Are We Meeting the Promise of Endotypes and Precision Medicine in Asthma?

Abstract

While the term asthma has long been known to describe heterogeneous groupings of patients, only recently have data evolved which enable a molecular understanding of the clinical differences. The evolution of transcriptomics (and other ‘omics platforms) and improved statistical analyses in combination with large clinical cohorts opened the door for molecular characterization of pathobiologic processes associated with a range of asthma patients. When linked with data from animal models and clinical trials of targeted biologic therapies, emerging distinctions arose between patients with and without elevations in type 2 immune and inflammatory pathways, leading to the confirmation of a broad categorization of type 2-Hi asthma. Differences in the ratios, sources, and location of type 2 cytokines and their relation to additional immune pathway activation appear to distinguish several different (sub)molecular phenotypes, and perhaps endotypes of type 2-Hi asthma, which respond differently to broad and targeted anti-inflammatory therapies. Asthma in the absence of type 2 inflammation is much less well defined, without clear biomarkers, but is generally linked with poor responses to corticosteroids. Integration of “big data” from large cohorts, over time, using machine learning approaches, combined with validation and iterative learning in animal (and human) model systems is needed to identify the biomarkers and tightly defined molecular phenotypes/endotypes required to fulfill the promise of precision medicine.

Clinical relevance of molecular phenotyping:

1. Asthma is a heterogeneous disease, in which identifying phenotypes can lead to better responses to both nonspecific and specific targeted therapies.

2. Patients who are not responding to broad anti-inflammatory therapies should be evaluated for elevations of currently available biomarkers, including blood eosinophils and fraction exhaled nitric oxide.

3. Severe asthma patients in whom comorbidity and adherence issues have been addressed should be considered for targeted type 2 (T2)-biologic therapies.

4. Not all patients will respond equally well to every T2-targeted biologic, and more studies are needed to determine which biologic for which patient.

5. T2-Lo asthma remains poorly defined and requires rule out of T2 activity/biomarkers over time.

I. INTRODUCTION

Although “asthma” arose from the Greek term for “short of breath, ” it likely was not associated with a broadly recognizable entity, or disease, until the 19th century. Henry Hyde Salter, in his treatise “On asthma and its treatment,” specifically described asthma as “Paroxysmal dyspnoea of a peculiar character with intervals of healthy respiration between attacks.” This definition, with some adaptation, remains recognizable today, as asthma is classically defined as reversible airflow limitation (or airway hyperresponsiveness) in the face of variable symptoms, which include wheeze, dyspnea, and cough. Spirometric measurement of airflow limitation with reversible improvement of at least 12% (and 200 ml) was recognized as increasing the likelihood of diagnosis in the 1980s (1). In the absence of reversible airflow limitation, hyperresponsiveness can be measured by inhalational challenges with various stimulatory agents (or exercise), which, in asthmatic patients, induce bronchoconstriction at a lower threshold than individuals without asthma. Additionally, most guideline-associated definitions include inflammation as a key characteristic, but without specific inflammatory biomarkers. This current definition of asthma includes nonspecific characteristics that allow inclusion of a wide range of patients. In fact, many patients receive a clinical asthma diagnosis without meeting the physiological definitions When asthma is defined by doctor diagnosis and asthma medication use, as in most large epidemiologic and genetic population studies, up to 30% may not in actuality have asthma (2, 3). Thus asthma is ripe for investigations which refine the understanding of the mix of clinical, physiological, pathobiologic, and genetic characteristics to enable identification of molecular phenotypes, their biomarkers, and precision medicine therapies.

II. BACKGROUND

A. Heterogeneity of Clinical Asthma

A phenotype is defined as the observable characteristics of an organism that result from the interaction of its genotype (total genetic inheritance) with the environment (https://www.britannica.com/science/phenotype). While the term phenotype has only recently been popularized, asthma has long been recognized as a heterogeneous disease, with the first clinically defined phenotypes those of intrinsic versus extrinsic asthma, described by Francis Rackemann in the 1940s (226). From the early 20th century, asthma was considered an allergic disease, often treated with allergy immunotherapy (71). However, using his clinician’s recognition of overlapping characteristics, Rackemann (226) identified differences between patients who typically developed their disease in childhood, associated with allergic sensitization (extrinsic asthma), and those who developed their disease in adulthood (age 40 and above) (intrinsic asthma), whose disease was less associated with allergies and more likely associated with infections and lower lung function. Similarly, a clinical phenotype of aspirin exacerbated respiratory disease (AERD) was described in 1922, but it was not until 1968 that the phenotype of aspirin sensitivity, asthma, and nasal polyposis became widely recognized (167, 234). These early clinical observations have now been reproduced in multiple hypothesis-driven and unsupervised approaches (184, 186, 190).

In the ensuing era of broadly effective anti-inflammatory/corticosteroid (CS) treatment (280), it eventually became recognized that not all asthma patients responded optimally to this nonspecific immunosuppression. This recognition reignited the concept of heterogeneity in the late 1990s (28, 282). This clinical recognition of asthma heterogeneity, combined with therapeutic needs unmet by CSs, contributed greatly to the evolution of molecular phenotyping.

B. Breaking Down the Characteristics of Asthma

The definition of asthma encompasses multiple clinical and physiological characteristics, which contribute to the heterogeneity of patients’ presentations (280). In the last several years, efforts have been made to deconvolute asthma into these essential characteristics or “treatable traits” (5, 213). These traits include clinical, physiological, and pathobiologic variables such that patients can be identified as having airway obstruction which normalizes with treatments or by having a specific eosinophil-associated airway process, cough, or repeated infections. Numerous other traits could also be considered (TABLE 1). While this simplifies approaches to some degree, in isolation, it does not account for overlapping or conflicting traits. In fact, “treatable traits” typically exist in relation to other characteristics. Traits also have varying definitions and underlying mechanisms, with eosinophilia and airway obstruction often defined in different ways. Similarly, multiple mechanisms could drive a trait like airway obstruction. These include excess mucus, smooth muscle contraction, fibrotic changes, or even alveolar changes, each of which could be driving obstruction in some patients more than others and all with differing treatment implications. Thus, although deconvoluting a disease into its simplest parts clearly offers guidance (see next section), a more integrated approach may be more beneficial.

Table 1.

“Treatable” traits and their nonspecificity

Trait	Definition	Related Factors	Overlap with Other Traits	Range of Potential Treatments
Eosinophilic inflammation	Defined by blood/sputum eosinophils with variable thresholds	T2-related inflammation, eosinophils in blood/lung	Associated with worse airflow limitation, exacerbations, leaner BMI	Corticosteroids, T2-targeted biologics
Airflow limitation	Obstruction: various definitions of low FEV/FVC. Reversibility: 12% and 200-ml improvement in FEV1. Full reversibility: FEV1 improves to normal range.	Obstruction: smooth muscle hypertrophy/hyperplasia and contraction, fibrosis, loss of alveolar attachments, mucoid obstruction, edema. Reversibility: contraction, T2 inflammation.	Reversibility associated with eosinophilia. Airway obstruction associated with infections.	Bronchodilators, corticosteroids, mucus modifying agents, possibly T2 biologics. Comment: smooth muscle targeted bronchodilators only treat contraction, possibly mucus clearance.
Airway infection	Positive sputum cultures	Bronchiectasis, bronchiolectasis, microbiome dysbiosis		Antibiotics, possibly macrolides long term
Cough	Cough hypersensitivity to challenges, truncated inspiratory flow volume loops, direct examination of inspiratory vocal cord closure	Neuropeptide abnormalities, gastroesophageal reflux, postnasal drip, vocal cord dysfunction, asthma		Treatment directed to cause
Obesity	BMI, % body fat, CT	High-dose corticosteroids, inactivity from disease, metabolic dysfunction, microbiome, differences by age at onset	Low eosinophils	Reduction in corticosteroid dose, improved asthma, exercise and diet
Psychosocial issues	Psychiatric diagnosis?	Poor adherence, access to meds, socioeconomic, cultural, psychological, very severe asthma		Counseling, understanding of social issues, better asthma treatment
Age at onset	Variably defined cut-points from 12 to 40	Presence of allergic features, sinus disease, eosinophilia	Eosinophilia, airway obstruction, reversibility	Anti-IL-5 may do better in later onset

BMI, body mass index; CT, computed tomography; FEV, forced expiratory volume; FVC, forced vital capacity; IL, interleukin; T2, type 2.

III. LESSONS FROM CLINICAL TRAIT-TARGETED AND CLUSTERING STUDIES

Renewed interest in identifying new subgroups of asthma patients emerged in the late 1990s. Three factors contributed to this renewed interest, including that 1) responses to CSs were highly variable, 2) pathological heterogeneity was increasingly appreciated (28, 97, 282), and 3) studies of Th2 (now type 2)-pathway targeted biologics were negative across “all-comers” with asthma (83).

A. Targeted, Hypothesis-Driven Approaches and Application of Precision Medicine

Early trait-based studies often targeted single characteristics. Age at onset was consistently identified as a distinguishing trait, confirming and expanding on the concepts of extrinsic and intrinsic asthma (184, 190, 226). Importantly, these studies also shed light on eosinophilic inflammation, measured in blood or sputum. Despite earlier biases, it was not present in all patients with asthma or even severe asthma, but was more common in exacerbation-prone, CS-responsive, and late-onset asthma (28, 101, 184). This recognition was critical to completion of the first successful precision medicine trials utilizing eosinophil targeted biologic therapies only in patients with eosinophilic exacerbation-prone severe asthma (110, 199). Similar hypothesis-driven studies identified interleukin (IL)-6 and related pathways, particularly in blood, in association with more severe obese and exacerbation-prone asthma (218). Finally, fungal-associated asthma has also been suggested as a clinical phenotype, based, among other things, on the biologic association of fungi with extrinsic proteolytic activity (169). It was included on an early list of asthma endotypes as well (174). Clinically, fungal asthma is variably defined based on levels of fungal specific IgE-associated sensitivity and/or fungal presence. Three small double-blind placebo controlled trials of antifungal therapies have been performed. Of the two larger studies, one, which studied voriconazole in patients sensitized to Aspergillus fumigatus, was negative, while the other, which studied itraconazole in those sensitized to any of a number of fungi, was positive (4, 62). Both utilized asthma quality of life as the primary end point, suggesting that molecular approaches may be needed to better define a fungal-driven asthma phenotype. However, these hypothesis-driven studies all identified single pathways (i.e., eosinophils, fungi, IL-5, IL-6) in relation to certain traits (exacerbations/obesity/age at onset, fungal sensitivity), initiating the application of precision medicine to biomarker-identifiable patients.

B. Less Biased/Supervised Clinical Clustering Approaches

As statistical methods evolved integrating increasing amounts of data into single analyses, less biased (or less hypothesis-driven) approaches emerged. One early study used only seven selected/hypothesis-driven variables (including sputum eosinophil percentages, symptoms, and peak flow variability) and a K-means clustering approach (see analytic approaches) on two separate clinic populations (111). Four reproducible asthma patient clusters were identified, on the basis of symptoms, eosinophilic inflammation, and age at onset. While simple in design, these clusters (or phenotypes) are broadly reproduced in more complex studies, including 1) a benign early-onset asthma phenotype; 2) a symptom predominant, minimally obstructed obese phenotype; 3) an early-onset increasingly severe phenotype; and 4) a late-onset more severe eosinophilic phenotype.

