When compared with non-asthmatics, whole-genome expression in peripheral blood of severe asthmatics is different such that a severe asthma disease signature comprised of nearly 1,700 genes was identified by Bigler et al. (38). Within severe asthma, distinct gene signatures associated with eosinophilia, mast cells, and group 3 innate lymphoid cells have been identified in patients with adult-onset asthma when compared with patients with childhood-onset asthma (32). Beyond asthma versus non-asthma stratification, numerous studies (summarized in Table 1) have aimed to define transcriptomic endotypes of asthma by analyzing differential gene expression in bronchial epithelium (27) and induced sputum samples (28, 30, 31, 33). Woodruff and colleagues were the first to identify two evenly sized “Th2-high” and “Th2-low” subgroups of mild, steroid naive asthma based on unsupervised hierarchical clustering of the expression levels of three Th2 cytokine induced genes (POSTN, CLCA1, and SERPINB2) in bronchial epithelial brushings. These subgroups were different with respect to their expression levels of IL-5 and IL-13 within the airway, airway hyper-responsiveness, IgE, blood and airway eosinophilia, and reticular basement membrane thickness. Not surprisingly, Th2-high gene expression was predictive of corticosteroid responsiveness as clinically relevant improvements in FEV₁ following 8 weeks of fluticasone use were only observed in the Th2-high group (27). Wilson et al. (39) identified seven genes (COX-2, ADAM-7, SLCO1A2, TMEFE2, and TRPM-1, and two unnamed genes) in bronchial brushing samples with expression levels that were moderately correlated with submucosal eosinophils, suggesting that they may also predict corticosteroid responsiveness. Given the limited clinical applicability of invasive bronchoscopic samples, Woodruff and colleagues went on to evaluate the variable expression of 14 genes relevant to Th2 inflammation in induced sputum samples (30). Generating a quantitative composite metric of IL-4, IL-5, and IL-13 gene expression, termed the “Th2 gene mean,” the study population was again dichotomized into Th2-high (70%) and Th2-low (30%) subgroups (30). When compared with the Th2-low cluster, the Th2-high cluster had higher eNO levels as well as sputum and blood eosinophil counts (30). It is notable that sputum (AUC = 0.89) and peripheral blood (AUC = 0.89) eosinophil counts alone, but not eNO (AUC = 0.76), performed well as biomarkers of Th2-high asthma as assessed by the sputum cell “Th2 gene mean” (30). Acknowledging the limitations of analyzing pre-selected genes, Baines et al. (28) subjected whole-genome gene expression profiles from induced sputum of adults with stable asthma to unsupervised hierarchical cluster analysis. Three distinct transcriptional asthma phenotypes (TAPs) were identified that had similarities to previously defined sputum inflammatory phenotypes of eosinophilic (TAP1), neutrophilic (TAP2), and paucigranulocytic (TAP3) asthma (28). In fact, 92% of the differentially expressed genes between the TAPs overlapped when the population was grouped according to sputum quantitative cell count (eosinophilic, neutrophilic, and paucigranulocytic). The same investigators subsequently identified a sputum gene expression signature comprised of six genes (CLC, CPA3, DNASE1L3, IL1B, ALPL, and CXCR2) that was able to discriminate asthmatics according to their inflammatory endotype and predict ICS response (29). Of most interest, the six-gene expression signature outperformed the ability of sputum eosinophil count to predict corticosteroid response (FEV₁ responder vs. non-responder; AUC = 91.5) (29). Somewhat contradictory to the findings of Baines et al., when unbiased hierarchical clustering was performed on 508 genes that were differentially expressed between severe asthmatics with and without eosinophilic airway inflammation confirmed by sputum cytometry, one Th2-high and two non-Th2 transcriptome-associated clusters (TACs) were defined (33). When examining the distribution of the three TACs according to sputum inflammatory endotype, two TACs were associated with eosinophilic (TAC1 or TAC3) and neutrophilic (TAC2 or TAC3) inflammation, suggesting that two distinct transcriptional signatures are associated with both sputum eosinophilia and neutrophilia (33). Similarly, Yan et al. (31) did not detect significant between cluster differences in sputum cell differentials suggesting an imperfect association with Th2 pathophysiology. Taken together, conflicting evidence surrounds the association between transcriptomic endotypes and sputum quantitative cell count.

