The influence of fine particulate matter on the association between residential greenness and ovarian reserve.

After applying exclusion criteria, the final analytical sample included 565 women with complete data on greenness and AFC ( Figure 1 ). Peak NDVI ranged from 0.07 to 0.92 ( Table 1 ) with a mean of 0.59. Participant age varied by quartile of peak greenness with an average age of 36.0 years among women in the highest quartile of greenness compared to 34.8 years among women in the lowest quartile of greenness. Average PM 2.5 exposure also differed by greenness quartile. The highest average PM 2.5 exposure (9.4 μg/m 3 ) was among women in the second quartile of greenness whereas women in the first quartile had the lowest exposure (8.9 μg/m 3 ); however, the Spearman correlation between exposures to peak greenness and average PM 2.5 indicated little linear association (ρ=−0.01). Greenness also varied slightly according to year of the antral follicle scan (ρ=−0.18). All other demographic and reproductive characteristics were similar across quartiles of peak greenness. We did not observe evidence of a nonlinear relationship between peak greenness and AFC (P for non-linearity: 0.26). After multivariable adjustment, a one SD increase in peak greenness was associated with a 1.1% (95% CI −3.3, 5.7) higher AFC; however, this association was not statistically significant ( Table 2 ). In contrast, a 2 μg/m 3 increase in average PM 2.5 exposure was associated with a 6.2% (95% CI −11.8, −0.3) lower AFC. When we used a model with a cross-product term between average PM 2.5 exposure and peak greenness, we found evidence of a statistically significant multiplicative interaction ( P interaction: 0.03). Among women with an average PM 2.5 exposure of 7 μg/m 3 , a SD increase in peak greenness was associated with a 5.6% (95% CI −0.4, 12.0) higher AFC. Conversely, among women with an average PM 2.5 exposure of 12 μg/m 3 , a SD increase in residential peak greenness was associated with a 5.8% (95% CI −13.1, 2.1) lower AFC ( Figure 2 ). In general, women with high exposure to greenness and low exposure to PM 2.5 tended to have the highest AFCs while women with high exposure to both greenness and PM 2.5 had the lowest AFCs ( Figure 3 ). There was a suggestion that women with low exposure to greenness and high exposure to PM 2.5 also had higher AFCs although as indicated by the lack of dots in this corner of the contour plot, but this was based on very sparse data. The addition of self-reported physical activity as a covariate to the multivariable model did not change the associations between peak greenness and AFC or average PM 2.5 exposure and AFC ( Supplemental Table 1 ). We also found no evidence of effect modification physical activity ( P interaction: 0.93). When we used average greenness exposure in the year prior to antral follicle scan as our primary exposure, we observed similar associations ( Supplemental Figures 1 and 2 ). This was likely due to the high correlation between peak and average greenness exposures (ρ=0.90). For example, among women with an average exposure of 7 μg/m 3 PM 2.5 , a SD increase in average greenness exposure in the year prior was associated with a 5.2% (95% CI −2.5, 13.5) higher AFC. Conversely, among women with a PM 2.5 exposure of 12 μg/m 3 PM 2.5 , a SD increase in residential average greenness exposure in the year prior was associated with a 6.9% (95% CI −15.2, 2.2) lower AFC. We found no evidence effect modification by age ( P interaction: 0.67) or infertility diagnosis ( P interaction: 0.35) ( Supplemental Table 3 ).