The maturation of large asthma networks allowed incorporation of increasing numbers of characteristics into clustering approaches of larger populations. Using only clinical and physiological characteristics collected as part of the National Heart, Lung, and Blood Institute’s Severe Asthma Research Program (SARP) network, Moore et al. (118) clustered 34 asthma characteristics, derived using a random forest approach, to reveal 5 different patient clusters, 3 of which were severe, and differed primarily by lung function variables and age at onset. As described in earlier hypothesis-driven studies, later onset disease was less likely to be allergic, but was also more obese and female, while the most severe disease had the most exacerbations, highest systemic CS use, and the worst airflow limitation which failed to normalize after bronchodilators. This study was followed by an analysis of 378 asthmatic and healthy controls (HC), which incorporated lung-specific inflammatory characteristics [i.e., bronchoalveolar lavage (BAL) cell differentials] from those who had undergone bronchoscopy (186). This study included a greater number (and range) of variables into the clustering solution, as well as HCs. With the use of hierarchical clustering, 10 variable clusters and 6 clinical clusters were identified: 1 healthy cluster, 3 severe, and 2 milder asthmatic clusters. Validating previous studies, a late-onset, nasal polyposis-prone highly eosinophilic cluster was recognizable, as were several early-onset more allergic groups. Similar to the Moore SARP clusters, a very severe cluster was identified with persistent obstruction, despite a large bronchodilator response. Inclusion of inflammatory characteristics showed that this most severe cluster associated with high levels of the asthma biomarker, fraction exhaled nitric oxide (FeNO), as well as a mixed neutrophilic/eosinophilic inflammatory process. These variable associations were validated in separate studies where FeNO levels were identified as the strongest independent predictor of oral CS use and where a mixed granulocytic process was observed in sputum from the most severe patients (191, 291). Finally, a large two-way cluster analysis was performed on over 300 patients in the British Thoracic Society Severe Asthma Registry, using factor analysis to first do initial dimension reduction (203). With the use of 23 variables including allergy, obesity, blood eosinophils, treatment, exacerbations, and lung function, late-onset eosinophilic, late-onset obese, and early-onset allergic clusters again emerged. Although lacking molecular input, consistent clinical patterns were emerging, laying the groundwork for the approaches to molecular phenotyping, endotyping, and precision medicine outlined in FIGURE 1.

FIGURE 1.

Generalized paradigm for the implementation of precision medicine. Asthma presents in distinctly different ways based on integration of genetics and environmental exposures. Granular clinical and molecular phenotyping can generate novel target pathways that can be validated over time and through the use of animal models. These approaches should lead to development of biomarker targeted therapeutic trials which improve the overall efficacy and safety of drugs, bringing the best drug option to the correct patient.

IV. EVOLUTION OF MOLECULAR PHENOTYPING AND ENDOTYPING

The term molecular phenotyping alludes to identification of specific molecular pathways in relation to clinically distinguishable traits or phenotypes. Molecular phenotyping requires identification of consistent plausible molecular pathways in relation to disease, association with clinical outcomes, and, ideally, the improvement of outcomes when the identified pathways are specifically targeted (TABLE 2, FIGURE 1). Endotyping is the penultimate phenotype, or even disease. It is a condition which is defined by a distinct functional or pathobiologic mechanism (or treatment response) (10). This implies that endotypes as defined above require identification of a specific pathway(s) that controls most aspects of the disease. Application of this definition to asthma is controversial as interpretations of the overall importance of a specific pathway to a molecular phenotype may vary. Some might suggest an endotype includes all T2-Hi phenotypes. However, a more literal interpretation is that a molecular pathway (or treatment) truly defines the disease. While T2 (or associated) inflammation is present and relevant to mild to severe early-onset allergic asthma, the data per se do not yet fully support T2 inflammation as the defining pathway for those molecular phenotypes. In fact, it is increasingly recognized that T2 biologics (treatment responses related to IL-4, -5, and -13 pathways) are not as effective in early-onset disease (23, 29). However, considerable data now support the concept that a T2-Hi, adult-onset eosinophilic asthma endotype is emerging, with the dramatic responses to anti-IL-5/5R and perhaps anti-IL-4Rα as well, supporting involvement of ILC2 (or other) cells. As noted, treatment response could also eventually define an endotype, if, for whatever reason, a patient had a dramatic response to a specific pathway blocker but did not fall into any previously defined clinical phenotype. However, ideally additional clinical characteristics would associate with that response. Thus newly defined and characterized endotypes could emerge either as outshoots of the current molecular phenotypes, or alternatively as distinctly new entities.

Table 2.

Postulates that define molecular phenotype/endotypes

1. Identification of plausible biologic pathway

2. Consistency of expression of the pathway over time

3. Correlation of biologic pathway expression level with relevant clinical outcomes

4. Identification of measurable biomarker for the pathway

5. Reduction (molecular phenotype) or resolution (endotype) of clinical manifestations of disease when the pathway is targeted

The development of multiple/’omic platforms, capable of measuring hundreds to even millions of analytes, typically of a specific type, on a single specimen, opened the flood gates for molecular phenotyping by vastly increasing the complexity and granularity of molecular data (FIGURE 1). These platforms included microarrays, RNA sequencing, two-dimensional gels, multiple-antibody bead-targeted proteomic assays (luminex and others), mass spectroscopy-based multi-antigen cell sorting techniques (CyTOF and others), as well as mass spectroscopy studies of proteins, lipids, and metabolic profiles. Most ‘omic data have been transcriptomic, with newer approaches including single cell sequencing, CITE-sequencing, ribo-sequencing, which identifies mRNA sites undergoing active ribosomal translation and ATAC-seq which identifies open chromosomal regions linked with gene transcriptomics (See TABLE 3 for a short list of current and evolving approaches). In all cases, vast amounts of data are generated on a single subject’s samples, or in some cases thousands of sequences on a single cell from a single individual (single cell RNA sequencing). As of yet, there is no gold standard approach. In all cases, validation is necessary, ideally at the protein, pathway level, with model systems and with additional support through alternate computational approaches.

Table 3.

Emerging ‘omics platforms

Platform/Array	Description	Reference Nos.
Single cell RNA sequencing (scRNA-seq)	Manual or FACS isolation of single cells which are then lysed, undergo poly-A selection, cDNA preparation, and amplification and sequencing similar to bulk NGS (see above). Allows single cell resolution, but can be hampered by quality control limiting ability to compare cells.	Stegle et al. (251)
Imaging flow cytometry	Combined antibody fluorescence with bright- and dark-field microscopy. This allows protein quantification and subcellular visualization. Limited number of channels can limit the ability to investigate large numbers of markers.	McGrath et al. (181a)
Mass cytometry (CyTOF)/HD flow cytometry	High-dimensional flow cytometry techniques to assess large panels of antibodies. CyTOF utilizes metal isotope labels and mass spectroscopy, while HD cytometry uses photomultiplier tubes across multiple channels to allow a large number of fluorophores.	Han et al. (113), Mair and Prlic (178)
High-dimensional scRNA-seq	Cell populations can be analyzed and sorted by CyTOF or HD cytometry. Isolated cells are then lysed and undergo single cell RNA-seq. Expense and similar issues associated with scRNA-seq can limit ability to apply in large cohorts.	Stegle et al. (251)
Assay of transposase-accessible chromatin sequencing (ATAC-seq)	Tn5 transposase cleaves DNA fragments from accessible regions of the cell’s DNA. These sequences are then read and mapped to the chromosome. Areas with frequent reads show up as peaks, and these peaks are used to determine accessible regions of chromatin.	Chang et al. (44)
Chromatin immunoprecipitation sequencing (ChIP-seq)	DNA binding proteins are crosslinked in vivo using formaldehyde. Chromatin is then digested and antibodies against the DNA binding proteins are used to precipitate out these proteins with their associated chromatin. The recovered chromatin is then sequenced and mapped.	Davies et al. (57)
Cellular indexing of transcriptomes and epitopes by sequencing (CITE-seq)	Antibodies with coupled oligonucleotide barcodes are able to bind to cell surface proteins. These cells then undergo scRNA-seq, and the unique barcode allows protein quantification in addition to RNA sequencing in the same readout.	Stoeckius et al. (253)
MethylC-seq	Bisulfite treatment converts nonmethylated cytosine to uracil, providing a sequencing marker for methylation status of cytosine residues. This allows assessment of potential epigenetic changes in tissues and cells	Urich et al. (266)

In almost all ‘omic studies, millions of data points are generated. Broad analysis of lung diseases requires sampling of multiple different organ compartments, including 1) site-directed bronchoscopic approaches (i.e., bronchial brushings, primarily for epithelial cells), 2) BAL fluid and cells (luminal distal airway and alveolar cells), and 3) endobronchial biopsies (large airway structural and inflammatory cells). Sputum (large airway luminal/primarily immune-inflammatory) cells and nasal brushings, in addition to more accessible peripheral compartments, including blood and urine, are also targets for study. Each of these compartments can be collected at the same or at different time points and yield different, sometimes complementary information, from the same or different platforms. Interdigitating multiple compartments, cell types, time elements, as well as the large numbers of data points from each measurement poses enormous statistical challenges. Few studies have satisfactorily combined (or harmonized) these differing data sets, or even more importantly, different data types, such that this remains an ongoing challenge.

A. Analytic Approaches to Phenotyping and Endotyping

While clinical phenotyping often utilizes less than 100 variables, ‘omics studies require approaches capable of integrating thousands if not millions of variables. Not surprisingly, the use of computational pattern detection is integral to making sense of these big data (FIGURE 1). Machine learning, the ability of software to learn from data without human direction, allows patient classification based on underlying similarities in the data set variables in a manner too complex for traditional assessments.

1. Dimension reduction

Clustering approaches using supervised or unsupervised techniques were developed to group large numbers of similar variables. Several steps are required before clustering analysis. Dimension reduction utilizes approaches to decrease variable numbers thereby facilitating analysis. In many cases, input variables are selected based on subjective or objective measures. Less biased approaches include principal component analysis (PCA) and t-stochastic neighbor embedding (tSNE) (171). PCA identifies small numbers of uncorrelated variables from large data sets, is computationally fast, and generally maintains large differences between individuals. However, it performs less well with nonlinear data. tSNE (often used with ‘omics studies of mixed cell types) maintains similarities between data points but is less informative at describing magnitudes of differences (268a). Regardless of method, dimension reduction involves optimizing distance measures among data points but is confounded by increases in variance at the extreme end of observations in big data sets. Algorithms such as rLog transformation of RNA sequencing data can stabilize variance to facilitate meaningful analysis (175).

2. Clustering

Following dimension reduction, popular clustering methods include hierarchical, partitioning (K-means)- and model-based or probabilistic approaches. Both hierarchical and K-means clustering have been used in asthma phenotyping studies (186, 290). Hierarchical clustering can be performed utilizing a bottom-up approach where each data point is iteratively merged with its neighbor until all have been joined to a single cluster (58), or as a top-down approach where objects are recursively divided (105). Hierarchical clustering is dependent on input distance measure and linkage criteria used to merge data points. Distance measures include Euclidean or squared Euclidean, which represent “ordinary” straight line distances in coordinate space (or its squared value). Less often, Manhattan, representing absolute distance between vectors, or Maximum distance between vectors are used.

In single-linkage hierarchical clustering, cluster relationships are determined by the similarity of their most alike members (195). The merge criterion is local, meaning only the area where two clusters come closest is taken into account, while the overall structure is not. This can result in a straggling chain of points representing local relationships without providing information on higher level structure. In contrast, complete-linkage clustering utilizes similarity between the most dissimilar members of a cluster and takes the entire clustering structure into account (37). Perhaps the most popular, Ward’s method, which considers the distance between two clusters as how the sum of square errors increases when they are merged, was developed to minimize error within clusters (272).

K-means clustering algorithms belong to a class of approaches known as partitioning methods (117). K-means is considered the fastest and simplest method for clustering data starting with a user defined number of data points that defines the initial clusters to which all other points are assigned (138). The centroid, or resultant mean from the newly generated cluster, is then updated and cluster assignment processing is repeated until a stopping-criterion such as distance minimization or maximum iterations is reached. K-means clustering is sensitive to outliers and tends to perform poorly on data where attribute variance is unequally distributed. To address these issues, the partitioning around medoids (PAM) and CLustering LARge Applications (CLARA) algorithms utilize the most centrally located point in a cluster, or medoid, for dissimilarity calculation, rather than the mean value of all the data points in the cluster (117, 268).

Model-based clustering, also known as latent class analysis or mixture modeling, assumes that each cluster present in the data set corresponds to a different model (182). Although this allows for nonuniform cluster shapes as well as fuzzy cluster assignments, the process is relatively complex and requires user input based on Bayesian information criteria.

A common feature of these methods is that they require user specified cluster assumptions to define groups. Although hierarchical clustering provides a dendrogram for visual inspection of relationships, the level to cut the tree to define groupings/clusters is subjective. Methods for estimating the number of clusters in a data set include optimization of criterion via the elbow or silhouette approaches, as well as calculation of the gap statistic (263). No approach is perfect, and each requires assumptions regarding what the right solution would represent. One way to estimate veracity is through cross-validation, whereby the rules for assignment from a training set are used to classify a testing set, giving a sense of how well a clustering solution can be generalized. However, in asthma studies, this is often problematic, as parallel data sets may not be available.

