Serum Concentrations of Emerging Vitamin D Biomarkers and Detection of Prevalent High-Risk HPV Infection in Mid-adult Women

High-risk human papillomavirus (hrHPV) infections are causally linked to anogenital and oropharyngeal cancers (1). While approximately 90% of hrHPV infections become undetectable within 1–2 years, the minority that persist are associated with increased risk for cervical cancer (2). Understanding factors that may contribute to acquisition and persistence of infection (i.e., vitamin D deficiency) are critical to efforts to prevent HPV infections and HPV-related carcinogenesis.

Vitamin D sufficiency is associated with improved bone and intestinal health, reduced risk and severity of respiratory and influenza-like infections (3), and lower incidence of certain cancers (4). However, data collected by the population-based National Health and Nutrition Examination Survey (NHANES) in 2005–2006 showed 42% of adults in the United States are vitamin D deficient or insufficient with 25(OH)D ≤20 ng/mL (5). Previous studies have investigated the potential biological pathways by which vitamin D acts as an immune modulator in the body (6, 7). Lower concentrations of serum 25(OH)D may lead to diminished innate immune function and increased susceptibility to and duration of infection. Specifically, insufficient and deficient levels of vitamin D may limit the body's ability to produce and regulate the expression of antimicrobial peptides (AMP; ref. 8), which inactivate pathogens and boost the innate immune response by attracting phagocytes to the site of infection (9). AMPs are produced in endothelial cells in the vagina and the cervix, where they can specifically defend against sexually transmitted infections (10, 11). In response to HPV, certain AMPs can eliminate the virus before it establishes infection in the host (12). Optimal AMP production and activity (associated with sufficient serum concentrations of 25(OH)D; ref. 13) may be critical in bolstering the innate immune response to HPV and, thus, defend against infection with a virus capable of evading adaptive immune response (14).

Metabolism of vitamin D occurs in the bloodstream, kidney, and liver; this process generates several biomarkers indicative of an individual's vitamin D status (15, 16). Most epidemiologic studies have estimated vitamin D status by measuring serum concentrations of total 25-hydroxyvitamin D, or 25(OH)D, but this biomarker is only one means (albeit a stable one; ref. 17) of characterizing vitamin D status (18). Additional biomarkers, including (i) calcitriol [the hormonally active metabolite of D₃ otherwise known as 1,25(OH)₂D], (ii) 24,25(OH)₂D₃ (the most concentrated product of 25(OH)D₃ catabolism), (iii) vitamin D–binding protein (DBP), and (iv) free vitamin D (circulating unbound to DBP or albumin; refs. 17, 18), are present at different stages along the vitamin D metabolic pathway, and have potential to characterize vitamin D status more fully than with 25(OH)D alone (18). However, the absence of established cut-points of sufficiency for these additional biomarkers makes comparisons across biomarkers difficult. Additional vitamin D biomarkers have historically not been factored into recommendations (19).

To our knowledge, only one study has evaluated the association between serum 25(OH)D and HPV. Using cross-sectional data from NHANES, investigators reported that concentrations of serum 25(OH)D were inversely associated with HPV infection (20). To expand on these findings, we evaluated the cross-sectional associations between five different vitamin D biomarkers and hrHPV infection among mid-adult women in Washington State (21). Given previous literature suggesting differences in vitamin D metabolism and bioavailability of the micronutrient among racial groups (22, 23), we also evaluated whether this association varied by race.

We used baseline data and stored specimens from a 6-month longitudinal cohort study of HPV infections in healthy mid-adult women in Seattle, WA. Details of the parent study design were described previously (21). Briefly, women aged 30–50 years who were affiliated with the University of Washington (Seattle, WA; faculty, staff, or student) were enrolled from March 2011–January 2012. Women who were pregnant, had a hysterectomy, or had a current medical condition that prohibited participation were excluded. At enrollment, participants self-collected vaginal swabs for HPV DNA testing and provided venous blood specimens; residual serum specimens were stored at −80°C. Demographic (age, race/ethnicity, education level, marital status), health [body mass index (BMI), alcohol intake, smoking status, oral contraceptive use, HPV vaccination history], and sexual behavior characteristics (history of genital warts and/or other sexually transmitted infections, history of pregnancy, and number and recency of male sex partners) were collected through a combination of an interview with a study coordinator and a self-administered online questionnaire. Information on vitamin D supplementation was not collected. The protocol was reviewed and approved by the University of Washington Institutional Review Board.

