Background
Kawasaki disease (KD) is an acute vasculitis that affects infants and children and is the leading cause of acquired pediatric heart disease in the U.S. and Japan [
1]. The cause of KD remains unknown, although epidemiologic and clinical observations suggest that an infectious agent(s) may trigger the inflammatory process in genetically susceptible hosts, who then manifest the clinical syndrome [
2]. The diagnosis of KD is currently based on clinical signs and supportive non-specific laboratory testing [
3,
4]. There is no specific diagnostic test for the disease. If not diagnosed and treated promptly, patients with KD may develop coronary artery dilatation or aneurysms. The cardiovascular damage can be largely prevented by timely administration of intravenous immunoglobulin (IVIG). Thus, there is an acute need for a sensitive and specific diagnostic test or panel that can facilitate diagnosis and permit timely treatment.
We postulated that specific patterns of blood leukocyte gene expression and plasma or urine protein excretion patterns are associated with KD. Identification of these biomarkers could provide insight into the pathophysiology of KD, and even give clues to its etiology. Investigators have taken both genomic and proteomic approaches to biomarker discovery in KD. Transcriptional profiling of blood leukocytes has identified disease-specific expression patterns [
5‐
9]. Protein biomarker studies [
7,
10‐
14] have revealed elevated levels of cytokines, chemokines, and acute phase reactants, but none are uniquely elevated in KD.
Our previous analysis [
8,
9] of peripheral whole blood gene expression compared acute KD and febrile control (FC) patients, revealing increased relative abundance of transcripts associated with innate immune and proinflammatory responses and decreased abundance of transcripts associated with natural killer cells and CD8+ lymphocytes. Expression analyses of separate blood cell type would be more informative. However, the isolation of peripheral blood subsets is cumbersome and may alter gene expression. We, therefore, compared KD and FC whole blood gene expression using cell type-specific significance analysis of microarrays (csSAM [
15]) to analyze differential gene expression for each cell type in a biological sample based on microarray data and relative cell-type frequencies.
Urine is a rich source of proteolytically cleaved proteins cleared from plasma by the kidneys. Profiling analysis of the urinary proteome/peptidome is highly informative for both uro-genital and systemic disease classification [
16,
17]. Using this approach, we previously described urine peptide biomarkers associated with renal transplantation rejection [
18,
19] and systemic juvenile idiopathic arthritis (SJIA)[
20]. We, therefore, performed mass spectrometric analyses of urinary peptides in KD and FC patients.
In this study, we applied an ensemble data-mining approach [
21] integrating either blood cell type-specific gene expression or urine peptidome profiling with clinical multivariate analysis to improve the diagnosis of KD.
Discussion
We have identified three different biomarker panels (7 clinical parameters, 32 blood lymphocyte-specific genes, 13 urine peptides) and developed an integrated algorithm to accurately diagnose KD.
The clinical data we used in the multivariate analysis are routinely obtained during the evaluation of fever. However, clinicians have not used scoring systems derived by multivariate techniques for KD diagnosis. Although the clinical score correctly classified only 80% of febrile patients, patients with either low or high KD clinical scores were diagnosed as FC or KD respectively with 95% accuracy. For febrile patients with the confident diagnosis of KD, timely administration of IVIG can thus be feasible to prevent the development of coronary artery dilatation or aneurysms. For febrile patients with intermediate clinical scores for whom confident diagnosis is not feasible, we developed a sequential algorithm, integrating clinical and molecular findings to improve KD diagnosis. Both the peripheral blood cell type-specific analysis and the urine peptidome biomarker analysis yielded sensitive and specific classifiers, which performed well in the diagnosis of KD. Prospective testing of these biomarker panels will be necessary to confirm their diagnostic utilities.
The csSAM-derived lymphocyte-specific gene markers and their mapped canonical pathways, for example PI3K signaling in B cells and T cell receptor signaling, provide insight into the host response in KD. Confirmation of our de-convolution observations on independent samples will establish the role of these genes as KD biomarkers. Validation of these markers may help to focus the search for the etiology of KD on agents that suppress specific lymphocyte gene expression.
