Comparison of Liquid Chromatography Mass Spectrometry and Enzyme-Linked Immunosorbent Assay Methods to Measure Salivary Cotinine Levels in Ill Children

Objective: Cotinine is the preferred biomarker to validate levels of tobacco smoke exposure (TSE) in children. Compared to enzyme-linked immunosorbent assay methods (ELISA) for quantifying cotinine in saliva, the use of liquid chromatography tandem mass spectrometry (LC-MS/MS) has higher sensitivity and specificity to measure very low levels of TSE. We sought to compare LC-MS/MS and ELISA measures of cotinine in saliva samples from children overall and the associations of these measures with demographics and TSE patterns. Method: Participants were nonsmoking children (N = 218; age mean (SD) = 6.1 (5.1) years) presenting to a pediatric emergency department. Saliva samples were analyzed for cotinine using both LC-MS/MS and ELISA. Limit of quantitation (LOQ) for LC-MS/MS and ELISA was 0.1 ng/mL and 0.15 ng/mL, respectively. Results: Intraclass correlations (ICC) across methods = 0.884 and was consistent in sex and age subgroups. The geometric mean (GeoM) of LC-MS/MS = 4.1 (range: < LOQ to 382 ng/mL; 3% < LOQ) which was lower (p < 0.0001) than the ELISA GeoM = 5.7 (range: < LOQ to 364 ng/mL; 5% < LOQ). Similar associations of cotinine concentrations with age (β^ < −0.10, p < 0.0001), demographic characteristics (e.g., income), and number of cigarettes smoked by caregiver (β^ > 0.07, p < 0.0001) were found regardless of cotinine detection method; however, cotinine associations with sex and race/ethnicity were only found to be significant in models using LC-MS/MS-derived cotinine. Conclusions: Utilizing LC-MS/MS-based cotinine, associations of cotinine with sex and race/ethnicity of child were revealed that were not detectable using ELISA-based cotinine, demonstrating the benefits of utilizing the more sensitive LC-MS/MS assay for cotinine measurement when detecting low levels of TSE in children.


Introduction
Biochemical verification is the gold standard used to validate levels of tobacco smoke exposure (TSE) in children and the tobacco use status of smokers in clinical trials [1][2][3][4][5][6]. Since pediatric TSE

Subjects and Biological Samples
Participants were 218 children of parental smokers enrolled in a randomized controlled trial (RCT) of a smoking cessation intervention who presented to the Pediatric Emergency Department (PED) or Urgent Care (UC) of a midwestern Children's Hospital; details are described elsewhere [31]. Adult participants were parents or legal guardians of children 0-17 years of age who presented with a potentially TSE-related complaint (e.g., cough); enrollment occurred for a 28-month period beginning in April 2016. We included a convenience sample of participants from this study whose saliva samples were previously analyzed with ELISA techniques as part of the RCT. Leftover saliva samples from these participants were analyzed using LC-MS/MS. A total of 218 saliva samples were analyzed from children obtained at either the baseline PED/UC visit (T0; 203 participants) or during a 6-week home follow-up visit (T1; 15 participants); a single sample was tested per participant at either timepoint.

Measures
Parents completed electronic assessments that included demographics, TSE patterns, and type of home (e.g., multiunit housing). This study was approved by our hospital's Institutional Review Board. Parental consent and child assent on children 11 years of age or older was obtained.

Chemical Analyses
Saliva was tested for cotinine and by the University of Minnesota using LC-MS/MS with isotope dilution and by Salimetrics LLC using ELISA techniques. The LOQ for the ELISA assay was 0.15 ng/mL [32]. For LC-MS/MS, salivary cotinine was analyzed as previously described [33] with the exception that the analyses were performed on a Luna C18 column. Briefly, the methods used were 96-well plate-based liquid chromatography-tandem mass spectrometry assays. Saliva cotinine was quantified by the area ratio of the analyte to the deuterated standard using specific MS/MS transitions. Calibration curves were established before each set of LC-MS/MS analyses. The same sample of deuterated analyte was used for constructing the calibration curve as that added to each sample as internal standard. Lower LOQ was calculated based on three times the noise measured in an extracted blank and was 0.1 ng/mL.

Statistical Analyses
Summary statistics of cotinine levels obtained with LC-MS/MS assays were compared to levels obtained with ELISA. Geometric means (GeoM) and standard deviations were calculated to account for the right skew of the data.

