Introduction to Precision Medicine
The goal of precision medicine is to provide the right treatment to the right patient at the right time. Inherent in this framework is the concept of disease heterogeneity, such that patients with similar symptoms or diagnoses may have different underlying pathophysiologies. Jameson and Longo [1] have provided a useful structure for precision medicine. It starts with a heterogeneous disease, which is classified into subtypes based on a diagnostic test or biomarker. The subtypes may be targeted with new or existing treatments. In the current era, precision medicine usually encompasses genetics or other omics technologies (e.g., metabolomics or microbiome) for disease subtyping [2]. Precision medicine has been widely adopted across multiple types of cancer, but implementation in non-malignant diseases has been slower.
In any new field, one must first understand the terminology. Adapted from classical genetics, the term ‘phenotype’ indicates any characteristic that may be observed [3]. Examples in respiratory diseases include lung function tests, Immunoglobulin E levels, physician diagnosis of idiopathic pulmonary fibrosis, or patient reported outcomes of symptoms. A ‘subtype’ or ‘subphenotype’ is a clinical division of a disease, such as extrapulmonary acute respiratory distress syndrome or rheumatoid arthritis associated interstitial lung disease. An ‘endotype’ is a disease subtype with a shared biological mechanism, such as T2 high asthma or emphysema due to alpha-1 antitrypsin (AAT) deficiency. The U.S. Food and Drug Administration defines a biomarker as an ‘indicator of normal biological processes, pathogenic processes, or biological responses to an exposure or intervention’ [4].
Current State of Chronic Obstructive Pulmonary Disease Precision Medicine
Chronic obstructive pulmonary disease (COPD) is a clear example of a heterogeneous disease that may fit into the precision medicine framework. In fact, the Global Initiative for Chronic Obstructive Lung Disease (GOLD) includes heterogeneity in their definition of COPD [5]. COPD is diagnosed based on airflow limitation on spirometry. Multiple pathologies may lead to airflow limitation, including emphysema and small and large airway disease [6]. In contrast to cancer, pathologic analysis of lung tissue is rare in COPD, so non-invasive means, most commonly chest computed tomography (CT) scans, may be used to assess these heterogeneous pathologies [7,8].
The distinction between emphysema and chronic bronchitis has been established since the 1950s [9], with the classic COPD subtypes of the ‘pink puffer’ (thin, emphysema, without hypoxemia) and the ‘blue bloater’ (overweight, chronic bronchitis symptoms, and resting hypoxemia) [10]. Modern COPD subtyping, which began in the 2000s, envisions a potential multitude of COPD subtypes [11,12].
GOLD recommends a multidimensional assessment of COPD, with treatment decisions derived from the combination of symptoms (based on modified Medical Research Council dyspnea score or COPD assessment test score) and exacerbation risk (based on prior exacerbation and hospitalization history) [5]. The first line treatments for COPD are bronchodilators, though several precision medicine treatments for COPD are already in clinical use (Table 1). Patients with frequent exacerbations are managed based on blood eosinophil counts. Those with low blood eosinophils may be treated with daily azithromycin [13] or roflumilast [14] to decrease the risk of future exacerbations. Patients with elevated blood eosinophils may be treated with several medications targeting type 2 inflammation, including inhaled corticosteroids [15] and biologic therapies dupilumab [16] or mepolizumab [17].
Patients with severe resting hypoxemia, determined by pulse oximetry or arterial blood gas analysis, have a survival benefit when treated with supplemental oxygen [18,19]. Selected patients with emphysema on chest CT scan and hyperinflation on pulmonary function tests may see an improvement in exercise capacity with lung volume reduction, either surgical [20] or by bronchoscopic placement of endobronchial valves [21,22].
Emphysema due to severe AAT deficiency may be considered a COPD endotype, based on the shared biology, even though the mechanism is not fully elucidated [23,24]. In populations of European ancestry, this is usually due to the protease inhibitor ZZ genotype. AAT deficiency is rare in East Asian populations [25], and when present, other alleles may be involved, such as Siiyama [26]. Augmentation therapy with intravenous infusions of pooled human AAT protein has been shown to slow emphysema progression in a randomized clinical trial [27] and may improve survival based on observational data [28].