Participants were part of the Environment and Reproductive Health (EARTH) study, 23 a prospective cohort designed to evaluate how the environment affects fertility. 23 Women, 18 to 45 years old, attending the Massachusetts General Hospital Fertility Center for infertility evaluation and treatment were invited to enroll in the study. Upon entry, participants completed a detailed questionnaire concerning their demographics, medical history, environmental exposures, diet, lifestyle, and reproductive health. Physical activity was self-reported on this questionnaire using a validated assessment 24 and was summarized as total hours per week. Height and weight were measured by a research assistant at study entry and used to calculate BMI (kg/m 2 ). Participants provided their residential address for reimbursement purposes which were then used for geospatial analysis. To be eligible for this analysis, women had to have undergone an antral follicle scan on or before December 2014, when greenness data was available (n=757). From there, we excluded women with incomplete scans, women on Lupron (a Gonadotropin-releasing hormone antagonist), women with polycystic ovary syndrome, or women missing air pollution exposure information ( Figure 1 ). Most of the women underwent one antral follicle scan, thus we also excluded repeated scans from women. The institutional review boards at Massachusetts General Hospital and the Harvard T.H. Chan School of Public Health approved this study. Participants provided written informed consent. We used AFC as a maker of ovarian reserve. AFC was measured by a trained reproductive endocrinologist using transvaginal ultrasonography. Scans took place on the 3 rd day of an unstimulated menstrual cycle or the 3 rd day following a progesterone withdrawal bleed. Only follicles above 2mm in diameter were included in the count. We examined AFC as continuous outcome but to reduce the influence of very high counts, we truncated AFC at 30 (12 women, 2.1% of population). We geocoded the women’s residential address information using ArcGIS StreetMap USA (ESRI; Redlands, CA), a nationwide street network for map visualization, geocoding, and routing. Match scores (on a scale of 0–100) were reviewed to determine the reason for scores <100. In all of the women with match scores below 90%, we manually checked and confirmed their address. Most often, lower scores were the result of slight misspellings of street names or incorrect street type abbreviations. To assess residential greenness, we used remote sensing data from the Moderate Resolution Imaging Spectroradiometer (MODIS) aboard the Terra and Aqua satellites operated by the United States National Aeronautics and Space Administration (NASA). Normalized Difference Vegetation Index (NDVI) values, the most widely used satellite-derived indicator of greenness, were derived from these data at a 250 m 2 resolution. 1 The NDVI quantifies vegetation by measuring the difference between near-infrared (which vegetation strongly reflects) and red light (which vegetation absorbs). NDIV measures greenness on the ground ranging between −1 and 1, with 1 representing maximal vegetation, 0 representing barren areas of rock, sand, or snow, and −1 indicating bodies of water. We chose 250 m 2 resolution to capture local and accessible residential greenness. We selected a single buffer closest to the residence because a prior study observed the strongest relationship between greenness and ovarian reserve hormones at the closest buffer (100 m around the geocoded residential address). 22 Since NDVI reaches its maximum and highest level of geographic variation during the height of the summer, we used the greenness estimate from the most recent July prior to the woman’s antral follicle scan date as our main exposure of interest and we referred to this exposure as peak greenness. Using a time-varying NDVI dataset containing measurements from January, April, July, and October of each year, we also calculated a woman’s average greenness exposure in the year prior to scan. Using the participant’s residential address, we estimated daily PM 2.5 exposure starting 3 months prior to the date of AFC scan. A 3-month window was selected given the antral follicle development window is approximately 2 to 4 months. 