Mixed Model
We used a one-way random effect mixed model comparing the final LC-MS/MS and ELISA-based cotinine measurements to calculate the intra-class correlations (ICCs) and to assess the variability attributed to the two methods. We also added the median relative percent difference (RPD) value to evaluate reproducibility in the presence of duplicate samples. Following the Environmental Protection Agency's RPD guidelines, a median RPD less than 30% (or 0.3) indicates high quality of reproducibility [34]. For these mixed model analyses, only those samples with cotinine measurements above the LOQ for both methods were included (n = 203). For the LC-MS/MS and ELISA-based cotinine measurements, we had batch-to-batch replicate measurements for 20 and 55 participants, respectively, that were utilized to calculate the internal ICC and RPD for each method. For all mixed models, the cotinine concentrations were natural log (ln)-transformed to account for the right skew of the cotinine data. In follow-up analyses, models were stratified by sex and age group (0-6 and 7-17 years old).

Paired Sample t-Test
A comparison of GeoMs for the LC-MS/MS and ELISA detected cotinine measures was performed by way of a paired sample t-test. As with the mixed models, we only included samples with both LC-MS/MS and ELISA cotinine measurements >LOQ (n = 203). All cotinine values were ln-transformed prior to analysis.

Linear Regression Models
Associations of cotinine at T0 with explanatory variables were determined by way of linear regression models for each detection method. In order to ensure an accurate comparison of methods, only those samples with cotinine measured in both LC-MS/MS and ELISA at T0 were included in the models (n = 203); samples taken at the six-week timepoint (n = 15) were excluded from these analyses. Continuous explanatory variables included age of child (years), number of cigarettes smoked per day by primary caregiver, and number of cigarettes smoked per day around child by all household members. Categorical explanatory variables included child sex and race/ethnicity, type of home, and caregiver's income. To account for the lower number of subjects in the Hispanic White and Black groups, race/ethnicity categories were collapsed to only include 'non-Hispanic White', 'non-Hispanic Black', and 'other'. Similarly, categories for caregiver's income were collapsed for income groups above $75,000. A total of six samples were dropped from the model due to missing explanatory variable data.
In contrast to the mixed model analyses, for the linear regression analyses, all values below LOQ were imputed as LOQ/ √ 2 using the detection method-specific LOQ values for both LC-MS/MS and ELISA-detected cotinine (3% and 5% of values imputed, respectively). Cotinine measurements were natural ln-transformed to account for the right skew of the data. An alpha of 0.05 was the criterion for statistical significance. All statistical analyses were conducted with SAS 9.4.

Demographics and TSE Patterns
Participants (N = 218) were mostly non-Hispanic Black (55.1%) followed by non-Hispanic White (36.4%), and other race/ethnicity (8.4%); sex was evenly represented (51.4% female); mean age (SD) of child was 6.1 (5.1) years. A total of 37.3% of caregivers had an annual income of ≥ $15,000 and 47% lived in a single-family home. The mean (SD) number of cigarettes smoked by the primary caregiver and the total number of cigarettes smoked around the child was 9.8(6.0) and 9.1(16.7) cigarettes, respectively.

Cotinine Measurements and Comparisons of Analyses by LC-MS/MS and ELISA Distribution of Cotinine by LC-MS/MS Compared to ELISA
Cotinine was detected in 97% (n = 211) of saliva samples by LC-MS/MS and 95% (n = 208) of samples by ELISA. The LC-MS/MS assay showed higher sensitivity than the ELISA assay, with LOQs for each assay of 0.1 ng/mL and 0.15 ng/mL, respectively. Of the seven measurements < LOQ for LC-MS/MS, five were detectable by ELISA. Of the 10 measurements below LOQ by ELISA, eight were detected by LC-MS/MS. The GeoM for the LC-MS/MS assay was 4.1; median (Mdn) = 4.3 ng/mL; Q1 = 1.3 ng/mL; Q3 = 10.3 ng/mL; range <LOQ to 382 ng/mL. The GeoM for cotinine measured by ELISA was 5.7 ng/mL; Mdn = 5.1 ng/mL; Q1 = 2.2 ng/mL; Q3 = 12.2 ng/mL; range < LOQ to 364 ng/mL (Table 1). Based on samples with both LC-MS/MS and ELISA measured cotinine values > LOQ (n = 203), the GeoM for ELISA measured cotinine was significantly higher than LC-MS/MS measured cotinine (p < 0.0001). The distributions of the ln-transformed cotinine measurements were similar across method of detection ( Figure 1). Scatter plots of LC-MS/MS-measured cotinine and ELISA-measured cotinine for samples with valid measurements in both methods showed that the measurements were largely in agreement (slope = 1.03, intercept = −0.36, R-square = 0.839; Figure 2). The negative intercept and slope >1 indicate that ELISA values tended to be higher than LC-MS/MS values, and this is more notable at lower levels. Stratification by age group and sex showed a similar trend that was more pronounced in in older subjects (7-17 years of age) and females. notable at lower levels. Stratification by age group and sex showed a similar trend that was more pronounced in in older subjects (7-17 years of age) and females. Given the high range of cotinine values, we ran a sensitivity analysis excluding the top 5% values which was greater than 41.5 ng/mL for LC-MS/MS or greater than 41.2 ng/mL for ELISA. This reduced the sample size from 203 to 190. Of note, eleven of these thirteen children were under age five, thus, they were nonsmokers. There were no differences in the distributions or significance levels for Figure  1 or Figure 2 when we excluded these values.   Given the high range of cotinine values, we ran a sensitivity analysis excluding the top 5% values which was greater than 41.5 ng/mL for LC-MS/MS or greater than 41.2 ng/mL for ELISA. This reduced the sample size from 203 to 190. Of note, eleven of these thirteen children were under age five, thus, they were nonsmokers. There were no differences in the distributions or significance levels for Figure 1 or Figure 2 when we excluded these values.