The overlapping sequences of the two COL1A1 and four UMOD peptides suggests that these peptide biomarkers reflect differential activities of disease-related proteases or their inhibitors such as TIMP1 or matrix metalloproteinases in KD [
6,
27‐
32]. Serum peptide biomarker analysis of cancer subjects [
33] has demonstrated overlapping peptide biomarkers generated by disease-specific exo-peptidase activity. We have also observed tight clusters of urine peptide biomarkers in renal allograft dysfunction [
19] and SJIA [
19]. Therefore, the discovery of multiple overlapping collagen and uromodulin peptides suggests that the pathophysiology of KD involves the active degradation of proteins including collagen and uromodulin.
With respect to the concern regarding incomplete KD cases hidden among the FC, we agree that inaccurate diagnosis is always one of the limitations in the absence of a gold standard diagnostic test. However, FC in this study included only patients whose illness resolved within three days of blood sampling OR for whom a definite diagnosis was established (for example osteomyelitis, JIA). None of the FC included here had peeling in the convalescent phase. As for the KD patients, we have maintained a stable rate of coronary artery aneurysms from year to year (approximately 9%) suggesting that our diagnostic practices are stable. All the KD patients in this study were evaluated by one of two experienced clinicians at a single medical center. In this study, most of the FCs were enrolled by our team member, thus assuring consistency in diagnosis and sample collection. Our study is unique in focusing on a clinically relevant control group of children with fever who were actually being evaluated to rule in or rule out KD. All FC were evaluated with a standardized set of clinical laboratory tests that was also used to evaluate our KD patients. Our study also differs from many previous investigations on KD that used samples collected from a large number of hospitals that cared for only a few KD patients each. Therefore, a big problem with consistency in these studies was expected for comparative studies between KD and FC.
Although all FC subjects in this study had laboratory testing for KD as recommended by the American Heart Association (AHA), very few FC had echocardiographic studies done. This is indeed a limitation. Although we acknowledge the potential inaccurate diagnosis of incomplete KD, our status as the sole freestanding children's hospital, sole KD referral center, and sole pediatric emergency department in San Diego County (catchment area of 5 million people) maximizes the likelihood that FC with persistent or progressive illness confused with KD would be captured during a return visit.
We recognize several limitations to the current molecular study for future translations of these biomarkers into bedside practice. First, the small sample sizes limit the power of our biomarker analyses to validate statistically significant associations and to avoid spurious discovery. Future prospective studies with larger sample sizes will be needed to validate our cell type-specific gene expression and urine peptide biomarkers. A second limitation of our study was the lack of formal assessment by clinicians of the pre-test probability of KD in the subjects included in this study. While a large proportion of the febrile controls were referred to our emergency department by physicians for evaluation of possible KD, this was not uniformly true as some febrile controls likely had a low pre-test probability of KD. Since the pre-test probability is an important consideration in evaluating the performance of a diagnostic test, collection of this information will be critical in the next testing phase of a KD diagnostic test. Third, the application of both the cell-specific transcript patterns and urine peptide biomarkers for the diagnosis of KD will require development of technology for the rapid identification of both whole blood transcripts and urine protein fragments in the clinical laboratory.
Our flexible clinical scoring metric is amenable to automation to develop data-driven predictive systems. Consistent with the current mandate to improve electronic medical record (EMR) use [
34] and future interoperability between the hospital EMR and our predictive algorithm based applications consisting of demographic, clinical and genomic/proteomic data can serve an effective platform to allow interfacing between interdisciplinary teams (bed and bench side; what is known and what is practiced) for productive translational medicine.
Acknowledgements and funding
The authors thank the participating patients and their families who donated blood and urine for these studies. We also thank Joan Pancheri for sample collection and DeeAnna Scherrer for specimen processing. The authors thank our colleagues at the Stanford University Pediatric Proteomics group for critical discussions, and the Stanford University IT group for Linux cluster support.
This work was supported in part by a grant (HL69413) from the National Institutes of Health, National Heart, Lung and Blood Institute to JCB, Stanford University Children's Health Initiative (CHI program) Grant to XBL, KL, and JS.
Competing interests
The authors declare that they have no competing interests.
Authors' contributions
XBL and HJC are the major contributors responsible for data analysis and project management. XBL, HJC, JCB and JS contributed to overall experimental design and assay platform setup. ZP, SP and JJ contributed to the pathway data analysis. GGL and YS contributed to the patient demographic analysis. KL contributed to the urine peptidome profiling analysis. KL, TTSY and JCW contributed to the sample storage and process. XBL, JTK, JCB and HJC contributed to the manuscript writing. All authors read and approved the final manuscript.