Role of untargeted omics biomarkers of exposure and effect for tobacco research

Abstract

Tobacco research remains a clear priority to improve individual and population health, and has recently become more complex with emerging combustible and noncombustible tobacco products. The use of omics methods in prevention and cessation studies are intended to identify new biomarkers for risk, compared risks related to other products and never use, and compliance for cessation and reinitation. to assess the relative effects of tobacco products to each other. They are important for the prediction of reinitiation of tobacco use and relapse prevention. In the research setting, both technical and clinical validation is required, which presents a number of complexities in the omics methodologies from biospecimen collection and sample preparation to data collection and analysis. When the results identify differences in omics features, networks or pathways, it is unclear if the results are toxic effects, a healthy response to a toxic exposure or neither. The use of surrogate biospecimens (e.g., urine, blood, sputum or nasal) may or may not reflect target organs such as the lung or bladder. This review describes the approaches for the use of omics in tobacco research and provides examples of prior studies, along with the strengths and limitations of the various methods. To date, there is little consistency in results, likely due to small number of studies, limitations in study size, the variability in the analytic platforms and bioinformatic pipelines, differences in biospecimen collection and/or human subject study design. Given the demonstrated value for the use of omics in clinical medicine, it is anticipated that the use in tobacco research will be similarly productive.

Introduction

The worldwide burden from the use of combustible and non-combustible tobacco products remains a substantial public health issue. There are known methods to foster cigarette smoking cessation, which have broad applicability to other tobacco products. Biomarkers in tobacco studies are intended to elucitate pathways for disease risk, and more recently for the comparison of emerging tobacco products compared to conventional tobacco products and never use. The primary use of biomarkers in smoking cessation studies focus mainly on compliance of specific tobacco-smoke constituents, such as cotinine, anatabine, anasabine and nicotelline [1,2]. However, with a rapidly expanding tobacco marketplace, mostly with non-combustible products (e.g., electronic cigarettes [e-cigs], heat and not burn tobacco products [HTPs] and oral nicotine pouches), there is a need for additional study of the relative harms for tobacco users switching to other products rather than complete cessation of all nicotine products. There are additional needs for biomarker studies to assess compliance of some of these new products. A summary of commonly used biomarkers has been published following a 2015 Food and Drug Administration (FDA) Center for Tobacco Products (CTP) workshop for tobacco-related biomarkers of exposure and harm [3,4], and other reviews [5,6]. However, this workshop only superficially addressed omics biomarkers and appropriately referred to them as novel.

Biomarkers are generally classified in several ways, with some overlap, as biomarkers of exposure and biomarkers of effect (or harm). The biomarkers of exposure are typically chemically-specific measurements of smoke constituents and their metabolites, such as for tobacco-specific nitrosamines and tobacco alkaloids that only can come from tobacco leaves [6]. Other chemically-specific biomarkers are not unique to tobacco exposure, but also derive from other sources (e.g., diet and environment). Biomarkers of effect represent changes in cell function and cannot be specific to any particular exposure, and includes those derived from endogenous exposures, such as markers of DNA damage, proliferation, cell death, alterations in xenobiotic metabolism, and changes in gene expression, methylation, proteins, and endogenous metabolites.

Biomarkers may be targeted, e.g., the measurement of a specific protein or panel of proteins, carcinogen metabolites, or specific type of DNA damage. Untargeted biomarkers, which include omics platforms, are broad screen approaches that are generally agnostic to specific hypotheses and are used for biomarker discovery (including profiles and unique phenotypes) with the intent to be useful as clinical assays, or for large scale studies impacting policy. They also are used to discover effects on disease pathways and networks. There is a desire for validated tobacco-related omics biomarkers that would be clinically relevant as predictive of future harm.

Omics strategies focus on specific biological types of molecules, including proteins, gene expression (including non-coding RNAs such as micro-RNAs), gene methylation, small molecules, and lipids [7–9]. These all yield complex datasets ranging from 100′s to 100,000 features (e.g., chromatographic mass spectroscopy peaks that have not yet been identified) that require deep understanding of data processes (pipelines) and analysis incorporating systems biology approaches [10,11]. They can be done using focused arrays that offer ease of use through manufacturer supplies reagents and software analysis, or they can be more comprehensive such as through next generation sequencing of RNA transcriptomics, or mass spectroscopy methods for proteomics, metabolomics and lipidomics.

The focus of this manuscript is on the use of untargeted screening strategies through omics assay for biomarker discovery and correlative assessments for the biological impacts of tobacco products relative to each other, complete cessation, smoking reinstatement and relapse prevention. (Epigenomics and genomic risks will not be discussed herein as they are addressed in a separate article in this monograph.) The various omics are summarized in Table 1, including a ranking for the number of available studies and applications. This manuscript reviews general principles for validation and use of omics biomarkers.

Table 1.

Omics Approaches For Tobacco Research: Ranked by extent of application, from numerous studies to a state of infancy.