COPD Endotyping in the COPDGene Study
There are several possible pathways to identifying endotypes of common chronic diseases such as COPD. Some authors have proposed disease subtyping based on multiple omics [2,29]. Machine learning and artificial intelligence may play an important role in analyzing large datasets with multiple clinical or omics variables [30]. The concept of ‘treatable traits’ recognizes that multiple phenotypes may be relevant in an individual patient, and these should be measured and addressed, rather than trying to fit patients into subtype categories [31].
However, most precision medicine efforts still start with the recognition of relevant clinical phenotypes and subtypes of COPD [32]. Once the clinical problem is identified, the next step is to understand the epidemiology of the subtype, such as the prevalence and the clinical, physiologic and imaging characteristics. Incorporating genetics and other omics assays is necessary to start to determine biological mechanisms and to find biomarkers. These steps require large studies of well characterized subjects with appropriate biological specimens.
The Genetic Epidemiology of COPD Study (COPDGene) is an observational study of over 10,000 current and former smokers (at least 10 pack years) with and without COPD, along with additional non-smoking controls, enrolled at 21 centers across the United States [33]. One-third of COPDGene subjects are African Americans. Subjects were ages 45 to 80 at the initial visit. Subjects with cancer within the prior 5 years and with other lung diseases except asthma were excluded. The baseline study visit (2007 to 2012) included questionnaires on symptoms and medical history, spirometry pre- and post-bronchodilator, 6-minute walk test, high-resolution chest CT scan with computerized analysis for emphysema and airway disease, and a blood sample for genetic analysis. A 5-year follow-up visit included the same assessments and added a complete blood count with differential and diffusion capacity for carbon monoxide. Ten-year visits used a similar protocol, though some subjects only had virtual visits due to the coronavirus disease 2019 (COVID-19) pandemic. Fifteen-year follow-up visits started in 2024.
Additional blood samples for omics analysis were collected starting at the 5-year visit including RNA-sequencing [34], proteomics [29], and metabolomics [35]. These omics data types have different strengths and weaknesses for precision medicine, compared to DNA variation as assessed by genome-wide association or whole genome sequencing studies. Since genetics are determined at conception (assuming no somatic mutations), it needs to be measured only once, and in general, DNA is easier to collect and store. These other omics data types are context dependent, meaning they vary with development, exposures, and disease, and therefore may more closely reflect pathophysiology. They also differ by cell and tissue type. Ideally lung tissue would be available for precision medicine studies, though in contrast to cancer, lung biopsy is not part of routine COPD care. Blood samples may capture inflammatory phenotypes of COPD, and as demonstrated in COPDGene, are more suitable for large scale epidemiology studies. In the next sections, we will provide examples of research within COPDGene that is generating the evidence base needed for precision medicine in COPD.
Airway-Predominant COPD
As mentioned above, historical subtyping divided COPD patients into pink puffers and blue bloaters, but there are no objective definitions in the literature. Based on expert consensus, we defined pink puffers as having greater than 10% emphysema (percent of lung voxels with attenuation ≤-950 Hounsfield units) on quantitative chest CT scan, a body mass index (BMI) ≤20 kg/m2, and oxygen saturation ≥90% on pulse oximetry [36]. Conversely, blue bloaters had chronic bronchitis symptoms, BMI >25 kg/m2, and oxygen saturation <90%. Both subtypes increased in frequency in more severe COPD, but even in GOLD stage 4, they captured less than 15% of subjects (Figure 1); pink puffers were more common across all GOLD stages.
Since the pink puffer-blue bloater subtyping would not be clinically useful, we went further to define emphysema-predominant COPD in subjects with airflow obstruction who had >10% emphysema, and non-emphysematous COPD as <5% emphysema [36]. The latter is presumed to represent airway-predominant COPD, though chest CT scan airway measurement may be more challenging than emphysema assessment. This definition captured a much larger proportion of the COPD subjects. Examining the association between these subtypes and other clinical features, we did not find a difference in chronic bronchitis symptoms or exacerbations. However, there was an association between non-emphysematous COPD and diabetes and the metabolic syndrome. This may reflect differences in systemic inflammation in airway disease compared to emphysema.
In a subsequent study, we used blood RNA-sequencing to investigate inflammatory pathways in airway disease [37]. In 2,396 subjects, type 1 interferon pathway genes were associated with three quantitative airway measures on chest CT scans, most strongly with airway wall thickness. A summary score comprised of these genes was associated with lower forced expiratory volume in 1 second (FEV1) and increased exacerbation risk and was validated in bronchial brushings from 23 COPDGene subjects. The interferon response genes may be a reaction to viral infections that are detrimental in COPD or may be an aberrant inflammatory process driving airway disease.
