Total Exposure Study Analysis consortium: a cross-sectional study of tobacco exposures

Scientists employed by the tobacco company sponsor have published analyses in peer-reviewed scientific journals using data from the TES pilot study [8] and the TES main study [2, 3, 9‐16]. Analyses included population estimates of BOE levels for smokers and non-users [2], estimates of levels of BOPH in smokers and non-users [3], the relationships between machine-derived tar yields of cigarette products and BOE in smokers [9], models of BOPH [11], the impact of menthol-containing cigarettes on selected BOE in White and Black smokers [10], and the relationships between selected BOE and BOPH in smokers [14]. These scientists have also reported on the relationships between BOE and nicotine dependence [13] and between nicotine and carbon monoxide BOE and other factors, including smoking topographical variables [12]. These authors utilized TES data to examine the relationships between smoking mentholated cigarettes or non-mentholated cigarettes and glucuronide metabolite ratios [15], and with measures of nicotine dependence [16]. We review Roethig et al. [2] and Frost-Pineda et al. [3] here to introduce TES BOE (Additional file 2: Table S1) and BOPH (Additional file 3: Table S2).

Roethig et al. published estimates of BOE (Additional file 2: Table S1) in smokers and non-users and, within smokers, within different age, sex, BMI, and self-identified racial strata [2]. Mean levels of BOE were weighted by age, sex and BMI variance estimates from the U.S. Behavioral Risk Factor Surveillance System (BRFSS), an annual telephone-based behavioral survey established in 1984 [17], to produce weighted estimates of BOE reported and described by Roethig et al. as population estimates [2]. The BRFSS used post-stratification weighting based on United States Census data from the 1980s until 2011 [17]. Lee and Messiah criticized the application of weights extracted from a nationally representative sample to a sample for which inclusion rates at recruitment sites were not known or not reported [18]. In response, Sarkar and Liang noted that the weighted means were similar to or unchanged from unadjusted means [19]. Weighted estimates of tobacco-specific biomarkers [nicotine, cotinine and trans-3′-hydroxycotinine and their glucoronides (nicotine equivalents, NE), serum cotinine, and total 4-(methylnitrosamino)-1-(3-pyridyl)-1-butanol and glucuronide (total NNAL)] suggested that the younger participants (21–34 years) and female participants had the lowest tobacco-specific exposures, and that individuals with BMIs < 25 kg/m² compared with individuals with BMIs ≥ 25 kg/m² had higher serum cotinine levels and lower total NNAL levels, suggesting reduced cotinine clearance and NNK metabolism in heavier individuals [2]. Significant differences in serum cotinine by BMI similar to those reported in the TES have been previously observed in the National Health and Nutrition Examination Survey [20]. In the TES, self-identified White smokers smoked significantly more cigarettes per day and had greater NE and total NNAL exposure over 24 h, but lower NE and total NNAL exposure per cigarette, and lower serum cotinine exposure, than self-identified Black smokers [2]. It has previously been observed that White smokers smoke more cigarettes per day than Black smokers and that nicotine intake per cigarette measured by serum cotinine is higher in Black smokers than in White smokers [21, 22], which is related to significantly reduced nicotine clearance in Blacks compared to Whites [23, 24].

Frost-Pineda et al. [3] reported mean values for 29 BOPH (Additional file 3: Table S2) in both smokers and non-users. The BOPH represented various physiological functions: cardiovascular, endothelial, hematologic, inflammation, lipid, hepatic, renal, respiratory, metabolic, and oxidative stress [3]. The effects of multiple BOE [cigarettes per day (CPD), NE, and smoking duration] on the BOPH in current smokers versus non-users were evaluated in two stepwise regression models (model A with CPD and smoking duration, and model B with NE and smoking duration) with age, sex, BMI and self-identified race as additional independent variables [3]. The three most elevated mean BOPH in current smokers versus non-users were those reflecting oxidative stress, platelet activation and inflammation. The oxidative stress biomarker 8-epi-prostaglandin F_2α exhibited the largest difference between smokers and non-users (+42 %), while BMI and age, and BMI and NE, were the most important correlates in models A and B, respectively. The platelet activation biomarker 11-dehydrothromboxane B₂ exhibited the second largest difference between smokers and non-users (+29 %), and sex and BMI, and sex and NE were the most important correlates in models A and B, respectively. The inflammation biomarker white blood cell count exhibited the third largest difference between smokers and non-users (+19 %) and BMI and self-identified race were the most important correlates in both models. Overall, BMI and sex were the first and second most common significant correlates reported by Frost-Pineda et al. [3].