18 To estimate daily PM 2.5 , we used a validated hybrid model of satellite and land use data 25 with a 1 km 2 spatial resolution. These models used satellite-derived aerosol optical depth data, Multi-Angle Implementation of Atmospheric Correction algorithms from MODIS, land use data (e.g. measures of population density, elevation, traffic, percentages of land use, normalized difference vegetation index, and point and source pollutant emissions) from the US Geological Survey National Land Cover dataset, as well as meteorological conditions (e.g. air temperature, wind speed, daily visibility, sea land pressure, and relative humidity) to estimate ground-level exposure. We then averaged the daily values for the 3-month window prior to the participant’s scan date. Based on our previous work showing that PM 2.5 had a negative, linear association with AFC in this population, 21 we modelled this variable as a continuous variable per standard deviation (SD) increase (approximately 2 μg/m 3 ). We selected covariates a priori based on biological relevance. Covariates included age at scan (continuous), body mass index (continuous), smoking status (never vs. ever smoked), education (less than a college degree, college degree, vs. graduate/advanced degree), year of the scan (continuous) and season (Jan-Mar, Apr-Jun, Jul-Sept, vs. Oct-Dec). Because air pollution exposure is often negatively correlated with greenness, average PM 2.5 exposure was evaluated as both a confounder and effect modifier of the relationship between residential greenness and AFC. We used basic descriptive statistics, frequency distributions and means and standard deviations (SD), by quartiles of peak greenness, to describe the sample. We were concerned with potential overdispersion in AFC. We tested for overdispersion by fitting a negative binomial distribution and testing to see if the negative binomial dispersion parameter was equal to zero. Our test indicated overdispersion in AFC. To account for overdispersion, we used unadjusted and adjusted Poisson regression models with robust standard errors to assess the association between peak greenness and AFC. Non-linearity was assessed using restricted cubic splines. A likelihood ratio test was used to compare the model with the linear term to the model with the linear and the cubic spline terms. 26 We examined effect modification (e.g. multiplicative interaction) by average PM 2.5 exposure on the relationship between peak greenness and AFC by including a cross-product term in the adjusted models. To visualize the interaction effects, we used a contour plot and a slice plot. We also provide the percent change in mean AFC for a one SD increase in peak greenness at the 10 th and 90 th percentile of PM 2.5 exposure (7 μg/m 3 and 12 μg/m 3 respectively). We conducted several sensitivity analyses to test the robustness of the results. First, we included physical activity as a confounder in the adjusted model because physical activity could be used as a proxy for time outdoors which may influence exposure to greenness and PM 2.5 . Physical activity was not included in the main analysis because, while the relationship between physical activity and exposure to greenness is clear, we were concerned that physical activity might represent more of a mediator or pathway through which greenness affects health rather than a true confounder. We also examined effect modification by physical activity (stratified by the median: <4.5 vs. ≥4.5 hours/week). Next, we evaluated the associations between the average greenness exposure in the year prior to the scan (instead of peak greenness prior to scan) and AFC. We also examined effect modification by age (<35 vs. ≥35 years) and infertility diagnosis (female, male, vs. unexplained) as these are two of the strongest predictors of AFC. All analyses were conducted using SAS version 9.4 (SAS Institute; Cary, NC).

Higher exposure to residential greenness was associated with higher ovarian reserve but only in the context of low PM 2.5 exposure. These preliminary results suggest that residential greenness, a vital aspect of a woman’s built-environment, may play a small but potentially important role in dictating the pace of reproductive aging in women. If confirmed by other studies, these findings could provide valuable information to women when choosing where to live to enhance their reproductive health and to city planners on the best ways to design a built environment to maximize reproductive longevity in a rapidly urbanizing world. Additional longitudinal studies with repeated assessments of exposure and outcome are needed to confirm this association, further evaluate measures of greenspace quality and utilization, and clarify the mechanisms between greenness and health.