Omics	Description	Techniques	Advantages	Limitations
Genomics	Genetic susceptibility Somatic mutations in tumors	Arrays Next generation sequencing	Well-developed technologies and analytic pipelines Low error rates Stable at room temperature Unbiased discoveries of tobacco-related genetic alterations	Biological consequences of variants unknown Ethical issues about heritable traits Tumor heterogeneity for somatic tumors Computational cots and resources for the whole genome
Transcriptomics	Coding (i.e., mRNAs) and non-coding RNAs (i.e., long non-coding RNAs, small non-coding RNAs including microRNAs) Alternative splicing RNA editing RNA modification	Arrays Next generation sequencing	Assessing post-translational modifications better reflects active proteins Assesses coding and non-coding RNAs Well-developed methods and analytic pipelines Unbiased discoveries of tobacco-related genes or transcripts	Levels of transcription might not reflect protein levels or functions Many transcripts give rise to more than one protein (e.g., alternative splicing) RNA instability, quantity, and quality in biospecimens Potential batch effects Lack of assessment of ancestral markers, and considerations of racial/ethnic differences in outcomes
Epigenomics	Methylation modification of DNA Histone modification MicroRNAs	Arrays Next generation sequencing Mass spectroscopy	Assesses gene regulation Assesses reversible effects Well-developed technologies Stable at room temperature for DNA methylation Genome-wide analysis with resolution at locus Low cost per locus Quantitation Unbiased discoveries of tobacco-related epigenetics	Unclear biological impact in untranslated regions Potential direct influences by genetic factors for DNA methylation Potential batch effects Lack of assessment of ancestral markers, and considerations of racial/ethnic differences in outcomes
Proteomics	Proteins including post translational modifications	Arrays Mass spectroscopy	Closer to biological inference than transcriptomics and epigenomics Well-developed methods and analytic pipelines Relative and absolute Quantification Unbiased discoveries of tobacco-related proteins	Does not provide information about biochemical function Sample preparation that may impact the downstream analysis (e.g., incomplete lysis or solubilization) Potential batch effects
Metabolomics	Small molecules	Mass spectroscopy Nuclear magnetic resonance	Closest link to phenotype Assesses Exogenous metabolites (tobacco toxicants, nutrients), endogenous metabolites (amino acids, hormones, fatty acids and phospholipids, signaling) and lipids Precise quantitation Cost per assay Unbiased discoveries of tobacco-related metabolites	Fewer analytic software Less developed databases for feature identification Challenging biochemical verification Potential batch effects Lack of standard operating procedures
Lipidomics	Phospholipids, glycerides, sterols, fatty acids	Mass Spectroscopy	Cost per assay Unbiased discoveries of tobacco-related lipids	Quantitation Many lipids have the same molecular weight Fewer analytic software Less developed databases for feature identification Various sensitivity and difficulties for quantification and identification based on types of mass Spectroscopy Potential batch effects

Methods:

systematic review of tobacco use and untargeted omics in healthy individuals

This systematic review reports studies of tobacco use and cessation only in healthy individuals that therefore avoid disease bias (e.g., not studying lung cancer or COPD patients) and provide data for toxicological changes prior to the onset of disease. Most available studies were about the latter and only a few cessation studies have been conducted. A systematic search for all papers was conducted through PubMed and Google Scholar. Search terms including smoking and either transcriptomics, proteomics, metabolomics and lipidomics. Studies must have used untargeted methods, and those with specific a priori hypotheses assessing individual or groups of related biomarkers were excluded. Studies that did not provide data for healthy individuals were excluded. Additionally, cited papers in those publications were reviewed, as well as the related articles and publications citing the study as listed in PubMed. As this manuscript shows, there are only limited studies across many types of biospecimens (serum, plasma, urine, saliva, sputum and lung samples). The cited studies are summarized in Supplementary Table 1. Cited studies in this review are those that provided data for healthy individuals of any age (often referred to as “controls”) using untargeted omics methods and provided data specifically about the healthy individuals.

General principles for validation and use of omics biomarkers

Omics approaches for use in epidemiology and prevention intervention studies is an evolving field, and the application to tobacco research has been quite limited. The use of omic’s derived signatures or profiles has been somewhat validated for some applications in the clinical setting for susceptibility, early detection, diagnosis, treatment decision making and prognosis of various illnesses [12–15]. Thus, there is proof-of-principle applications for the use of omics in tobacco risk and cessation studies (and relapse). Validation of omics biomarkers such that they can be applied to the clinical setting or inform policy include both technical (laboratory) validation and clinical validation [7].