Our search of the UCSF LTDL identified multiple documents we had received from the VTHRR, including: the Amended Final Research Protocol, dated 19 August 2002, that describes the clinical protocol, laboratory testing and biospecimen banking procedures to be conducted by the primary clinical and laboratory CRO [1]; the TES Adult Smoker Survey [25]; and the TES Adult Non-Smoker Survey [26]. We found no differences between the documents we had received from VTHRR and those available on the UCSF LTDL. We also identified summary documents that provided information on the design and analysis goals of the TES, including: a Statement of Work for data management and analysis to be conducted by the primary data analysis CRO dated 30 April 2004 [27]; a draft version of the Statistical Analysis Plan dated 1 September 2004 [28]; and a PowerPoint presentation dated 7 February 2005 that presented TES pilot results, and design and initial analyses of the TES [29]. Review of these documents enriched our understanding of the design and conduct of the study and confirmed study parameters, e.g., numbers of individuals recruited within design strata. With the assistance of a UCSF Industry Documents Digital Librarian, we identified SAS datasets available in the Philip Morris collection but these refer to an unrelated study [30].

The original enrollment goal of the TES [1] was 1000 smokers among four strata defined by FTC tar levels of the smoker’s usual cigarette (≤2.9, 3–6.9, 7–12.9, and ≥ 13 mg), and 1000 non-users [1]. The distribution of evaluable subjects in the five categories (504, 953, 1066, and 1062 smokers, and 1077 non-users) was significantly different from the design (Pearson χ²_4d.f. = 159.9, P < 0.0001). The distribution of participants with clinical data by enrollment strata and by demographic strata is shown in Table 1.

The demographic composition of TES participants with clinical data (N = 4662) was 57.9 % female, with mean (standard deviation, SD) age 42.1 (13.2) years and mean (SD) BMI 27.9 (6.7) kg/m² (Table 2). Self-identified race distributions were 77.1 % “Caucasian or White”, 16.5 % “African American or Black”, and four other self-identified race categories comprising 6.5 % of participants. Only a small fraction of TES participants self-identified as Hispanic ethnicity (3.8 % of total participants). Most (76.9 %) TES participants were current smokers with mean (SD) CPD of 16.0 (8.9). Age, self-identified race, and education distributions differed significantly by smoking status (current smokers were significantly more likely to be older, self-identify as Black, and significantly less likely to have a college degree), while sex, BMI, and self-identified ethnicity (Hispanic versus Not Hispanic) did not differ by smoking status.

Two-thirds (66 %) of TES participants have banked biospecimens. PBMCs are the most common biospecimen type while urine is the least common (Table 3). Among participants with banked biospecimens, and among the four biospecimen types, there are no significant differences in sex, age and BMI proportions, but there are significant differences in smoking status (Table 4). Compared to participants with banked PBMC biospecimens, participants with banked urine biospecimens are significantly more likely to be smokers (OR = 1.26, 95 % CI 1.11–1.44, P = 0.0004).

Participants with banked biospecimens are significantly older (age, continuous or categorical), have significantly increased BMI (continuous), and are more likely to self-identify as White compared to those without (Table 2). When stratified by ethnicity and race, self-identified non-Hispanic Black participants with banked biospecimens are significantly older and have significantly increased BMI than those without biospecimens [mean (SD) age 40.8 (10.7) vs 38.8 (11.1) years, t = 2.45, P = 0.0144, N = 755; mean (SD) BMI 30.5 (8.0) vs 28.5 (6.9) kg/m², t = 3.58, P = 0.0004, N = 755, data not shown]. Significant differences in age and BMI among all participants and stratified by self-identified ethnicity and race are small (Cohen’s d-values = 0.10, 0.09, 0.18 and 0.27, respectively). Smoking duration, CPD (continuous and categorical), and usual cigarette FTC tar level (categorical) are significantly increased in those with banked biospecimens compared to those without, overall, and when stratified by self-identified race (Table 5). Significant differences are small; d-values for smoking duration overall, and among self-identified non-Hispanic Blacks and Whites are 0.13, 0.22 and 0.08, respectively, and d-values for CPD among self-identified non-Hispanic smokers, and among self-identified non-Hispanic White smokers, are 0.14 and 0.09, respectively.

