Using omics approaches to understand pulmonary diseases

DNA sequence variation is known to cause, or confer risk for, various rare and common diseases, and genetic testing is increasingly being integrated into medical practice [18, 19]. Starting in 2005, as initial cataloging efforts of common DNA single nucleotide polymorphisms (SNPs) led to the design of commercially available microarrays [20], many investigators sought to relate SNPs to disease presence via genome-wide association studies (GWAS). Since then, over 2600 GWAS have been completed for a wide range of phenotypes [21]. With the advent of NGS in the late 2000s, attempts to relate DNA sequence variants to diseases have extended to whole-exome and whole-genome sequencing [22]. Because an individual’s DNA sequence in non-cancerous somatic cells is relatively stable over time and equal in all such cells, obtaining high quality DNA via a single peripheral blood or saliva sample at any point in a person’s life can be used to measure an individual’s sequence variation across the genome [23]. The availability of cost-effective technology, ease of sample collection, and commonality and stability of DNA sequence across a person’s cells and lifetime, have resulted in genomics studies’ sample sizes outpacing that of all other omics approaches.

Characterizing a transcriptome, or all transcripts expressed in a cell or tissue, entails capturing a static measure of a dynamic process that depends on many factors, including developmental stage, health status, time of day, and recent exposures [70]. As such, the goal of most transcriptomic studies is to compare cells or tissues under controlled conditions or disease states to identify major changes in gene or transcript expression that lead to specific functional hypotheses or biomarker development. In parallel with genomics, transcriptomes were first characterized using microarrays, and shortly after the advent of NGS, their measure was extended to RNA-Seq. Although microarrays have remained a cost-effective and widely used technique [71], a primary microarray supplier (Illumina, Inc., La Jolla, CA) discontinued its expression arrays in Dec. 2016, suggesting that sequencing will soon replace arrays entirely.

Because cell and tissue types have their own characteristic transcriptomes [72], selecting the proper site to collect RNA for a study is critical. For pulmonary disease transcriptomic studies, RNA is often extracted from 1) blood, either as whole blood, peripheral blood mononuclear cells (PBMCs), neutrophils, CD4+ T cells, or CD8+ T cells; 2) lung and airway tissues, either as whole lung, endobronchial biopsies, airway smooth muscle cells, or bronchial epithelial cells; 3) induced sputum; and 4) bronchoalveolar or nasal lavage fluids. Several disease-relevant genes and pathways have been identified via transcriptomic studies, and transcriptomics results have been used to identify disease sub-phenotypes – so called endotypes [73]. Although most results from transcriptomics have not yet yielded diagnostic tests or new drugs for the five diseases discussed, given the growing sample size and careful design of ongoing transcriptomic studies, translation of more results into actionable clinical insights may occur in the near future.

An epigenome refers to genome modifications that regulate gene expression activity and downstream phenotypes, but do not involve DNA sequence variation per se. Epigenomic information can be heritable, but it also varies considerably according to cell type, developmental stage, and environmental exposures [117]. Most epigenomic studies are designed to address the question of whether epigenome states are different in samples with disease or other phenotype vs. those without it. Commonly studied epigenomic phenomena are DNA methylation and histone modifications. In human methylation studies, “methylation” usually refers to the presence of covalently attached methyl groups to 5′ cytosine DNA positions in CpG dinucleotides, a change that typically represses gene expression [118]. Global profiling of DNA methylation can be achieved using quantitative molecular assays such as bisulfite treatment [119] and methylated DNA immunoprecipitation (MeDIP), followed by genotyping or sequencing of DNA [120]. Although whole-genome bisulfite sequencing (WGBS) is the most accurate way to measure DNA methylation, it is an expensive technique. Hence, the considerably more affordable DNA methylation arrays have been widely used. Histone post-translational modifications (PTMs), including methylation, phosphorylation, acetylation and ubiquitynation that occur on N-terminal tails of histones, modulate gene expression by affecting chromatin condensation and the ability of transcriptional proteins to access DNA [121]. To create global maps of genomic locations where histones with modifications of a specific type are present, chromatin immunoprecipitation (ChIP) to select DNA bound to histones with specific modifications is used followed by microarray analysis [120] or sequencing (ChIP-Seq) [122]. More broadly, techniques such as ChIP-Seq can be used to characterize the cistrome, that is, the genome-wide catalog of all short DNA sequences where a transcription factor binds [123]. Because epigenomes are cell-specific, pulmonary epigenomic studies use DNA extracted from disease-related tissues and cells, including blood, PBMCs, lung, and airway. While epigenomic findings related to most pulmonary diseases have not yet yielded diagnostic tests or therapies that are used in clinical practice, some of the modifications identified have provided insights into disease mechanisms that with further study may yield actionable insights.

