An Exploratory Analysis on the Influence of Genetic Variants on Weight Gain among Etonogestrel Contraceptive Implant Users

Abstract

Objective:

To identify genetic variants associated with weight gain related to etonogestrel contraceptive implant use.

Study Design:

We conducted a retrospective analysis from a parent pharmacogenomic study of healthy, reproductive-aged women using etonogestrel implants. We reviewed medical records to calculate objective weight changes from implant insertion to study enrollment and asked participants about subjective weight gain (yes/no) during contraceptive implant use. We used genotyping data (99 genetic variants) from the parent study to conduct backward-stepwise generalized linear modeling to identify genetic variants associated with objective weight changes.

Results:

Among 276 ethnically diverse participants, median body-mass index (BMI) was 25.8kg/m² (range 18.5-48.1). We found a median weight change of +3.2kg (range −27.6 to +26.5) from implant insertion to study enrollment. Report of subjective weight gain had minimal agreement with measured weight gain during implant use (Cohen’s kappa=0.21). Our final generalized linear model contained two variables associated with objective weight change that met conservative statistical significance (p<5.0x10⁻⁴). Participants with two copies (homozygous) of the ESR1 rs9340799 variant on average gained 14.1kg more than all other participants (p=1.4x10⁻⁴). Higher enrollment BMI was also associated with objective weight gain (β= 0.54, p=9.4x10⁻¹²).

Conclusion:

Genetic variants in the estrogen receptor 1 (ESR1) do not have known associations with obesity or metabolic syndrome, but there is physiologic plausibility for a progestin-mediated genetic association between ESR1 and weight gain. Additional genetic research is needed to substantiate our findings and elucidate further advances in individualized counseling on the risk of weight gain with exogenous steroid hormones.

1.1. Introduction

Weight gain remains a rare primary reason for etonogestrel contraceptive implant (Nexplanon®, Merck & Co., Whitehouse Station NJ) discontinuation [1, 2]. However, etonogestrel implant users experience wide inter-individual variability in weight changes [3–7]. The largest study with a non-hormonal comparison group found etonogestrel implant users had a greater mean weight increase (3.0kg versus 1.1kg) after 36 months of use, but similarly wide variability in individual weight changes (−24kg to +25kg versus −18kg to +20kg) [3]. Although the average weight gain associated with etonogestrel implant use may not be clinically significant, reproductive-aged women experiencing a 25kg weight gain likely face additional health risks [8].

We currently lack means to determine the risk of weight gain for an individual considering an etonogestrel implant or other hormonal contraceptive methods, though individual patient characteristics like race can affect this risk [7]. Pharmacogenomics, the study of how genetic variants affect drug metabolism and side effects, may help elucidate if and how individual genetic differences contribute to variability in hormonal drug-related weight gain [9]. For example, studies using pharmacogenomic methods have identified genetic variants associated with antipsychotic drug-related weight gain [10].

To explore potential pharmacogenomic associations between hormonal contraceptive use and weight gain, we used a candidate gene approach to identify genetic variants associated with measured objective weight change among etonogestrel contraceptive implant users. We hypothesized that variants in genes encoding proteins involved in progestin and estrogen pharmacokinetics and pharmacodynamics would be associated with increased measured weight from time of implant insertion.

1.2. Materials and Methods

This was a retrospective analysis of a parent pharmacogenomic association study [11]. The candidate gene methodology for this study has been previously published [11]. Participants were reproductive aged women (18-45 years old) with an etonogestrel contraceptive implant in place for at least 12 months and no more than 36 months at the time of enrollment (the steady-state period of the implant’s pharmacokinetics) [12, 13]. Of particular pertinence to this study, the parent study excluded women with a measured body-mass index (BMI) less than 18.5kg/m² due to concerns for altered etonogestrel metabolism among underweight women; there was no upper BMI limit. The hypotheses tested in this analysis were pre-specified secondary outcomes of the parent study [11]. The protocol was approved by the Colorado Multiple Institutional Review Board and all participants gave written informed consent before study initiation. We recruited participants through community advertising and contraceptive clinics at the University of Colorado Anschutz Medical Campus.

