Background
Asthma is heterogeneous, and more refined phenotypic subgroupings will be required to better guide clinical management [
1]. For example, serum IgE levels are used to successfully target the anti-IgE monoclonal antibody omalizumab [
2], and sputum eosinophils have been used to target the anti-IL5 antibody mepolizumab [
3‐
5]. Nonetheless, most asthma guidelines do not account for underlying biological mechanisms in the decision to treat with the most commonly used anti-inflammatory therapy, inhaled corticosteroids (ICS). Indeed, randomized controlled trials have shown that 30-45% of patients with asthma fail to have an improvement in FEV
1 even to high dose ICS [
6,
7], and 25% are not well-controlled based upon symptoms [
8]. These data suggest that biomarkers are needed to determine who will benefit from ICS or from progressive increases in ICS dosing. Furthermore, as targeted therapies such as anti-IgE and anti-interleukin-13 become more widely available, the identification of the patients who are most likely to respond will be critical.
Our group has previously identified airway epithelial gene expression markers of a Th2-high molecular phenotype of asthma [
9]. In a clinical trial of ICS, only this Th2-high subgroup had the expected increase in FEV
1. However, one practical limitation is that Th2 status was established through unsupervised hierarchical clustering of microarray data [
9]. Disadvantages inherent to clustering of microarray data include instability in phenotype assignment, cost, inefficient use of RNA, and only a dichotomous high vs low output. These limitations have been partially addressed by the development of a quantitative gene expression signature from airway biopsies, but this metric still was constrained by the use of microarray data [
10]. Another limitation of our prior work is that we did not concurrently measure the fraction of exhaled nitric oxide (FeNO), an alternative biomarker of Th2 inflammation. Our overarching goal in the current study was to develop a continuous metric of Th2 inflammation based on qPCR that avoids the limitations of array- and cluster-based phenotyping and that can be used to standardize the measurement of Th2 inflammation across bronchoscopy-based asthma studies. Our specific goals were to determine the relationship between our qPCR-based Th2 metric and other accepted markers of Th2 inflammation including FeNO, and ascertain how well this metric predicts lung function and symptom response to ICS when compared to other candidate biomarkers.
Discussion
This qPCR-based three-gene-mean metric contributes to our understanding of Th2 inflammation in asthma by: 1) demonstrating that previously published epithelial gene expression markers correlate well with FeNO, 2) identifying a strong association between Th2 inflammation and smooth muscle dysfunction (measured by AHR) among patients with asthma, and 3) demonstrating that epithelial gene expression markers have good predictive capacity for both lung function and symptomatic improvement with ICS. Since it is obtained through bronchoscopic evaluation, the major application of this three-gene-mean metric of Th2 inflammation will be in clinical and translational research studies. However, development of a quantitative and reproducible gold-standard for clinical research applications is a necessary pre-requisite for the assessment of non-invasive biomarkers.
This qPCR-based metric has several advantages over our prior classification which used hierarchical clustering of microarray data. Most notably, microarrays and hierarchical clustering cannot be repeated in a standardized way in future studies. This new metric of Th2 airway inflammation requires as little as 20 ng of RNA from airway epithelial cells using qPCR, a widely available technique. With the use of inter-run calibrators, three-gene-mean values can be compared across different studies, platforms, reagents, and laboratories [
20]. Finally, a continuous measurement of Th2 airway inflammation provides more statistical power to detect associations with other clinical and molecular features of asthma, without losing the ability to apply a threshold to make dichotomous Th2-high vs low classifications for practical applications. An important finding from application of this continuous metric is that regardless of where the threshold is made, Th2-low asthma subjects have low rather than absent Th2 inflammation. Analysis of microarray-based gene expression in airway biopsies from our earlier study also showed that a continuum of Th2 inflammation is present even among those with Th2-low inflammation [
10].
To place our metric into context, we correlated the three-gene-mean with a range of other markers of Th2 inflammation and ICS responsiveness. We found a strong correlation between the three-gene-mean and blood eosinophils, FeNO, and PC
20 methacholine, but not with total serum IgE. In particular, FeNO was well correlated with the three-gene-mean both before and after a course of ICS, suggesting that FeNO measurements are a good surrogate for the three-gene-mean. Nitric oxide production is increased in the lungs of asthma patients in part due to increased transcription of the NOS2 gene encoding inducible nitric oxide synthase in epithelial cells [
21]. We have previously found airway epithelial expression of periostin, CLCA1, and serpinB2 to be correlated with NOS2 expression [
10,
22]. However, we have not previously directly compared our epithelial expression markers to FeNO. FeNO's reported ability to identify a subgroup of severe asthmatics on high-dose ICS that selectively respond to the Th2-targeted, anti-interleukin-13 monoclonal antibody lebrikizumab, also supports its strength as a Th2 marker [
22,
23]. Notably, however, FeNO has not performed well in clinical studies as a measure by which to modulate ICS dose, which may be from poor reproducibility and other technical issues [
22,
24]. Therefore, the identification of additional non-invasive markers of Th2 inflammation may be valuable. Sputum eosinophils have also been proposed as a marker of Th2 airway inflammation, with studies showing that modulating ICS therapy by sputum eosinophils counts can lead to a reduced number and severity of exacerbations [
24,
25]. Unfortunately, we did not have sputum eosinophil counts in the present study, but certainly its relationship to the three-gene-mean will be of interest in future studies.