Technical validation

The technical validation of a biomarker addresses the variability of assay results due to methodological issues and instrument performance, with the intent of improving accuracy and reproducibility at levels observed in human biofluids [3,7,14]. Consideration should be given to each step of the process, namely how biospecimens are collected, time to processing, quality and consistency of reagents, instrument variability, choice of control samples, data pre-processing for analysis (transformation and normalization), and data analysis (Table 2) [16–18]. In the research setting, samples should be blinded to laboratory staff in the context of experimental design, subject characteristics and intervention. Although not typically available for omics assays, “gold standards” or well-defined standards are used to develop and monitor the assay performance. Errors could be random or systematic. Some assays may be less informative for disease pathways when study subjects have some concurrent disease (e.g., high blood lipids or bilirubin), at the levels that features are found in human biospecimens (e.g., below a limit of detection or very high), or might be affected during sample collection and transport. Several statistical approaches should be used, including the analytic sensitivity, coefficient of variation (and for each feature), precision, accuracy, interferences (including from specimen containers), and allowable total error [16]. These can vary based on the type of biospecimen and can be different based on the detectable levels, e.g., more low levels of the biomarkers might have higher coefficients of variation and higher biomarker levels may lead to cross-contamination of samples. Quality control procedures are put into place to monitor the assay performance over time, for example using representative control samples (e.g., pooled subject samples with high and low results), spiked samples and blank reagent samples. It should be noted that quantitative biomarker levels are only applicable to the laboratory and method being used, and should not be extrapolated to results from other laboratories, unless a cross-laboratory validation has been done, or assays are developed with the same biospecimen standards. Manufacturer’s claims for assay performance should be independently validated by the investigator. In the clinical laboratory, or for research where the biomarker results are used to influence decision making no matter how trivial (e.g., subjects receive an intervention based on an omics assay result needs to be developed under the Clinical Laboratory Improvement Amendments (CLIA) methodology as “Laboratory Developed Tests” as high complexity assays. The CLIA process does not ensure laboratory validation, only that the laboratory has addressed this, has standard operating procedures and regularly follows them.

Table 2.

The Many Steps of Omics Research (after hypothesis formulation, research strategy and statistical analysis plan [including power analysis]).

Steps	Examples
Standardization and understanding the research subject	Fasting, age, gender, medications, diet, occupation, recent changes in body weight or lifestyle, health status
Biospecimen collection	Biological matrix (e.g., urine, serum, sputum, lung, etc.), time from collection to processing, conditions of collection (temperature), transport media, storage media, storage temperature, avoiding freeze-thaw cycles
Determine quality control procedures	Assess inter-experimental quality variations, batch effect, filtering thresholds, replicates, use of chemical standards, pooled controls, assess sample quality
Sample preparation and assay	Determine the technology and what is needed to generate data, and the experimental conditions (e.g., columns, flow conditions, depth of coverage)
Data storage and pre-processing	Where will data be stored, archived and processed, quality control, method for filtering, alignment and retention time correction, peak detection
Main analysis	Parameters of machine learning, supervised or unsupervised, Bayesian approaches, choice of software (e.g., quality and validity), background correction or normalization of data, visualization tools, types of linear and non-linear models, multiple testing correction, characteristics of features to be studied (e.g., fold change, p value, feature present in X% or samples, high coefficient of variation), dimension reduction, network analysis and pathway analysis
Replication	Training and validation procedures within a dataset or replication across independent datasets
Annotation	Identification of features and validation of results by an independent technology

The laboratory validation for tobacco-focused omics studies is not commonly done, although it should be. The above procedures to validate assays for the clinical arena are described by the FDA for therapeutic drug development. The FDA CTP does not have similar guidance for the analysis for biospecimens, but does have a draft guidance for the validation of chemical analytic testing for tobacco products that includes many of the same principles that would be used for tobacco biomarker assays (https://www.regulations.gov/docket/FDA-2021-D-0756). Other resources that are available is a summary of a CTP workshop for biomarkers of exposure and harm, which does not address the validation of omics assays, but also has principles that can be applied to omics biomarkers [3,4].

Clinical validation (Fig. 1)

Fig. 1.

Clinical Validation of Omics Biomarkers.

Clinical validation refers to the biological implications for assay results [7]. The intent of clinical validation is to justify use of the biomarkers in the clinical setting that are directly predictive of future disease risk. Biomarkers of harm generally reflect a surrogate endpoint, e.g., a urine or blood test predicting lung or heart disease [3]. However, there are currently no validated tobacco-related biomarkers of harm of any type, although some epidemiology studies indicate the potential. To become validated, the best evidence shows predictivity in long term observational studies where human disease develops after many years of use. The challenge of these studies is the limited availability of biospecimens, but these are available when there is sufficient preliminary data to justify the use of the precious biospecimens, including for omics research. While study designs and human interventions for omics biomarkers might be used to assess the biological impact of tobacco cessation, or switching from combustible to non-combustible tobacco products, the data analysis requires different approaches than typically used for behavioral or other interventions. The large number of features increases the risk of false positives, and so statistical analysis plans include the use of statistical significance testing using false discovery rates, or other methods such as Bonferroni corrections. As with all human studies, replication of omics studies is critical, which is often initially done within a study as internal replication, or across studies as external replication [7,12]. Internal replication is done within a single data set with a two-step process for splitting the sample set into two groups as an initial discovery set as an unbiased approach and then a validation set seeking to replicate the results of the discovery set [10]. The best replication would come from external datasets in different populations and using different study methods showing the same results (e.g., the perturbation of the biological pathway in two study sets associated with the same omics profile). During this process of replication, consideration also should be given to differing results by gender, race, ethnicity, and/or age [17]. An important research gap for any omics method is the understanding of the variability of results over time unrelated to changes in tobacco use, for example day to day variation in diet, environmental exposures, season, health, and physical activity. It should be noted that with the use of omics, perturbation of biological systems (e.g., increased xenobiotic metabolism) alone may not necessarily infer an adverse toxic effect increasing disease risk, but also can be a normal host response to reduce the harm from a toxic exposure. It also is important to note that statistical significance may not equate to clinical significance, and a substantial limitation of omics studies is the lack of clinical correlates to statistically altered features or pathways [12]. Thus, the determination of a biomarker’s clinical utility is a function of how well a biomarker is a surrogate for a risk or disease outcome [7].

