Workers were recruited from 13 free-stall dairies with over 1000 cows in California’s San Joaquin Valley. Each farm was visited for 2 to 7 consecutive days. The control facility was a vegetable processing plant in Gilroy, CA. The control was chosen because it had similar socio-demographic characteristics as the dairy workers and lacked exposure to animals. Single cross-shift measurements of each worker were collected from June to October 2008. Further details of participant recruitment and health measurements are described in [28] .

Following UC Davis Institutional Review Board approval, subjects were recruited with the following eligibility: age 18+ years, male, able to wear the samplers for their entire work-shift (more than 6 hrs), speak English or Spanish, and able to perform Pulmonary Function Tests (PFTs). Detailed health measurements, demographics, study design, and analysis can be found in [27] and [28] . This study had a high recruitment rate at 93% of the workers consenting to take part in the study. A total of 226 dairy workers and 52 control workers were enrolled. In the present paper 185 dairy workers and 45 control workers were used in the analysis. Subjects were excluded for the following reasons: 29 dairy workers and 4 control workers did not have usable spirograms, 12 dairy and 3 controls workers with PM_2.5 concentrations falling more than 1.5 inter-quartile ranges above the 75^th or below the 25^th quartile in boxplots, as in [29] . As mentioned in [28] , participants excluded from analysis because of spirometer values that were not acceptable were defined as those greater than 8% variability in pre- or post-shift between maneuvers. Pre- and post-shift questionnaires were administered by trained interviewers. For dairy workers, the pre-shift questionnaire included demographics, health history, current health, work history, and the post-shift questionnaire asked about tasks performed on the dairy the day they wore the personal monitor and symptoms, as well as other behaviors, such as Personal Protective Equipment (PPE) use. PPE included use of mask, N95 respirator, or cartridge respirator. Use of a bandana was not included in the definition of using PPE. For control workers, a simplified version of the dairy questionnaire was given. This included demographics, health history, current health, and work history, but did not include tasks.

Pulmonary function was measured using an Easy One Portable Spirometer (NDD Medical Technologies, Andover, MA). The National Institute for Occupational Safety and Health (NIOSH) protocol was followed and technicians were NIOSH certified. Statistical modeling of pulmonary function changes examined a mixed effects model that jointly examines pre- and post-FEV₁ and FVC by including a shift factor (before the start of the shift and at the end of the shift). FEV₁ is the volume of air forcefully expelled from the point of maximum inhalation in the first second and is an indicator of obstructive conditions reducing exhalation speed. FVC is the maximum volume of air exhaled forcefully from the point of maximum inhalation and indicates restrictive conditions and very severe obstructive diseases.

A single cross-shift measurement was performed on each enrolled worker. Workers wore personal monitors for the duration of their work-shift, including lunch and breaks. Sampling methods and analysis can be found in further detail in [29] . Briefly, workers were fitted with a personal monitor backpack. The backpack included one Leland Legacy personal sampling pump (SKC, Inc., Eighty Four, PA) connected to a GK2.05SH (KTL) cyclone sampler and a SKC button sampler (SKC, Inc. 225 - 360, Eighty Four, PA) via Tygon tubing. The GK2.05SH (KTL) cyclone sampler was fitted with a Teflon 37 mm, Millipore filter with a pore size of 0.45 μm (Fisher, FHLP03700). The SKC button sampler was fitted with a 3.0 μm pore size, Teflon 25 mm, Millipore PTFE filter (Fisher, FSLW02500). Airflow to both samplers was adjusted through a Swagelok needle valve (Swagelok, Solon, OH). Airflow for both personal and area samplers were calibrated using a Defender-series (Bios International, Butler, NJ) electron piston volumetric gas flow meter at 3.5 L/min for the cyclone sampler and 4.0 L/min for the button sampler.

Filter samples were analyzed for biologically active endotoxin using recombinant Factor C (rFC) assay (Lonza, Inc., Walkersville, MD). Samples were shipped over night on dry ice to Colorado State University for rFC analysis. The method was described in detail in [29] . Any result below the limit of detection (LOD) was assigned a value of LOD divided by the square root of 2.