At the time of enrollment, we measured participants’ height and weight and calculated their BMI. For all participants, we performed electronic medical record review to capture their weight at the time of implant insertion if available. We then calculated the weight change from time of implant insertion to study enrollment for participants with documented weight in the electronic medical record for the initial time point. Participants also completed a brief questionnaire to obtain self-reported demographics and side effect information, including if participants had experienced “weight gain” (yes/no) while using their contraceptive implant.

The full genotyping methodology for this study has already been published [11]. See Appendix 1 for the full list of single nucleotide polymorphisms (SNPs) selected for this study and their reference (rs) ID numbers. Allele and genotype frequencies and Hardy-Weinberg equilibrium calculations were also included in this publication [11]. We ultimately included 99 genetic variants for statistical analysis after excluding nineteen variants not present in this study population and two variants with significant deviation from Hardy-Weinberg equilibrium [11]. Our methodology for serum etonogestrel concentration analysis is previously described [11, 13].

We used IBM SPSS® version 25 statistical software for our analyses. We performed descriptive frequencies and utilized Cohen’s kappa coefficient to determine agreement between objective and subjective weight gain. We then conducted simple linear modeling to identify genetic variants associated with objective weight change. For our multivariable analysis of objective weight change, we used a generalized linear model due to the non-normal distribution of this outcome [14]. In our generalized linear modeling, we included genetic variants significantly associated with weight change in simple linear modeling (p-value cut-off of 0.05 for inclusion) and the continuous variables of age, BMI, duration of implant use, and serum etonogestrel concentration, and the categorical variables of self-reported race and ethnicity. We utilized a backward-stepwise approach to create our final linear model based on Akaike’s Information Criterion [15]. Given the multiple hypothesis testing performed using simple logistic regression (99 independent analyses), we determined a Bonferroni corrected p-value cut-off of 5.0x10⁻⁴ We used this corrected p-value to determine overall significance for the variables included in our final generalized linear model. The sample size (N=350) was determined a priori based on the primary outcome of the parent pharmacogenomic study [11].

1.3. Results

We reviewed the available medical records for all 350 participants from the parent study for inclusion in this analysis [11]. Weight at the time of implant insertion was available for 276 participants (78.9%). Table 1 shows pertinent patient characteristics and demographics for these 276 participants. Participants’ median age was 22.3 years (range 18.0–39.1) and median duration of implant use was 27.0 months (range 12-36). Approximately 57% reported their ethnicity as Hispanic/Latina and 43% self-reported their race as White or Caucasian. The median BMI at enrollment was 25.8kg/m² (range 18.5–48.1). Table 1 demonstrates the BMI categories of participants [8]. The median duration of implant use and median BMI at the time of enrollment were not statistically different between participants from the parent study with or without insertion weight data (Independent-samples median test, p=0.42 and p=0.36, respectively).

Table 1:

Characteristics and demographics of etonogestrel implant users included in this exploratory analysis (N=276)

	Median (Range)
Age (years)	22.3 (18.0 – 39.1)
Months of implant use	27.0 (12.0 – 36.0)
BMI (kg/m2)	25.8 (18.5 – 52.0)
Serum etonogestrel concentration (pg/mL)	136.1 (61.0 – 434.1)
Weight change*(kg)	3.2 (−27.6, 26.5)
	n (%)
BMI category†
Normal (18.5 to <25)	116 (42.0)
Overweight (25.0 to <30)	85 (30.8)
Obese (30.0 or higher)	75 (27.2)
Class 1 Obesity (30 to <35)	47 (17.0)
Class 2 Obesity (35 to <40)	20 (7.2)
Class 3 Obesity (40 or higher)	8 (2.9)
Race
White or Caucasian	119 (43.1)
Black or African American	33 (12.0)
Asian or Pacific Islander	15 (5.4)
Native American or Alaskan	6 (2.2)
More than one	34 (12.3)
No response or Unknown	69 (25.0)
Ethnicity
Hispanic or Latina	158 (57.2)
Non-Hispanic	118 (42.8)

^*Calculated by: (measured weight at time of study enrollment) minus (measured weight at time of implant insertion)

^†Based on the Centers for Disease Control and Prevention definitions [8]

The median weight change from implant insertion to study enrollment was +3.2kg (range −27.6kg to +26.5kg) with an interquartile range of 8.3. Figure 1 shows the distribution of weight changes. Though only 43.1% (119/276) of participants with objective weight data reported subjective weight gain with the implant, the majority of participants had objective weight gain since implant insertion (73.9%, 204/276). A report of subjective weight gain had only minimal agreement with measured weight gain during implant use (Cohen’s kappa = 0.21).