3. Pathway analysis

The importance of these clustering approaches to biologic or clinical reality is often unclear. To this end, identifying patterns to provide meaningful biologic context is not only important but also controversial. As examples, differential gene expression analysis from RNA sequencing/microarray data can be queried for meaning using databases such as protein analysis through evolutionary relationships (PANTHER); Database for Annotation, Visualization and Integrated Discovery (DAVID); or Kyoto Encyclopedia of Genes and Genomes (KEGG) for enrichment in functional terms (132, 145a, 196, 293). Patterns of upregulated genes may also be used to infer transcription factor activity by querying data sets such as ENSEMBLE, TRANSFAC, or MotifMap (181, 292, 298a). Patterns have also been identified from previously annotated and targeted mouse or human gene expression studies by gene set variation analyses (GSVA) (121, 122, 231). Gene Set Enrichment Analysis (GSEA) compares gene expression using sets of transcripts more broadly associated with biologic processes or diseases (217, 254). Weighted gene coexpression analysis (WGCNA),unlike GSEA, does not rely on curated or annotated lists, but rather constructs correlated gene modules that are then compared with a clinical trait or annotated through one of the platforms described above (161, 185, 217). WGCNA also facilitates extraction of key “hub” genes that connect networks or summarize the behavior of modules (eigengenes). These approaches all promote development of molecular hypotheses which must be tested in vitro and in vivo before results can be fully validated, a bar which most studies fail to reach.

Although tools available for evaluating complex analyses grow every day, there are still limitations. New and emerging tools improve integration of multiple data sets (i.e., data-harmonization) across cellular compartments or differing time points, including multi-view clustering recently applied to identify corticosteroid response patterns in asthma (289). Expanded analytic approaches are still needed.

B. Molecular Phenotypes: Linking Molecular Pathways to Clinical Traits

Breaking a disease into its components allows for a more simplified analysis linking broad ‘omic profiles with disease or specific characteristics. Three general approaches have been used. The most common uses the trait as the dependent (fixed) characteristic, with transcriptomic (or other ‘omic) data, either individual genes/proteins or gene signatures, evaluated in relation to that characteristic only (16, 17, 121, 122, 156). Others have identified all genes related to a specific trait, then clustered those genes to identify expression patterns in relation to individual patients (156, 186). Finally, unbiased approaches have clustered genes and then related the gene clusters to specific defined characteristics/traits (185, 217).

Multiple transcriptomic studies have evaluated differentially expressed genes between two or more categorical variables (traits), from persistent airflow limitation to inflammatory phenotypes, to clinically defined clusters. The simplest approaches identified differentially expressed genes between categorical variables, such as asthma versus health (287). Others have taken a second step and evaluated the differentially expressed genes for the presence of predefined, typically human in vitro or mouse-model related, gene clusters using GSVA, to give insight into the functional aspects of the differential gene expression patterns (121, 122, 231). Clustering of GSVA gene sets has also been used to identify clinical phenotypes (156). Limitations to these approaches begin with using an arbitrary/fixed definition of the comparator groups or traits, often based on categorical cut-points with limited biologic validation. This biased approach simplifies continuous characteristics into single prespecified categorical variables, which for statistical analysis are now the dependent variable. In some cases, statistically identified clinical clusters have been used as the anchors for the molecular analysis (166). Depending on the accuracy of these clinical clusters, their predictive characteristics may have less value. Additionally, GSVA gene sets can give limited or even misleading information on pathway analyses, especially when mouse comparator gene sets are included. It is likely that deficiencies in both the independent and dependent variables contribute to the modest associations between GSVA and clinical characteristics.

Despite these limitations, these studies have consistently produced relationships of eosinophilic or type 2/IL-4-IL-13 gene sets with traits such as persistent airflow limitation and eosinophilic asthma phenotypes (FIGURE 2) (121, 156). Additional differences have been observed in inflammasome-related pathways, although its presence is complex, with certain elements (IL1RL1, the IL-33 receptor) upregulated in association with eosinophilic inflammation and others (IL-1-related genes) more likely associated with neutrophilic processes (FIGURE 3) (149, 159, 231). Inconsistencies in associations of these pathways are also apparent when comparing multiple compartments and cell types. In cases where mixed cell types are included (sputum or biopsy tissues in particular), controlling for background differences in cell types, each with their own specific gene signatures, is often not done, leading to difficulty in interpretation. However, this approach confirmed an overall association of T2 immune pathways with asthma characteristics or eosinophilic phenotypes and suggested complex relationships to inflammasome-related pathways.

FIGURE 2.

Complex interplay between type 1 and type 2 immune pathways contributes to differing asthma phenotypes. Repeated exposure to allergens in genetically susceptible individuals induces the development of Th2 cells, which produce interleukin (IL)-4, IL-5, and IL-13. These type 2 cytokines can be also produced by alternate mechanisms involving cells of the innate immune system. In this regard, pathogens including bacteria, viruses, and fungi as well as allergens can cause epithelial damage in which proteases encoded by these agents play a prominent role. Epithelial damage leads to increased expression and release of IL-33 and thymic stromal lymphopoietin (TSLP), which stimulate ILC2 cells to produce IL-5 and IL-13. Tissue resident mast cells and basophils recruited from the periphery or generated from in situ differentiation of progenitors can also generate these cytokines. IL-5 released from circulating Th2 cells and other growth factors such as IL-3 and granulocyte-macrophage colony stimulating factor (GM-CSF) stimulate eosinophil differentiation, proliferation in the bone marrow, which are then recruited to the airways under the influence eosinophilic chemokines. Mast cells also produce prostaglandin D₂ (PGD₂) that binds its cognate receptor CRTH2 on Th2, ILC2 cells, triggering cytokine release. IL-4 and IL-13 produced by these different cell types augment mucus production via expression of mucin genes including MUC5AC. In certain individuals, gene-environment interactions also promote a type 1/interferon (IFN)-γ response with or without IL-17 production from T cells (not shown). IFN-γ in turn stimulates production of the chemokines CXCL9 and CXCL10 from both airway epithelial cells and resident macrophages with the potential to create a positive feedback loop promoting recruitment of Th1 cells and eosinophils via interaction with the receptor CXCR3. Increased expression of the cytokine IL-18 in macrophages can also promote development of Th1 and Th2 cells. Synergism between T1 (IFN-γ) and T2 (IL-13) cytokines augments expression of the enzymes inducible nitric oxide synthase (iNOS) and dual oxidase 2 (DUOX2) in the airway epithelial cells driving production of nitric oxide (NO) and H₂O₂, respectively, with increases in nitrative and oxidative stress. APC, antigen presenting cell; FeNO, fraction exhaled nitric oxide.

FIGURE 3.

“Inflammasome” linked interleukin (IL)-1β, IL-18, and IL-33 pathways in relation to downstream pathway activation and cellular responses. Both genetic and transcriptomic studies have strongly supported involvement of these cytokines in asthma phenotypes. These pathways are all activated by danger signals/receptors, with responses likely dependent on initiating signal, cell type, and perhaps genetic factors. IL-1β activation, through a classical inflammasome process, leads to broad activation of many cell types, generally linked to neutrophilic inflammation, perhaps through interferon (IFN)-γ or IL-17 pathways. Genetically associated IL-33 activation, arising from epithelial damage and augmented by inflammatory cell protease activity, stimulates ILC2 cells and mast cells/basophils to generate type 2 cytokines increasing type 2 immune responses. IL-18, strongly linked genetically with asthma through IL-18R1, can augment either type 1 or type 2 immune responses depending on the presence of IL-12. Interestingly, all 3 receptor pathways signal through MyD88, NF-κB, and Jun kinase (JNK) pathways. IL18RAP, IL-18 receptor activating protein; IL-18BP, IL-18 binding protein; sST2, soluble ST2/IL1RL1.

1. Identifying new patient clusters based on trait-based molecular profiles (limited molecular phenotypes)

The second approach utilizes gene sets to identify clinically distinguishable clusters exclusively derived from molecular characteristics. Thus “biased” molecular pathways drive clinical outputs. This approach was applied to 154 participants in the SARP with bronchoscopic microarray data from fresh human airway epithelial cells (HAEC) (186). Over 500 genes were identified in the brushings, which correlated strongly with the asthma biomarker FeNO. A K-means clustering approach identified five clinical clusters, three with high FeNO and two with low levels. The three high FeNO clusters included a very severe T2-Hi gene expression group, associated with low epithelial growth and repair expression, an older-onset group, with more complex inflammatory profiles, as well as a younger, milder, more traditionally allergic cluster. The two low FeNO clusters, which contained nearly all the HCs, varied by atopic status, asthma severity, and disease duration. A similar approach was used on ~500 genes which differentiated patients with ≥1.5% eosinophils compared with those with less (156). Hierarchical gene clustering identified three different “transcriptome associated clusters” (TACs). Sputum eosinophils strongly differentiated the TACs, identifying one high, one low, and one sputum eosinophil intermediate. Using GSVA, potential pathways were identified, including T2/IL-13, ILC2, neutrophilic inflammation, inflammasome, and Th17. Pathways linked to ILC1 and aging were also identified. Unfortunately, clinically or statistically significant differences across the three clusters were small. Thus the ability of this approach to identify clinically meaningful clusters may vary by study, depending on biomarker/associated genes and compartment chosen.

2. Molecularly defined phenotypes (FIGURE 1)

The most unbiased approach is to start with expressed genes and allow them to identify molecularly defined clinical phenotypes. The earliest (and biased) example was the original identification of what was then defined as “Th2-High” asthma (now T2-Hi), based around a highly selected signature of three genes upregulated by the addition of IL-13 in cultured HAECs (287, 288). Evaluating a group of mild, CS naive patients and HCs, ~50% of asthmatic patients had a broad increase in expression of these genes in fresh HAECs, establishing a standard for a HAEC T2 gene signature. These T2-Hi patients were more allergic and had higher eosinophils and more bronchial hyperreactivity than patients with low expression. Importantly, this was the first time response to an anti-inflammatory therapy was evaluated by molecular cluster, with Th2/T2-Hi patients having robust improvement in forced expiratory volume in the first second (FEV1) to inhaled CSs, while those with low expression did not. Confirming the CS responsiveness of this gene signature in mild asthma, follow-up bronchoscopies showed a reduction in the T2 gene signature expression in response to the inhaled CS.

This initial identification of the two T2 molecular phenotypes was based on a very small number of biologically identified T2 signature genes. A less biased approach was undertaken on microarray data from sputum samples utilizing KEGG pathways to identify samples, which differed in relation to the presence (numbers and percentages) of genes from these identified pathways (293). Three transcriptomic endotypes of asthma (TEA) clusters were identified which integrated the expression of the enriched KEGG pathways across the samples. Several KEGG pathways distinguished the clusters, including pathways linked to epithelial and neuronal function. These three TEA clusters differed by overall severity and biomarkers of T2 inflammation but, surprisingly, were not enriched for T2-related genes, perhaps because of low T2 genes in sputum cells. There was also little overlap of the differentially expressed genes from this study compared with results from other approaches.

Identifying molecularly defined clinical phenotypes has also been limited by the number of genes evaluated, primarily due to lack of approaches to analyze many more than a thousand variables. Microarrays and more recently RNA-sequencing now provide access to differential expression of thousands of genes. WGCNA first clusters genes in relation to each other and then weights these clustered genes in relation to specific categorical and continuous trait variables (see sect. IVA). With the use of this approach on HAEC microarray data, over 50 different “related gene” modules were identified, with plausible gene relationships by pathway analysis (185). Genes in these related modules identified cohesive functional, immune, or structural cell pathways, including those related to T2 signature genes and epithelial dysfunction (FIGURE 2). This less biased approach also identified new relationships between epithelial growth and repair, neural processes, and mitochondrial gene modules with severity traits, which could also contribute to the pathogenesis of severe asthma.