The CTP provides regulations, guidance and premarket approvals based on a population health standard, and therefore assessing human impacts are central to their considerations. In 2019, the CTP recommended for product applications seeking premarket approval of e-cigs and other electronic nicotine delivery devices to include biomarkers of exposure and harm (https://www.fda.gov/regulatory-information/search-fda-guidance-documents/premarket-tobacco-product-applications-electronic-nicotine-delivery-systems-ends). There is no guidance, however, about how to assess the clinical implications of biomarkers.

Omics studies utilize the full range of tissues. Urine, blood, sputum, and nasal epithelia, for example, are considered surrogates for more difficult tissues to collect, such as in the lung and bladder epithelia. How well surrogate studies represent target organs are generally unclear and need further study. Target organ analysis has typically had limitations as these tissues may require invasive procedures (e.g., bronchoscopy) and are typically collected from unhealthy subjects with cancer or COPD. These studies are less informative because the presence of the disease may affect assay results rather than a pathogenic process, even if surrounding “normal” tissues are used [19]. Biospecimens such as sputum are predicated on the concept that there will be a combination of broncho-epithelial cells and inflammatory cells, or that proteins in sputum are secreted from lower airways; optimally this is done with sputum induction techniques that are often not successful and has some risk [20]. Nasal epithelium is considered to have similar transcriptomic changes as broncho-epithelium within the context of field cancerization [21]. While there is a strong rationale to provide greater emphasis on interpretation for studies using target organ tissues, databases that provide correlations for target to surrogate tissues enhance the value of surrogate studies, although there is limited availability for the omics assays discussed herein for tissues from healthy individuals (in contrast projects such as the Genotype-Tissue Expression (GTEx) Project).

Analytics

Omics approaches can focus on individual features and their biological significance, establish profiles or molecular phenotypes related to a particular trait (e.g., smokers v. quitters), and/or identification of perturbation of disease pathways. Profiling can be focused on a priori hypotheses, for example in tobacco research this can assess pathways known to be involved in carcinogenesis as hallmarks of cancer [1], or xenobiotic metabolism of carcinogens and other toxicants. For each omics type bioinformatic methods and databases are available to associate features with pathways and assess the magnitude of impacts on the pathway [2].

Pre-processing of data readies the data for analysis by assessing quality, removing assay artifacts, summarizing raw data into analysis ready features, normalization, and transformation of the data [3]. Further processing can include QC based filtering (e.g., samples or features with a high coefficient of variation, low signal, or excessive missingness) and feature selection for analysis (e.g., pathway specific feature analysis)

Analytic approaches have been summarized by Kaur and coworkers [4], Kim and Tagkopouos [3] and Reel and coworkers [2]. They note advantages and disadvantages for analytic tools for machine learning. Frequently, principal components analysis is used to show the clustering of subjects or other characteristics by omics, check for batch effects, and identify outliers and the features driving the clustering. Correlation, regression, group comparisons, and other statistical methods are used to determine features associated with an outcome/measurement. Due to the large number of tests required for most omics analysis, the false discovery rate [22] or similar methods such as Bonferroni corrections are typically employed to identify features of interest. Analytic methods for modeling/prediction can be done with supervised (labeled data where the inputs or features are known, and their effects) or unsupervised learning (drawing inferences blind to data labels seeking associations without prior knowledge of feature function) [2–4,23]. As an example of the latter, ‘natural’ clusters (unsupervised) derived from the data may be indicative of different disease states and be associated with outcome measures of interest. Supervised learning methods can be more accurate and are simpler to apply but more upfront knowledge is needed. Methods incorporating both as semi-supervise learning also are available [3]. Pitfalls include overfitting and the curse of dimensionality both related to having complex models with too few data points, and imbalanced class size dues to skewed distributions of data [2,3]. Specific analytic applications are reviewed by Kaur and coworkers (2021) [4].

For most, if not all, of the omics types, databases and informatics approaches are available to relate omics features to specific molecules and biological pathways [5]. For many omics types there is also a plethora of previously analyzed datasets available for analysis or comparison of results. For example, there are more than 800,000 gene expression cancer-focused datasets in the Gene Expression Omnibus (GEO) database of the National Center for Biotechnology at NIH.