Stata 12 software was used for all statistical analyses. Exposure measures (i.e., IPM, PM_2.5, endotoxin in IPM fraction and PM_2.5 fraction) were log-transformed to correct for positive skew and mean-centered to reduce multicollinearity when examining interactions. Multiple linear regression models were estimated for exposure measures to examine differences between dairy and control workers. Mixed linear regression models were estimated to examine the association between worker exposure to pollutants (explanatory variables) and cross-shift lung function as measured by FEV₁ and FVC (response variables). Lung function changes and interactions by pollutant exposures during the shift were examined as time main and interaction effects, where the pre-shift lung function measure was coded as time 1 and the post-shift measure as time 2 (pre-shift minus post-shift lung function measures). Separate analyses were conducted for FEV₁ and FVC lung function measures. In each case, the lung function measure at pre- and post-shift was the response variable, and the four exposure measures, time (pre-shift and post-shift) and other covariates (age, race, height, years in the US, country or origin, education level, family income, smoking status, exposure to second-hand smoke, number of days since last day off, PPE use, years worked on a dairy, length of work-shift, hours spent in an enclosed cab, dairy facility, herd size, tasks, RH, and temperature) were entered as explanatory variables (Table 1). In the first stage of the analysis, all explanatory variables were entered as main effects. Interaction effects of exposure measures with shift and the other covariates were entered in stage two. Statistically nonsignificant interaction effects and nonsignificant main effects not involved in interaction effects were then dropped one at a time, and the model was re-estimated. This parsimonious final model was then further examined using Stata’s margin command to better understand the main and interaction effects supported by the model.

The male dairy workers averaged 32.9 years (SD = 10.9), mostly (94.2%) Latino immigrants from Mexico, and had spent on average 12 years in the US (Table 1). Participants in the control site were close in age (Mean = 34.8 years, SD = 11.6) and were also mostly Latino immigrants from Mexico (p < 0.001). Dairy workers reported a median yearly income between $20,000 and $30,000, and control workers’ median income was between $10,000 and $20,000. The majority of the dairy workers described themselves as being Indigenous (Central and South) American Indian (59%) and White (32%). About half of the dairy workers had six or fewer years of formal education, while half or fewer of the control workers had fewer than 9 years of education (Table 1). About 25.9% of dairy workers and 13.3% of controls were current smokers (Table 2). Among current and former dairy worker smokers, the average pack-years was 5.5. About 23% dairy workers reported being exposed to second-hand smoke at work, and about 23% reported being exposed to second-hand smoke at home (Table 2). Control workers had lower pack-years (3.72, p < 0.001) and 11% reported being exposed to second-hand smoke at home (Table 2).

Table 3 shows the work history and personal protective equipment (PPE) use for both dairy and control workers.

^aPack-years calculated among ever smokers, n = 79; 74 dairy workers and 15 control workers, 2 missing. ^bCigarettes per day among ever smokers. ^cCigarettes per shift among workers who smoked during shift, n = 25 dairy workers.

^aIncludes use of mask, N95 respirator, or cartridge respirator. ^bOnly 15 dairy workers who spent time in enclosed cab included.

Dairy workers had spent more time working in an agricultural setting than the controls with a mean of 19.2 years (SD = 12.6) and 8.3 years (SD = 6.5), respectively. None of the controls had previously worked on a dairy. PPE use was higher (p < 0.001) among the control workers at 60%. Only 22.7% of the dairy workers reported using PPE either up to half of the time or more than half of the time during their shift. During the observed shift none of the dairy workers had reported wearing PPE more than half of the time. On average dairy workers’ shifts were half an hour longer than the control workers, at 9.1 hrs (SD = 0.9) as compared to 8.6 hrs (SD = 0.2), respectively.

Personal exposures to IPM, PM_2.5 and endotoxin in both fractions are reported in Table 4. Comparison using t- test of the log transformed exposure means reveals statistically higher exposures for dairy workers versus control workers with respect to exposure to endotoxin in IPM (p < 0.001, r² = 0.487) and endotoxin in PM_2.5 (p < 0.001, r² = 0.240). Geometric mean personal exposure to endotoxin in IPM and PM_2.5 in dairy workers was 331.5 EU/m³ and 12.5 EU/m³, respectively. Mean exposure to endotoxin in IPM and PM_2.5 was over 26 times and 5 times, respectively, higher in dairy workers than in control workers. T-tests were also used to compare log-transformed exposures to particulate matter, which also were higher in dairy workers versus controls, IPM (p < 0.001, R² = 0.092) and PM_2.5 (p < 0.001, R² = 0.092) when controlling for tobacco use and second-hand smoke exposure. Personal exposure to IPM and PM_2.5 geometric mean was about 2 times (800 μg/m³) and 1.5 times (33 μg/m³) higher in dairy workers than in the control workers. None of the dairy worker PM_2.5 endotoxin samples were below LOD, but 15.5% of the control samples were below LOD. The correlations between IPM and PM_2.5, PM_2.5 concentration and PM_2.5 endotoxin, IPM concentration and IPM endotoxin for both dairy and control workers are demonstrated in Figure 1. Personal exposure to PM_2.5 endotoxin was correlated with PM_2.5 and IPM endotoxin (r² = 0.157, p < 0.001, and r² = 0.348, p < 0.001, respectively). An interaction with different linear relationship between PM_2.5 endotoxin and IPM endotoxin was observed (Figure 1(d)). Data were collected during the summer, with average daily temperature of 28˚C and average relative humidity (RH) of 38%. Bivariate temperature and RH were not correlated with PM_2.5 endotoxin (p = 0.307, r² = 0.025, and p = 0.749, r² = 0.032, respectively).