We observed that higher amounts of greenness surrounding a woman’s home was associated with higher AFC, but only when exposure to PM 2.5 was low (≤9 μg/m 3 ). At high levels of PM 2.5 exposure (≥12 μg/m 3 ), there was a suggestion that the relationship was negative, although this was based on sparse data as few women in our study had high exposure to PM 2.5 and low exposure to greenness. Only one other study has examined greenness and particulate matter exposure in relation to markers of ovarian reserve. In this study, serum concentrations of FSH and AMH were used to quantify ovarian reserve among 67 women in Sabzevar, Iran. 22 Women with higher levels of greenness in the 100 m buffer around their home had higher AMH levels. Moreover, higher annual exposure to PM 2.5 was negatively associated with AMH. Both associations persisted after adjustment for age, BMI, education, menstrual cycle regularity, parity and smoking. 22 FSH concentrations were unaffected by greenness and PM 2.5 exposure. Our results are somewhat in line with these findings, given that both studies observed negative associations between PM 2.5 exposure and ovarian reserve and beneficial associations between greenness and AFC (although ours was only among women with low exposure to PM 2.5 ). However, differences between the two studies are worth noting. While the demographics of women were quite similar, their environmental exposures were drastically different. The median (IQR) exposure to greenspace and PM 2.5 were 0.07 (0.01) and 42.2 (7.9) μg/m 3 in the study by Abareshi et al. compared to 0.59 (0.29) and 8.9 (2.2) μg/m 3 in our study. Therefore, the results are hard to directly compare since we had no women with similar environmental exposure profiles. Indirect evidence in support of our findings also comes from a prospective European study, which found that among 1955 premenopausal women at baseline, those with higher exposure to residential greenness during follow-up had an older age at menopause. 27 While there is conflicting evidence on whether markers of ovarian reserve correlate with timing of reproductive senescence, our study provides one potential biological pathway through which greenness may decelerate reproductive aging. Of note, neither of these past studies investigated the potential interaction between greenness and PM 2.5 exposure despite biological plausibility. In support of our finding of an interaction, several other studies have observed a similar interaction between air pollution and greenness. 28 – 31 For example, a recent prospective study from China found a beneficial association between residential greenness exposure and lower risk of gestational diabetes that was strongest among pregnant women with the lowest exposure to air pollution. 28 In a case-crossover study, the detrimental effect of PM 2.5 exposure on cardiovascular mortality progressively weakened with increasing exposure to NDVI in areas of lower socioeconomic status. 29 Similarly, in a case-crossover study from Hong Kong, elevated greenness around the patient’s home significantly attenuated the association between PM 2.5 exposure and increased risk of pneumonia mortality. 30 Several pathways have been proposed to explain the positive associations observed between greenness and a variety of health outcomes including; (1) reducing harm, (2) restoring capacity, and (3) building capacity. 11 In the reducing harm pathway, greenness exposure may potentially mitigates some of the negative influence of air pollution and other environmental/neighborhood exposures on ovarian reserve. 20 , 21 Our results support this hypothesis given the differing responses we observed of greenness at various PM 2.5 concentrations on mean AFC. For example, in our results, we observed that increasing greenness exposure was associated with higher AFCs but only when PM 2.5 exposure was low. However, additional studies are needed to confirm the differential associations between levels of greenness exposure and PM 2.5 exposure on ovarian reserve and other reproductive health outcomes. Restoring capacity, specifically through reduced stress, is another possible pathway mediating the association between greenness and ovarian reserve. 11 In previous studies, higher exposure to greenness has been linked to lower psychosocial stress and lower psychosocial stress has been also linked to higher ovarian reserve. 12 – 14 , 32 – 34 The third proposed pathway, building capacity though physical activity or social cohension, 11 could influence ovarian reserve but evidence is currently lacking for this mechanism. In our study, physical activity neither confounded nor modified the association between greenness and AFC indicating that building capacity through physical activity did not play a major role in this relationship. Future studies are needed to confirm these proposed pathways. The limitations of our study are worth noting. First, while our data came from a prospective cohort study, we only had a single measurement of AFC. Therefore, we were limited to evaluating the association between greenness and AFC across women rather than evaluating changes within a woman over time. We also only evaluated most recent peak greenness and past year average greenness exposure due to the lack of information on women’s residential history. Since women are born with all of their oocytes and this number declines over her lifetime, other time windows of exposure could be of interest. Ideally, future longitudinal studies with multiple assessments of greenness and ovarian reserve will be conducted to further this area of research. Second, while NDVI is the most commonly used metric to assign greenness exposure, it is a limited measure as it does not provide information on the specific types of vegetation such as trees, grass, or shrubs, nor does it describe the quality or accessibility of green spaces. 