As part of the parent study, self-collected vaginal swabs were HPV genotyped using the Roche Linear Assay (Roche Diagnostics; ref. 21). Our outcome of interest was detection of one or more of 19 hrHPV strains: 16, 18, 26, 31, 33, 35, 39, 45, 51, 52, 53, 56, 58, 59, 66, 68, 73, 82, and IS39 (24, 25).

Vitamin D biomarker testing was conducted on stored sera at the University of Washington's Nutrition and Obesity Research Center Analytic Core. Sera specimens were thawed and 750 μL aliquoted and transported on dry ice. Serum concentrations of albumin, calcium, and phosphate were determined using an automated clinical chemistry analyzer (Beckman AU5812, Beckman Coulter). Concentrations of 1,25(OH)₂D₃, 1,25(OH)₂D₂, 24,25(OH)₂D₃, 25(OH)D₃, and 25(OH)D₂ were determined using immunoaffinity enrichment LC/MS-MS, as described previously (26, 27). Concentrations of vitamin D binding globulin and the haplotypes present in each sample were determined using trypsin digestion–LC/MS-MS, as previously described (28, 29).

Specifically, we measured as exposures serum concentrations of (i) 25(OH)D, (ii) 1,25(OH)₂D, (iii) 24,25(OH)₂D₃, (iv) free/unbound vitamin D, and (v) 25(OH)D adjusted for DBP and stratified by DBP haplotype. 25(OH)D and 1,25(OH)₂D are composite measures and are calculated as the sums of 25(OH)D₂ and 25(OH)D₃, and 1,25(OH)₂D₂ and 1,25(OH)₂D₃, respectively. We calculated free vitamin D using an established equation that factors in affinity constants for albumin and DBP (30). Free vitamin D was measured in 1,000 nmol/L given its presence in low concentrations.

Our primary exposure was 25(OH)D, analyzed as a continuous increase of 10 ng/mL of serum 25(OH)D. As a secondary analysis of 25(OH)D, we generated two binary exposure variables using clinically meaningful cut-points of sufficiency: ≥20 ng/mL and ≥30 ng/mL (31, 32). These cut-points were based on recommendations from the Institute of Medicine (IOM; ref. 31), the Endocrine Society (32), and previous literature (20), although clinical cut-points for sufficiency remain contested (33). The other four biomarkers were secondary exposures and were each analyzed as a continuous variable.

To explore the potential implications of modeling 25(OH)D as a linear versus categorical measure, we conducted a post hoc analysis using additional cut-points. First, we evaluated hrHPV prevalence among five subgroups of women categorized by 25(OH)D serum concentration: <12 ng/mL (severe deficiency, per IOM), 12–19 ng/mL (deficiency, per IOM), 20–29 ng/mL (sufficiency, per IOM), 30–49 ng/mL (sufficiency, per Endocrine Society), and ≥50 ng/mL. Distinguishing concentrations around the 50 ng/mL cut-point was also of interest because serum concentrations at or above this level in Seattle, WA typically indicate vitamin D supplementation (34). Second, we generated a binary exposure variable (≥50 ng/mL, <50 ng/mL).