This approach was recently applied to BAL cell microarray data from 154 SARP subjects, 104 of whom overlapped with previous epithelial brushing data (185, 273). Due to the complex nature of BAL cells, the data were first “deconvoluted” by cell differentials and then by race, age, sex, and body mass index (BMI), all of which influenced gene relationships. All expressed genes were then related to various clinical outcomes, while controlling for cell type, demographics, and medication use. BAL cell genes associated with the cAMP pathway inversely linked to severity outcomes, including amount of β2-agonist use. WGCNA then identified 49 gene networks, only one of which, regulation of cAMP-dependent protein kinase activity, related to disease severity parameters, and in particular to β2-agonist use. Additional genes in this pathway include several growth and repair factors, as well as dendritic cell markers, suggesting β2 agonist treatment, especially in excess, could also contribute to molecular phenotypes, and confirming the potential for large effects of treatments on molecular patterns. Using the hub genes identified by the WGCNA approach to prioritize hypothesis-driven investigations, the authors expanded and validated the β2-agonist hypothesis using in vitro studies. Unlike β2 agonist use, CS-related gene expression patterns did not relate to disease severity.

V. MOLECULAR PHENOTYPES: TYPE 2 AND BEYOND

A. Identification of T2-Hi Molecular Phenotypes: Biologic and Clinical Validity

Type 2 asthma has clearly reached the status of a broad molecular phenotype. The asthmatic inflammatory response was first associated with the Th2 arm of the adaptive immune system in the early 1990s (230), a few years after mouse immune system studies helped establish the Th1/Th2 T-lymphocyte-focused paradigm (192) (FIGURE 4). Subsequent research led to identification of GATA-3 as the master regulator of Th2 cell development (299, 302). GATA-3 was also associated with allergic asthma in an experimental mouse model (300), and increased expression of GATA-3 was described in the airways of asthmatics (201). Connections of Th2 immunity to human asthma were driven by associations with IgE/allergic type 1 hypersensitivity reaction, eosinophils, and CD4 lymphocytes (192, 230, 270). Human studies reported IL-4, -5, and -13 expression in both peripheral blood cells and airways (54, 135, 153). Transgenic animal models of allergic asthma followed in the 1990s, which confirmed the importance of the Th2 cytokine IL-13 to allergic responses in mice (103, 285). Unfortunately, although it appeared obvious that Th2 immunity was critical for asthma, human trials of targeted therapies failed. At the same time, human studies of Th2 cytokine expression and eosinophilic inflammation revealed wide variability across asthma (153, 282, 295). It also became clear that the canonical Th2 cytokines (IL-4, IL-5, and IL-13) were produced by many cells beyond Th2 cells, including ILC2, mast cells/basophils, and even eosinophils (reviewed in Zhu, Ref. 304) (FIGURE 2). This led to a broad acceptance that Th2 immunity, especially in relation to asthma phenotypes, is more properly referred to as type 2 or T2 immunity, encompassing the broader cellular sources and potential immune mechanisms. While these variations were originally believed to be “noise” associated with human studies, nearly all ‘omic studies to date have identified subsets of asthma patients who manifest consistent evidence for T2-related immune processes, without specific relation to Th2 or other immune cells. Simultaneously, new tools led to increasing appreciation of the reach of the immune system beyond Th2 cells to the importance of other arms of the adaptive immune system, including Th1 and Th17 (or less well appreciated T1 and T17) (FIGURE 2). Thus, while T2 immunity was apparent in human asthma, complexities in its levels, cell sources, and accompanying immune pathways likely contribute to differing clinical manifestations and importance.

FIGURE 4.

Timeline in establishment of the Th1/Th2 paradigm, identification of master regulators of T helper cells, and association with asthma. Naive CD4+ T cells differentiate in response to antigens and costimulation to different subsets, each characterized by its expression of specific cytokines, a paradigm established in the mid 1980s. Increased numbers of Th2 cells were identified in the airways of asthmatics. A surge of interest in molecules that are essential for lineage development led to the association of GATA-3 with interleukin (IL)-5 gene expression and as a master regulator of Th2 cells and subsequently of ILC2 cells as well. Likewise, T-bet and RORγt are essential for the development of Th1 and Th17 cells, respectively. Increase in GATA-3 expression was observed in human asthma, and expression of a dominant-negative (DN) mutant of GATA-3 in mice blocked allergic eosinophilic airway inflammation in a mouse model. IFN, interferon; BAL, bronchoalveolar lavage; ConA, concanavalin A; TCR, T-cell receptor; HSC, hematopoietic stem cell; CLP, common lymphoid progenitor.

1. Consistency and complexity of T2-Hi molecular phenotypes (FIGURES 2 and 5)

FIGURE 5.

Schematic of the current understanding of T2-Hi molecular and clinical phenotypes. Type 2 inflammation is characterized by enhanced activity of interleukin (IL)-4, -5, and -13 which commonly leads to eosinophilic inflammation, mast cells, mucus, and elevated fraction exhaled nitric oxide (FeNO), all of which are likely to overlap to varying degrees. However, depending on the mix of cytokines and these accompanying immune processes, at least 4 distinct type 2 molecular phenotypes are recognizable, with varying response to therapy, ranging from simple mild allergic early-onset asthma to very severe, T2-Hi+ asthma. IFN, interferon; CS, corticosteroid.

A T2-molecular phenotype is identifiable in a large percentage of the overall asthma population (280). This phenotype is associated with evidence for the presence or downstream activation of IL-4, -5, and/or -13, in varying lung compartments (tissue, fresh HAECs, sputum) (219), and with tissue, blood and sputum eosinophils, sputum T2 gene expression, FeNO in exhaled breath and blood biomarkers, like periostin (141, 186, 219). Population-based identification of T2-Hi patients is limited by the availability, sensitivity, and specificity of these downstream biomarkers. Unfortunately, measuring the cytokines themselves in any compartment is limited by their very low and highly variable levels. Thus current biomarkers are rather nonspecific for T2 immune processes, with other factors, including treatments, contributing to their elevation and variability over time. Severe asthma patients with elevations in these T2 biomarkers also generally respond to T2-targeted biologics, confirming their overall relationships (41, 42, 52, 110, 199, 277). The best therapeutic responses to the T2/IL4Rα targeted drug dupilumab occur in patients with dual elevations in both FeNO and blood eosinophils (40). However, significant clinically meaningful responses to T2-targeted therapies also occur in patients with smaller elevations in FeNO or blood eosinophils, suggesting that widely used cut-points for T2-Hi and -Lo may be inaccurate. In fact, although most T2-targeted biologics used 300 eosinophils/μl as the initial threshold, responses are apparent at cut-points as low as 150 cells/μl (40). Thus, until better biomarkers are identified, assignment to a T2 molecular phenotype could also include a positive response to a specific T2 targeted therapy.

Consistent pathobiologic processes and clinical features are generally observed across this T2-Hi phenotype, such that, like asthma, it should also be thought of as an umbrella definition (FIGURE 5). As expected, T2 molecular phenotypes are consistently associated with eosinophilic inflammation in the lung/blood, elevations in FeNO, elevated IgE/atopy, and bronchodilator responsiveness (TABLE 4). Active T2 immune processes generate profound changes to the airway epithelium, driving goblet cell differentiation, ciliary dysfunction, and mucus abnormalities (163, 240, 303), linked to increased mucus plugging in T2-Hi asthma (66). Placebo arms from T2 targeted studies, as well as large-scale epidemiologic studies, show an increase in exacerbation rates in biomarker defined T2-Hi phenotypes compared with T2-Lo (22, 40, 61, 70, 224, 276). Clustering of inflammatory and physiological variables (together) report bronchodilator reversibility more strongly correlates with eosinophilic inflammation than FEV1 or forced vital capacity (FVC), measures of static lung function (186). As original definitions of asthma were based on clinical observations, followed by physiological definitions, it is conceivable the original perceptions/definitions of asthma were built around patients with T2-immune responses, and that some of the heterogeneity observed today relates to loosening of these original definitions, particularly in large-scale studies (see sect. I). Finally, most studies support stability of T2 inflammation over time (157, 176, 197, 271). The few long-term studies also suggest an association of persistent T2 inflammation with loss of lung function and persistent exacerbations (150, 197, 271).

Table 4.

Biologic and clinical characteristics consistently observed in T2-Hi molecular phenotypes

1. Elevated blood and/or sputum eosinophils

2. Elevated FeNO

3. Atopy/elevated IgE

4. Bronchodilator reversibility (may be less in later-onset T2-Hi phenotype)

5. Higher exacerbation rates

6. Increased mucus on CT imaging (silent mucus)

7. Greater response to corticosteroids

CT, computed tomography; FeNO, fraction exhaled nitric oxide.

Despite many consistencies, patients and pathological processes under the broad T2-Hi molecular umbrella also vary dramatically one from another (FIGURE 5). The reasons for this are not clear, but likely include an oversimplification of T2 immune pathways (FIGURE 2). As is increasingly clear, the canonical T2 cytokines can be expressed by different cell types (Th2 cells, ILC2 cells, mast cells/basophils), in different proportions, in tissue compartments, and under varying regulatory control. In addition, accompanying innate [thymic stromal lymphopoietin (TSLP), IL-33] and adaptive [interferon-γ (IFN-γ), IL-17] immune pathway activity modifies these cytokines and their downstream pathways. Thus these variations may explain the spectrum of clinical subphenotypes associated with T2 immune activity.

2. Mild early-onset/allergic asthma (FIGURE 5)

Perhaps the earliest identified T2-Hi molecular phenotype is mild allergic asthma. Allergic responses have long been linked to Th2/T2 immunity in mouse models and human disease (153, 229), with the majority of transgenic mouse studies relying on Th2-skewed allergic models varyingly dependent on IgE. IgE, the allergic immunoglobulin, is produced by IL-4/13-induced isotype switching of B cells. It binds to high-affinity IgE receptors (FcεR1) on mast cells and basophils where crosslinking by allergen leads to activation (FIGURE 2). Human allergic/allergen challenge systems have consistently been inhibited by monoclonal antibody blockade of IgE (27, 68), T2 associated cytokines/receptors, including IL-13 (alone), the IL4Rα receptor (IL-4 and -13), and more recently by TSLP (90, 92, 279), confirming the importance of these pathways to both the immunologic and human physiological process. Additionally, inhalation of a DNAzyme against the canonical T2 transcription factor GATA-3, which controls expression of IL-4, -5, and -13, was also effective in a human allergen challenge setting (154). This allergic T2-Hi phenotype typically begins in childhood (before age 12), and the majority of patients are responsive to CS therapy (75, 107, 288). It is also associated with other T2-related diseases, including allergic rhinitis and dermatitis/eczema, as noted years ago in relation to extrinsic asthma (184, 226).

While eosinophilic inflammation is present in these patients, its relation to pathophysiology is less clear, as unlike the IL-4/-13 and IgE approaches, monoclonal antibodies towards IL-5 do not decrease allergen challenge responses, despite eliminating blood eosinophils (164). In contrast, the efficacy of anti-IgE in this subphenotype supports a role for mast cells and/or basophils (FIGURE 2). Mast cells are found in large numbers in proximal airway epithelium and submucosa where they can be activated by inhaled allergen, even in patients with atopic disease and no asthma (20, 91, 281). Mast cells are often considered the front line for innate immune defense/epithelial repair, such that movement of mast cells into the epithelium could be the initial response to the epithelial damage signal observed in many transcriptomic studies (186, 265, 275), even initiating mild allergic asthma. Exercise induced bronchospasm (EIB) is also associated with increases in and activation of mast cells in the epithelium, perhaps in response to epithelial damage induced by activation of TSLP and/or IL-33, or even by the overall vulnerability of the epithelium to hyperventilatory damage (98, 114, 160). It should be noted that the strongest and most consistent genetic associations with asthma are in the 17q12–21 asthma susceptibility locus (208). The region is specifically linked with childhood-onset, allergic asthma, increasing the rationale for identifying these early-onset patients as a separate T2-subphenotype, and questioning the overall importance of Th2 immunity.