Integrative analysis of multi-omics data holds the promise of a better understanding of the biological impact of conventional and new tobacco products [6]. The combined analysis will, for example, employ machine learning methods to profile individuals using a combination of gene expression, epigenetics, proteins and small molecule metabolomics to provide a comprehensive molecular picture of a disease state or the involvement of particular biological pathways [7]. Analytic methods and deep learning methods have been summarized by Correa-Auila and coworkers [8] and Grapov and coworkers [9]. Integrative analysis can include, for example, the combination of different omics data for cluster analysis, clustering of clusters, and interactive clustering, which has been summarized recently by Zhang and coworkers [6]. However, there are substantial challenges that must be met before the promise of integration for better predictions is realized. It is important to note the challenges of using single omics for both laboratory and clinical validation are not reduced by integrative analysis, rather they are amplified [9]. Integrating omics platforms will require larger datasets, novel analytics not yet developed, do not obviate the need for good study design, and have limitations because omics platforms frequently do not have overlapping features to integrate related genes, proteins etc.

Omics methods and examples of applications (Table 1)

Transcriptomics

Transcriptome-wide analysis through mRNA profiling can be done by both array methods and next generation sequencing (e.g., RNASeq) [24]. Both coding mRNA and non-coding mRNA have been studied (e., miRNAs and long non-coding RNA). A note of caution for mRNA based assays is that mRNA is not stable and needs to be processed and collected in media that stabilize the mRNA; are subject to sample quality; and, mRNAs may have different stability biasing results, with miRNAs being the most stable because of their small size. Kopa and Pawliczak (2018) summarized various human studies for transcriptomics associated with smoking and e-cig use [25], while Silva and Kamens (2020) published a transcriptomics review for smoking [26]. A recent review by Kopa-Stojak and Pawliczak (2022) summarized reported differential miRNA expression in smokers and e-cig users, in experimental animal models and in vitro cell cultures noting that e-cig exposures had lower impacts compared to smoking, but there were some unique effects in the experimental models compared to air controls [27]. Devadoss and coworkers (2019) provided a summary of long non-coding mRNA associated with smoking and the pathogenesis of COPD, including a summary of available databases [28].

Blood testing comparing smokers to never-smokers use white blood cells and sometimes mononuclear cell subsets for DNA. This may or may not be corrected for the relative numbers for white blood cell subtypes, which can vary greatly among individuals as does the gene expression for the different subtypes. A meta-analysis of whole blood transcriptomics in 6 cohorts of smokers, former smokers and never smokers yielded associations with platelet and lymphocyte activation, inflammatory response, and protein biosynthesis [29]. Interestingly 12 genes including LRRN3, GPR15, and CLDND1, remained differentially expressed in former vs. never-smoker 30 years after quitting. Consistently, blood studies show differences for xenobiotic metabolism, immune-related pathways and AHRR pathways in both cross-sectional studies and switching studies [29–33].

Lung sampling by bronchoscopy in a cross-sectional study of smokers, never smokers and long-term e-cig users demonstrated clear differences between smokers and never-smokers, and the e-cig users had similar gene expression patterns as the never-smokers, indicating an attenuation of smoking effects [34]. In a very small study using nasal epithelium, the e-cig users for 6 months or more did not show the same attenuation, indicating that in this case the nasal epithelium is not a surrogate for the lung [35]. In a separate lung study, never-smokers were provided nicotine- and flavor-free e-cigs for one month, where a small increase in inflammatory cells and immune cytokines was noted, but there was no change in gene expression [36]. A smaller study with pre-and post-bronchoscopies following only 10 e-cig puffs, but still there was some genes affected within the immune pathways, but only minimal effects were reported likely due to the limited exposure [37]. The brief exposure is a limitation of this study and whether the lack of changes was due to a lesser impact of e-cigs or the duration of exposure is unclear.

Biopsies have been used to assess transcriptomics in the oral mucosa of current smokers and never smokers, and similar to other tissues, expression differences were found for xenobiotic metabolism, especially CYP1A1, although not effects were observed for AHRR (there were changes reported for AHRR methylation) [38].

Sputum is another surrogate biospecimen used to assess lung effects. For example, Titz and coworkers (2015) induced and processed cell free sputum from smokers with and without COPD, and never smokers [39]. They found transcriptome differences for xenobiotic metabolism and oxidative stress response, and former smokers profiling was the same as never smokers. The former smokers were on average 8.85 years since quitting, and so it is unclear how long it takes after cessation to attenuate the smoking effect.