While gene expression studies have dominated the omics landscape, it is their translated products, the proteins, which mediate airway inflammation in asthma by regulating cell recruitment. Numerous studies, limited by sample size, have investigated the “proteomic profile” of asthma in bronchoalveolar lavage fluid (BALF) (40, 41), bronchial biopsy (42), and sputum supernatants (43, 44). One relatively large study by SARP investigators (34) focused on 18 cytokines detectable in BALF and discriminated mild-to-moderate and severe asthmatics into four groups based on cytokine expression. The groups were independent of corticosteroid use and phenotypically distinct with respect to BALF cellularity and lung function. In another study, Hastie et al. (25) investigated the hypothesis that sputum inflammatory granulocytes define phenotypic subgroups of asthma with different patterns of inflammatory proteins. Protein microarray data of induced sputum were stratified by sputum cell differential (eosinophilic: ≥2% eos and <40% neuts; neutrophilic: <2% eos and ≥40% neuts; mixed-granulocytic: ≥2% eos and ≥40% neuts; paucigranulocytic: <2% eos and <40% neuts) and revealed differentially increased inflammatory proteins between the groups (25). Of note, 19 inflammatory mediators were significantly elevated in those asthmatics with neutrophilic bronchitis, a subset of which were correlated with neutrophil counts.

Metabolomics, the exploration of biochemical molecules derived from metabolic processes, was recently reviewed for its applications in asthma (36). Studies to date strongly suggest that metabolic profiles measured in exhaled breath, urine, plasma, and serum may be applied as a point-of-care tool to discriminate asthma endotypes (36). Of most interest, “breathomics” (45) profiles volatile organic compounds in exhaled breath using an electronic nose and has demonstrated ability to discriminate asthmatics from healthy controls (46–48). Electronic nose “breathprints” are correlated with the percentage of sputum eosinophils (48, 49). One highly relevant study concluded that exhaled breath analysis outperformed the ability of both eNO and the percentage of sputum eosinophils to predict corticosteroid response in asthmatics from whom corticosteroids had been withdrawn (48).

There are numerous limitations to clustering techniques applied in the studies discussed above; therefore, these results should be interpreted with caution. One such concern is cluster stability. Specifically, clusters identified at a specific point in time may not be reproducible at subsequent time points. Longitudinal stability assessment of the clusters that have been described is required to understand how they behave over time and in response to treatment and/or exacerbations. Similarly, different clusters may be identified across different asthma populations and therefore validation of cluster classification in independent cohorts of asthmatic patients is necessary to understand generalizability. These important considerations were evaluated for the first time by Loza et al. (6) who defined four severe asthma phenotypes in two independent severe asthma cohorts (ADEPT and U-BIOPRED), performing both external validation and longitudinal stability assessments. Cluster number and composition may also be influenced by the clustering methodology applied. Finally, it is also important to consider statistical robustness as the majority of studies to date define clusters comprised of a small number of asthma patients and hence have limited statistical power.

None of the omics-signatures discussed above have been translated to clinical practice as prognostic or predictive tests. In fact, the clinical utility of these signatures is currently unknown as the majority of studies to date have been observational and hypothesis generating. The ultimate potential of a biomarker-based clinical test is dependent on its analytical and clinical validity in addition to its clinical utility. A 30-point checklist of criteria, developed by McShane et al., should be considered to gauge the potential clinical utility of the omics-based signatures discussed in this review (50). Important considerations include those related to the specimen and assays used, and the appropriateness of the statistical methods used to develop and validate the signature (50). Studies are now required to address the paucity of evidence concerning (1) the longitudinal variability of omics data and endotypes and (2) the responsiveness of omics to therapies. Randomized clinical trials (RCTs) will be necessary to definitively confirm the clinical utility of novel omics-based signatures and design consideration for biomarker RCTs have been proposed (51). Specifically, intervention studies will be necessary to shed light on the capacity of these signatures to direct personalized therapy and to adjust doses of drugs during exacerbations in severe asthma. The qualitative nature of omics is an obvious limitation, therefore the development and application of qualitative metrics are certainly desirable. Similar to that described by Hinks et al. (52), multi-dimensional clinicopathobiologic clustering should also be considered to maximally leverage the measurements available to tertiary care clinicians. It is clear however that omics-based endotypes have similar molecular, physiological, and clinical characteristics to the inflammatory endotypes of eosinophilic, neutrophilic, mixed-granulocytic, and paucigranulocytic asthma.

Although most commonly referred to as Th2-high, type 2 (T2)-high has emerged as a more appropriate and inclusive term given the involvement of numerous cell types (including type 2 innate lymphoid cells and natural killer cells) that are outside the originally described Th2 cell population (54). Eosinophilic bronchitis indicates a T2-driven mechanism and is usually steroid responsive. When bronchitis is eosinophilic in nature with a differential cell count of more than 3% (or eosinophil-free granules are observed), inhaled corticosteroids should be initiated and the dose titrated to keep the sputum eosinophil count below 3%. In situations where high-dose inhaled corticosteroids do not control sputum eosinophilia, oral corticosteroids are required. In corticosteroid-treated patients, absent eosinophils may suggest that the current steroid dose is unwarranted and therefore should be reduced to avoid over-treatment. In RCTs in adults, moderate-to-severe asthmatics managed by normalizing induced sputum eosinophils had significantly reduced exacerbations and hospital admissions when compared with those managed by national asthma guidelines (23, 55). When this strategy was applied clinically in 52 OCS-dependent asthmatics, maintained symptom control, reduced exacerbations, and preserved lung function was observed over 5 years (56). Novel biologic therapies should be considered for their steroid sparing effect, also for the minority of patients who are corticosteroid-insensitive (unresolved sputum eosinophilia despite high doses of oral corticosteroids) (57).