3. Moderate to severe early-onset allergic asthma (FIGURE 5)

Patients with more severe allergic/early-onset asthma are found in most cluster analyses, often in association with higher levels of allergic sensitization (186, 203). These patients respond to both anti-IgE and anti-IL-4Rα therapies (40, 116, 276), although responses to anti-IgE appear less robust as disease severity increases, suggesting this allergic phenotype involves more than IgE (34, 115, 134). Interestingly, recent data suggest anti-IL-5/5R approaches may be less effective in early-onset allergic asthma, even when matched for baseline blood eosinophils, and consistent with the lack of efficacy of anti-IL-5 in human allergen challenges (23). In fact, treatment of moderate, likely predominantly allergic, asthmatic patients with anti-IL5R only modestly reduced exacerbations and led to only small improvements in lung function (69). Little is understood regarding the mechanisms which drive development of this more refractory T2-Hi allergic subphenotype, or its longitudinal trajectory, with studies suggesting this more severe disease may arise early in childhood, as opposed to progressing over time (237). The subphenotype may also transform rather acutely from milder to more severe disease in adulthood. Although ‘omic studies show consistent increases in epithelial T2 signatures, additional differences in epithelial differentiation, repair, mast cell signatures, and innate pathways may also determine more severe disease (140, 186, 187, 273).

4. Adult-onset T2-Hi asthma (FIGURE 5)

Another well-defined T2-Hi subphenotype is that of adult-onset, eosinophilic/T2-Hi asthma, typically associated with sinus disease, nasal polyposis, sometimes with AERD, and much less with allergies. Despite similar elevations in T2-biomarkers, numerous factors differentiate adult from childhood-onset asthma (TABLE 5). This phenotype has origins as intrinsic asthma and is identified in most clusters that include inflammatory variables (111, 186, 203). Unlike T2-Hi early-onset asthma, this subphenotype is less CS responsive, often requiring treatment with systemic CSs. Acute severe exacerbations requiring invasive or noninvasive ventilation (except in response to nonsteroidal anti-inflammatory drugs in susceptible patients) are less common. However, the T2 component of this phenotype is confirmed by robust responses to IL4Rα blockade, both clinically and biologically in nasal polyp tissue, where marked reductions in T2 chemokines were observed (15, 142). Importantly, IL-5 is a critical cytokine for this phenotype, as two reports now support greater efficacy of anti-IL-5 pathway inhibition in patients with this adult-onset T2-Hi eosinophilic phenotype as compared with early-onset disease (23, 29). Data with anti-IL-5R revealed the presence of nasal polyposis and systemic CS dependency as additional response predictors (22). Thus it is conceivable that an IL-5 dominant, adult-onset, nasal polyposis eosinophilic asthma endotype has emerged, with strong historical links to phenotypes identified by astute clinicians and more recently in cluster analyses. The high degree of responsiveness to anti-IL-5/5R (as relates to asthma), and blockade of IL-13 through anti IL-4Rα therapy, support the proposed mechanistic link to ILC2 cells, but identification of these cells in asthma remains controversial (31). While mast cells and/or basophils are also thought to play a role in these patients, particularly in relation to AERD, specific serum IgE levels are often low, weakening the link with traditional allergy (9, 186). Leukotrienes also play a role. They are abundantly generated in these patients, and 5-lipoxygenase inhibition improves outcomes, especially in those with aspirin sensitivity (136).

Table 5.

Differences in early childhood- versus adult-onset T2-Hi asthma

Characteristic	Early Childhood Onset	Adult Onset
Age at onset	Typically <12 yr old	Begins to increase late 20s and beyond
Sex	In adults, women > men	Proportionally more men than early onset
Allergies (rhinitis, eczema)	Prominent	Often absent
Sinusitis	Often absent	Prominent
Eosinophilia	Modest	Prominent
Nasal polyps	Typically absent	Prominent
Genetic elements	ORMDL3, GSDMA/B, IL33/ST2, TSLP, IL18R/RAP	Less clear, possibly ALOX15
Corticosteroid response	Moderate to good at lower doses	Often requires high/systemic doses
Biologic response	No known relation to anti-IgE response; lesser response to anti-IL-5/5R approaches	May be better response to anti-IL-5/5R approaches

The epithelial and eosinophilic/T2-linked enzyme 15 lipoxygenase-1 (15LO1) also appears important to this phenotype. A recent single cell RNA-seq study supports a primary association with upper airway/nasal polyp epithelial cells. These mRNA data were confirmed at the protein level where 15LO1 was essential for the expression of the eosinophilic chemokine eotaxin-3/CCL26 (172, 209). In further support, a meta-analysis of genome-wide studies identified a coding single nucleotide polymorphism (SNP) in 15LO1, associated with a loss-of-function mutation, as highly protective against polyp development (152). Finally, local autoantibodies have been identified in some patients, but the relation to systemic immunity is less clear (193, 258).

5. Complex T2-Hi asthma (T2+) (FIGURE 5)

T2 immunity can be persistent despite increasingly severe CS-treated disease. In these cases, the T2 immunity may be impacted by additional immune pathways, different cell sources, and more distal lung inflammation. Distal lung studies of T2-Hi asthma are few. However, a persistent small airway signature of T2 immunity has been reported in distal airway epithelial cells, beyond the reach of most large particle inhaled CSs and only accessible to systemic CSs (243). Distal lung pathology also revealed increased small airway tissue inflammation, including the presence of mast cells in the inner and outer airway walls, in patients with more severe disease (19).

In patients with very severe systemic CS-dependent asthma, video-assisted thoracoscopic biopsy tissues have revealed noncaseating granulomas in the distal lung in a subset, without relation to eosinophilic granulomatosis with polyangitis. Thus, in addition to persistent T2 and mast cell presence, the granulomas suggest immunity beyond T2 alone (284). While asthma is traditionally associated with allergic as opposed to autoimmune inflammation, many of these very severe patients have a personal or family history of an autoimmune disease, supporting potential involvement of autoimmunity. BAL fluid/cell IFN-γ and downstream chemokines CXCL9 and -10, as well as the macrophage/dendritic cell transcription factor IRF5, in association with type 1 inflammation, have all been reported to be increased in severe asthma (89, 210, 227, 269). However, upstream processes that augment IRF5 or other factors in potentiating type 1 inflammation in severe asthma are unknown.

IL-6 pathways have also been reported to be activated in association with increased blood eosinophilia and airway T cell/macrophage infiltration in patients with more severe and exacerbation-prone disease, along with increases in innate immune pathways (IL-1β, Toll-like receptors) (140). Importantly, in contrast to studies linking obesity, asthma, and elevated serum IL-6 patients, this complex T2-Hi-IL-6 phenotype had no evidence for increases in systemic IL-6.

Some complex T2-Hi+ patients also have evidence for very high/‟ultra-high” levels of T2 gene expression in sputum, associated with older age, worse lung function, and more severe disease (217). This “ultra-Hi” T2-Hi phenotype was seen in association with increases in sputum CD11b+/CD103-/IRF4+ dendritic cells. If and how they are antigenically driving the “ultra-Hi” T2 phenotype and associated disease severity remains unclear.

While three of the current T2-directed biologics have shown efficacy in systemic CS-dependent (and refractory) asthma, supporting an important role for T2 immunity, some of these most severe patients remain refractory to these therapies, despite elevations in traditional T2 biomarkers (21, 200, 225). Inducible nitric oxide synthase (iNOS), the primary enzyme contributing to nitric oxide (NO) generation in human airways, is increased by IFN-γ, as well as IL-4/13 (48, 108), suggesting that elevated FeNO may not be due to T2 inflammation alone (269). Many additional factors cause blood eosinophilia, suggesting other non-T2 drivers may elevate these biomarkers, supporting the need for more specific T2 biomarkers, especially given the high cost of a trial of a T2 biologic.

6. Does the microbiome play a role in T2-Hi asthma phenotypes?

There is increasing evidence for a gut-lung or lung-microbiome axis in relation to disease and health. Current studies can generally be divided into those addressing disease development and those addressing persistence or severity. While the relationship of gut microbiota studies to T2-Hi (or Lo) asthma phenotypes is limited, many have addressed its role in development of atopy/allergy (FIGURE 6). It has been observed for years that birth by Caesarian section increases the child’s risk for atopy and asthma (reviewed in Sandall et al., Ref. 235), which led to enhanced Staphylococcus-associated microbiota in the infant. The presence of fecal Ruminococcus gnavus has also been associated with occurrence of allergy in children at 3 yr of age (51). Mouse models suggested this may occur through migration of gut ILC2 cells to the lung, but human studies are lacking. Having a dog in the home has also been associated with less risk of atopy/asthma. Mice exposed to dog-associated house dust developed increases in gut Lactobacillus johnsonii, which was protective against T2-lung responses (87). The presence of many other bacteria have been variably related to development of asthma and allergies (reviewed in Budden et al., Ref. 32), thus emerging data to support an association between the gut microbiome and human T2 (allergic) responses, and thereby asthma as well. However, cause and effect in humans remains unproven. In addition to gut microbiota, several studies have identified differences in the lung microbiota in relation to asthma phenotypes (FIGURE 6). Intriguingly, high diversity of organisms, in general, is observed to be protective in relation to disease. In patients with mild asthma, the presence of T2 inflammation (as measured by lung T2 biomarkers) was associated with nearly unmeasurable 16S rRNA for bacterial species in most cases, while modest differences existed between atopic asthmatic patients and those with atopy alone (67). Importantly, CS responsiveness appeared related to the presence of certain bacteria 16S rRNA, including Hemophilus, which was associated with low T2 (high neutrophilic) inflammation in a second study (261). Similar potentially pathogenic species (Streptococcus) have been identified in sputum from severe asthma, in association with sputum eosinophilia (T2-Hi) (301). Whether these studies truly represent the lung microbiome remains unclear, as sputum is highly contaminated with upper airway bacteria. While most studies have focused on bacterial rRNA, a recent bronchoscopic study also addressed fungal rRNA. Somewhat similar to bacterial rRNA, less fungal diversity was observed in asthma patients with evidence of T2-biomarker elevations, with Trichoderma species found in greater abundance in these patients (239). Yet, relations between fungal species and disease severity were more complex, including a positive relationship between Trichoderma and FEV1, but negative relationships with more traditional fungal allergens, including Cladosporium and Alternaria. How lung measurements of these fungal RNA relate to traditional (specific) allergen sensitization remains to be studied. While many associations have been observed with both gut and lung microbiota in relation to T2 (allergic and eosinophilic) asthma, further work (including ongoing studies of probiotics to alter gut microbiota) is required to determine whether these are causal or reactive relationships.

FIGURE 6.

Simplified schematic of potential microbiome-immune interactions in relation to molecular and inflammatory asthma phenotypes. Emerging data support the importance of gut microbiota/dysbiosis in the development of atopic-allergic disease, with certain bacteria including staphylococcus and ruminococcus associated with atopy, while Lactobacillus johnsonii has been considered protective. These alterations could promote the development of mild-moderate T2-Hi allergic asthma. In all cases, loss of diversity of the microbiome is believed to be a disadvantage and drives development of ILC2 and/or ILC3 cells which, under certain circumstances, can migrate to the airways. Abnormalities in the lung microbiome are more controversial and difficult to confirm. An abnormal airway epithelium could predispose to bacterial contamination, colonization, and infection with pathogenic bacteria being found more commonly in T2-Lo/neutrophilic asthma. Interestingly, in mild asthma, the presence of T2-Hi immune processes is associated with significantly less evidence for bacterial 16S RNA, suggesting the T2 environment/phenotype may inhibit bacterial presence. Similarly fungal RNA has also been observed in lower abundance.

7. Is T2 immunity the primary molecular abnormality?

Interestingly, a T2 transcriptomic signature has only been reported once in a fully unbiased manner (187). In contrast, unbiased clustering of genes reveals several other potentially relevant pathways. In fact, using the same WGCNA which identified the T2 gene signature by unbiased gene clustering, three other gene modules, including epithelial repair and mitochondrial and neuronal processes, were identified which more strongly linked to asthma diagnosis and severity than the T2 gene module. This study is complemented by genetic/genome-wide studies of asthma compared with healthy populations which have consistently identified asthma susceptibility loci in relation to epithelial genes (17q12–21), inflammasome related genes (2q), as well as other immune and tissue repair signals, but without relation to atopy-related genes (148, 188). No prominent and over-arching T2 or Th2 genetic signal has emerged from genetic studies. Importantly, a phase 2b trial of an anti-TSLP antibody in moderate to severe asthma, of mixed T2 background, showed reductions in T2 biomarkers, as well as improved clinical outcomes, even in patients without elevations in background T2 biomarkers (53). If confirmed in phase 3, a secondary, as opposed to primary, mechanism for T2-immunity in asthma would be further supported.