In lung tissue, Szymanowska-Narloch, et al., (2013) reported in a small number of subjects that gene expression of untreated lung cancer patients differed by smoking status in the lung tumors, but not surrounding non-cancer tissues, indicating that the impact of smoking may have happened after the development of cancer [40]. Morissette et al., (2014) studied adjacent non-cancer tissues in lung cancer patients, comparing smokers to never smokers [41]. They concomitantly identified differences in gene expression for the cancer patients and used a mouse model for cigarette smoke exposure (BALB/c female mice, 8-week whole body exposure). It was reported that 17 genes, 11 pathways, and 58 affected biological functions were identified in both. Included were previously known effects such as the AHRR, CYP1B1, and NQ01, Separately, there were 6 new associations with smoking, namely ACP5, ATP6V0D2, BHLHE41, NEK6, DCSTAMP and LCN2. Using Ingenuity Pathway Analysis, the impacted pathways were xenobiotics response/detoxification, phospholipids metabolism/degradation and oxidative stress defense/generation. As biological functions affected in both humans and mice, there were associations with immune processes, metabolism and cell functions, diseases, tissue injury and repair, and organ development. Han and coworkers (2021) used publicly available datasets from the GEO Database that had transcriptomics data for smokers and heathy non-smokers for alveolar macrophages collected by lavage [42]. They reported by pathway analysis effects on myeloid leukocyte migration, cytokine activity, and leukocyte chemotaxis and migration. They also studied smokers with COPD, but the relationship of the healthy subjects’ differences to the development of COPD and its transcriptomics was unclear.

Proteomics

The study of proteins can be more attractive than gene expression studies because gene expression may not correlate with protein expression and function [43]. The study of proteins is somewhat more complex than gene expression because their biological function may be modified by post-translational modifications, so that protein levels might not reflect actual biologically relevant levels [43]. Proteomics assays typically use mass spectroscopy or arrays. The work flow for assay development and implementation, and assay procedures was recently described by Nakayasu and coworkers [17], Khan and coworkers [44] and Smith and coworkers [43]. Normalization of results, for example per total protein, can be complicated for some biospecimens such as broncho-alveolar lavage or sputum because recovery can be different by smoking status, but is critical to do for any type of biospecimen.

A literature search for the use of proteomics to study tobacco use in blood identified only a few studies. While urine has been used for proteomic analysis, no studies were identified assessing smokers and non-smokers. Methods and studies for plasma analysis have been summarized by Smith and Gerszten [43]. Several large cohorts have been utilized for cross-sectional analysis plasma studies by mass spectroscopy. This includes the assessment of 897 Framingham Heart Study participants by Corlin and coworkers (2021) to find smoker and never-smoker comparisons with statistical differences for cytokine–cytokine receptor interaction, Th1 and Th2 cell differentiation, interleukin‐17 signaling pathway, chemokine signaling pathway, and complement and coagulation cascades [45]. The authors used a second Framingham dataset for replication and were able to assess exposure-response relationships (pack-years). Iglesias and coworkers (2021) studied 1005 subjects from the population-based Swedish Cardiopulmonary Bioimage Study reporting somewhat different results with statistical significance, such as differences for ERG, VWF, RHOJ, GABRE and GucY1A3. Bortner and coworkers (2010) used plasma in 14 smokers and non-smokers noting differential levels of proteins in immunity and inflammatory responses. A single pilot study of 5 smokers who smoked one cigarette in the laboratory comparing pre- and post-smoking saliva; the saliva was polled and compared to 4 control subjects [46]. In this limited study, changes were found for stress response including fibrinogen alpha, cystatin A and sAA.

Some proteomics studies have been conducted examining bronchoalveolar lavage and sputum in smokers and non-smokers. Bronchoscopies were performed on 25 smokers and 17 never-smokers reported by Yang and coworkers (2018) selected from the Karolinska COSMIC cohort [47]. They found 610 significantly altered proteins and 15 molecular pathways (i.e., oxidative phosphorylation, citrate cycle, ribosomal and antigen presentation, phagosome pathway and lysosomal pathway). Ghosh and coworkers (2018) conducted a cross-sectional study of 22 smokers, non-smokers and e-cig users [48]. They found some unique effects for e-cig users, namely an increase in For example, CYP1B1 (cytochrome P450 family 1 subfamily B member 1), MUC5AC (mucin 5 AC), and MUC4 levels were increased in vapers. Franiosi, et al., (2014) conducted a broncho-alveolar lavage study of occasional smokers who were abstinent for 2 days and then smoked 3 cigarettes followed by bronchoscopy [49]. The context was that acute responses are predictive of future COPD risk, and so they studied these occasional smokers with and without a family history of COPD (susceptible and non-susceptible). While differences were found for the two groups, the protein changes were not consisted with those reported for COPD. Titz and coworkers (2015) assessed the proteome in the sputum for smokers and early-stage COPD, using cell free material (so as to assess secretory proteins from the lung lining), reporting alterations in mucin/trefoil proteins and a prominent xenobiotic/oxidative stress response [39]. Importantly, former smoker’s sputum were more similar to never-smokers than current smokers, indicating the utility of biomarkers to assess changes in biomarkers for smoking cessation studies. Baraniuk, et al., (2015) also conducted a study of 20 smokers and non-smokers using sputum, and identified increases in Mucin 5A consistent with other studies, and also SGNB1A1, Prr4, AZGP1, DEF1&3 and S100A8 [50].