Using logistic regression with robust SEs (to account for potential model misspecification), we estimated ORs with 95% confidence intervals (CI) for associations between serum vitamin D and hrHPV detection. Because serum concentration of vitamin D fluctuates during the year, we adjusted for season of enrollment using a B-spline with five degrees of freedom (to account for 10 months of enrollment). We considered additional variables a priori for inclusion. In an effort to present a reliable, precise estimate of an unconfounded relationship, we performed our modeling in a phased approach. For descriptive purposes, we produced a minimally adjusted model adjusting only for age (continuous), self-reported race [White, African American, Asian, Other (American Indian or Alaska Native, Native Hawaiian or Other Pacific Islander, other, or multiple races)], and season. Our moderately adjusted model is our primary model. This model incorporated additional biological and behavioral risk factors: oral contraceptive use (never, former, current), highest level of education (less than college, bachelor's degree, masters/doctoral degree), smoking status (never, former, current smoker), BMI (<18.5, 18.5 to <25, 25 to <30, or ≥30), and continuous serum concentrations of calcium (ng/mL) and phosphate (mg/dL). Finally, we ran a fully adjusted model that also included sexual behavior risk factors evaluated as precision variables: lifetime number of male sex partners (quintiles) and number of male sex partners in the 6 months prior to enrollment (none, non-new only, one new, ≥2 new).

We evaluated potential effect modification by race/ethnicity. Comparisons were made between White and non-White women due to small sample sizes of non-White race subgroups.

We ran generalized additive models for each exposure to assess evidence of non-linearity in the relationship between hrHPV and our predictors.

Using logistic regression with robust SEs to estimate ORs and 95% CIs, we measured the association between serum 25(OH)D ≥50 ng/mL and hrHPV. We conducted this analysis using the three stages of modeling outlined above.

All analyses were conducted in R version 3.4.3 (35).

One woman was excluded from analyses due to a specimen that was insufficient for HPV testing. After completing laboratory results quality assurance, four additional women were excluded due to uncertainty in patient-to-specimen matching attributable to specimen labeling errors. The mean age of the remaining 404 participants was 38.3 years (SD = 6.1 years; Table 1). The study population was predominantly White, with a mean BMI of 25.9 (SD 5.8). One third of participants reported current hormonal contraceptive use, and 26% were current or former smokers. The median lifetime number of male partners was seven (IQR 3–15). hrHPV prevalence was 22%.

Median and interquartile ranges for all vitamin D biomarkers evaluated were within normal ranges (ref. 36; Table 2). Mean serum concentrations of calcium (37) and phosphate (38) were normal, at 9.5 ng/mL (SD 0.4 ng/mL) and 3.5 mg/dL (SD 0.5 mg/dL), respectively.

Associations between serum concentrations of 25(OH)D and hrHPV detection were inconsistent and not statistically significant (Table 3). Controlling for season of enrollment, age, race, BMI, smoking status, education level, and oral contraceptive use, each 10 ng/mL increase in 25(OH)D was not associated with significantly increased odds of hrHPV infection (aOR 1.08; 95% CI, 0.83–1.41). Estimates produced using minimally and fully adjusted models were also not significantly different from 1.0.

Secondary analyses of categorical exposure variables designed to classify 25(OH)D as sufficient or deficient also yielded non-significant results. After adjustment under our primary model, compared with women with deficient serum 25(OH)D concentrations below 20 ng/mL, those with serum concentrations ≥20 ng/mL (indicating sufficiency) had estimated odds of hrHPV infection of 0.85 (95% CI, 0.38–1.89). Under the same model, the estimated odds of infection for a higher 25(OH)D cut-point of 30 ng/mL was also 0.85 (95% CI, 0.50–1.43).

Upon examination of 1,25(OH)₂D, free vitamin D, and 25(OH)D adjusted for DBP and DBP haplotype, we found no significant associations between serum concentrations of vitamin D biomarkers and odds of hrHPV infection. However, in a minimally adjusted model, serum 24,25(OH)₂D₃ was significantly associated with hrHPV infection. Each 1 ng/mL increase in 24,25(OH)₂D₃ was associated with higher odds of infection (aOR 1.28; 95% CI, 1.03–1.58), although the significance of this association waned upon further adjustment (under primary model, aOR 1.22; 95% CI, 0.97–1.52).

Analyses examining potential interaction between 25(OH)D levels and race did not reveal significant differences in odds of hrHPV infection between White and non-White women. After adjustment with our primary model, each 10 ng/mL increase in 25(OH)D suggested higher odds of infection among both White (aOR 1.05; 95% CI, 0.77–1.41) and non-White women (aOR 1.23; 95% CI, 0.66–2.28); however, these results were not statistically significant.