Protein expression levels reflect the metabolic state of, and physical processes experienced by, cells. In addition to measures of protein levels themselves, assays to measure critical aspects of protein function, including localization, protein-protein interactions, and post-translational modifications (PTMs) have also been developed. Characterizing the full proteome is still challenging, but technological innovations are improving our ability to obtain cross-sectional time and space snapshots of protein levels. These snapshots reflect observed phenotypes more closely than those of genomic, transcriptomic, or epigenomic techniques [3]. Protein microarrays designed based on known proteins or peptides were the first to increase high-throughput capacity to discover protein biomarkers, analogous to the microarray technologies that were first used to measure global gene expression and variation [155]. Beside array-based assays, high-throughput technologies for proteomics include mass spectrometry (MS)-based techniques (e.g., tandem-MS (MS/MS)) and gel-based techniques (e.g., differential in-gel electrophoresis (DIGE)) [3].

Pulmonary disease proteomics studies have been carried out using induced sputum, pulmonary epithelial lining fluid, bronchoalveolar and nasal lavage fluids, exhaled breath condensate, and blood plasma and serum [156]. Most proteomic studies of pulmonary diseases have case-control designs, with sample sizes ranging from tens to hundreds of samples [156]. While some of the statistical issues relevant to other omics techniques apply to proteomics, the technologies used for proteomics are quite different than those used for next-generation sequencing and arrays, necessitating tailored analytical approaches and introducing new limitations. Additionally, the greater diversity of technologies used to obtain proteomics data has resulted in a slower adoption of standards to identify and report findings. While the creation and use of a public repository analogous to GEO or SRA to store and catalog proteomics data contributed by researchers has lagged relative to other omics data types, the European Bioinformatics Institute’s PRoteomics IDEntifications (PRIDE) database, a standards-compliant repository that now contains data from over 70,000 assays, is a comprehensive and widely utilized resource [157].

Metabolites are small molecules (<1 kDa) that participate in chemical reactions within living organisms, and they include endogenous (amino acids, nucleic acids, vitamins) and exogeneous (drugs, toxins) chemicals. Metabolomics refers to the measure of all metabolites in a biological system [175]. Similar to proteomics, metabolomics provides a snapshot of the active physiological status of a cell or tissue. The most common techniques used to characterize the metabolome are nuclear magnetic resonance (NMR) and high-resolution MS [176, 177]. While metabolomics studies are unbiased, the identification of specific metabolites relies on having references to link measured spectra to them. The public repository Human Metabolome Database (HMDB) provides curated metabolomic data, currently listings 42,000 metabolites [178]. This resource allows users to search metabolites by categories (e.g., metabolite name, disease, and biofluid), and it expedites the process of mining metabolomic data. A wide range of biological samples have been used for pulmonary disease metabolomics studies, including blood serum and plasma, induced sputum, exhaled breath condensate, bronchoalveolar lavage fluid, and lung tissue [179]. Due to the non-invasive and convenient nature of acquiring urine in a clinical setting, urine is also sought as a target to identify metabolomics biomarkers.

The integration of diverse omics datasets provides an opportunity to gain insights beyond those that are possible from individual datasets. Thus far, most integrative omics analyses involve the pairing of two data types, but increasingly complex analyses have been conducted as novel methods have been developed [201]. Genome-wide expression quantitative trait loci (eQTL) studies, the most common paired approach, involve the measurement of associations between genetic variants and gene expression levels. An assumption of eQTL studies is that differing levels of mRNA result in different observable phenotypic outcomes. Thus, if a GWAS hit is also an eQTL, one can hypothesize that the SNP modulates an associated phenotype via changes to level of expression of a specific gene. Due to the cell- and exposure-specific nature of gene expression, eQTLs are most relevant to a GWAS if their results were obtained in a disease-relevant context. The most comprehensive eQTL database currently available is the Genotype-Tissue Expression (GTEx) resource, which includes data for 1641 samples across 43 tissues from 175 individuals [202]. Although this resource is helpful, it does not ideal for pulmonary disease studies, as results for lung are derived from lung tissue, rather than cell-specific lung components. The design of eQTL studies has been extended to relate genetic variants to protein and methylation levels, resulting in so-called protein quantitative trait loci (pQTL) [203] and methylation quantitative trait loci (meQTL) [204] studies. Approaches that consider more than two data types often involve multi-staged analyses, where relationships among datasets are sought in a hierarchical fashion [205, 206]. Prior to integration, individual datasets must be analyzed carefully to reduce propagation of errors, as these compound when performing integrated analyses [11]. Additionally, overfitting errors are more prominent because the number of observations in integrative analyses are much greater than the number of individual samples [11].