Figure 1 Legend:

Histogram of weight change from implant insertion to study enrollment for all 276 participants with available medical records.

We performed simple linear regression to test for genetic associations with objective weight change from contraceptive implant insertion to study enrollment. Seven genetic variants had significant associations (p<0.05): CYP2C19 rs11568732, AKR1C3 rs12529, ESR1 rs2077647, ESR1 rs2228480, CYP2C19 rs7088784, ESR1 rs9322335, and ESR1 rs9340799. Table 2 shows the unadjusted beta-coefficients for all patient characteristics and demographics and Table 3 contains the unadjusted beta-coefficients for these seven genetic variants. Using generalized linear modeling BMI at time of enrollment and two genetic variants (CYP2C19 rs7088784 and ESR1 9340799) remained in the final model (Table 4). For every 1kg/m² increase in enrollment BMI, the average weight gain from contraceptive implant insertion to study enrollment increased by 0.54kg.

Table 2:

Associations between patient characteristics and demographics and objective weight change in a study of etonogestrel implant users

	Objective weight change
	β	95% CI
Age (per year)	−0.16	−0.44, 0.12
Duration of implant use (per month)	0.04	−0.06, 0.13
BMI (at enrollment, per 1kg/m2)	0.54†	0.40, 0.67
Serum etonogestrel concentration (per 1pg/mL)	−0.01	−0.02, 0.01
Race*
White or Caucasian	−0.43	−2.15, 1.28
Black or African American	−2.5	−5.09, 0.11
Asian or Pacific Islander	−0.75	−4.50, 2.99
Native American or Alaskan	3.51	−2.29, 9.32
More than one	−0.78	−3.36, 1.80
Hispanic or Latina	2.37†	0.68, 4.07

Performed using simple linear regression. Weight change defined as measured weight at time of study enrollment minus measured weight at time of implant insertion.

^*Each race category was dummy coded for comparison to all other participants

^†Statistically significant using p-value cut-off <0.05

Table 3:

Associations between genetic variants and objective weight change in a study of etonogestrel implant users

	Objective weight change
	β	95% CI
Genetic Variants
CYP3A5 rs15524*	0.75	−0.71, 2.21
CYP3A4*1G§	0.38	−1.39, 2.16
CYP3A4 rs2246709§	−0.93	−2.09, 0.24
CYP3A4 rs4646440§	1.53	−0.05, 3.09
ESR1 rs9322335*	1.64†	0.27, 3.01
CYP2C19 rs11568732§	−2.82†	−5.62, −0.02
AKR1C3 rs12529*	1.17†	0.03, 2.31
ESR1 rs2077647*	1.29†	0.05, 2.53
ESR1 rs2228480*	2.68†	1.13, 4.24
CYP2C19 rs7088784§	−2.99†	−5.46, −0.52
ESR1 rs9340799*	1.95†	0.34, 3.55

Performed using simple linear regression

^†Statistically significant using p-value cut-off <0.05

^*Genetic variant grouped as participants with two copies of variant allele (homozygous variant) versus all others

^§Genetic variant grouped as participants with at least one copy of the variant allele (carriers) versus all others

Table 4:

Final associations between patient characteristics and genetic variants and objective weight change in a study of etonogestrel implant users

Variable	β	95% CI	p-value
BMI (at enrollment)†	0.54	0.39, 0.70	9.39 x 10−12
CYP2C19 rs7088784 carrier	−3.74	−6.12, −1.35	0.002
ESR1 rs9340799 homozygous variant†	14.10	6.86, 21.35	1.36 x 10−4

Performed using generalized linear modeling with a backward-stepwise approach

^†Statistically significant using Bonferroni corrected p-value cut-off < 5.0 x 10⁻⁴