Although results of a generalized additive model did not provide strong evidence of a nonlinear effect of continuous 25(OH)D (Supplementary Fig. S1), our post hoc analysis indicated that hrHPV prevalence varied by serum concentrations of 25(OH)D. There were no hrHPV-positive women in our sample with 25(OH)D <12 ng/mL. hrHPV prevalence was nearly identical among women with serum concentrations 12–19 ng/mL (24.4%; 95% CI, 12.9–39.5) and 20–29 ng/mL (24.1%; 95% CI, 17.3–32.0). Prevalence was lower among women with 25(OH)D concentrations 30–49 ng/mL (18.3%; 95% CI, 13.0–24.6). However, hrHPV prevalence was highest in women with the highest serum 25(OH)D concentrations (≥50 ng/mL; 43.5%; 95% CI, 23.2–65.5). Finally, we evaluated the odds of prevalent hrHPV among women with 25(OH)D ≥50 ng/mL (Table 3). Under our primary model, the odds of hrHPV infection among women who likely supplement [25(OH)D ≥50 ng/mL] were significantly higher than the odds among women in whom supplementation is unlikely [25(OH)D <50 ng/mL; aOR 2.74; 95% CI, 1.14–6.59].

To our knowledge, this is the first study to analyze an association between serum vitamin D status and hrHPV prevalence using five different biomarkers of vitamin D status, four of which are novel in this context. We aimed to understand the relationship between serum concentrations of vitamin D and detection of hrHPV through evaluation of not only 25(OH)D, but other forms of the micronutrient along the metabolic pathway as well. These biomarkers included byproducts of 25(OH)D metabolism tasked with keeping cellular proliferation and differentiation in check [1,25(OH)₂D and 24,25(OH)₂D₃; refs. 8, 17] and a measure of circulating vitamin D unbound to binding proteins and other nutrients (free vitamin D). We did not observe significant associations between 25(OH)D and prevalent hrHPV in our analyses of 25(OH)D as both a continuous and categorical variable, as defined a priori. Post hoc analysis of 25(OH)D revealed women who likely supplement [25(OH)D ≥50 ng/mL] had significantly higher odds of prevalent hrHPV compared with other women. Apart from our findings on 24,25(OH)₂D₃, we did not detect significant relationships between additional biomarkers and prevalence of hrHPV. Similarly, our analysis of race did not yield evidence of effect modification.

Higher serum concentrations of 24,25(OH)₂D₃ were significantly associated with higher odds of hrHPV infection. Although statistical significance waned in the moderately and fully adjusted models (which may be related, in part, to including a large number of covariates in models with relatively few outcomes), this hint of an association warrants closer examination. Low concentrations of 24,25(OH)₂D₃, the most plentiful byproduct of 25(OH)D metabolism by the enzyme CYP24A1 (found in most tissues throughout the body), have previously been associated with chronic kidney disease (39). Unsurprisingly, serum concentrations of 24,25(OH)₂D₃ are positively correlated with concentrations of 25(OH)D (40). Thus, the direction of the association we observed between increasing serum concentrations of 24,25(OH)₂D₃ and detection of hrHPV is in the opposite direction than we would expect.

Our hypothesis of an inverse association between serum concentrations of 25(OH)D and detection of hrHPV was motivated by prior research of the relationship (20) and suggestions of vitamin D's immunomodulating properties in the context of chronic (4) and acute disease (3). However, our selection of cut-points for our categorical analyses, though clinically motivated, perhaps oversimplified the association and resulted in attenuated estimates. By grouping very deficient and deficient women together [all with 25(OH)D <20 ng/mL or <30 ng/mL, depending on the analysis], our original a priori categorical analyses generated comparison groups of women who looked very similar with regards to hrHPV prevalence. Among women whose 25(OH)D concentrations were ≥30 ng/mL, two distinct subgroups exist; those whose serum concentrations suggest supplementation (≥50 ng/mL) had the highest odds of prevalent hrHPV in the study population, and this association was statistically significant. It is possible women who (likely) supplement were previously diagnosed as deficient, and an association between hrHPV status and 25(OH)D level prior to supplementation might better reflect a relationship between vitamin D status and detection of hrHPV.