35 Additional measures of greenness exposure could be utilized by future studies including the Modified Soil-Adjusted Vegetation Index (MSAVI2), Vegetation Continuous Field (VCF), or distance to major greenspaces. By using 250 m 2 as the spatial resolution, we also assumed that pathways through which greenness influences AFC were limited to the women’s activity in the immediate area surrounding their residence. While findings from the sole prior study on this topic suggest that greenspace in more proximate residential buffers were better predictors of ovarian reserve, additional studies are needed to help determine the optimal scale. 22 Next, because this is an observational study, there is the potential for residual confounding. Even though we included several important demographic, lifestyle, and reproductive characteristics in our adjusted models and performed several sensitivity analyses to address this concern, the potential still remains. Another potential source of bias is reserve causality but the likelihood of this bias occurring in this study is low for several reasons. Specifically, the EARTH study was prospective in nature and women are unlikely to know their AFC prior to assessment, and even if it was known, there is limited evidence on environmental exposures influencing ovarian reserve so it is unlikely women would move to areas with high levels of greenness or low levels of PM 2.5 to improve their AFC. Finally, due to the sole inclusion of subfertile women undergoing infertility treatment at a single fertility clinic in New England, who were predominantly of older reproductive age, White, and of high socioeconomic status, it may not be possible to generalize our findings to all women. While the homogeneity of our study participants does potentially restrict the generalizability, it also reduces the potential for confounding by these characteristics. Additionally, because the study participants were recruited from an infertility clinic, there is a concern for selection bias. However, for this type of bias to be present, selection into our cohort would have had to been associated with both exposure and outcome, which we do not think is likely. Previous work has demonstrated that AFCs are similar between women who do and do not report a history of infertility. 36 There is also little reason to believe characteristics of a women’s residential built environment are strongly associated with her likelihood of seeking infertility treatment. Despite these limitations, our study has several strengths including its prospective design, large size, validated air pollution models, an objective measure of greenness, and gold standard assessment of ovarian reserve. 18 Additionally, we had the ability to account for a comprehensive set of reproductive and lifestyle factors as potential confounders and effect mediators.

The association between greenness, a measure of vegetation surrounding a residence or occupational setting, and health is a growing area of scientific interest. 1 Over the past 10 years, evidence has mounted demonstrating a potentially beneficial effect of exposure to residential greenness on chronic diseases such as type II diabetes, cardiovascular disease, and metabolic syndrome, 2 – 5 as well as pregnancy and birth outcomes. 4 , 6 – 10 The proposed mechanisms underlying the positive health effects of greenness include dampening the negative impacts of other environmental exposures (such as heat, noise, and air pollution), providing relief for mental and physiologic stress, and promoting physical and social activities. 11 Although these pathways are also implicated in the etiology of reproductive aging, 12 – 14 less research has focused on this outcome. By 2050 an estimated 66% of the world’s population will be living in urban areas, 15 with limited access to natural vegetation. During this same time period, an increasing number of women are expected to delay pregnancy until their later reproductive years, 16 when a sharp decline in ovarian function occurs. Therefore, understanding environmental factors, such as greenness, that may accelerate or slow ovarian aging is becoming increasingly relevant. Ovarian reserve is a marker of reproductive aging and fertility in women 17 and can be measured either directly via an antral follicle scan or indirectly via serum anti-Müllerian hormone (AMH) or serum follicular stimulating hormone (FSH) levels. 17 , 18 Previous epidemiological studies have linked a variety of environmental containments to reduced ovarian reserve including higher exposure to endocrine disrupting chemicals such as bisphenol A, parabens, and phthalates 19 and air pollutants such as nitrogen dioxide and fine particulate matter (PM 2.5 ). 20 , 21 In addition, a small cross-sectional study (n=67) from Iran found a positive association between the amount of greenness surrounding a woman’s home and ovarian reserve; 22 however, given the low levels of greenness (and high levels of particulate matter air pollution) observed in this study, it is unclear how generalizable these results are to high-income countries. 22 Therefore, we sought to further evaluate the association between greenness and ovarian reserve using a cohort of women in the United States where antral follicle count (AFC), a direct marker of ovarian reserve, was measured. Given the complicated interaction between greenspace and air pollution, we were particularly interested in exploring potential effect modification by PM 2.5 exposure, 18 as this may shed insight into the potential mechanisms of action and the best ways to maximize any potential health benefits of residential greenness.