Cord lymphocyte distributions varied with exposure to ambient pollutant levels over the nine gestational months. Overall, exposures to total PAHs or PM2.5 during early gestation were associated with altered distributions of lymphocyte immunophenotypes in cord blood: increases in CD3+ and CD4+ lymphocyte percentages and decreases in CD19+ and NK cell percentages. In contrast, exposure during late gestation was associated with decreases in CD3+ and CD4+ fractions and increases in CD19+ and NK cell fractions. There was no significant association between any of the lymphocyte distributions and either air pollutant during the middle of gestation, generally months three to six.
Studies have reported first trimester exposures to be harmful for fetal development. First trimester exposures to lead  or the pesticide dichlorodiphenyltrichloroethane  have been reported to have an adverse association with neurodevelopment and exposure to organic solvents increased the risk of congenital malformations . In this study we observed that exposure during the early months of pregnancy was associated with an increase in the percentage of T cells (CD3+ and CD4+) and a decrease in the percentage of B cells (CD19+), which may influence the Th1/Th2 homeostatic balance or create an environment that may lead to autoimmunity. Maintaining a Th2 skewed environment for much of the gestation is necessary for a successful pregnancy , but higher levels of air pollution exposure during the first trimester may alter this balance. However, this mechanism does not depend on percentages of the T helper cells alone but also on circulating cytokine levels.
The associations we observed during late gestation between lymphocyte phenotype percentages and air pollution exposure were similar to the findings we reported earlier in relation to pollutant exposures 14 days prior to birth  and for more chronic exposures resulting from residence in a high air pollution area . Possibly the reduced T-cell percentages and increased B- and NK cell percentages are associations of more than the 14 days prior to birth, reflecting a longer period of late prenatal gestational exposure. Adjusting our models for short-term air pollution exposure (14-days prior to birth) yielded similar point estimates, but because of the correlation between the monthly and 14-day average we could not attempt for a tight control using a continuous scale. Thus, it remains uncertain whether the effects observed for the last months of gestation were confounded, with the primary impact actually arising from the last few weeks, or whether the 14-day results were confounded by associations from a somewhat longer relevant time period of late gestational exposures. In other words, the effective exposure may have been a few weeks or a few months. In either case, air pollution appears to be associated with a shift in lymphocyte production or survival based on fractions observed at birth.
Our findings indicate that timing of exposure seems to determine the direction of associations between air pollutants and lymphocyte distribution in cord blood. Alternatively, these associations might be an artifact related to seasonality in air pollution, whereby PAHs and PM2.5 peak in the winter and reach a nadir in the summer. The strongest negative correlations for the pollutant levels were those that were six months apart, e.g., months one vs. seven, two vs. eight,and three vs. nine. This was reflected to some extent in the magnitudes of the pollutant associations with lymphocyte proportions. For instance, PM2.5 and CD4+ or CD3+ cells showed the most extreme opposite associations in months one and seven, but months two vs. eight did not show such extremes. Had the inverse associations with lymphocytes comparing first and third trimester exposures been due to seasonality of air pollutants one might expect comparable magnitudes of associations for all these pairings.
Rather, the pattern for many associations appeared to be virtually monotonic (or close to it) over the course of gestation: PAHs and CD3+ cells, PAHs and CD4+ cells, PAHs and CD19+ cells, PAHs and NK cells, and PM2.5 and CD19+ cells. Thus, it is possible that the associations are truly a function of the timing of gestational exposure. This interpretation is supported by the robustness of the overall findings to adjustment for season. It also supports the notion that the previously reported finding for exposure in the 14 days prior to birth represented an association with a late gestational exposure longer than 14 days. In short, the pattern of results demonstrates that seasonality alone cannot account for the qualitatively opposite associations for first and third trimesters.
Some of the monthly air pollution averages had a moderate to high correlation. Although we fit separate models for each of the gestational months, one must be cautious not to infer that one specific month is critical while another is not. The high correlations precluded adjustment of any specific month of exposure for the surrounding months.
It has long been known that the timing of fetal exposures can affect developmental outcomes. Similar to our findings for lymphocyte distributions, a biomarker for in-utero organophosphate exposure showed the opposite associations with fetal growth when exposure occurring during four to seven weeks of gestation versus 25-28 weeks . Cohen et al. reported third trimester exposure to fluoxetine (a commonly prescribed antidepressant) was associated with three-fold higher rate of newborn complications including admission to a special care nursery, compared to early exposures . Research on in-utero exposure to maternal tobacco smoking has established that the strongest associations on birth weight occur with late gestational smoking, but several studies have also reported associations with early exposure [30, 31]. An iron-deficient maternal environment during the first trimester, but not the second and third trimesters has been reported to be a risk factor for low birth weight . To our knowledge, no previous studies have examined pollution exposures in specific time windows of gestation and immunophenotype outcomes at birth.
We examined the relative distribution of immune phenotypes with each lymphocyte subtype presented as a percentage of all cells counted and not absolute counts. If the percentage of one subfraction decreases, the percentage of at least one other will increase. Although desirable, measurement of absolute counts was not feasible due to the conditions of data collection and the need to transport samples to Prague. Nevertheless, the redistribution of lymphocyte immunophenotypes can potentially alter susceptibility to infection or to inflammatory responses. Schultz et al. reviewed life-time longitudinal patterns of lymphocytes and concluded that the developmental trajectory of absolute counts differs from that of relative distributions . They further pointed out both absolute and relative sizes of lymphocyte subpopulations should be considered relevant variables for assessing maturation of the immune system.
Immunological development is intimately connected to the interactions between the organism and the environment via antigenic challenge; thus, a role for environmental pollutants in influencing such development is highly plausible. Studies have reported reduced thymus size at birth and effects on fetal immune cell counts following prenatal polychlorinated biphenyls or dioxin exposure [34, 35]. In fetal mice, hypocellularity and atrophy of the thymus were associated with PAH and BaP exposure respectively [36, 37], making our findings all the more relevant given that the thymus plays a critical role in differentiation, maturation and selection of T lymphocytes. Whether alterations in developmental patterns of lymphocytes have clinical implications remains to be established, but the results of our study suggest an initial link in what might be a chain of causation. Implications for childhood respiratory morbidity and atopic conditions will require further investigation.
A few data quality issues deserve consideration. Exposure data for some months, especially in the summer, had to be imputed because of the pattern of monitoring in the air-quality field measurement program, but since we focused on monthly averages, the impact of the imputations was likely to have been small. Moreover, the probability that missing data were related to the actual pollutant levels, after conditioning on daily measurements for SO2 and NOx and on the existing data for PM10, PM2.5 and PAH, is likely to be very small, supporting the assumption that these data were missing-at-random.
There was only one air monitor in operation in each district, which likely resulted in exposure misclassification. Area-wide measures of exposure to air pollution such as those obtained from fixed-site monitoring stations generally produce smaller estimates of pollutant associations compared to those determined by personal sampling . Accordingly, nondifferential measurement error would tend to attenuate associations, such that the true associations may be larger than observed, but we are unable to assess the extent of this problem.
Participants of this study were randomly selected from the "pregnancy outcome study" (as described in the methods). Also, the refusal rate for this study was five percent, supporting the external validity of the results. The study sample was similar to the full cohort with respect to most of the variables; differences remaining for several variables were controlled for in the analysis through model selection and through adjustment for the sampling design by using sampling weights. Thus the results were assured generalizability to the full population. Furthermore, these results could potentially be generalized to other populations exposed to comparable levels of similar pollutants within the Czech Republic , Europe , and some cities in the USA .