In RNA-sequencing data, there is a proportion of the RNA reads that do not map to the human genome and are usually discarded. We took these unmapped reads and mapped them to viral genomes [38]. The most common virus identified was Epstein Barr virus, but we also discovered RNA from clinically relevant viruses such as human immunodeficiency virus (HIV) and hepatitis C virus (HCV). Expression of interferon alpha and gamma pathway genes were higher in subjects with HIV or HCV detected. Usually research results from large scale omics analysis are not provided to study participants. The medical and ethical implications of returning HCV results was considered in a separate study [39]. Many of these subjects had already been diagnosed with these infections, but some had not.
Viral and bacterial infections are major causes of COPD exacerbations [40], yet the frequent exacerbations phenotype (two or more per year) appears to be a stable trait within an individual [41]. We identified blood gene expression associated with three exacerbation variables: history of exacerbation in the past year, prospective exacerbation rate, and persistent exacerbators (at least one per year) [34]. The largest number of genes was associated with prospective rate; 214 genes overlapped across all three measures.
Standard ‘bulk’ RNA-sequencing measures total gene expression in a tissue like blood or lung, but every tissue is a mixture of cell types. Single cell sequencing technologies can assay gene expression in individual cells and cell types. However, single cell sequencing remains costly and requires specialized sample collection and processing, which may not be possible in large epidemiological studies. Cellular deconvolution is a machine learning method which uses reference single cell or purified cell type RNA-sequencing to estimate the frequency of various cell types in a complex tissue. Applying deconvolution to blood RNA-sequencing [42], we found that CD8+ T cells, naïve CD4+ T cells and resting natural killer cells were negatively associated with exacerbation rate [34]. The naïve CD4+ T cell result was replicated in the Evaluation of COPD Longitudinally to Identify Predictive Surrogate Endpoints (ECLIPSE) study, another large observational study of COPD [43,44]. These findings suggest that lymphocyte deficiencies may increase susceptibility to COPD exacerbations, including persistent exacerbations.
Asthma-COPD Overlap
Patients carrying the diagnosis of both asthma and COPD are common in clinical practice. Asthma-COPD overlap (ACO) has been reported worldwide [45], with prevalences ranging from 15% to more than 50% of COPD cases in database studies [46]. However, these numbers should be considered cautiously, since there are not consistent diagnostic criteria across studies. Regardless, these subjects may have more severe disease manifestations, such as symptoms, medication use, exacerbations, and hospitalizations, and increased healthcare costs [47-52]. Despite the prevalence and impacts, ACO has been poorly studied. The Dutch hypothesis, formulated in the 1960s, proposed that there are common host factors for asthma and COPD [53], suggesting the concept of ACO as an endotype.
In COPDGene, we devised a clinical definition of ACO [54]. COPD was defined by post-bronchodilator FEV1 <80% with FEV1/forced vital capacity <0.7, corresponding to GOLD spirometry stages 2-4. Asthma overlap was determined by self-report of a doctor’s diagnosis of asthma before age 40, since COPD would be usual at that age. This definition was subsequently adopted in a consensus report [55]. In the initial analysis, we confirmed that subjects with ACO had higher exacerbation rates than COPD alone [54]. In a subsequent paper, we showed that 12% of COPD subjects had ACO, and they were more likely to be women or African Americans [56]. ACO subjects had higher exacerbation rates, lower disease related quality of life, and a higher frequency of allergic rhinitis and family history of asthma. They had less emphysema but more airway thickening on quantitative analysis of chest CT scans. Furthermore, we found that bronchodilator responsiveness did not distinguish between ACO and usual COPD.
We performed a genome-wide association study (GWAS) in 450 subjects with ACO compared to 3,120 with COPD, a sample size that would be considered very small by current standards. Not surprisingly, there were no genome-wide significant associations. The top gene in a combined analysis of non-Hispanic white and African American subjects was G protein-coupled receptor 65 (GPR65), which is involved in eosinophil activation [57]. One of the top genes in non-Hispanic white subjects was SRY-box transcription factor 5 (SOX5), which we had previously associated with COPD and shown to play an important role in lung development in a mouse model [58].
