A qPCR-based metric of Th2 airway inflammation in asthma

We studied two cohorts that were recruited sequentially at UCSF. Our primary analyses were performed on the more recent cohort (hereafter referred to as the primary cohort) which has not previously been studied by mRNA microarray or qPCR for Th2 phenotyping (miRNA data are published elsewhere [11]). Additional analyses were performed in a secondary cohort in which we have previously performed microarray-based Th2-phenotyping [9]. See Additional file 1: Figure S1 for additional details on these cohorts. The primary cohort comprised adults aged 18-70 years recruited via community-based advertising into two groups: 1) 22 healthy controls and 2) 32 subjects with mild-to-moderate asthma not on ICS. Healthy controls had no history of asthma or allergic rhinitis. Subjects with asthma had a participant-reported history of asthma. Subjects with asthma were also required to not have used inhaled or oral steroids for at least 6 weeks prior to enrollment, to have airway hyper-responsiveness (AHR) to methacholine (PC₂₀ to methacholine < = 8.0 mg/mL), and at least one of the following: asthma symptoms on at least two days per week, beta agonist use on at least two days per week, or FEV₁ < 85% of predicted. All participants were non-smokers defined as never smoker or former smokers with no smoking for at least 1 year and total pack-years < = 15. Additional exclusion criteria were pregnancy, upper respiratory tract infection in the 4 weeks prior to enrollment, and the use of certain other medications [11].

In this report, the secondary cohort contributes 25 healthy controls and 38 subjects with mild asthma pre-ICS, of which 18 received 8 weeks of ICS. The research protocol is similar to that for the primary cohort [9], with the following relevant differences: 1) FeNO and symptoms were not measured; 2) ICS treatment was with fluticasone 500 μg inhaled twice daily; 3) a second bronchoscopy was performed at the end of 1 week after the initiation of ICS, rather than at the end of 8 weeks as in the primary cohort.

Baseline characterization included detailed medical history and examination, spirometry, PC₂₀ to methacholine (highest concentration used: 10 mg/ml), measurement of FeNO (NIOX MINO, Aerocrine), complete blood count and differential and measurement of serum total IgE. On the second study visit (1-2 weeks following the baseline visit) subjects underwent bronchoscopy and subjects with asthma were then treated with inhaled budesonide 180 μg twice daily for 8 weeks; subjects maintained diaries of symptoms (shortness of breath, chest tightness, wheeze, cough, or sputum) and returned for spirometry at 2, 4 and 8 weeks. At 8 weeks, subjects underwent repeat bronchoscopy and blood draw for IgE and eosinophil measurements. All clinical studies were approved by the University of California at San Francisco Committee on Human Research, written informed consent was obtained from all subjects, and all studies were performed in accordance with the principles expressed in the Declaration of Helsinki. The study was registered at Clinicaltrials.gov (NCT00595153).

RNA extraction and two-step, nested-primer qPCR was performed as previously described [12]. Epithelial brushings obtained during research bronchoscopy were stored in RLT, with subsequent RNA extraction using column-based purification (Rneasy Micro, Qiagen). RNA quantity was determined by a Nanodrop spectrophotometer (Thermo). RNA quality was assessed on an Agilent Bioanalyzer. 20 ng of RNA was reverse transcribed with random hexamer primers (Superscript/VILO, Invitrogen), and then amplified in multiplex for POSTN, CLCA1, SERPINB2, and endogenous control genes (EEF1A1, PPIA, RPL13A, DNAJA1, and ACTB in the primary cohort; EEF1A1, PPIA, RPL13A in the secondary cohort) with custom primers (Biosearch Technologies; Additional file 1: Table S1) and the Advantage polymerase and reaction mix (Clontech). The resulting amplified cDNA underwent uniplex qPCR with custom primers and Taqman-based probes (Biosearch Technologies; Additional file 1: Table S1), with samples run across separate batches (on a 7900 or Viia 7 from ABI). Potential batch effects were corrected by adjusting cycle threshold (Ct) values on a per-gene basis using samples that were replicated across all three batches, using methods described previously [13]. For each gene, log₂-transformed, normalized, relative expression values were generated by subtracting the Ct value for a particular sample from the mean Ct of the control genes (only EEF1A1, PPIA, RPL13A were used for comparability of qPCR data between the primary and secondary cohorts) for that sample. This approach assumes equal PCR efficiency for all genes of 100%, and produces a result that is mathematically equivalent to normalization of expression values by the geometric mean expression of the control genes [14]. Relative, normalized, transcript quantities were log₂-transformed and then centered and scaled. The Th2 metric is the mean of the centered and scaled values for periostin, serpinB2, and CLCA1 and is labeled the “three-gene-mean”.

