use of an MMP-8 inhibitor protects mice against endotoxic shock [
50]. There are no previous reports of MMP-8 involvement in the progression of kidney injury. Neutrophil elastase (elastase-2) is a serine protease secreted by neutrophils in response to bacterial infection, is a marker of inflammation, and has been documented as a biomarker of neonatal sepsis [
51]. Interestingly, inhibition of neutrophil elastase reduced progressive renal injury seen in experimental direct endotoxin-induced AKI [
52].
Our study has several limitations. Since in many cases neither a baseline creatinine nor a height was available, the definition of SSAKI in our population was based on a referenced median-for-age creatinine [
34,
35]. Importantly, the initial pediatric RIFLE study also identified AKI in patients based on
assumed estimated creatinine clearance when baseline creatinine levels were unavailable [
53]. Second, some of the performance characteristics (negative and positive predictive values) are influenced by the relatively low prevalence of AKI in our cohort. This is probably secondary to the stringent
a priori definition used for SSAKI. The intentional use of such a definition allowed us to identify the most severely affected patients with fluid-unresponsive AKI [
22]. Third, the data for MMP-8 and elastase-2 continue to demonstrate the relatively low specificity that many candidate biomarkers carry for SSAKI. Still, when all 21 upregulated gene probes are analyzed using the leave-one-out procedure, the specificity increases to 80%, a level not yet seen for SSAKI biomarkers. Fourth, neither MMP-8 nor elastase-2 has been studied formally in AKI disease pathogenesis, from sepsis or other causes. Since MMP-8 and elastase-2 are neutrophil-derived proteases, it is tempting to speculate that they contribute to neutrophil tissue invasion and end-organ damage. In support of such speculation, our microarray data indicate that septic shock is associated with increased expression of several basement membrane-related genes (for example, laminin-family genes) that contribute to neutrophil diapedesis. Finally, our approach relies on extrapolation of mRNA expression data for the derivation of serum protein-based biomarkers. This can be problematic in that mRNA expression does not always correlate with protein expression. What will be required in the future is exploration into the feasibility of measuring the other candidate biomarkers, such as IFNα-inducible protein 27, which demonstrated a more than twofold mRNA expression level between patients with and without SSAKI.
Competing interests
The authors declare that they have no competing interests.
Authors' contributions
RKB assisted with data analysis and wrote the manuscript. SWS assisted with data analysis and database mining. NZC, GLA, NJT, RJF, NA, KM, PAC, RL, TPS, and MTB contributed patient samples and clinical data to the database. They also reviewed the manuscript prior to submission. DSW, PD, and SLG assisted in the initial study design, assisted with the data analysis, and edited the manuscript. HRW conceived the study, performed the majority of the analyses, and assisted with the writing of the manuscript. All authors read and approved the final manuscript.