Gene transcript abundance profiles distinguish Kawasaki disease from adenovirus infection

Abstract

Background: Acute Kawasaki disease (KD) is difficult to distinguish from other illnesses that involve acute rash or fever, in part because the etiologic agent(s) and pathophysiology remain poorly characterized. As a result, diagnosis and critical therapies may be delayed.

Methods: We used DNA microarrays to identify possible diagnostic features of KD. We compared gene expression patterns in the blood of 23 children with acute KD and 18 age-matched febrile children with 3 illnesses that resemble KD.

Results: Genes associated with platelet and neutrophil activation were expressed at higher levels in patients with KD than in patients with acute adenovirus infections or systemic adverse drug reactions, but levels in patients with KD were not higher than those in patients with scarlet fever. Genes associated with B cell activation were also expressed at higher levels in patients with KD than in control subjects. A striking absence of interferon-stimulated gene expression in patients with KD was confirmed in an independent cohort of patients with KD. Using a set of 38 gene transcripts, we successfully predicted the diagnosis for 21 of 23 patients with KD and 7 of 8 patients with adenovirus infection.

Conclusions: These findings provide insight into the molecular features that distinguish KD from other febrile illnesses and support the feasibility of developing novel diagnostic reagents for KD based on the host response.

Figure 1

Interindividual variation in patients with Kawasaki disease or febrile illnesses with similar clinical presentation. Transcripts with a ≥3-fold change from the median expression level in ≥2 of 41 samples are shown. Genes and samples are organized using hierarchical clustering; each row represents a single transcript (complementary DNA clone), and each column a sample. Color scale at bottom reflects range of expression relative to median expression for each transcript. Black indicates median level of expression; red, greater expression; green, less expression; gray, missing data. Horizontal bars at bottom indicate sample clusters referred to in the text.

Figure 2

Comparison of whole-blood transcription profiles from patients with Kawasaki disease (KD) and control groups. Correlation coefficients were calculated for each KD sample and each control sample in the data set, and they were plotted by comparison group. The central horizontal line in each box represents the median correlation coefficient, the upper and lower borders of each box indicate the 25th–75th percentiles, and the whiskers indicate the 10th–90th percentiles. Open circle, data point outside the 90th percentile. P values were calculated with the nonparametric Spearman rank sum test; NS, not significant. GAS, group A β-hemolytic Streptococcus infection; Rxn, reaction.

Figure 3

Kawasaki disease (KD)–associated transcripts. Transcripts with significant differences in abundance were organized using hierarchical clustering. Color scale at bottom reflects range of expression relative to median expression for each transcript; black indicates median level of expression; red, greater expression; green, less expression; gray, missing data. C1–C4 and vertical bars denote gene clusters discussed in the text. A, Samples sorted by diagnosis and day of illness. B, Diagnosis-specific differences in transcript level. From left to right, columns represent patients with KD versus all control subjects and patients with adenovirus infection, group A Streptococcus (GAS) infection, or drug reactions (Rxn). Red indicates statistically significant difference compared with patients with KD; black, indicates no statistically significant difference. C, Whole-blood transcript levels from 20 patients with KD sampled in both acute and convalescent phases of disease [4].

Figure 4

Average expression of gene clusters associated with Kawasaki disease (KD). A, Average expression of gene clusters identified in figure 3. B, Estimated number of samples (both KD and control) needed to identify an existing difference in expression levels between patients with KD and a specific control group, given the average difference in expression observed in this data set. Cluster numbers correspond to clusters identified in figure 3. The type I error (α) was set at .05; power (β) at .8.

Figure 5

Interferon-responsive gene transcript levels. Transcript levels in whole blood were measured using TaqMan real-time reverse-transcription polymerase chain reaction and normalized to levels of TAF1B transcript. KD, Kawasaki disease. *P < .05.

Figure 6

Prediction of Kawasaki disease (KD) and adenovirus infection using transcript abundance levels. The prediction analysis of microarrays algorithm was used to identify 38 transcripts that predicted KD infections in a 10-fold cross-validation scheme. A, Predicted and actual classification. B, Gene expression patterns for the 38 transcripts used to predict infection class. Gene symbols or Unigene cluster identification numbers are presented for each transcript. Samples from patients with KD and subjects with adenovirus infection are sorted by study identification number. Color scale at bottom reflects range of expression relative to median expression for each transcript; black indicates median level of expression; red, greater expression; green, less expression; gray, missing data.