Most tobacco-specific (NE, serum cotinine and total NNAL) and non-specific BOE are significantly higher in smokers with banked biospecimens than in smokers without, except for serum cotinine, 4-ABP and MHBMA (Table 6). Metabolites of acreolein and 1,3 butadiene are significantly greater in non-users with banked biospecimens than in non-users without. All statistically significant differences in BOE by banked biospecimen availability have small effect sizes, ranging from 0.10 to 0.24. When stratified by self-identified ethnicity and race, more BOE differ significantly by biospecimen availability among non-Hispanic Whites than among non-Hispanic Blacks (Tables 7 and 8). The effect sizes of the two BOE differences in self-identified non-Hispanic Black smokers are small, and the effect size of the one BOE difference in self-identified non-Hispanic Black non-users is a medium effect size (d = 0.47). Among self-identified non-Hispanic White smokers, NE, total NNAL, carboxyhemoglobin and an acreolin metabolite, and among self-identified non-Hispanic White non-users, a 1,3 butadiene metabolite, exhibit significant differences. All these significant differences are of small effect size.

The distribution of BOPH by banked biospecimen availability is shown in Table 9. Six of 29 BOPH measures have nominally significantly higher levels in TES participants with available banked biospecimens versus those without, while the respiratory function measure FVC and hemoglobin remain significantly different after false discovery rate correction (q-values = 0.0128 and 0.0496, respectively) [31]. After excluding individuals with implausible FEV₁ values < 35 % or > 125 % of predicted, as suggested by Frost-Pineda et al. [3], and then stratifying by self-identified ethnicity and race, and then by smoking status, we observed that self-identified non-Hispanic White smokers with banked biospecimens have significantly increased FVC compared to those without [93.9 (23.2) vs 90.8 (17.9), t = 3.81, P = 0.0001, N = 2584]. The statistically significant increase in % predicted FVC among self-identified non-Hispanic White smokers with available biospecimens is unexpected because multiple BOE are significantly increased in self-identified non-Hispanic White smokers with banked biospecimens and lung function is expected to be reduced in individuals with increased measures of exposure. Evidence for the influence of current smoking on longitudinal decline in FEV₁ and FVC suggests that current smoking influences longitudinal FEV₁ decline more than FVC [32], though this would not explain an increase in FVC. We constructed another regression model including education and household income, but these potential confounders [33] had no effect on the observed differences in FVC within non-Hispanic White smokers (data not shown). Further analyses of lung function measures and other variables in the TES may identify possible explanatory factors or confounders. After stratifying by self-identified race and ethnicity, and then by smoking status, we observed that self-identified non-Hispanic White non-users exhibit a significant difference in hemoglobin by banked biospecimen availability [14.50 (1.42) vs 14.31 (1.25), t = 1.84, P = 0.033, N = 828]. Statistically significant differences in FVC and hemoglobin in these strata are small (d-values are 0.15 and 0.14, respectively).

Information on participant’s usual cigarette brand is available from 606 and 1336 self-identified non-Hispanic Black and non-Hispanic White smokers, respectively. The top 20 brands account for 66.0 and 49.4 % of the brand information available from self-identified non-Hispanic Black and non-Hispanic White smokers, respectively (Table 10). Usual cigarette brand distributions do not differ significantly among self-identified non-Hispanic Black or among non-Hispanic White smokers by the presence or absence of banked biospecimens (Table 10).

Sample sizes among self-identified non-Hispanic Black and White smokers with complete data and with imputed data for the progressively more complex models were 3236 and 3318 (2.5 % of participants had missing data in Model 1), 2317 and 3318 (30.2 % of participants had missing data in Model 2), and 1090 and 3053 (64.3 % had missing data in Model 3), respectively. However, while a large fraction of the population was missing one or more variable values, on average they were only missing a single value out of a large number of independent variables. The number of missing values that were imputed was relatively small; 0.3 % of all values required imputation in Model 1, 1.9 % in Model 2, and 2.5 % in Model 3. Significant demographic variables, BOE and BOPH in progressively more complex multivariate models of banked biospecimen availability with imputed data were: Model 1) BMI, self-identified race, age and age squared; Model 2) self-identified race, age, age squared, and NE/24 h; and Model 3) self-identified race, age, age squared, NE/24 h, serum cotinine, MHBMA and FVC (Table 11). The mean (SD) predicted probabilities of banked biospecimen availability, in progressively more complex multivariate models without and with imputed data are: Model 1) 0.647 (0.069) and 0.665 (0.060); Model 2) 0.643 (0.077) and 0.668 (0.070); and Model 3) 0.625 (0.102) and 0.669 (0.095). Explanatory power estimates (r²) of the anthropometric, demographic, BOE and BOPH variables in progressively more complex multivariate models with imputed data among self-identified non-Hispanic Black and White smokers to predict banked biospecimen availability are 0.018, 0.024, and 0.037, respectively. In permutation analyses of self-identified non-Hispanic Black and White smokers with imputed data in Model 3, the mean (95 % confidence interval) r² was 0.020 (0.016 - 0.023) suggesting that about half of the explanatory power of variables is due to random variability (0.020/0.037 = 0.54).