Statistical analyses were performed in R version 2.15.1 [15]. Correlations used Pearson's sample correlation coefficient. Univariate tests used a Welch's t test (for unequal variances) unless otherwise noted. Total serum IgE, blood eosinophil concentrations, FeNO, and PC₂₀ to methacholine were log-transformed prior to analysis. ROC curves were computed using the pROC package [16]. All plots were created with the ggplot2 package [17]. p < 0.05 was taken as statistically significant, without corrections for multiple testing.

Of 22 eligible healthy controls and 32 eligible subjects with asthma, 20 and 26, respectively, underwent bronchoscopy; 23 subjects with asthma underwent repeat bronchoscopy at the end of 8 weeks. Data in this report are from the 19 healthy controls, 24 subjects with asthma pre-ICS, and 21 subjects post-ICS from whom epithelial brushings and qPCR data were obtained (Table 1). FEV₁ and the ratio of FEV₁ to FVC were lower in subjects with asthma compared to healthy controls but both groups had FEV₁ > 80% predicted (Table 1). Subjects with asthma had increases in multiple markers of inflammation and atopy: FeNO, total serum IgE, and blood eosinophils.

In our prior study, Th2-high vs low phenotype was assigned through hierarchical clustering which separated groups on the basis of the microarray expression of three epithelial-derived markers of Th2-driven inflammation: CLCA1, periostin, and serpinB2 [9]. In the current study, we developed a PCR-based assay for the same three genes in epithelial brushings. The expression levels of these three genes were intercorrelated across healthy subjects and subjects with asthma (Additional file 1: Figure S2). Given our aim to create an accurate measure of the degree of Th2-driven inflammation present in the airway, this intercorrelation suggested that the effect of variation in individual genes (e.g. changes not in response to Th2 cytokines) would be mitigated by combining their expression values into a single metric. Thus, the expression values for these three genes were averaged into a single quantitative metric we call the “three-gene-mean”, as described in the Methods. The distribution of the three-gene-mean for subjects with asthma not on ICS across both primary and secondary cohorts has two peaks suggesting a natural separation between subjects with low and high levels of Th2 inflammation (Figure 1); nonetheless, it is clear there are a considerable number of subjects with asthma with intermediate levels of Th2 inflammation as well. The peak corresponding to lower values is slightly shifted to the right of the healthy distribution, indicating that as a group, even asthmatic airways with low Th2 inflammation have more Th2 activity than healthy airways. In subjects with asthma on ICS, there is an incomplete reduction of Th2 airway inflammation towards healthy values as measured by this metric (Figure 1).

The three-gene-mean correlates extremely well with FeNO both in the absence of ICS (r = 0.75, p = 0.0002), and in the presence of ICS (r = 0.53, p = 0.024) (Figure 2A). The three-gene-mean is also significantly correlated with blood eosinophils in subjects with asthma not using ICS (r = 0.58, p = 0.003) but this relationship is not significant in subjects on ICS (r = 0.40, p = 0.07) (Figure 2B). Whether on or off ICS, the relationship between Th2 airway inflammation and total serum IgE was not significant (r = 0.33, p = 0.1 off ICS; r = 0.42, p = 0.07 on ICS; Figure 2C).

We found a highly significant correlation between the three-gene-mean and PC₂₀ methacholine measured at baseline off ICS, with 43% of the variance in PC₂₀ explained by the degree of Th2 inflammation (r = -0.65, p = 0.0006; Figure 2D). This suggests that in people with asthma, the degree of Th2 inflammation is a significant determinant of AHR, the functional abnormality of airway smooth muscle which is characteristic of asthma.