Metabolomics

Untargeted metabolomics is an emerging technology that is identifying new biomarkers of tobacco use exposure [51–59]. The assay, using mass spectroscopy or nuclear magnetic imaging methods, can be used to identify thousands of small molecules reflective of exogenous exposures (e.g., tobacco smoke) and cellular responses to those exposures (e.g., lipids hormones, nutrients, and signaling mediators} [15,60–64]. A recent review by Jendoubi (2021) outlines methods and challenges for the integration of metabolomics analysis with other omics [65]. Henglin, et al., (2022) examined various bioinformatic approaches across several datasets, noting sparse multivariate models demonstrated greater selectivity and lower potential for spurious relationships, especially for smaller studies [66]. Metabolomics is now being widely applied to evaluate disease and disease causation [60,67–70], with a variety of bioinformatics workflows [71]. Urine, blood and exhaled air have been assessed [72,73]. Early studies established the feasibility for studying metabolomics in urine and plasma studies that have identified biomarkers related to smoking (e.g., markers of effect such as glycophospholipids and pathways related to inhibition of cAMP), and identified the presence of new menthol metabolites [52,53 74–77].

There are numerous methodological studies about sample collection and processing, for example summarized by Bi and coworkers [78], and Dominguez and coworkers [79]. Serum and plasma (and by type of anticoagulant) may not yield the same results [80–82]. Some methods studies have assessed time of day of blood collection and season, season, hours of fasting, physical activity, NSAID use, tobacco use [and time since last cigarette], and alcohol consumption had variable results, and so given the large metabolic impact of these parameters and this did not apply to all types of features, careful study design should assess these as possible confounders [52,83–85]. Gender, race, ethnicity, and age are other confounders that should be assessed (because nicotine metabolism differences by race, metabolism decreases with aging, and gender differences may be related to hormonal metabolism) [84,86]. Specimen and laboratory processing should be standardized within studies to increase the accuracy and validity of metabolite measurements [87,88]. Repeated freeze-thaw cycles should be avoided, although this might only affect <3% of features [89]. Another study indicated that storage temperature can affect some features, and that a serum glutamate/glutamine ratio greater than 0.20 as a biomarker of storage at −20 °C vs − 80 °C [90].

Two studies for smoking cessation were identified. Goettel and coworkers (2017) assesses the urine, plasma and saliva among 39 males before and after 3 months of smoking cessation [91,92]. There were 26 altered features in plasma, 20 in saliva, and 12 in urine, including those in fatty acid and amino acid metabolism. Liu and coworkers (2021) studied smokers switching to an e-cig for 5 days, assessing plasma and urine (the number of subjects was unclear from the manuscript) [93]. A 30-metabolite signature was reported that could distinguish smoking from e-cig use, many of which were related to smoking xenobiotics or nicotine-derived metabolites. No non-smoking controls were used and so it is unknown how other features reflect smoking cessation and reversal of smoking effects or features that result from e-cig use.

The blood of ex-smokers and never-smokers were compared in a cross-sectional study of 252 subjects by Liang, et al. [94]. No features were statistically different among the groups, suggesting the application of metabolomics in tobacco cessation studies. A larger study of 1252 participants incorporating the assessment of smoking-related weight loss where eight plasma xenobiotics were associated with former smoking and 22 xenobiotics and 94 endogenous metabolites were significantly associated with current smoking, including α-ketobutyrate, homoarginine, β-cryptoxanthin, 1-linoleoyl-GPE (18:2), 1-Palmitoyl-GPE (16:0), 1-Stearoyl-2-arachidonoyl-GPE (18:0/20:40 and 2 sphingomyelins [95].

Du Toit et al., (2022) studied the urine of healthy blacks and whites aged 20–30 (n = 363 smokers and 166 controls) in a cross-sectional analysis reporting statistical differences for arginine, asparagine, glycine, serine, glutamine and other amino acids [96]. Dator and coworkers (2020) conducted a small urine study of 60 African American and white smokers reporting global differences for the metabolism of carbohydrates, amino acids, nucleotides, fatty acids, and nicotine, with known differences for nicotine degradation pathway (cotinine glucuronidation) [97].

Lipidomics

The study of lipids and lipid homeostasis by lipidomics is another emerging omics approach [98–100]. Lipids are the major component of the biological membranes serving as a barrier to the cell body, and are critical to membrane protein functions, storing energy, signal transduction, cell growth, differentiation, and apoptosis. Altered lipid homeostasis has been implicated in carcinogenesis and respiratory disease pathways [101–104]. Surfactant, which is more than 90% lipids, plays a role in lung immune regulation [102]. There are eight major eight major categories of lipids: a) fatty acyls, b) glycolipids, c) glycerophospholipids, d) sphingolipids, e) sterol lipids, f) prenol lipids, g) saccharolipids, and h) polyketides [101,105,106]. Like other omics, workflow assays and analytics can be challenging and newer methods are being evaluated [107–114]. Hoffman and coworkers (2022) has published a comprehensive review of available methods [115], and Zullig and coworkers (2020) has summarized sample preparation methods [116]. Liakh and coworkers (2020) also reviewed methods for lipidomics in studies of obesity as a unique consideration given the impact on lipids by obesity [117]. Given the newness of this approach, a major limitation is lack of common nomenclature, and robust databases for feature identification standardized reporting [115,118].

