Clearing the Air on Reproductive Health: Unpacking the Impact of <scp>PM₂</scp>_.5 Constituents on Fecundability

Infertility remains a pressing public health challenge, affecting 10%–17% of couples globally in their lifetime [1]. While characteristics such as age and lifestyle behaviours are well-documented risk factors, emerging evidence points to environmental exposures—specifically, air pollution—as significant contributors to fertility outcomes. Previous studies examining the association between particulate matter (PM) with a diameter less than or equal to 2.5 μm (PM2.5) and fertility provided mixed results in spontaneous conceptions and IVF populations [2-4]. One of the possible explanations for these mixed results is the difference in PM2.5 composition, which varies over both time and space [5]. In this issue of Paediatric and Perinatal Epidemiology, Wesselink and colleagues [6] offer a critical advancement in this area by investigating how specific chemical constituents of PM2.5, rather than total PM2.5 mass alone, are associated with fecundability, a robust marker of fertility. They used data from an internet-based preconception cohort study of 5905 Danish couples trying to conceive linked with PM2.5 constituents from spatiotemporal models across each menstrual cycle. Among seven constituents, primary organic aerosol (POA) had the strongest association with fecundability (fecundability ratio (FR) 0.93, 95% confidence interval [CI] 0.87, 0.99). Notably, the study adds to the ongoing discussion regarding whether the health effects of PM2.5 are driven primarily by particle size, chemical toxicity, or both. This commentary will focus on the additional considerations and analytical challenges to infer causality beyond those encountered in research focused solely on PM2.5 mass. The data used to assign exposures to PM2.5 constituents are subject to measurement errors, generally larger than those for total PM2.5 mass, potentially influencing point estimates and standard errors in health outcome analyses. The extent of measurement error can also vary between pollutants, which may lead to stronger and more precise associations for certain constituents simply because they are measured more accurately than others. When a constituent's daily levels are extremely low, they may fall below the detection limit, leading to unstable concentration estimates. Additionally, PM2.5 mass is often correlated with both constituent concentrations and health outcomes, making the estimation of the constituents' causal effect complex (Figure 1). If a constituent makes up a substantial portion of PM2.5 mass or shares a similar exposure pattern with PM2.5 (e.g. due to a common source), it may appear more strongly linked to health effects—not because of its inherent toxicity, but due to its association with PM2.5. Since constituents often serve as indicators of pollutant mixtures from the same source, findings related to a specific constituent reflect not only its effects but also those of other co-occurring pollutants with similar exposure patterns. Although many constituents contribute minimally to total PM2.5 mass, associations with fecundability may still be observed because, despite their small mass, they can exhibit high toxicity individually or in combination with other co-pollutants. Additionally, the concentrations of many constituents are influenced by the same meteorological conditions, leading to strong correlations in their daily fluctuations. Some constituents also act as tracers for specific pollution sources. As a result, health outcome estimates for a given constituent reflect both its direct effects and those of related constituents with similar sources or exposure patterns. Exposure misclassification in pollution studies mainly stems from two types of measurement error [7]. Berkson-type errors occur when pollutant concentrations averaged to the available spatial scale (e.g. county, zip code) are used as proxies for individual exposures. While these errors typically introduce minimal or no bias when the true dose–response is linear, they can decrease statistical power. In contrast, classical measurement errors arise from discrepancies between an individual's exposure and the actual ambient pollutant concentration, often biassing associations toward the null [8]. This attenuation is more pronounced when the surrogate measure's error variance is high relative to the variance of the true exposure. When two correlated pollutants are included in a linear model, where only one of them has a true harmful effect, the second (non-harmful) pollutant may seem harmful if there is a strong negative correlation between their measurement errors. These complexities highlight the challenges of accurately estimating pollutant and constituent-specific health effects in the presence of exposure misclassification. Multicollinearity, interaction effects and nonlinearity present major challenges in estimating the combined health effects of PM2.5 constituents [9]. Multicollinearity, which arises from shared sources and exposure pathways among PM2.5 constituents, can result in unstable parameter estimates or model convergence issues [10]. Interaction effects are often assessed using product terms of constituent pairs, but this method can reduce statistical power. Moreover, traditional linear models may yield misleading or biassed estimates of interaction effects when applied to constituents with nonlinear relationships [9]. Advanced methods such as quantile-based g-computation (QGC) and Bayesian kernel machine regression (BKMR) have been developed to address these issues. QGC, based on weighted quantile sum regression, simplifies the analysis by converting constituents' concentrations into quantiles, though this transformation can result in some loss of information [9]. In contrast, BKMR uses Gaussian kernel functions to model exposure-response relationships, allowing for flexible estimation of nonlinear and non-additive effects [11]. BKMR also mitigates multicollinearity through hierarchical variable selection, grouping highly correlated constituents, and has been successfully applied in studies to evaluate the combined effects of pollutant mixtures. Wesselink et al. utilised proportional probabilities regression models to estimate fecundability ratios (FRs), allowing for interpretable and policy-relevant findings. Furthermore, after adjusting for all PM2.5 constituents to account for their complex interactions, the association with POA became even stronger (FR = 0.90, 95% CI: 0.78–1.04). As research continues to uncover the relationship between PM2.5 constituents and reproductive health, understanding the specific chemical constituents of PM2.5 is crucial for identifying potential risks to fertility. This study by Wesselink et al. contributes to this growing field by highlighting the importance of differentiating between PM2.5 mass and its constituents when evaluating health effects. However, significant challenges remain, including exposure misclassification, statistical complexities and the need for more refined analytical methods to disentangle these pollutants' independent and interactive effects. Future studies should continue to integrate advanced statistical approaches, such as BKMR and QGC, alongside biological evidence from human and animal models to strengthen causal inference. Additionally, further exploration of source-specific PM2.5 constituents and their mechanisms of toxicity could inform targeted regulatory interventions to mitigate reproductive risks. By addressing these challenges, research in this area will advance scientific understanding and contribute to public health policies aimed at reducing air pollution's impact on fertility and overall reproductive well-being. Stefania Papatheodorou declares no support from any organisation for the submitted work, no financial relationship with any organisations that might have an interest in the submitted work in the previous 3 years, and no other relationships or activities that could appear to have influenced the submitted work. The author declares no conflicts of interest.