Small but Mighty: Prenatal Ultrafine Particle Exposure Linked to Childhood Asthma Incidence.

Research demonstrating a role for air pollution in the prenatal programming of asthma has largely considered exposures to criteria pollutants (pollutants routinely monitored to assess air quality), especially particulate matter with an aerodynamic diameter of 2.5–10 μm or fine particles ≤2.5 μm (PM2.5) and ambient nitrates (NO3) (1). Although air quality regulations currently do not address ultrafine particles (UFPs) ≤0.1 μm, submicron-sized particles may exert greater toxic effects compared with larger molecules because of their larger surface area/mass ratio, chemical composition, deeper lung penetration, and enhanced oxidative capacity and ability to translocate to the systemic circulation (2–4). A recent workshop identified the lack of studies differentiating the effects associated with UFP exposures from effects related to other particle size fractions and gaseous copollutants as a significant gap in the evidence needed to move toward regulation of UFPs (5).

In this issue of the Journal, Lavigne and colleagues (pp. 1487–1495) address this gap directly, as well as contributing further innovations to this field (6). The retrospective cohort combined data from a province-wide birth registry in Toronto, Ontario, Canada, with health administrative data identifying incident asthma cases during a 9-year period. They assigned pollution exposure levels during each week of pregnancy and each month of childhood at the centroid of the postal code for individuals (approximating a city block). A particular strength of the analysis is that the researchers were able to adjust for other components of ambient pollutants from similar sources, including PM2.5 and nitrogen dioxide (NO2). This is the first large-scale epidemiologic study to demonstrate independent risk associated with in utero UFP exposure, as well as corroborating previous findings linking prenatal PM2.5 and nitrate exposures to asthma incidence (7, 8).

Because fetal development occurs through sequential biologic events, toxins that disrupt these processes can have a variable effect, depending on the nature of the pollutant, as well as timing and/or exposure level. The authors use distributed lag functions, an effective way to model critical windows of exposure in a more objective fashion (8). The focus on critical windows makes the temporal aspects of the exposure modeling central to their findings. The authors base their exposure modeling of the spatiotemporal variability in UFP exposures in the study region from 2006 to 2015 on 3 weeks of exposure data collected as part of a mobile monitoring campaign conducted over the course of 2 weeks in summer and 1 week in winter. They build a land use regression model that includes characteristics of a given location (distances to major roads, bus routes, etc.) to capture variation over time. An immediate question is how well this model trained on data from a highly targeted monitoring campaign conducted on the order of weeks extrapolates to the scale of nearly a decade. They report that the land use regression model explained 67% of the variation in mean UFPs. To better address the extrapolation of these model predictions to longer-term timescales, they use a temporal adjustment that computes ratios of weekly averages of PM2.5 and NO2 concentrations at each ground monitor location to the long-term estimates derived from their exposure models, spatially smooth these scaling factors on a weekly basis, and then apply these predicted scaling surfaces to their long-term estimates to obtain exposure measures at each residential location. To our knowledge, this scaling approach has not been reported previously. It would be interesting to know the out-of-sample predictive performance of this overall procedure, calculated on the basis of data from monitors held out of the modeling exercise. Moreover, it would be interesting to calculate these prediction statistics at timescales of interest: weekly, monthly, and trimesters. This would allow one to assess how well the proposed exposure assessment captures variation in UFP monitoring data at the scales of greatest interest in the health effects study.

The group implemented an approach that incorporates the estimation of critical windows of exposure via a distributed lag function in a multilevel Cox proportional hazards framework that accounts for clustering among children by family and by postal code. Another advantage of the distributed lag model is that in settings in which there is a strong seasonal trend in a pollutant, such as that observed for PM2.5, the distributed lag function avoids the temporal confounding that arises in some simpler modeling approaches, such as separate models that use a single trimester-averaged exposure (9). Because trimesters are approximately the same length as a season, seasonal trends can result in unique correlation patterns among trimester-specific exposures. Notably, in Lavigne and colleagues (6), the trimester-specific UFP exposures do not exhibit such patterns, with all pairwise correlations around 0.6. It may be that for UFPs, the spatial variability is much higher than the temporal variability, so that seasonal variability is a smaller fraction of overall variability in UFP concentrations as compared with that for PM2.5, highlighting that considerations for modeling critical windows of exposure can differ according to the specific pollutant under study.

When considering effect modification, it was unclear whether they included interactions in the distributed lag models or simply in the aggregated pregnancy averaged models. In distributed lag models, the lagged function can vary in magnitude, in the location of the critical window, or both. Thus, it can be useful to parameterize the distributed lag function as an overall effect times a weight function that characterizes the critical window of exposure, and to allow either or both of these features to vary by subject characteristic (10).

Notably, local and systemic induced oxidative stress is central to toxic effects of PM, with UFPs having even greater oxidative potential than other particle fractions. As optimal placental oxidant balance is critical for normal fetal development (11), signaling mechanisms operating at the level of the placenta are receiving increased attention with regard to the prenatal programming effects of air pollutants (12). Epigenetic mechanisms are already being examined in this context (13, 14). Recent studies focus on placental extracellular vesicles (EVs) as targets of environmental toxins. Placenta-derived EVs (i.e., 0.05- to 1-μm membrane-bound vesicles released into maternal and fetal circulations over gestation) prime maternal and fetal tissues via signaling molecules encapsulated in EVs (mRNA, microRNAs, DNA, lipids, and protein mediators), which interact with adjacent or distal cells to reprogram their phenotype and regulate their function. Although we cannot sample placenta during pregnancy to measure biological changes induced by particle exposures, we can readily assess the placental EVs and their content in a defined critical window via a maternal blood draw, conceptualized as a placental liquid biopsy (15). Figure 1 depicts a future strategy that would allow researchers to more comprehensively interrogate environmentally responsive mechanisms operating in the placenta, using a system biology approach that may provide markers of early risk and further elucidate underlying mechanisms.

Figure 1.

Placenta as a target of prooxidant pollutant effects: potential enhanced toxicity of ultrafine particles. Ultrafine particles penetrate deeper into the lungs, have greater ability to induce oxidative stress, and more readily enter the systemic circulation in mothers, all features that may enhance toxicity in this schema. Programming effects may result from pollutant-induced shifts in a number of molecular, cellular, and physiological states and their interacting systems in mothers and children. Specific key regulatory systems susceptible to programming may influence vulnerability to respiratory diseases including both central and peripheral components of neuroendocrine pathways and autonomic nervous system functioning, which, in turn, influence the immune system. ANS = autonomic nervous system; HPA = hypothalamic–pituitary–adrenal; miRNAs = microRNAs. Illustrated by J. Gregory (Mount Sinai Health System).