To date, all biologic therapies that have been approved and the majority of those in development aim to target T2 inflammation [recently reviewed (58)] and are therefore directed toward the T2-high eosinophilic asthma endotype. Detailed illustrations of asthma pathobiology and the mechanism of action of targeted therapies are provided in recent reviews (59, 60). Anti-IgE [omalizumab (61)] therapy was the first approved monoclonal antibody and is effective in patients with allergic asthma, confirmed by a positive skin prick test and serum IgE levels ≥30 IU/mL. Two anti-IL-5 therapies are approved [mepolizumab (62) and reslizumab (63)] and one is in phase 3 development [benralizumab (64, 65)] for use in severe eosinophilic asthma. With the potential to block both IL-4 and IL-13, one anti-IL-4 receptor alpha therapy is currently in phase 3 development (dupilumab) following successful phase 2b trials (66, 67). The efficacy of strategies targeting IL-13 alone [lebrikizumab (68) and tralokinumab (69)] is inconclusive as only modest clinical benefit has been shown. Drugs that target both IL-4 and IL-13 signaling (e.g., dupilumab) have reported more clinically relevant effects in phase 2 clinical programs. The reason for this difference is not immediately obvious. Perhaps one of the reasons may be related to the lack of accurate endotyping to identify patients in whom IL-13-mediated biology was not the dominant pathobiology of asthma. Selecting patients with significant airway hyper-responsiveness and mucus secretion may have demonstrated greater clinical effect. Other novel therapies that are currently being investigated include anti-thymic stromal lymphopoietin (70), IL-33 blocking agents (71), and prostaglandin antagonists (CRTH2) (72). A general scheme to choose the appropriate monoclonal antibody based on simple clinical features (e.g., clinical history of allergy, severity of asthma based on the dose of corticosteroids, and readily available biomarkers such as blood and sputum eosinophils and fraction of eNO) is shown in Figure 2. This is based on our clinical experience, the lack of evidence supporting omalizumab in prednisone-dependent asthmatics (73), the lack of efficacy of mepolizumab 100 mg subcutaneous in patients with persistent sputum eosinophilia despite high-dose inhaled and oral corticosteroid use (74), and the efficacy of benralizumab (75) and reslizumab (76) in severe prednisone-dependent patients. However, this approach needs to be prospectively validated in clinical trials.

Neutrophilic bronchitis with a raised total cell count is suggestive of a non-T2-driven disease and is not usually steroid-responsive, but instead a predictor of response to antibiotics. Macrolides, including clarithromycin (77) and azithromycin (78), have demonstrated effectiveness in non-eosinophilic asthmatics. To identify the causative pathogen that may help to direct antibiotic therapy, molecular microbiology, and extended cultures, including 16 s deep sequencing, should be considered in those patients with recurrent infective exacerbations. Logically, neutrophilic bronchitis may be a predictor of therapies targeted at pathways that lead to neutrophil recruitment such as TNF, IL-1, IL-6, IL-8, IL-23, and IL-17. Several small molecule anti-neutrophilic biologics have been developed, although there are currently no active phase 3 trials. Such molecules include CXCR2 antagonists (79, 80), 5-lipoxygenase-activating protein inhibitors (81), anti-IL-17 (82), and anti-TNFα (83). It is evident that few therapeutic options exist for these patients; therefore, intense study of the underlying mechanisms contributing to the T2-low endotype is urgent. Similarly, despite the severity of their disease, currently there are no treatment options for patients with mixed-granulocytic asthma. In fact, no interventions have been evaluated for this specific inflammatory endotype although preliminary evidence suggests therapies targeting the IL-6 pathway may be beneficial (84).

Asthmatics with paucigranulocytic bronchitis may not warrant anti-inflammatory therapy as symptoms in these patients may be driven solely by smooth muscle dysfunction (airway hyper-responsiveness). Therefore, these patients may benefit from smooth muscle-directed therapies such as additional bronchodilators and long-acting anti-muscarinic antagonists, mast-cell directed therapies, or in the most severe cases, bronchial thermoplasty. Our clinical experience suggests that bronchial thermoplasty is indicated when severe airway hyper-responsiveness (PC₂₀ < 0.25) and frequent exacerbations persist despite absent or controlled airway inflammation (85); however, clinical trials are required to confirm this hypothesis. Bronchial thermoplasty after therapy has been optimized to control eosinophilic and/or neutrophilic inflammation has been previously described by Cox et al. (85) for those patients with persistent symptoms. Although the mechanism of action is uncertain, bronchial thermoplasty aims to attenuate airway smooth muscle mass through the delivery of localized thermal energy (86).