Internal Consistency of LC-MS/MS and ELISA
Despite these areas of discrepancy, the calculated ICC indicates strong agreement of the LC-MS/MS and ELISA cotinine measurements with an overall ICC of 0.884 across methods and an ICC >0.82 and a median RPD <0.16 for all sex and age subgroups both across and within methods ( Table 2).

Internal Consistency of LC-MS/MS and ELISA
Despite these areas of discrepancy, the calculated ICC indicates strong agreement of the LC-MS/MS and ELISA cotinine measurements with an overall ICC of 0.884 across methods and an ICC >0.82 and a median RPD <0.16 for all sex and age subgroups both across and within methods ( Table  2).

Differences in Cotinine Results by Demographics, TSE, and Home Characteristics
Next, we assessed associations of cotinine in saliva at T0 with explanatory variables for samples assessed for cotinine by both LC-MS/MS and ELISA (n = 197) and compared the results (Table 3). Both models showed significant negative associations of age (β < −0.09, p < 0.0001) and positive associations of caregiver's reported daily cigarette intake with cotinine (β > 0.07, p < 0.0001). Looking at the overall effect of categorical variables while controlling for covariates, there were significant effects of caregiver's income with measured cotinine in saliva (p < 0.02), regardless of detection method. Plotting the residuals of a covariates-only model against income category indicated that the cotinine concentration broadly decreased as the categorical income level increased, regardless of detection method. Differences appeared between the models with LC-MS/MS-measured cotinine and ELISA-measured cotinine when looking at the effects of sex and race/ethnicity, where these effects were only significant in the model utilizing LC-MS/MS-measured cotinine. In the LC-MS/MS-based cotinine model, cotinine was shown to be lower in females than males and higher in non-Hispanic blacks than non-Hispanic whites, based on the covariate-adjusted plot. In separate models testing the possible interaction of race/ethnicity and age group, there were no indications of significant age by race/ethnicity interactions, regardless of cotinine detection method.