^aInhalable particulate matter (IPM). ^bParticulate matter less than 2.5 μm in diameter (PM_2.5).

No significant difference was observed in baseline FEV₁ and FVC between dairy workers and control workers (p = 0.180, R² = 0.524 and p = 0.218, R² = 0.543, respectively) after controlling for age and height; smoking and second-hand exposure were not significant. The subgroup of dairy workers observed in the current paper experienced a crude-change cross-shift decline in FEV₁ and FVC of −44.3 mL and −35.6 mL, respectively, in comparison to controls, who had cross-shift levels in FEV₁ and FVC of +18.6 mL and +60.8 mL, respectively. There was a significant difference between dairy and control workers in cross-shift FEV₁ and FVC when controlling for age and height (p = 0.029, R² = 0.021 and p = 0.006, R² = 0.037, respectively). Cross-shift FEV₁ and FVC in dairy workers declined significantly (t = −3.53 (184) p = 0.001, and, t = −2.75 (184), p = 0.006, respectively). In controls, cross-shift FEV₁ (t = 0.79 (44), p = 0.436) did not change and FVC increased (t = 2.12 (44), p = 0.04). Findings in the current paper are consistent with results reporting lower pulmonary function in dairy workers than in controls in a larger set of subjects from the same facilities [28] .

A mixed linear regression was performed for FEV₁ among dairy and control workers. Statistically significant covariates included: age, race, number of days since last day off, PPE use, height, and temperature (Table 5). Neither current or former smoker status nor exposure to second-hand smoke were significant in the models in either dairy and control facilities. Age-related decrease was much higher in dairy workers versus control workers

^aData centered.

(p < 0.001), decreasing 1 ml instead of 0.2 ml per 10 years. Older control workers also had lower pre- and post-shift FEV₁ (p < 0.001), however, for every 10 year increase in age, FEV₁ declined by 0.2 ml. With respect to race, dairy workers who identified as being “Other” had lower FEV₁ than those who identified as being White (p = 0.023). However, workers who identified as being Indigenous Indian/S American Indian and those who declined did not differ from those who identified as being White (p = 0.596). There was no significant difference in FEV₁ among control workers who identified as being Indigenous Indian/S American Indian, those who declined and “Other” (p = 0.824, p = 0.293, and 0.273, respectively). Workers wearing a PPE showed better cross- shift FEV₁ by 0.3 ml (p < 0.001) than those who did not. There was no significant change observed in FEV₁ among control workers and use of PPE (p = 0.090). Finally, the more days dairy workers had worked in a row, was marginally significant in decrease of cross-shift FEV₁ (p = 0.075). The more days workers had worked in a row was not significant in cross-shift FEV₁ among control workers (p = 0.309).

Analyses showed significant interactions between PM_2.5 endotoxin exposure and dairy worker height and time of FEV₁ measure (pre-, post-shift) and temperature of sampling day. There was no significance observed between PM_2.5 endotoxin exposure and control worker height and time of FEV₁ measure and temperature of sampling day (p = 0.112 and p = 0.409, respectively). The observed interactions for dairy workers indicate that the effect of exposure on FEV₁ differed by worker height and that cross-shift changes depend on temperature. To further study these interactions, we examined the conditional relationships for workers at different heights (mean ± 1SD: 163 cm, 169 cm, 175 cm) and for sampling work-shifts with different temperatures (mean ± 1SD: 16˚C, 22˚C, 28˚C). For the endotoxin exposure-by-height interaction, findings show a statistically significant negative association among taller workers (p = 0.006) and no association among workers of average or shorter height (p = 0.268 and p = 0.274, respectively). Further investigations of cross-shift by temperature interaction showed a significant cross-shift decline in FEV₁ when temperature was ≥28˚C (p = 0.019 and p = 0.002, respectively). When workers worked during shifts when temperature was relatively low (16˚C) there was no significant cross- shift change in FEV₁ (p = 0.820).