Most integrative omics studies in pulmonary diseases are eQTL studies [207, 208]. A large-scale asthma eQTL study of 1111 human lung tissues identified an eQTL within GSDMA that was a risk allele for asthma in the GABRIEL GWAS study, and created a network of relationships with eQTL and GWAS data that identified SOCS3 as a key asthma pathway [209]. A large COPD eQTL study that used whole blood and sputum gene expression data from 121 ECLIPSE subjects found eQTLs near previously reported GWAS loci (FAM13A, CHRNA3/5, HHIP), suggesting hypotheses for how association signals are functionally related to COPD [210]. A COPD pQTL study based on expression levels of 88 blood proteins and a COPD GWAS, found 527 pQTLs, including surfactant protein D, vitamin D binding protein, and TNFRSF10C [211]. Despite not using unbiased proteomic measures, this study showed that pQTLs provide helpful functional links that were absent at the level of eQTLs for most of the genes in question. A broader integrative study of longitudinal FEV₁ in children with asthma combined GWAS, RNA-Seq, and ChIP-Seq data to identify seven nominally significant variants that could be related to this phenotype [212]. Although this study did not fully integrate datasets and it suffered from having a small sample size, it demonstrated that leveraging multiple datasets can lead to helpful disease insights.

Another approach taken in integrative omics studies is to use networks to model higher-order interactions among biological, environmental, and clinical characteristics of patient groups to identify hypotheses regarding disease mechanisms. A common network approach is to represent molecules (e.g. genes, proteins) and diseases as nodes, and assign edges between nodes according to known or measured relationships. From such networks, various structural measures are made, including the identification of the most highly connected nodes (i.e., the key nodes) and of modules consisting of highly connected groups of nodes that are thought to participate in specific biological or pathogenic processes or share other commonalities [213, 214]. Such measures are used to prioritize relationships for further validation studies. Network-based analyses focused on transcriptomic data are most common, and have been applied to identify clinically distinct asthma and COPD subgroups [94, 215, 216].

In contrast to the omics techniques already discussed that use cells or tissue in bulk as starting materials, newer assays are enabling the characterization of single cells [217]. Such approaches offer a more thorough understanding of physiological processes, as single-cell resolution omics data enables the characterization of intra-cellular populations, cell states, and cell transitions that are not observable with population-averaged cell data [218, 219]. The workflow for single-cell techniques is similar to that for their bulk counterparts, but with additional key steps to isolate single cells and amplify the genome component of interest [220]. Novel computational and statistical methods have arisen to deal with challenges related to these techniques. For example, analysis of single-cell RNA-Seq (scRNA-Seq) data, the most widely used single cell approach, involves dealing with a large number of undetected transcripts (i.e., an abundance of zeros), increased variability, and multi-modal expression distributions [221].

Although the number of scRNA-Seq studies is quickly expanding, few related to pulmonary diseases have been published. Mouse model scRNA-Seq studies have been used to 1) reconstruct cell lineage hierarchies of lung epithelial cells, unveiling progenitors of two alveolar type cells and novel cell-type specific markers [222], and 2) identify innate lymphoid cell precursor subsets using mouse bone marrow progenitors [223]. An IPF scRNA-Seq study found that IPF cells often coexpressed markers of alveolar type 1, alveolar type 2, and conducting airway cells, and were thus in “indeterminate” states of differentiation, in contrast to normal lung epithelial cells that expressed alveolar type 2 markers [224]. Future single-cell omics studies will be especially helpful to understand cell populations that are drivers of disease and characterize developmental processes.