Discussion
To our knowledge, this is the first study conducted in a pediatric population to examine and compare salivary cotinine levels obtained with LC-MS/MS assays with levels obtained using ELISA assays. In this study of 218 children who were exposed to varying levels of tobacco smoke from SHS and THS in their environments, we observed that overall levels of cotinine were high.
The main objective of this study was to compare whether LC-MS/MS-based measures of salivary cotinine were comparable to the ELISA-based measures. Given the large differences in costs, turn-around time, and sensitivity and specificity between the two methods, investigators need to be able to weigh the potential trade-offs with each method so that they can assess which method may be more suitable for their planned research outcomes. For example, if investigators need to assess levels of TSE to broadly differentiate nonsmokers from smokers, then highly sensitive measures may not be needed. However, if cotinine levels are going to be used to differentiate those who are exposed to low levels of tobacco smoke, then more sensitive techniques are needed [1,23]. We observed good overall agreement with LC-MS/MS and ELISA in the relative ranking across the entire range of exposure. There were significant differences in the mean cotinine levels we observed with the GeoM for the LC-MS/MS assay of 4.1 compared to the ELISA GeoM of 5.7 ng/mL. These differences may be due to a number of reasons including the cross-reactivity of the ELISA assay with 3HC [12], but not LC-MS/MS, or because the ELISA calibration curves used with the ELISA assay were created based on studies conducted on adult participants [32]. It is possible that the ELISA calibration curves may need to be recalibrated for children.
The results show that both methods are internally highly reliable as demonstrated by ICCs > 0.98. The between method ICC, however, is substantially lower (0.82-0.90) and RPDs below 0.16, indicating that the two methods are subject to different sources of error that affect the observed cotinine levels. Using LC-MS/MS as the gold standard, ELISA cotinine measures may be affected by 10-18% of variance unrelated to actual interindividual differences in salivary cotinine levels. This is further illustrated by inconsistencies between the methods at the low exposure range. Of the seven measurements below LOQ for LC-MS/MS, five were detectable by ELISA. Of the 10 measurements below LOQ by ELISA, eight were detected with LC-MS/MS. Taking the LC-MS/MS measures as the gold standard, ELISA results have high rates of false positive results (five out of seven) at the low exposure range. Overall, these finding support the benefits of LC-MS/MS because of its superior specificity and invariance to cross-reactions and as the method of choice for populations with relatively low levels of exposure to tobacco smoke.
Our secondary objectives were to determine how differences in cotinine values based on child demographics, TSE patterns, and housing types compared between the two methods. Overall, our results were consistent with prior research in that we observed higher cotinine levels using both methods in children who were: younger [35][36][37]; had lower household incomes [36,38,39]; and had higher numbers of cigarettes smoked by their primary caregiver or around them [24,39]. We did, however, observe differences in cotinine associations with sex and race/ethnicity that were only significant in models using LC-MS/MS-derived cotinine levels. We found that cotinine levels were lower in females and higher in non-Hispanic blacks than non-Hispanic whites. Other studies have not reported differences in cotinine by gender [24,35,37,40] and it is unclear why this is the case. However, similar to our findings, several other studies have observed differences by race/ethnicity with higher cotinine levels seen in non-Hispanic blacks. These differences may be due to differences in nicotine metabolism due to genetic variation in the CYP2A6 enzyme [8,11,19]. It appears that a relatively high proportion of African American smokers (15% compared to 1% in Caucasian smokers) have a particular variant in their metabolic pathway that contributes to significantly higher cotinine levels per cigarette [41]. However, it is not clear how these racial/ethnic differences in nicotine metabolism translate to TSE and cotinine levels in children.
Our findings are not without limitations. Our sample consisted of ill children recruited from the PED/UC. While we chose to collect and analyze saliva in our population due to the ease of collection and tolerability compared to collection of plasma or urine, especially in young children, we acknowledge that there are limitations to assessing levels of TSE using saliva. Salivary cotinine concentrations can be affected by age, sex, race, oral pH, type of diet, dehydration, or drug treatment [1]. All of these factors were highly variable in our population of ill children. Nevertheless, other studies have found salivary cotinine levels to be comparable to plasma cotinine levels in adolescent and adult smokers [42,43], although levels are approximately 10% to 40% higher [1,[42][43][44]. Additionally, our saliva collection method (i.e., the use of cotton swabs as recommended by Salimetrics [32]) may have resulted in differential salivary flow rate which also could have affected cotinine levels [42]. Further, the timing of saliva collection may have lowered cotinine levels if saliva was obtained after the child had been in the PED/UC for a while, since cotinine has a mean half-life of 16-18 h [8]. Since we enrolled a convenience sample of participants, our results cannot be generalized to different racial/ethnic groups or other sociodemographic subgroups. Moreover, since our population was highly exposed to tobacco smoke, using ELISA assays on children who had lower levels of exposure or who were predominantly exposed to THS may have resulted in more results below the LOQ. Finally, since cotinine levels obtained using LC-MS/MS on urine samples yield lower LOQs, urine samples may be a better measure of TSE when there are low levels of exposure [1] and studies are needed to compare these salivary cotinine results with urinary cotinine results obtained with both methods.

Conclusions
In conclusion, our results indicate that ELISA is a cost-effective alternative to LC-MS/MS for detecting TSE to classify children into highly exposed versus not exposed. However, LC-MS/MS is a superior method with which to measure cotinine in children with lower levels of TSE. This is evident given our results that the associations of cotinine with sex and race/ethnicity were detected only when cotinine was quantified by LC-MS/MS and because we had fewer values that were below the LOQ with LC-MS/MS. These results demonstrate the benefits of utilizing the more sensitive LC-MS/MS assay for cotinine measurement in children when detecting TSE, especially when low levels of exposure to nicotine via THS or SHS are to be measured. Future research should include validation and calibration studies of ELISA-based cotinine methods for pediatric populations and studies of LC-MS/MS-based measures with a broader range of TSE to better understand the reliability and validity of measuring TSE in age and racially/ethnically diverse populations.