B. Molecular Phenotypes in the Absence of Prominent T2 Immune Pathways

1. Definition

Non or low T2 phenotypes are defined by the absence of T2 signatures as opposed to the presence/activation of other immune signatures. Unfortunately, this approach is highly dependent on definitions used for T2 signatures/immunity. Measuring sputum eosinophilia, blood eosinophils or FeNO are more likely to reveal elevated T2 levels as more measurements are collected or biomarkers analyzed. Furthermore, using cut points of ≥3 versus ≥1.5% for sputum eosinophils or FeNO >19 versus >25 or even >35 ppb provides very different percentages of T2-Hi vs Lo patients. Thus the percentage of asthma, which exists in the absence of any T2 biomarker elevation, or even granulocytic inflammation, over a period of time is unknown.

2. Associated characteristics (FIGURE 7)

FIGURE 7.

Schematic of the current understanding of T2-Lo molecular and clinical phenotypes. T2-Lo phenotypes are characterized by lack of clear biomarkers, obesity, and poor corticosteroid (CS) responsiveness, often despite high doses of CSs. Unlike the overlapping cellular and molecular characteristics which defined T2-Hi phenotypes, the characteristics of T2-Lo asthma are more variably overlapping. The subphenotypes include a broad range of severity from the most mild to severe patients, including elements of long disease duration (suppression of T2 immunity), a bronchitic, perhaps neutrophil associated phenotype, and a late-onset obese phenotype. Precision approaches to treating these patients remain poorly understood. IL, interleukin.

Non (or T2-Lo) asthma generally has been associated with later age at onset, long-term disease/high-dose CS use, milder and more stable disease, and, perhaps most consistently obesity (64, 118, 129, 264). Lung function is generally better, often with less obstruction as measured by FEV1/FVC, perhaps contributing to the lesser bronchodilator response typically seen in these patients (204). Patients may have a high degree of symptoms, but typically fewer life-threatening exacerbations (111, 118, 186). Most uniformly, they have a poor response to CSs, even when given systemically. Despite this, they are often treated with these drugs for long periods of time, only worsening obesity and metabolic dysfunction.

3. Obesity-associated T2-Lo asthma (FIGURE 7)

Obesity in association with asthma is a trait and not, by itself, a phenotype. While obesity almost certainly worsens symptoms and lung function in the majority of patients, the characteristics associated with this trait define the phenotypes. In many patients with severe asthma, obesity is associated with persistent T2 inflammation. This persistent T2 inflammation is treated with increasing doses of CSs, typically with modest improvements, but with increasing appetite, lower levels of physical activity, and more weight gain. In these patients, obesity is a confounder, not a cause of disease (64, 129) (FIGURE 8). Whether obesity worsens the T2 inflammation or drives a second metabolic or immune pathway, which additively or synergistically interacts with T2 inflammation, is poorly understood (63).

FIGURE 8.

Overview of the intersection of asthma and obesity. Obesity and asthma enjoy a complex relationship. Top left: persistence of severe, type 2 (T2) associated asthma, poorly responsive to standard treatment can lead to increasing obesity, worsening symptoms, and more exacerbations, which respond to improved asthma treatment (often with concurrent weight loss). Bottom right: however, an obesity-associated T2-Lo asthma arising in adulthood also exists in association with both metabolic and mechanical dysfunction. While balancing certain metabolic pathways may improve outcomes, weight loss appears to be the most effective therapy. FeNO, fraction exhaled nitric oxide; SOB, shortness of breath.

However, a second group, typically females with late-onset asthma and little evidence of T2 inflammation (111, 129), also is identifiable, who have profound symptoms, modest reductions in the FEV1/FVC (as well as FEV1), and low β2-agonist reversibility. Increasing data identify metabolic dysfunction in some patients which tracks with C-reactive protein and IL-6 levels in blood (218) (FIGURE 8). Alterations in the gut microbiome could additionally contribute to this metabolic dysfunction (reviewed in Shore et al., Ref. 241). Hypertension and diabetes are common, with lower lung function and exacerbations more likely in obese patients with elevated IL-6 levels. There is no association with T2 inflammation and no evidence for increased IL-6 expression in sputum cells. However, other studies have noted increases in IL-6 pathway-associated genes in epithelial brushings, such that details of this relationship are still unclear (140, 186). Finally, elevations in IL-6 also occur in nonobese patients and have been associated with similar worse outcomes. Thus, in addition to a pathological driver of disease, IL-6 could also be a biomarker for metabolic dysfunction. Trials are needed to determine whether specific blockade of this pathway in patients with elevated levels improves the clinical scenario.

Other studies have suggested T2-Lo obese asthma is associated with abnormalities in arginine metabolism, such that the traditional T2 biomarker FeNO may be spuriously low. In patients with adult-onset obese asthma, FeNO is inversely related to BMI (130). Low arginine levels associated with high asymmetric dimethyl arginine (ADMA) levels shift iNOS from generating NO to generating reactive oxygen species (12). In obese asthma, plasma ADMA levels are high relative to arginine, consistent with this hypothesis. When controlling for this ratio, the relationship between FeNO and obesity disappears. In vitro supplementation with l-citrulline, a precursor of l-arginine, restores the ability of iNOS to generate NO (286), but only in the presence of an active iNOS enzyme. Thus it is not clear whether this low arginine associated T2-Lo obese phenotype is, in reality, T2-Lo, or just traditional T2-biomarker low. In line with this, supplementing with l-citrulline should increase FeNO levels. This increase in FeNO could improve asthma outcomes concomitant with its known smooth muscle relaxation properties. However, increases in FeNO would also move these biomarker T2-Lo patients into a T2-Hi category, suggesting molecular studies of these patients are needed. In fact, it has been suggested that at least some obese asthma patients may have more tissue eosinophils and higher serum IL-5 than controls, consistent with a T2-Hi phenotype (63).

Other possible mechanisms for T2-Lo obese asthma include various mechanical abnormalities, such as premature small airway closure in relation to the increased fat-volume loading. Interestingly, waist circumference is associated with absolute FVC, but not methacholine-induced airway closure, while BMI is more strongly related (220). Thus some of this late-onset obesity-driven asthma may represent a relative fringe of the asthma umbrella. The associated drop in FVC during methacholine challenges, believed to be driven by airway closure, may not be reactive bronchoconstriction in the traditional sense, but rather compressive, metabolic, or even fatigue-related processes. When the 1980s ATS asthma definition was written, obesity represented a small section of society such that these patients were rarely encountered in asthma studies. Given the lack of T2 inflammation, it is unlikely these patients will respond to current asthma medications, with poor responses to both CSs and β2-agonists. Effective therapy appears to primarily be weight loss (64) (FIGURE 8). Thus some obese phenotypes could be considered obesity-induced thoracic disease with secondary airway effects, as opposed to a primary airway process more traditionally associated with asthma.

4. Other non-T2 molecular pathway phenotypes

One of the most consistent molecular pathways identified through various analyses are those related to the inflammasome, particularly in relation to IL-1β and IL-18 pathways (FIGURE 3). Genes in these pathways are consistently upregulated in both epithelial brushings and sputum, and related SNPs are consistently identified on genome-wide studies of asthma (208). However, some studies have shown IL-18 receptor upregulation in association with T2 gene expression (186, 283), while others have not (16, 156, 231), perhaps due to the complex relationship of IL-18 with type 1 or type 2 immunity, dependent on the associated IL-12 levels (296). Clustering genes related to sputum eosinophils and then using subsequent GSVA to identify pathways, the U-BIOPRED consortium identified a cluster associated with inflammasome expression, moderate T2 gene expression, and neutrophilic inflammation (231). This inflammasome cluster was associated with lower FEV1 and moderately high OCS use. They had the earliest age at onset and the highest C-reactive protein levels, suggesting T2 immune pathways may have been suppressed by CSs. A recent study from the SARP identified elevated sputum levels of extracellular nuclear DNA in 13% of its cohort (159). This increase in DNA associated with older, more severe disease, and bronchitic-like symptoms, and similar to the U-BIOPRED inflammasome cluster, more OCS use. Importantly, these higher DNA levels associated with identification of neutrophilic extracellular traps, sputum neutrophils, and apparent caspase-1 activation in the presence of higher sputum IL-1β levels, consistent with inflammasome activation. Unlike the U-BIOPRED study, there was no relation to type 2 biomarkers.

Neutrophilic asthma has also been proposed as a phenotype of asthma (97, 283). It remains one of the most frequently studied cellular phenotypes and is widely cited in review articles (137, 228). However, the specificity of this phenotype remains unclear. Cut-points for sputum neutrophilia depend on method, with high neutrophils ranging from >40 to >65% of total. Its consistency over time is also less than that of sputum eosinophilia (157). Importantly, two studies of a CXCR2 antagonist, the receptor for the neutrophilic chemokine IL-8, in neutrophilic asthma were negative on all asthma outcomes, despite requiring starting sputum neutrophilia, in one of the studies (198, 206). Target engagement occurred, however, as blood neutrophils decreased, with some subjects experiencing potentially dangerous reductions. Despite this lack of efficacy signal, the concept of neutrophilic asthma persists (17).

Increases in airway or sputum neutrophils also fostered interest in IL-17 in relation to asthma phenotypes. Two studies of an antibody to the IL-17RB, which neutralizes activation of the receptor by any IL-17 isoform, including IL17E/IL-25, considered a pro-type 2 innate cytokine have been completed. Despite an early suggestion that IL-17R may have therapeutic effects in those with a high degree of bronchodilator reversibility, a second study, enriching for that specific phenotype, was completely negative (35). However, sputum neutrophilia was not used as inclusion criteria in these studies. IL-17 pathway genes have occasionally been identified in transcriptomic studies, although overall their expression has been inconsistent (156, 227), with one study suggesting complete dissociation of T2 and Th17/type 17 inflammation (50). In that study, endobronchial IL-17 pathway gene expression was not associated with neutrophilia and was more closely linked to eosinophilia, particularly in the tissue, without difference from patients with T2-Hi asthma. The authors suggested that inhibition of one pathway could give rise to activation of the other. While the utility of dual blockade of IL-13 and IL-17 was suggested in a mouse model, validation of this approach is needed in humans, particularly in light of the negative trials with IL-17 specific therapies.

Smoking, or a history of smoking, has also been associated with neutrophilic (and IL-17-associated) asthma (47, 242). Whether the neutrophilic inflammation is any more clinically relevant in smokers as compared with non- or ex-smokers is unclear. Overall, the data to support a consistent neutrophilic driven (or IL-17-associated) phenotype of asthma are weak. Whether it could still be a biomarker for other pathogenic factors requires further study.

VI. PHENOTYPES AND MOLECULAR PHENOTYPES OF PEDIATRIC POPULATIONS: COMPARISONS WITH ADULTS AND SPECIFIC CHALLENGES

Important clinical differences are obvious in childhood as compared with adult asthma (TABLE 6). Whereas adults are likely to exhibit persistent symptoms, children with asthma tend to have more episodic disease. Many children have severe exacerbations triggered by viral infections and/or allergen exposure that result in healthcare utilization, but then remain asymptomatic between these episodes (88, 94). Depending on age cut-point, almost all childhood asthma is early onset, and the majority of children (>80%) have sensitization to aeroallergens (73, 81, 189, 262). Significant obstruction with air trapping is relatively uncommon (262), and lung function is often normal despite symptom burden and medication use (14, 139, 249). Instead, distal lung resistance may be increased in the absence of significant large airway involvement (128), and other measures such as FEV1/FVC, FEF25–75% predicted, or airway hyperresponsiveness to bronchodilators may relate better to asthma severity (14). While there are subsets of children with more hyperinflation, children tend to have less airway resistance than adults (250) and are generally reversible with bronchodilation (248).

Table 6.