Importantly, sputum eosinophil measurements are difficult to perform in clinical practice; an easier alternative is protein levels of serum periostin, which like FeNO, identified a subgroup of severe asthmatics responsive to lebrikizumab [
22,
23]. Although serum periostin was not measured in the current study, future studies that relate the three-gene-mean to serum periostin will potentially help optimize the threshold for serum periostin that best identifies individuals that will respond to ICS or Th2-targeted therapies. Such an analysis will also provide a better understanding of the relationship between serum periostin and Th2 airway inflammation, especially in the setting of other, non-asthma epithelial disorders that can potentially alter serum periostin levels such as allergic diseases of the gastrointestinal tract, skin, and upper airway.
Using ROC curves, we compared the performance of the three-gene-mean to other markers of Th2 inflammation in predicting response to treatment with ICS. The more consistent performance for the three-gene-mean across our two datasets compared to its components supports the concept that the mean may be resistant to individual gene-level variation that is less reflective of Th2-driven inflammation. The three-gene-mean had a larger AUC than FeNO, eosinophils, total serum IgE, PC20, and a combination of three non-invasive measures (FeNO, eosinophils, and PC20) for FEV1 improvement. Overall, this comparative performance analysis suggests that non-invasive measures can accurately predict ICS response, and that the three-gene-mean is a valid research tool by which to establish a lung-based gold-standard for evaluating other biomarkers of Th2 inflammation.
Airway epithelial cells are increasingly recognized as a source of key initiators of allergic inflammation including IL-33, IL-25, and TSLP [
26]. Therefore, it is possible that transcript levels of these genes may also serve as useful biomarkers of allergic inflammation. Although we did not measure the expression levels of these genes by qPCR in this study to test this particular hypothesis, genome-wide microarray expression profiling on the cohorts in this study shows no difference between subjects with asthma and healthy controls for IL-33, IL-25, and TSLP (data not shown). Whether these negative results are due to limitations of the probes on the microarray or true lack of transcript-level differences can be addressed in a future study.
As an alternative to biomarkers, published reports have identified lung function measures as predictors of steroid response in asthma. Baseline values for FEV
1 as percent of predicted, the FEV
1/FVC ratio, and bronchodilator percent reversibility were strongly correlated with improvement in FEV
1 in response to ICS [
6,
7]. All three of these spirometry-based predictors use baseline FEV
1, which is a parameter used in the calculation of the outcome, improvement in FEV
1. Therefore, the predictive value of these lung function measures for lung function improvement is in part tautological. In addition, biological markers have the advantage over lung function measures that these markers can identify the presence of specific inflammatory pathways targeted by specific therapeutics. The most salient examples would be Th2-targeted biologics such as mepolizumab and lebrikizumab.
Th2 sub-grouping of asthma has broad clinical research applications. The presence of features of asthma despite low levels of Th2 inflammation raises the question of whether other underlying abnormalities are present in the airways of patients with Th2-low asthma. Hypotheses proposed for the pathology in Th2-low asthma include intrinsic airway smooth muscle dysfunction and non-Th2 inflammatory pathways such as Th17-driven inflammation [
27,
28]. The ability to characterize a patient with asthma as Th2-low and study airway specimens from this subgroup enables testing of these proposed underlying abnormalities. It is likely that some individuals with Th2-high asthma also have non-Th2 pathways of inflammation active given that Th2 status only partially explains asthmatic features such as airway mucin stores and AHR [
9]. Assignment of Th2 status to research subjects will also allow the study of abnormalities that co-exist with Th2 inflammation.
We recognize a number of limitations in the current study. The sample size did not prevent validation of the three-gene-mean's predictive capacity, but it did prevent us from making statistical comparisons of predictive capacity amongst competing markers of Th2 inflammation via logistic regression because of the risk of over-fitting the model. Although 8 weeks of ICS treatment may be considered a relatively short timeframe, a number of detailed studies of the effects of ICS on FEV
1 have shown that there is minimal additional response beyond 3-6 weeks [
6,
7]. Furthermore, reductions in exacerbations and symptoms are likely more relevant to patients and their health-care providers than improvements in FEV
1. Nevertheless, it was previously shown that a short-term, 6 week improvement in FEV
1 to ICS was significantly associated with reductions in exacerbations and symptoms over a longer term of 4 months [
7]. Ongoing and future longitudinal studies will better establish the performance of the three-gene-mean in predicting longer-term asthma control and exacerbation frequency. Our study also is limited to young adults with mild-to-moderate asthma. The performance of the three-gene-mean in a wider age range, and in patients with severe asthma on high doses of ICS requires further study.
Competing interests
The authors declare that they have no competing interest.
Authors’ contributions
NRB conceived and designed the qPCR assay, participated in the collection of samples and subject data in the primary clinical study, analyzed and interpreted the data, and drafted the manuscript. ODS conceived and designed the qPCR assay and analyzed data. CPN coordinated the primary clinical study and participated in the collection of samples and subject data in the primary clinical study. CNN carried out the qPCR and participated in sample processing from the clinical study. JRA conceived and designed the primary clinical study, and revised the manuscript critically for intellectual content. JVF conceived and designed the clinical studies, participated in the collection of samples and subject data in the clinical studies, and revised the manuscript critically for intellectual content. PGW conceived and designed the clinical studies, participated in the collection of samples and subject data in the clinical studies, analyzed and interpreted data, and drafted the manuscript. All authors read and approved the final manuscript.