The design of the present study, The Healthy Men Study, has been described previously (Olshan et al. 2007). The University of North Carolina School of Public Health’s Institutional Review Board approved the study protocol, and all study participants gave written informed consent. Briefly, The Healthy Men Study identified male partners of pregnant women who participated in a prospective study of drinking water disinfection by-products and spontaneous abortion [the Right From the Start (RFTS) study] (Promislow et al. 2004; Savitz et al. 2005, 2006). Men were prospectively identified from the RFTS study and recruited from the three RFTS study sites (Wake County, NC; Shelby County, TN; Galveston County, TX). Men eligible for this study were 18–40 years of age. Each participant provided a reproductive history, but only men who had undergone a vasectomy or chemotherapy were excluded.

A computer-assisted telephone interview was administered to each participant by experienced interviewers, with responses entered directly into a computerized database (Luben et al. 2007). The average duration of the interview was approximately 40 min. Questions covered the following topics: general lifestyle, health, reproductive history, environment, diet, stress, occupational exposures, hobbies, and demographic factors.

The methods for semen collection and analyses have been described in detail previously (Luben et al. 2007). Participants were asked to provide a single semen sample using a special kit designed to allow the man to collect a semen specimen in the privacy of his own home and at a time convenient to him (Royster et al. 2000). Before sending the kit to the participant, study staff confirmed by telephone the participant’s mailing address, gave brief instructions on how to use the kit, and asked the participant if he had previously used a similar collection method. Verbal instructions included the importance of doing the collection after 2–7 days of abstinence from sexual activity (Luben et al. 2007). A pilot study confirmed the stability of semen outcomes after simulated overnight shipping at 70°F or 40°F. Under all conditions, results were comparable except those for the sperm chromatin structure assay–DNA fragmentation index (SCSA-DFI), which increased significantly after shipping at 70°F but not 40°F (Perreault SD, unpublished data). Therefore, samples were shipped overnight with cold packs. The instructions accompanying the kit included photographs and instructions on how to properly collect the specimen; package the sample with cold packs and prepare it for shipping; and call to arrange the courier pickup (Luben et al. 2007). If the initial specimen volume was very low (< 0.5 mL), the man reported spillage or incomplete sample collection, shipping was delayed or the sample was not packaged correctly, or the participant’s abstinence interval was too far outside of the suggested 2- to 7-day range, participants were asked to provide a second or third specimen. This affected 10% (n = 20) of participants, and all 20 complied with the repeat collections.

All samples were processed upon receipt at a spermatology laboratory in the Reproductive Toxicology Division of the U.S. Environmental Protection Agency’s National Health and Environmental Effects Research Laboratory (Research Triangle Park, NC) by technicians trained in human semen analysis. All semen analysis protocols included quality control charts and competency review.

Immediately upon receipt, semen volume was measured and aliquots removed for determination of sperm concentration by IVOS-IDENT (Integrated Visual Optical System; Hamilton Thorne Research, Beverly, MA) (Zinaman et al. 1996) and calculation of total sperm count. Smears prepared from additional aliquots were air-dried and stored for later analyses of sperm morphology [World Health Organization (WHO) 1999]. Sperm motility, which declines over time and is therefore not a reliable measure for shipped semen, was not included in the statistical analysis. However, sperm motility and viability (using propidium iodide as a vital stain) were monitored. All samples retained motile and viable sperm, an indication that the sample had been collected and shipped according to instructions. Samples with low volume (< 0.5 mL) or evidence of spillage (or if the man reported incomplete collection) were discarded and another sample requested.

Additional aliquots (0.1 mL) were frozen and stored at −70°C for later analysis of chromatin integrity by the SCSA (Evenson and Jost 2000) and for chromatin maturity by chromomycin A3 (CMA) staining (Sakkas et al. 1995). For the SCSA, aliquots were shipped on liquid nitrogen to SCSA Diagnostics (Brookings, SD) for analysis according to established methods (Evenson et al. 2002). SCSA software calculates the percentage of sperm with fragmented DNA (%DFI). SCSA has been shown to be a highly reproducible test to measure %DFI, which compares favorably with other tests of sperm DNA damage such as the TUNEL (terminal deoxynucleotidyl transferase dUTP nick end labeling) and comet assays (Lewis and Agbaje 2008). The CMA assay is based on the stainability of sperm with CMA3, which detects sperm deficient in protamine as a characteristic of immaturity (Bianchi et al. 1993). We considered sperm to be CMA3 positive when at least 50% of the area of the nucleus fluoresced above background. Clinical studies have shown an association between relatively high percentages of CMA3 staining and subfertility or infertility (Bianchi et al. 1993; Esterhuizen et al. 2000). In both assays, aliquots of pooled semen were included in each run to serve as an internal standard.

We examined nine sperm outcomes reflective of testis function: a) sperm count (millions), b) sperm concentration (millions per milliliter), c) sperm morphology (percent normal sperm), d) percent of sperm cells with abnormal head, e) percent of sperm cells with abnormal midsection, f ) percent of sperm cells with abnormal tails, g) percent of sperm cells with cytoplasmic droplets, h) percent sperm with DNA fragmentation according to SCSA, and i) percent immature sperm according to CMA staining.