Differences between adults and children with severe asthma

	Adults with Severe Asthma	Children with Severe Asthma
Symptoms	Often persistent with significant burden	May be episodic (more common) or persistent (less common)
Exacerbations	Vary according to phenotypic cluster	More frequent and associated with greater healthcare utilization
Allergic sensitization	Varying degrees of atopy according to age of asthma onset and phenotypic cluster	Majority are sensitized
Airflow obstruction	Moderate to severe airflow obstruction often with incomplete reversal after bronchodilation	Typically mild to moderate airflow obstruction with near-complete reversal after bronchodilation; significant acceleration of airflow obstruction in some adolescents after puberty
Distal airways	Increased resistance that often accompanies large airway involvement, may not be reversible with bronchodilation	Less resistance than adults, may be increased in the absence of large airway involvement, largely reversible with bronchodilation

One of the bigger differences in children from adults is change with age, particularly in adolescents who may remit or have worsening asthma (56, 100) (FIGURE 9). Indeed, some adolescents assume a more adult-like phenotype after puberty and develop significant airflow obstruction (80) that increases the risk of chronic obstructive pulmonary disease in later adulthood (256, 257) (FIGURE 9). Yet, the pattern or mechanisms for the transition to increasing airflow limitation in children remain poorly understood. Sex hormones may influence phenotypes and natural history, as asthma shifts from male predominance before puberty to female predominance after puberty. However, the mechanisms behind this shift are poorly understood. Girls maintain hyperresponsiveness and symptoms longer than males, where androgens may have beneficial effects on asthma symptoms and lung function (60, 86, 260). Longitudinal studies of the incidence and temporal stability of childhood phenotypes are needed, as are studies of sex as a biological variable in the transition from childhood to adulthood.

FIGURE 9.

General paradigm for development and progression of childhood-onset asthma. It is likely over 60% of all asthma phenotypes have their origins in childhood. Genetic predisposition to childhood asthma combined with environmental exposures/changes in gut microbiota lead to atopy, asthma persistence, and potential progression. While an IgE-dependent, likely T2-Hi asthma phenotype is recognized, the association with cellular and molecular profiles is less clear. T2-Lo asthma likely also exists in childhood, although its trajectory is even less clear. Finally, the immune or hormonal factors that drive the resolution, remission, or persistence of disease into adulthood remain largely unknown.

A. Clinical Clusters in Children

Cluster analyses of asthmatic children across severities identified multiple phenotypic groups differentiated by degree of allergic sensitization, asthma duration, controller medications, BMI, and pre-bronchodilator lung function (79, 131, 143, 144, 236, 305). These clusters only marginally conform to definitions of asthma severity proposed by current treatment guidelines, since children with severe asthma were present across the clusters. Not surprisingly, the majority of asthma clinical clusters identified before puberty are most aligned with the early-onset T2-Hi phenotype in adults. However, like adults, clusters associated with minimal atopy have also been described. For example, in children 6–11 yr with difficult-to-treat asthma, Schatz et al. (236) identified a moderately-prevalent (31%) group of children without allergic rhinitis or atopic dermatitis with significant airflow obstruction. However, mean IgE levels were elevated in this group, suggesting that at least some children had underlying T2 inflammation. Similarly, a cluster analysis of inner-city asthmatic children identified a low-prevalent (15%) cluster with low blood eosinophil, FeNO, and little evidence of specific IgE by allergen screening (28% of the cluster) (305). Unlike adults with modest evidence for T2 biomarkers or atopy, no other specific clinical features have been identified. Interestingly, unlike adults, obesity does not appear to distinguish or identify T2-Lo pediatric clusters. In fact, a meta-analysis of pediatric studies did not find an association between obesity and poor asthma control, although a small increase in exacerbation risk has been observed (6).

A consistent theme among these studies is that the greater the severity of childhood asthma, the more prominent the T2/allergic features. In all large-scale pediatric cluster studies, the clusters with the highest degree of aeroallergen sensitization and/or history of other allergic diseases had the highest medication requirements, greatest airflow obstruction and airway lability (often not fully reversible), and the highest prospective risk of exacerbations requiring treatment with systemic CSs (79, 131, 305). Interestingly, these clusters often tend to have the best treatment responses to CSs (45). However, as in adults, not all T2-Hi asthma is CS-responsive (221), with some more severe atopic early-onset clusters having the least response to guideline-based asthma treatment (45). These observations suggest that response to CS therapy may be an important phenotypic variable in children. Yet the mechanisms for these differences are unclear, emphasizing the need to include more inflammatory and molecular phenotyping.

B. Inflammatory Phenotypes in Children

Inflammatory and molecular phenotyping is only emerging in children. Studies to date have focused on peripheral blood cell counts, with three groups described in children (95, 96): 1) T2-Hi/eosinophilic, 2) pauci-granulocytic, and 3) neutrophilic (11, 104, 106). Like adult asthma, the eosinophilic phenotype likely falls under the umbrella of T2-Hi asthma. Importantly, it can be identified in the preschool or early school-age years (33, 233) and is associated with increased symptoms, less controlled disease with frequent exacerbations, more aeroallergen sensitization, and impaired lung function measures with increased airway hyperresponsiveness (96, 120). Like atopic clusters, the eosinophilic/T2-Hi phenotype is generally more CS-responsive (75, 298). One study of severe asthmatic children refractory to systemic CSs noted decreased expression of inflammatory mediators including IL-2, IL-5, IL-10, and tumor necrosis factor-α (TNFα) (77). The lower IL-5 expression observed in children with severe asthma who failed to achieve control with systemic CSs was present despite the absence of differences in baseline blood eosinophils, and consistent with other studies (24, 74). However, adult studies have also not observed increases in airway IL-5 in severe asthma, such that the importance of this finding in children is unclear. Other T2 biomarkers, such as FeNO, are equally variable, often discordant with symptoms and airway eosinophils, and may vary with age (25, 26). Interestingly, in contrast to adults, incorporation of sputum eosinophils into management of school-age severe asthmatic children failed to reduce exacerbations or improve asthma control (82, 101). Nonetheless, subsets of children with severe asthma have persistent symptoms, exacerbations, and airway eosinophilia after supervised systemic CS administration (25, 151, 214), consistent with refractory T2-Hi disease. These same children also had increased reticular basement membrane thickening and airway smooth muscle mass (109, 207, 215) despite relatively low airway T2 cytokines (24). Whether they will respond to T2 targeted biologics remains to be proven and will be essential to confirming (or not) whether pediatric T2-Hi asthma is similar to adult T2-Hi phenotypes. However, anti-IgE therapy improves asthma exacerbations in children even with modest elevations in IgE and atopy (162) and has greater impact in children with allergic sensitization to perennial allergens, including cockroach and dust mite, consistent with its mechanism of action and supporting a relationship with an atopic, likely T2-Hi phenotype (36). Until more detailed studies of T2-targeted biologics are performed in well-characterized children including those <12 yr of age, defining T2-Hi molecular phenotypes will remain controversial.

As T2-Hi asthma in children is poorly defined, T2-Lo asthma is even more poorly defined as it relies on the absence of T2 inflammation. However, like adults, children with a neutrophilic inflammatory phenotype have been identified, which may more accurately reflect activation of innate immune defects or environmental exposures such as smoking, which promote airway neutrophil accumulation and IL-8 release (65). Increased IL-8 expression has also been reported in children with exacerbation-prone asthma (119, 179, 180), but this observation may relate more to viral respiratory infections (125). An airway neutrophil phenotype was also described in a subset of children with severe asthma independent of clinically recognized fungal or bacterial infection (104, 168, 205), but this phenotype has not been replicated (11, 104, 106). As noted previously, a T2-Lo obese childhood asthma phenotype has not yet emerged.

Dysregulation of the airway antioxidant glutathione, in associations with increased lipid peroxidation and DNA nucleoside oxidation byproducts, has also been reported in severe asthmatic children with marked airflow obstruction (72, 78). Severe asthmatic children had decreased histone deacetylase activity in airway cells (78), as well as global disruption of thiol redox signaling (76, 212) associated with posttranslational modification of the transcription factor, nuclear factor (erythroid-derived 2)-like 2 (78). Importantly, these redox changes have not yet been associated with T2 biomarkers such that it is unclear whether the patterns of redox imbalance are primary or secondary factors to T2 skewing.

C. ‘Omic Studies in Children

Transcriptomic studies of childhood asthma have been reported utilizing peripheral blood mononuclear cell gene expression and nasal brushings, but not lower airway samples. One analysis identified a cluster of children with elevated eosinophil counts, both eosinophils and neutrophils, and one with elevated neutrophil counts, The most severe cluster of children was the neutrophilic cluster which associated with expression of genes for glucocorticoid signaling pathways and Th1/Th17 immune pathways (294). Nasal brushing samples may offer insight into molecular phenotypes in children, as nasal brushings bear reasonable relation to lower airway inflammation (223). In nasal brushings of children, 70 genes, including IL-13, IL-5, periostin (POSTN), calcium-activated chloride channel regulator 1 (CLCA1), and serpine peptidase inhibitor, clade B (SERPINB2), were differentially expressed between asthmatic and healthy children (223). Children with T2-Hi nasal brushing samples were more likely to have had a history of exacerbations, allergy, and blood eosinophilia, consistent with adult phenotypes. Combining nasal and peripheral blood cell samples in children with asthma exacerbations (viral or not) identified a network of 46 genes associated with the transcription factor SMAD3 and transforming growth factor-β signaling as well as gene networks associated with eosinophil activation, extracellular matrix production, and epidermal growth factor receptor signaling, but with little evidence for lymphocyte involvement. These pathways were dynamically expressed in relation to exacerbations and more predominantly focused on epithelial growth and repair pattern as opposed to T2-associated, questioning the primary role of T2 immunity in pediatric asthma (8). Further molecular investigations, in coordination with studies of T2 targeted biologics, are warranted, particularly in children with more severe disease, incorporating trajectories over time as pediatric phenotypes transition into adulthood.

VII. ANIMAL MODELS OF MOLECULAR PHENOTYPES OF ASTHMA

Several animal species have been used to model immunological and pathophysiological features of human asthma, although mouse models remain the most common (13). Advantages of mouse models include availability of genetically altered mice (both transgenic and gene knockout) and mouse-specific reagents to study mechanisms of inflammation, airway hyperresponsiveness (AHR), and remodeling. Many reviews of this topic have focused on both usefulness and limitations of these models (59, 99, 155, 278). Limitations include the absence of spontaneous development of disease or disease exacerbations in mice, striking differences in airway structure and cell types including lack of eosinophil degranulation in tissue, as well as poor replication of airway remodeling features of human disease. However, mouse models remain potentially useful for modeling the full spectrum of asthma phenotypes, including their variable response to different treatment modalities.

A. Modeling Allergic Asthma

Mouse models have been used to study Th2-high asthma for years, beginning with ovalbumin (OVA) as a model allergen (30, 93). Since then, variations on the OVA-based model have been developed, all of which are essentially acute models of the mild allergic asthma molecular phenotype. These models variably induce an allergic Th2-dominated response with associated eosinophilic airway inflammation and AHR that responds to CS therapy, consistent with the human phenotype. BALB/c and C57BL/6 are the most common inbred mouse strains used, although the A/J strain, which is more prone to AHR, has also been used. These allergen-based mouse studies have collectively associated IL-4 as a driver of the Th2 phenotype, IL-5 as the key cytokine associated with eosinophilic response, and IL-13 as the one that promotes goblet cell hyperplasia, mucus production, and AHR. IL-4 and IL-13 both bind to the type 2 IL-4 receptor comprising IL-4Rα and IL-13Rα1, with IL-4Rα as the common signaling subunit mediating the response to both cytokines (123). The development of the IL-4Rα antibody dupilumab was influenced by studies using STAT6^−/− mice, which showed that lack of this key transcription factor downstream of IL-4Rα attenuated Th2 responses, airway inflammation, and AHR in an allergen-driven model (7, 158). Similarly, association of IL-5 with eosinophilic responses was demonstrated using IL-5^−/− (84) and IL-5 transgenic mice (165) in mouse models of allergic asthma prompting the development of monoclonal antibodies against IL-5 and its receptor. Although some studies reported a decrease in AHR using IL-5^−/− mice (84) or upon antibody-mediated neutralization of IL-5 (112), another study targeting IL-5 did not reduce AHR, consistent with human studies targeting IL-5 (55, 110). IL-5 also associated with airway remodeling using a mouse model (49). While many of these early studies investigating the various arms of Th2-induced allergic airway disease relied on OVA as a model allergen, Th2-driven immune response can be also induced in mice by allergens relevant to human asthma. These allergens include house dust mite (HDM) (102, 124, 210, 252), cockroach (38, 202, 211), ragweed (46, 255), and fungal allergens (127, 182a, 194, 247, 267) (TABLE 7). However, as described below, most of these allergens elicit a more complex immune response than just Th2 cells.