This study highlighted that GWAS may require a larger sample size than other omics. Therefore, we analyzed blood RNA-sequencing for ACO in COPDGene [59]. Enriched pathways included hematopoietic cell lineage, arachidonic acid metabolism, apoptosis, and sphingolipid metabolism. The first two pathways were interesting in that they contained genes whose products are targeted by currently approved asthma therapies, interleukin-5 receptor alpha (encoding IL5RA, target of benralizumab) [60] and arachidonate 5-lipoxygenase (encoding 5-LO, target of zileuton) [61], respectively. From a published study, we extracted a set of 100 genes representing type 2 inflammation in airway epithelium [62]. A score created from the 53 of these genes that were expressed in blood was elevated in ACO and associated with reduced FEV1, higher eosinophil counts and exacerbation frequency (Figure 2). Subjects could be clustered by the expression of these genes, but the clusters did not correspond to clinical diagnosis of asthma, COPD, or ACO, suggesting that there are COPD subjects with type 2 inflammation without the diagnosis of ACO and subjects with ACO without type 2 inflammation. We found similar results in a study using total and specific immunoglobulin E levels, which showed the presence of atopic and non-atopic subjects within ACO [63]. These studies dampened enthusiasm for ACO as a single endotype. Just as COPD and asthma are heterogeneous conditions, ACO is likely heterogeneous as well.
Eosinophilic COPD
Blood eosinophils are a readily available biomarker of type 2 inflammation in COPD, and eosinophilic inflammation is considered a treatable trait [31], with several medications already available (Table 1) and others in development. Blood eosinophils are reported numerically, but like any laboratory value, a threshold is usually desired for clinical applications. In COPDGene, we found an increase in exacerbation risk with increasing eosinophil counts (Figure 3) [64]. Receiver operating characteristic curve analysis showed that 300 cells/μL was the optimal threshold, which is the cutoff widely used in treatment decisions [16]. Similar to the RNA-sequencing study, type 2 inflammation defined by eosinophils ≥300 showed limited overlap with the clinical diagnosis of ACO.
A blood RNA-sequencing study in a COPDGene subset of 49 eosinophilic versus 182 non-eosinophilic COPD found 505 differentially expressed genes [65]. The top genes were known to be expressed in eosinophils, including IL5RA. The differentially expressed genes overlapped with results from genes expressed in eosinophilic asthma [66] and in COPD patients treated with benralizumab [67], as well as genes that predicted response to anti-interleukin-13 therapy in asthma [68].
Conclusions: Challenges and Future Directions
These studies in COPDGene provide examples of the pre-clinical evidence that is being assembled to support phenotypes, endotypes, and drug targets for precision medicine in COPD. However, there remain significant challenges for the clinical adoption of precision medicine. There is a need for additional large study populations with longitudinal data and biospecimens, and several such studies may provide that information, including the SubPopulations and InteRmedicate Outcome Measures in COPD Study (SPIROMICS) [69], Trans-Omics for Precision Medicine (TOPMed) [70], and the UK Biobank [71]. Most COPD cohort studies to date have included mostly participants of European ancestry, though COPDGene and other studies have targeted enrollment of other racial and ethnic groups. Several efforts are ongoing in Korea, such as the Korean Obstructive Lung Disease (KOLD) Study [72] and the Korean COPD Subgroup Study (KOCOSS) [73].
The COPDGene examples have shown the challenges of finding appropriate biomarkers in COPD, compared to various cancers where genetic testing of tumors tissue is widely performed. Plasma fibrinogen is predictive of COPD exacerbations and is qualified by the U.S. Food and Drug Administration to enrich enrollment in clinical trials [74]. However, this has not been adopted widely for clinical trials or patient management. There may not be a single blood biomarker in COPD like troponin for acute myocardial infarction; panels of more than one biomarker may ultimately be required. Chest CT scans provide valuable information about COPD subtypes and could be considered a biomarker. But chest CT is not part of routine care for COPD, due to limitations such as cost and availability. There is a large overlap between patients with COPD and those eligible for lung cancer screening [75], so the low dose screening CT scans may be a valuable source of data for COPD precision medicine, especially with the use of artificial intelligence for radiology interpretation [76].
Clinical trials will be needed to validate biomarkers and novel therapies, which requires acceptance of the potential value of COPD precision medicine by pharmaceutical companies and government funders. The growth of novel therapies targeting type 2 inflammation [77] and epithelial-derived cytokines [78] provide evidence of this interest in developing drugs for specific endotypes. The ultimate application of COPD precision medicine will have to be driven by physicians, health systems, and patients themselves [1].
PDF Links
PubReader
ePub Link
XML Download
Full text via DOI
Print
Download Citation