CYP2C19 rs7088784 had an overall variant allele frequency of 8.3% among our participants successfully genotyped for this SNP (n=296) with 15.9% (n=47) having one variant allele (heterozygous) and 0.3% (n=1) having two variant alleles. ESR1 rs9340799 had an overall variant allele frequency of 22.5% among our participants successfully genotyped for this SNP (n=331) with 40.8% (n=135) having one variant allele (heterozygous) and 2.1% (n=7) having two variant alleles (homozygous variant). Participants with at least one CYP2C19 rs7088784 variant allele (carriers) had on average less weight gain compared to participants homozygous for the wild-type genotype (mean weight change +0.97kg versus +4.28kg, respectively, Figure 2). Participants with two copies of the ESR1 rs9340799 variant (homozygous variant) on average gained 14.1kg more weight than participants with at least one wild-type allele (Figure 3). Using a conservative Bonferroni corrected p-value of 5.0x10⁻⁴, only the associations with BMI and ESR1 rs9340799 remained statistically significant.

Figure 2 Legend:

Box plots of weight change from implant insertion to study enrollment among participants with the CYP2C19 rs7088784 wild-type genotype (n=248) compared with CYP2C19 rs7088784 variant carriers (n=48), excluding those with unknown genotype for this single nucleotide polymorphism (n=54). The box represents the first and third quartiles with the band inside the box representing the median for each respective group. The whiskers represent the data within 1.5 interquartile range of the upper and lower quartile. Circles indicate outliers with values between 1.5 times and 3 times the IQR and asterisks indicate outliers with values greater than 3 times the IQR.

Figure 3 Legend:

Box plots of weight change from implant insertion to study enrollment among participants homozygous for the ESR1 rs9340799 variant allele (n=7) versus carriers of at least one wild-type allele (n=324), excluding those with unknown genotype for this single nucleotide polymorphism (n=19). The box represents the first and third quartiles with the band inside the box representing the median for each respective group. The whiskers represent the data within 1.5 interquartile range of the upper and lower quartile. Circles indicate outliers with values between 1.5 times and 3 times the IQR and asterisks indicate outliers with values greater than 3 times the IQR.

We also performed multivariable logistic regression to test for genetic associations with subjective weight gain during contraceptive implant use. Given the low agreement between our primary objective weight change outcome and subjective weight gain, these data are presented as a supplement.

1.4. Discussion

We undertook this study to assess whether a pharmacogenomic approach could help our understanding of contraceptive-associated weight changes. Among a large, diverse group of etonogestrel contraceptive implant users, we identified unique genetic variants with clinically significant associations for objective weight change. We found objective weight changes among our participants that closely match the findings of prior studies [2, 3]. As measured weight gain has only minimal agreement with reporting subjective weight gain, our study highlights that the perception of weight gain with the contraceptive implant may be inaccurate, which is consistent with previously published data [17]. BMI at time of enrollment was the only non-genetic variable we found significantly associated with objective weight change while controlling for all other patient characteristics and demographics.

We ultimately identified one genetic variant associated with objective weight change in contraceptive implant users that met our conservative threshold for statistical significance: ESR1 rs9340799. Estrogen receptor 1 (ESR1) found on chromosome 6, encodes an estrogen receptor involved in cellular hormone binding and DNA transcription when activated [18]. Though ESR1 rs9340799 is located in the intronic region of ESR1, prior studies have found pharmacogenomic associations between this variant and other medications. Postmenopausal women with the homozygous variant ESR1 rs9340799 genotype (GG) treated with conjugated estrogens and medroxyprogesterone had greater increases in spine bone mineral density as compared to similar women with the homozygous wild-type genotype (AA) [19]. In this same study, women with the homozygous variant genotype (GG) also had smaller decreases in spine bone mineral density when left untreated compared to women with the homozygous wild-type genotype (AA) [19]. These findings demonstrate an increased response to both exogenous and endogenous estrogen among women with the ESR1 rs9340799 variant genotype, supporting an association between this variant and either upregulation of ESR1 transcription or a more responsive estrogen receptor 1 protein. Additional pharmacogenomic studies further support that the ESR1 rs9340799 variant genotype confers an increased clinical response to estrogen agonists [20, 21].