Table 7.

Mouse models of asthma

Agent	Immune Response (Cell Type)	Airway Inflammation	AHR	Airway Remodeling	CS Response	Reference Nos.
OVA+alum (sensitization)/OVA (challenge)	Th2	Eosinophilic	+	+	+	30, 55, 84, 103, 285, 300
HDM	Mixed Th cell response (Th2, Th1, Th17) and ILC2	Mixed eosinophilic/neutrophilic	+	+	+	102, 124, 210, 252
Cockroach	Th2 and Th17	Mixed eosinophilic/neutrophilic	+	+	+	38, 202, 211
Aspergillus fumigatus	Th1, Th2, Th17	Mixed eosinophilic/neutrophilic	+	+	+	127, 182a, 194
Alternaria alternata	Th2, ILC2	Eosinophilic	+	+	+	247, 267
HDM+c-di-GMP	Th1, Th2, Th17	Mixed eosinophilic/neutrophilic	+	+ (unpublished)	Poor	210, 227
OVA+bacteria	Th2, Th17	Mixed eosinophilic/neutrophilic	+	Not studied	Poor	149

AHR, airway hyperresponsiveness; HDM, house dust mite; CS, corticosteroid; OVA, ovalbumin.

B. Non-Th2 Models of Eosinophilic Asthma

Cells other than T cells also secrete IL-4, IL-5, and IL-13. One such cell is the ILC2 cell, which upon activation secretes IL-5, IL-9, and IL-13 but likely not IL-4. Studies have associated ILC2s with human asthma (126). However, caution needs to be exercised in ILC2 identification by flow cytometry particularly because of their low abundance (126). If ILC2s have a role in eosinophilic inflammation, then developing mouse models to promote release of cytokines such as IL-33, TSLP, or IL-25 from lung epithelial cells, in turn causing release of T2 cytokines from ILC2s, may be useful to better understand the adult-onset eosinophilic phenotype. Many complex allergens present in HDM, cockroach, and fungi such as Aspergillus and Alternaria have protease activity, and instillation of these allergens into mouse airways triggers the release of IL-33 and TSLP from airway epithelial cells with downstream activation of ILC2s (18, 43, 145, 146, 222). Outcome measures were attenuated in ST2^−/− mice, although not by CS treatment, supporting a role for this pathway in persistent CS-resistant eosinophilic asthma phenotypes (39). Human trials are underway to assess ST2 as a target to attenuate disease. ILC2 cells have also been reported in nasal polyps associated with human late-onset eosinophilic asthma (29, 245), although no mouse models are available that recapitulate this association.

The epithelial-derived cytokines IL-33 and TSLP also activate basophils and mast cells, both of which have been implicated in asthma (98). While the growth factors stem cell factor and IL-3 promote mast cell and basophil development, respectively, the concept of in situ hematopoiesis (ISH) in driving mast cell and basophil expansion in tissues from hematopoietic CD34+ progenitor cells in response to IL-33 and TSLP is increasingly appreciated (FIGURE 10) (133). An increase in numbers of eosinophil-basophil progenitors was initially described during seasonal allergies with expansion of CD34+IL-5Rα+ cells in the nasal mucosa during the exposure season, and with allergen inhalation (173, 238). Mouse models suggest that IL-33, TSLP, and IL-25 may promote ISH (232, 244). Interestingly, allergen challenge induces recruitment of progenitor cells expressing TSLPR, CD127, and ST2 in the sputum of asthmatics, which was facilitated by preexposure to IL-33 and TSLP (245). Humans treated with anti-IL-5 showed reduced eosinophil maturation in the bone marrow as well as in the bronchial mucosa, suggesting that local tissue eosinophil development may also be partly IL-5 dependent (183). However, because of redundancy among the hematopoietic growth factors, IL-3, IL-5, and granulocyte-macrophage colony stimulating factor, targeting IL-5 alone may only achieve partial benefit where ISH drives local hematopoiesis. Promising results of anti-TSLP in reducing baseline eosinophil counts (92) begs for development of new animal models where the sole inciting stimulus is epithelial injury inflicted by pathogens or pathogen products such as (but not limited to) proteases that can induce release of TSLP and IL-33 from the damaged epithelium.

FIGURE 10.

In situ hematopoiesis (ISH) as a potential mechanism for expansion of T2 cytokine-producing cells in the airways of nonallergic asthmatics. The airway epithelium at the interface of the environment and the respiratory system is vulnerable to breaching by pathogens and allergens. Factors such as thymic stromal lymphopoietin (TSLP), interleukin (IL)-33, and TSLP released by disrupted and dysfunctional epithelial cells can be expected to induce IL-5 and IL-13 production by tissue-resident ILC2 cells. TSLP and IL-33 may also promote expansion of hematopoietic stem and progenitor cells (HSPCs) that are resident in the submucosa or are recruited from the periphery. Upon activation stimulation by these cytokines and potentially other mediators, the HSPCs would undergo differentiation into various myeloid lineages that have the ability to produce T2 cytokines. Eos, eosinophil; Baso, basophil; MC, mast cell.

C. Modeling Complex T2-Hi+ Asthma

Infections with bacteria, viruses, and fungi have been associated with asthma (246). Infections impact all arms of immunity and can rapidly activate one or more cells involved in innate immunity including macrophages, mast cells, basophils, and ILCs. To mimic infections, combinations of allergens and infectious agents or their components have been used in mouse models. These models have attempted to mimic more complex immune responses in asthma that include expression of T1, T2, and T17 cytokines, alone or in combination, as well as participation of innate immune responses. Identification of higher numbers of IFN-γ+ T cells in the BAL cells of severe asthmatic patients combined with the association of multiple bacterial species with more severe disease, prompted development of a mouse model involving combinations of the complex allergen HDM and the cyclic dinucleotide cyclic di-GMP (c-di-GMP), an intracellular messenger produced by bacteria. Furthermore, c-di-GMP activates the STING pathway, a target of type I IFNs induced subsequent to viral infections and implicated in asthma exacerbations. A complex immune response is elicited involving production of IFN-γ, T2 cytokines, and IL-17 from T cells as well as increases in airway mast cells. With the use of this model, a role for IFN-γ, but not IL-17, in increased AHR was demonstrated (227). Furthermore, airway inflammation and AHR as well as the level of inflammatory cytokines were only partially suppressed by CSs, similar to CS-refractory severe asthma. A different study also demonstrated the ability of IFN-γ to promote AHR when mice were exposed to OVA in the context of bacterial lipopolysaccharide (LPS) or respiratory syncytial virus (170). Additionally, an IFN-γ-mast cell axis was implicated in AHR in a model of chronic asthma in which mice were repeatedly exposed to intraperitoneal OVA (297). How mast cells and type 1 immune responses connect in the context of infections or otherwise will be interesting to examine further given a central role of mast cells in infections and the fact that once differentiated they are long-lived tissue-resident cells that are difficult to target. In similar models, a combination of OVA and bacterial infection resulted in inflammasome activation and neutrophil-dominated airway inflammation, which was poorly responsive to CS treatment (149). Thus many opportunities exist to iteratively model complex T2-immunity in mice, based on evaluation of reciprocating information from humans and mice.

D. Modeling T2-Lo Asthma

Limited attempts have been made to establish models that exclusively reflect T2-Lo driven mechanisms such as those associated with obesity and metabolic dysfunction. In one model, obesity was developed in mice by feeding a high-fat diet, which induced inflammasome activation, airway inflammation, and AHR, dependent on IL-1β and IL-17 production by ILC3s (147). The study did not report whether the model caused neutrophil accumulation in the airways. Models such as these will be helpful to understand mechanisms underlying T2-Lo, non-eosinophilic asthma.

1. Conclusions about mouse models

Use of overexpression and gene knockout approaches revealed specific roles for mediators controlling features of asthma. Most models have focused on CS responsive allergic asthma, with models of poor CS responsiveness only recently described. The mechanisms and treatment responses behind these newer models await further exploration, fueled by iterative information from human ‘omic studies.

VIII. CONTROVERSIES AND POTENTIAL SOLUTIONS

1. What percentage of the asthma population should be termed T2-Hi asthma, and how should it be defined? Broad integrated evaluation of elevations in known T2-biomarkers over time will be required in large cohorts or replicated in multiple cohorts. However, as these biomarkers are not sensitive or specific, additional biomarkers will need to be identified. Better predictive biomarkers should be identified in relation to responses to T2 biologics.

2. Will molecular phenotypes/endotypes be identified which respond differently to IL-4/13 targeted versus IL-5 pathway blockade? This question becomes increasingly important as more and more targeted expensive therapies become available. However, it is unlikely head to head comparative studies will be performed. Molecular understanding of responders and nonresponders, which utilize ‘omics data to refine the impact of the different interventions on known and unknown pathways and then relate them to clinical outcomes, could give important clues to better biomarkers for each therapy. With these improved biomarkers, it is conceivable that “smaller” biomarker-driven adaptive trial designs could be undertaken to confirm those results.

3. Will biomarkers be identified for T2-Lo asthma? Currently T2-Lo asthma is defined by the absence of T2 biomarkers. However, current treatments and cut-points are all likely to impact current T2 biomarkers, suggesting that specific biomarkers which track with pathobiology and clinical characteristics are needed to better understand T2-Lo asthma.

4. Will precision medicine in asthma be cost-effective? The goals of precision medicine are to optimize care for specific groups of patients with specific disease phenotypes, accomplished through improvement in disease outcomes such as hospitalizations, patient-reported outcomes (enabling return to work), and minimization of risks (side effects of drugs). In chronic, potentially life-long diseases, including patients with some severe asthma phenotypes, accomplishing these goals through selective use of targeted biologics is already cost-neutral, if not cost beneficial. However, markets cannot sustain broad use of medications with yearly costs of 30–50,000 USD. It is incumbent upon academia, industry, and regulators to rapidly develop predictive biomarkers for “at risk” patients with the most need. Additionally, as pathway importance is confirmed by expensive biologic therapies, newer technologies and platforms are needed to reduce costs of targeted therapies. Until precision medicine is delivered in the way it was envisioned, it will not be cost effective.

5. Will precise identification of molecular phenotypes/endotypes enable “curesˮ of asthma endotypes to occur? Development of targeted molecular therapies has reintroduced the concept of disease modification and even cure. As of now, no therapies have sustained impact on the underlying disease once the therapy is discontinued. However, as outlined in FIGURE 1, uncovering the complex networks underlying molecular phenotypes, “learning” the portions of the networks which targeted biologic approaches impact, “relearning” from improved/customized animal models, continuing trials of targeted human interventions, and finally “perpetually relearning” the data as it evolves, has the potential to eventually identify the core drivers of the disease and enable cures.

IX. CONCLUSIONS AND FUTURE NEEDS

Due to the findings of the last 10 yr, it is likely the term asthma will never be viewed in the same manner as it was even 20 yr ago. These findings have increasing application to the clinicians treating asthma patients (see call-out box). In fact, we would propose that “asthma” officially becomes relegated to a lowercase “a” noun, such that its use always requires a phenotypic modifier, as in T2-Hi asthma, similar in concept to terms such as rheumatoid arthritis and iron-deficiency anemia. At this point, the modifiers are limited and cumbersome (T2-Hi, early-onset mild asthma). Despite that, even our coarse ability to identity T2-Hi asthma phenotypes has led to marked improvements in treatments for some severely impacted patients. Unfortunately, most current subphenotypes only modestly meet criteria for molecular phenotypes, and all are associated (or not) with a broad definition of T2 inflammation. With the application of the molecular phenotyping approaches outlined in this review and summarized in FIGURE 1, more granularity (and precision) will be identified, linking the precise molecular pathways to clinical phenotypes and targeting them for intervention. Thus the impressive results achieved in only 10 yr of phenotyping these lowercase “asthmas” requires continued learning to fully achieve the goals of precision medicine.

GRANTS

This work was supported by National Institutes of Health Grants P01AI106684, HL69174, HL103453, and HL113956.

DISCLOSURES

S. Wenzel has consulted for and served as an investigator on multicenter clinical trials from the following companies: AstraZeneca, Boehringer Ingelheim, GSK, Novartis, and Sanofi Aventis. She has never personally received more than $10,000 from any particular company in a given year.