Very few studies are available for assessing lipidomics by smoking status. One study assessed sputum in 17 subjects after 2 months of smoking cessation [119]. They found for 17 smokers (with and without COPD) that 28 lipids were decreased after smoking cessation (3 ceramides, 6 dihydroceramides, and 17 GSLs). And they also identified solanesol and its esters in the sputum, which is derived from tobacco. This indicates that solanesol could be used as a marker of compliance in smoking cessation, although there are better methods with easier to obtain biospecimens.

Middlekauff and coworkers (2020) conducted a cross-sectional analysis of plasma in 119 smokers, e-cig users and non-users [120]. While the authors concluded that most features were no different among groups, cholesterol esters, ceramides and hexosylceramides were different for the smokers and the other groups, and e-cig users were much more similar to the non-users. T’Kindt and coworkers (2015) examined sputum in 20 heathy smokers and 14 never smokers [121]. While most lipids were not different by smoking status, several prenol lipids were associated with smoking, and they also detected solanesol esters.

Discussion

Biomarkers for prvention studies assess disease risk for individual tobacco products compared to others or those who have never used tobacco products. Tobacco cessation studies are frequently used to assess compliance and the reduction of tobacco toxicants. The only validated biomarkers for tobbacco use are those that are chemically-specific carcinogens metabolites as biomarkers of exposure. In contrast, there are no validated biomarkers of effect or harm (e.g., disease risk) for tobacco use, cessation, or later return to smoking. Omics approaches assess effect and harm, but the field is still early and research is applied only to small studies, and essentially not to intervention cessation studies. Also emerging for early detection and guiding therapeutics are “liquid biopsies”, for example with circulating DNA and extracellular vesicles. The potential for these as tobacco disease risk biomarkers or prediction of smoking cessation or later relapse has not been studied. Omics assays have the potential to uncover a much deeper understanding of biology and disease, but also have the considerable cost of added complexity in terms of both analytics and interpretation. While they are technically feasible to conduct in a variety of biospecimens, the analytics are complex and continuously developing. An additional challenge is the interpretation of alternations on future disease risk. For example, differential gene expression might be an adverse toxic response, a pro-active response to prevent toxicity, or neither.

Long term tobacco cessation, including for e-cigs, is a consistent goal to improve individual and population health. Pharmacological and behavioral methods to reduce tobacco use continue to evolve and predictive markers for tailoring cessation therapies are sorely needed. Also, biomarkers for future disease risk are needed. In both cases, omics methods have the potential to become validated biomarkers for cessation (and later return to smoking) and risk given their broad assessment of biological function, and are agnostic at the outset to specific biological pathways. Combining omics platforms as an integrated approach might be informative than using any single omics technology. However, the added complexity is substantial, becomes more problematic for small studies, and sometimes more data is not more helpful, especially if it is bad data. Having larger datasets for small studies may increase false discovery rates reducing chances for discovery of true associations and provide data for more features with unclear biological function.

There are well established compliance biomarkers for complete cessation for nicotine products by measuring nicotine and nicotine metabolites, and for tobacco products compared to non-tobacco products (e.g., tobacco alkaloids). For products such as e-cigs switching to other nicotine products, such as nicotine replacement therapy, there are no validated biomarkers of compliance. It may be that such can be discovered using a metabolomic approach. An important gap in the literature are studies for the omics discussed herein that may predict smoking reinitiation, and then consequently having tailored approaches to relapse prevention.

The various omics studies demonstrate wide interindividual variation among smokers and never-smokers, which are due to exogenous factors (diet, lifestyle, medications) and endogenous factors (age, gender, race, ethnicity, and comorbidities). Another important consideration for inter-individual variation are host and heritable traits. While for the general population, genetic predisposing genes are typically low penetrant, specific genetic traits might be important for omic analysis because of wide variation in xenobiotic metabolism. Some aspects of interindividual variation may be related to technical issues such as differences for sample collection (time of day, fasting, season), batch effects for sample process, and the reproducibility of the technologies.

While there has been a variable number of tobacco studies across the omics, currently there is a heterogeneity of results that preclude finding consistency in results for individual features or genes. This is due both to the variability in the analytic methods that identify features not consistently detectable across platforms or are identified in bioinformatic pipelines, small number of studies, limitations in sample size, differences in biospecimen collection or due to the human subject study design. Also, results from the same platform may vary by the bioinformatic methods and pipelines for data analysis that provide differing results. Heterogeneity also exists for study design, and the best evidence when applied to clinical trials for cessation are non-existent. Thus, there are major research gaps that may need to be addressed before any omics approach can be useful clinically for tailoring cessation therapies or prediction of future disease risk. The gaps need to address both the technical and clinical validation, as discussed above. Also, devoid in current studies are considerations of outcomes based on race and ethnicity (measured by self-report or through genetics), and thus there may be substantial health disparities that are not yet considered.

Data availability

No data was used for the research described in the article.