In addition to the difference in the distribution of subjects' 25(OH)D serum concentrations, differences in age distribution may also explain the difference in our primary findings compared with the population-based NHANES sample. While the NHANES study (20) included women aged 20–59, we restricted our analyses to a narrower age range of mid-adult women aged 30–50 years. In the NHANES study, younger women were the most likely, relative to all other age groups, to be severely 25(OH)D deficient and to have prevalent hrHPV. It is possible these younger women with low mean serum concentrations of 25(OH)D and relatively high prevalence of hrHPV drove the observed association between 25(OH)D deficiency and detection of hrHPV. However, because results of the NHANES study are not stratified by age, it is difficult to discern whether the association between vitamin D status and detection of hrHPV varies between age groups.

Our study had several strengths, including secondary analyses with four novel vitamin D biomarkers. Further research into these biomarkers, particularly in longitudinal analyses, would be useful in better describing potential associations in less healthy populations. In addition, our use of multiple stages of regression models enabled comparison across studies and allowed for transparent inclusion of confounders and precision variables, the latter of which can increase the precision of effect estimates and lead to tighter confidence intervals.

Our study is not without limitations. Previous literature (41) suggests low serum concentrations of 25(OH)D may be a consequence of chronic infection (for example, with hrHPV), rather than a contributing factor in the development or persistence of disease. Our cross-sectional study design precludes causal inference and prevents understanding of whether serum 25(OH)D influences hrHPV acquisition or if concentrations of the micronutrient are a marker of inability to clear an existing infection. In future analyses, we will use longitudinal data from the parent cohort of mid-adult women to evaluate associations between vitamin D status and persistent detection of hrHPV in monthly samples. In addition, information on vitamin D supplementation was not collected as part of the parent study, the focus of which was unrelated to vitamin D. Future longitudinal studies would benefit from capturing information on supplementation at baseline and could evaluate whether supplementation affects acquisition or clearance of hrHPV. Finally, while NHANES data from 2005–2006 estimated that 42% of US women were deficient at <20 ng/mL of serum 25(OH)D, our data revealed a healthy cohort in which only 13% of women had serum concentrations <20 ng/mL. Over half of our study population was 25(OH)D sufficient at ≥30 ng/mL and another third were sufficient at the cut-point of ≥20 ng/mL, reducing power to detect associations between 25(OH)D insufficiency and hrHPV. Future studies would benefit from larger sample sizes in which detection of an association would be possible at smaller effect sizes.

In conclusion, evaluation of the association between vitamin D status, as measured by five biomarkers measured along the micronutrient's metabolic pathway, and detection of prevalent hrHPV yielded mixed results in a population of healthy mid-adult women in Seattle, WA. Findings from longitudinal studies in populations with both more variability in vitamin D status and data on supplementation may be useful in motivating clinical recommendations for subpopulations of women.

A.N. Hoofnagle reports receiving a commercial research grant from Waters. No potential conflicts of interest were disclosed by the other authors.

Conception and design: A.N. Hoofnagle, R.L. Winer

Development of methodology: C. Troja, A.N. Hoofnagle, A.A. Szpiro

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A.N. Hoofnagle, R.L. Winer

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): C. Troja, A.N. Hoofnagle, A.A. Szpiro, J.E. Stern, R.L. Winer

Writing, review, and/or revision of the manuscript: C. Troja, A.N. Hoofnagle, A.A. Szpiro, J.E. Stern, J. Lin, R.L. Winer

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A.N. Hoofnagle, J.E. Stern, J. Lin

Study supervision: R.L. Winer

This work was supported by grants from the NIH [grant numbers R21AI129935 (to R.L. Winer) and P30 DK035816 (to M.W. Schwartz)].

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.