ESR1 rs9340799 has no known associations with obesity or metabolic syndrome and large studies have failed to find a consistent association between any ESR1 SNPs and obesity [22, 23]. The negative findings of these studies with the general population lend support to our potential discovery of a progestin-dependent effect between variants in ESR1 and weight gain. Though etonogestrel has no binding affinity to estrogen receptors [24], research on breast cancer has found that the activated progesterone receptor can directly modulate estrogen receptor 1 transcriptional activity [25]. Thus theoretically, these ESR1 intronic variants or an exonic variant in linkage disequilibrium may lead to an estrogen receptor 1 protein with greater sensitivity to this progesterone receptor-modulated relationship. As etonogestrel has a high binding affinity to the progesterone receptor (3 times higher than endogenous progesterone) [24], this ligand-mediated effect may be dependent on this increased binding affinity, thus potentially accounting for the lack of association between ESR1 variants and obesity in the absence of exogenous progestins. However, this research on the interplay between the progesterone and estrogen receptors primarily comes from tumor cell lines, which may not accurately reflect the signaling and expression patterns of normal cells [25]. More mechanistic research is needed to elucidate the functional impact of these genetic variants on the transcription and activity of the estrogen receptor 1 protein, particularly as related to interactions with the progesterone receptor.

The other genetic variant we found associated with objective weight change (CYP2C19 rs7088784) has less consistent physiologic plausibility. CYP2C19 encodes a metabolic enzyme that may play a role in steroid hormone metabolism, however, this variant was not associated with serum etonogestrel concentrations among our study participants [11]. CYP2C19 rs7088784 has no currently published pharmacogenomic associations, thus the associations found in this study may be a false positive.

The major strength of this study was our selection of SNPs from genes encoding proteins with known involvement in steroid hormone pharmacokinetics and pharmacodynamics. We were also able to recruit a diverse cohort of contraceptive implant users, and we had sufficient power to detect significant associations between genetic variants and our outcomes of interest. Our participants were also representative of the general population of etonogestrel implant users, being relatively young with a wide distribution in BMI [26].

The BMI cut-off for the parent study (≥18.5 kg/m²) and exclusion of underweight implant users limits the generalizability of our findings to such women. We excluded underweight women to avoid issues of altered metabolism for the parent study’s primary pharmacokinetic outcome [11].The lack of a control group also limits our ability to state that these genetic variants have an association with weight change dependent on etonogestrel exposure. As non-hormonal intrauterine device users also demonstrate wide variability in weight changes [3], more research on the genetic variants associated with weight change among both hormonal and non-hormonal contraceptive users is needed to substantiate a progestin-dependent effect. We also did not collect data regarding weight intention (i.e. gain, lose, maintain) from our participants, and could not control for potential confounding due to diet and exercise differences during implant use. We only enrolled participants with at least 12 months of contraceptive implants use, thus limiting our data on early implant discontinuers, however ensuring at least 12 months of opportunity for weight changes. Finally, our candidate gene approach prevented us from identifying novel genetic loci associated with progestin-mediated weight changes and did not allow us to control for population stratification, which is confounding by genetic ancestry.

In this exploratory analysis, variants in ESR1 accounted for four of the seven SNPs associated with weight change, all associated with increased objective weight gain. The ESR1 rs9340799 variant, in particular, was associated with both a statistically and clinically significant increase in average weight gain of 14kg among contraceptive implant users with two copies of this variant allele. Exogenous steroid hormone medications are used for various indications throughout a woman’s lifespan. It is imperative to better understand how individual genetic variation may influence a woman’s risk of adverse weight gain while using them. Our hypothesis-generating results support the pursuit of definitive mechanistic and clinical research to determine the actual influence of ESR1 variants on progestin-associated weight gain. As our understanding of pharmacogenomics in women’s health expands, we can develop individualized counseling that may reduce the incidence of hormone-related adverse effects, improve patient satisfaction, and help prevent future health risks associated with weight gain.

Implications.

Genetic variation in the estrogen receptor 1 gene may account for variability in weight gain among etonogestrel contraceptive implant users. If these findings can be replicated with other progestin-containing medications, we may be able to better individualize contraceptive counseling.