Correlations of clinical chemistry results in 47 plasma samples from the SRI CAL (2013) and those from serum reported by the CRO (2002–2003) were high and statistically significant [glucose (0.922), aspartate aminotransferase (0.993), alanine aminotransferase (0.997), total bilirubin (0.960), albumin (0.702), and total cholesterol (0.913), all p-values < 0.001] (Table 12 and Fig. 1). The lower correlation for blood albumin may be due to the two different matrices, the increased variance of some albumin clinical chemistry analysis methods [34], or the use of different methods in the clinical analyzers in the two different clinical chemistry laboratories.

Mean (SD) DNA and RNA from ~1 M cells was 4.63 (1.63) ug, and 2.10 (0.62) ug, respectively. We sent four DNA samples from Gentra Puregene extraction and 27 DNA samples from NORGEN extraction for Smokescreen Array genotyping at the Rutgers University Infinite Biologics facility. All DNA samples had genotype completion rates ≥ 97.5 % and passed the 97 % rate threshold; the mean genotype completion rate was 99.4 %. Mean (SD) RNA Integrity (RIN) scores from 28 RNA samples analyzed were 6.4 (2.2); 68 % of RIN scores were ≥ 6.0, a standard used in RNA sequence analysis [35]. There were no significant differences in sex, race, smoking status, or total nicotine equivalents between RNA samples with RIN ≥ 6.0 and < 6.0 (all p-values > 0.12). Thus, from PBMC pellets frozen at ultralow temperatures for over a decade, DNA quality and genotyping results were excellent, while RNA quality was good, but requires evaluation using transcriptome-wide methods.

Finally, we assessed statistical power to detect a priori genetic loci of interest from an example of a large-scale (1000 s) candidate gene association scan, and an example from a locus nominated by genome-wide association scans (GWAS), with genome-wide significance (GWS) as the statistical threshold. For self-identified non-Hispanic Black current smokers, we selected rs11187065 as an example, identified in the insulin-degrading enzyme gene as the gene-centric SNP most significantly associated with serum cotinine in the Coronary Artery Risk Development in Young Adults (CARDIA) study by Hamidovic et al. [36]. The influence of rs11187065 on serum cotinine was substantial with a β of −85.1 ng/ml, with mean (SE) of 236.5 (8.1) ng/ml from 365 African American smokers [36]. Mean (SD) CPD in the CARDIA sample was 10.5 (7.4), similar to that of the TES (Table 4). Using the sample size of self-identified non-Hispanic Black smokers with banked PBMCs in the TES (N = 340), there is 57 % power to detect the locus at genome-wide significance (and 83 % power to detect this locus at the original study’s Bonferroni adjustment level of 2.3 × 10⁻⁶) using an additive model, a one-sided test, the mean (SD) of serum cotinine among self-identified non-Hispanic Black smokers (Table 6), the rs11187065 minor allele frequency of 0.083 in the HapMap [37] African Americans in the Southwest sample, and the effect size from Hamidovic et al. For assessing power to detect a priori loci of interest among self-identified non-Hispanic White current smokers, we selected rs1051730 in the nicotinic acetylcholine receptor (nAChR) subunit gene cluster on chromosome 15q25.1, associated with smoking intensity and related phenotypes [38], including cotinine level [39], as an example. In an analysis of 2932 smokers with serum or plasma cotinine estimates, Munafo et al. estimated that each minor allele contributed to a mean increase in the unadjusted level of cotinine in European ancestry samples of 138.72 nmol/L [(95 % CI) 97.91 - 179.53 nmol/L, P = 2.7 × 10^-11] [39], or 24.42 ng/mL, although this was reduced 18 % upon adjustment for self-reported CPD. Using the sample size of self-identified non-Hispanic White smokers with banked PBMCs in the TES (N = 1840), there is 70–96 % power to detect this locus at a genome wide significance level (5 × 10⁻⁸) using an additive model, a one-sided test, the mean (SD) of serum cotinine among self-identified non-Hispanic White smokers (Table 7), the rs1051730 minor allele frequency of 0.385 in HapMap Utah Residents with Northern and Western European Ancestry sample, and estimated allele effect sizes of Munafo et al. (adjusted and unadjusted for CPD).