We obtained air pollution data from the U.S. Environmental Protection Agency (EPA) Air Quality System Data Mart (U.S. EPA 2008) for the period of exposure in each of the three counties where the study subjects resided. Air pollution data were originally obtained for PM₁₀, PM_2.5, O₃, nitrogen dioxide (NO₂), SO₂, and CO. However, in this study we focused on exposure to PM_2.5 and O₃ because the data for these pollutants were the most complete across all three counties for these pollutants (data for CO, NO₂, and SO₂ were monitored at only two of the three study sites during the study period). Meteorologic data were also obtained in the form of the daily minimum and maximum temperature readings.

For PM_2.5, the data represent the 24–hr average, which was collected at two monitoring sites in Wake County, North Carolina (recorded daily at one site and every third day at the other), four monitors in Shelby County, Tennessee (two recorded daily and two every third day), and one monitor in Galveston County, Texas (recorded every third day). For O₃ (reported as parts per billion), the data represent the maximum 8–hr average, which was collected at four monitors in Wake County and at two monitors each in Shelby County and Galveston County (all sites recorded daily data). For the daily maximum temperature, data were recorded at six monitors in Wake County, five monitors in Shelby County, and three monitors in Galveston County.

To allow for more complete exposure data, we interpolated the values for the 2 missing days between each 3-day reading for the monitors that measured PM_2.5 concentration every third day. We used the PROC EXPAND procedure (version 9.1; SAS Institute Inc., Cary, NC) in which the successive nonmissing values were connected with straight lines (using the JOIN method). However, there were several periods in which a 3-day reading was missing, which created periods of > 2 days of missing data. The missing values within these periods were left as missing because we determined that it was too long a period to interpolate the data.

To estimate the daily level of PM_2.5, O₃, and maximum temperature within each county, where possible, we calculated an average across the multiple monitoring sites within the county; otherwise, the daily reading was obtained from one monitor within that county (e.g., PM_2.5 in Galveston County). The estimated air pollution and temperature time series within each county was then linked to the 90–day period before semen sampling for each subject. For PM_2.5 and O₃, we then calculated an average exposure over the windows that represent important points of spermatogenesis (i.e., 0–9, 10–14, 70–90, and 0–90 days before sampling). We used the daily maximum temperature, rather than the average temperature, over each exposure period to facilitate comparison with the results presented by Sokol et al. (2006); the temperature variable in this analysis represents the number of days > 90°F within each exposure period.

We performed statistical transformations on several of the outcome variables to better approximate the normality assumption of the linear model. Specifically, we applied a natural log transformation to the sperm count and concentration variables, and an arc sine-root transformation to the percentages of normal sperm cells; sperm cells with abnormal head, midsection, or tail; and sperm cells with cytoplasmic droplets.

For interpretability, each of the outcome variables was standardized (after statistical transformation, if applied) such that both the SD and the variance were equal to 1. Thus, each regression coefficient provides an estimate of the effect in terms of a change in SD of the response variable.

We characterized the distribution of demographic, exposure, and other characteristics for all participants and by individual study site. We first examined differences in demographic characteristics across the three sites using a chi-square test of independence for the categorical variables and analysis of variance (ANOVA) on age when examined as a continuous variable. In addition, we conducted bivariate analyses for all covariates and exposure variables with each of the outcome variables.

We used linear regression to assess the association between each exposure variable and outcome, adjusting for potential confounders. We made an a priori decision to adjust for the same variables used in the model specified by Luben et al. (2007). Additionally, we included season and temperature in subsequent models to compare the results with those of Sokol et al. (2006).

For the main analyses, we used three linear regression models. The base model (model 1) consisted of the air pollution exposure metric adjusted for age (indicator variables with > 35 years as the reference), days abstaining (indicator variables with > 8 days as the reference), education (indicator variables with “some college” as the reference), and smoking (indicator variables with nonsmoker as the reference). The categories for the covariates are shown in Table 1. Model 2 included the base model plus season of the semen sample (indicator variables with “winter” as the reference). Model 3 was model 2 plus temperature (the number of days > 90°F during the exposure window). We chose the > 90°F cutoff arbitrarily to facilitate comparison with the results of Sokol et al. (2006). Several studies have suggested that testicular function is influenced by season (Gyllenborg et al. 1999; Levine et al. 1988, 1990, 1992); this may account at least in part for the reduction in spring births in regions with warm climates, although it is unclear if this effect is related entirely to temperature or if there may be some other seasonal component, such as photoperiod, that leads to this phenomenon. The air pollution exposure metric was entered into the models as a continuous variable, and the β–coefficients are presented for a 15–ppb increase in O₃ and a 10–μg/m³ increase in PM_2.5.

To be included in the final analyses, which examined the exposure period 0–90 days before sampling, the subject had to have at least 45 days (50%) of exposure data available. For PM_2.5, 80% (n = 183) of subjects had data for all 90 days, 14% (n = 32) had 70–89 days of data, and 6% (n = 13) had < 70 days (minimum number of days, 58). For O₃, 75% (n = 171) of subjects had data for all 90 days, 4% (n = 8) had 70–89 days of data, 7% (n = 16) had 50–69 days of data, 5% (n = 11) had 30–49 days of data, and 10% (n = 22) had < 30 days of data (6 subjects had zero days of data). For the inclusion criteria of 45 days, all subjects had at least 45 days of PM_2.5, and 87% (n = 199) had at least 45 days of O₃.

Similarly, for analyses that examined periods < 90 days (i.e., 0–9, 10–14, and 70–90 days before sampling), a subject was required to have at least 50% of available data. For the 0–9 day period each subject was required to have at least 5 days of data; for the 10- to 14-day period, 3 days; and for the 70- to 90-day period, 10 days.