Higher baseline three-gene-mean was associated with greater improvement in lung function with ICS treatment at 2 weeks (r = 0.68, p = 0.0004), 4 weeks (r = 0.79, p = 7×10-6) and 8 weeks (r = 0.47, p = 0.03) (Figure 3A). In further analyses, we dichotomized Th2 status at 2 standard deviations above the mean of the healthy distribution for three-gene-mean (Th2-high defined as > 0.1). The mean percent change in FEV₁ from baseline was significantly higher in Th2-high compared to Th2-low asthma at 2 weeks (difference of 9%, p = 0.004), 4 weeks (difference of 12%, p = 0.0002) and 8 weeks (difference of 8%, p = 0.038) (Figure 3B; individual subject responses in Additional file 1: Figure S3). As an alternative approach, we dichotomized based upon the natural separation in the three-gene-mean suggested by Figure 1 (Th2-high defined as baseline three-gene-mean > 0.5). With this alternate threshold, the mean change in FEV₁ was significantly higher in Th2-high asthma subjects at 2 weeks (difference of 10%, p = 0.006) and 4 weeks (difference of 13%, p = 0.0001), with a similar trend at 8 weeks (difference of 8%, p = 0.055).

To evaluate the sensitivity and specificity of the three-gene-mean for prediction of ICS response, we performed an ROC analysis. We defined a positive FEV₁ response as an increase of ≥12% and ≥200 ml, paralleling ATS recommendations [18]. Six out of 22 asthma subjects had a positive response to ICS at 8 weeks (7 out of 23 at 4 weeks). The area under the curve (AUC) of baseline three-gene-mean for prediction of FEV₁ response was 0.87 at 8 weeks (Figure 4A) and 0.94 at 4 weeks (Additional file 1: Figure S4A). Validation in the secondary cohort (4 responders out of 15 subjects at 8 weeks) confirmed a high AUC for three-gene-mean with a value of 0.91 at 8 weeks (Figure 4B) and 0.77 at 4 weeks (Additional file 1: Figure S4B). Table 2 summarizes the quantitative performance of the three-gene-mean by ROC analysis.

Given our aim to create a gold-standard of Th2 airway inflammation with the three-gene-mean, we next compared this metric's performance to that of classical markers of allergic inflammation and AHR, which as shown above and summarized in Additional file 1: Table S2, were well correlated with the three-gene-mean with the exception of total serum IgE. Although our datasets are underpowered to perform a multivariate comparison as with logistic regression, we found it informative to compare performance with ROC analysis. The AUC for three-gene-mean at 8 weeks was greater than the AUC for FeNO, blood eosinophils and IgE (Figure 4A). In addition, although AHR is not considered a Th2-specific characteristic per se, measurement of AHR by methacholine challenge has been shown to be a valuable guide in adjustment of ICS dosing as compared to a standard guideline-based algorithm in a prior study [19]. We found that the AUC for prediction of ICS-response was better for the three-gene-mean than for PC₂₀ methacholine (Figure 4A). Validation in the secondary cohort (4 responders out of 15 subjects at 8 weeks) showed that the three-gene-mean again had a higher AUC compared to other potential biomarkers of ICS response (Figure 4B at 8 weeks; Additional file 1: Figure S4B at 4 weeks).

Finally, whereas an individual gene component may have a higher AUC than the three-gene-mean in some instances, the three-gene-mean has more consistent performance across the datasets (Additional file 1: Table S3). This consistency in the three-gene-mean led us to examine whether the mean of the three markers with which it was best correlated with—blood eosinophils, FeNO, and PC₂₀—may also perform well in predicting ICS response. The results in Additional file 1: Table S3 show AUC's for this combined non-invasive metric that are comparable to, albeit lower than those for the three-gene-mean.

Using a baseline three-gene-mean cutoff value of 0.1, subjects with Th2-low asthma did not have significantly different baseline percentage of days with symptoms (defined as any shortness of breath, chest tightness, wheeze, cough, or sputum) compared to those with Th2-high asthma (82% for Th2-high, 78% for Th2-low, p = 0.7). However, subjects with Th2-high asthma did have a significantly greater improvement in the percentage of days with symptoms compared to Th2-low asthma subjects after 4 weeks of ICS (mean absolute decrease of 42% vs 10%, p = 0.03), and a trend at 8 weeks (mean absolute decrease of 53% vs 24%, p = 0.08). Without an accepted threshold to dichotomize the response in symptoms, we did not perform ROC analyses to compare predictors.