A mixed linear regression was also performed for dairy and control workers FVC results (Table 5). Statistically significant covariates included: age, PPE use, height, temperature, and PM_2.5 endotoxin. Older dairy and control workers had lower pre- and post-shift FVC (p < 0.001 and p = 0.003, respectively). As with FEV₁, FVC decreased by about 1 ml for every 10 years increase in age among dairy workers and 0.2 ml among control workers. Dairy and control workers who reported a history of PPE use showed better FVC than workers who did not (p = 0.031 and 0.041, respectively). Taller dairy and control workers had higher cross-shift FVC than short workers (p < 0.001 and p < 0.001, respectively). A 10 cm increase in height of dairy and control workers was associated with a 0.65 and 0.56 L/sec larger cross-shift FVC, respectively.

Similar to our findings for FEV₁, an interaction between dairy worker height and exposure to PM_2.5 endotoxin was observed. To further investigate this interaction, the conditional relationships between FVC and PM_2.5 endotoxin for workers at different heights (mean ± 1SD: 163 cm, 169 cm, 175 cm) were examined. These analyses showed a significant negative association among the taller dairy workers (p = 0.003) between PM_2.5 endotoxin exposure and FVC, a significant positive association between shorter workers (p = 0.028), and no association among average height workers (p = 0.102 and p = 0.652). There was no significant association between height and exposure to PM_2.5 endotoxin in control workers at shorter, average, and taller heights (p = 0.783, p = 0.510, and p = 0.512, respectively). The interaction between temperature and time was also further investigated to determine temperature effects on cross-shift FVC. These analyses showed a decline in cross-shift FVC (p = 0.021) at higher temperature and no association when temperatures were average or below average among dairy workers (p = 0.078 and p = 0.806, respectively). There was no significant interaction between temperature and time (pre- minus post-shift) with respect to FVC for control workers (p = 0.594).

Personal exposures to PM_2.5 endotoxin was weakly correlated with PM_2.5 concentration (r² = 0.16, p < 0.05) and inhalable PM endotoxin (r² = 0.35, p < 0.01), but not with inhalable PM concentration. Endotoxin in IPM in a study by Burch et al. [15] found moderate correlation with IPM measured in air (r = 0.4 to 0.6). A study that looked at total and respirable PM and endotoxin poultry farms found significant correlations between both size fractions and endotoxin in both size fractions [35] . Endotoxin concentration might vary from PM concentrations due to its dependence on particle source, composition, and season [36] [37] . The sampling environment (indoor vs. outdoor) also influences correlation between endotoxin in different size fragments and PM concentration. In two separate studies in indoor poultry facilities, endotoxin in IPM ranging from 230 to 284 EU/m³ and IPM correlated better than endotoxin in respirable PM ranging from 236.1 to 490.6 EU/m³ and respirable PM concentration [11] [38] .

In this study, we found that higher endotoxin exposure in the PM_2.5 size fraction appeared to be associated with reduced lung function, but only in taller workers. Specifically, we found significant negative associations between endotoxin exposure and FEV₁ as well as FVC among taller workers. There was a marginally significant exposure to PM_2.5 endotoxin among taller workers (p = 0.075). This novel finding, however, should be replicated with a larger sample size to further investigate the relationship between PM_2.5 endotoxin exposure and lung function. Smoking background and primary task was further explored to observe if there was any differences in regards to workers’ height (i.e. taller workers had a greater history of smoking). No significant differences were observed among workers height and smoking background. There was also no significant differences among height and tasks on the dairy, however, maintenance workers were a bit taller with an average height of 174.2 cm, but there were only 5 workers in this group. In this study, higher daily average temperatures were associated with the cross-shift declines in FEV₁ and FVC. Donaldson et al. [39] observed decreased FEV₁ (at 2.20 mL/˚C) and FVC (at 3.64 mL/˚C) during the warmest and coolest weeks in a one year study in a population of COPD patients living in London. Decrease in FEV₁ and FVC was also observed among children with asthma at 20˚C in Australia [40] . The warmer temperature causes strain in the airways, causing decrease in lung function [41] . Although there was no interaction observed between warmer temperature work environment and exposure to PM_2.5 endotoxin in the main effect model, the combination of both could further exacerbate lung function.

The number of days worked since the last day off was marginally significant with respect to cross-shift FEV₁. Workers who worked more days in a row had lower cross-shift FEV₁. Bonlokke et al. [42] found cross-shift FVC decrease of 3.1% among swine building workers returning to work after four days of either being off of work or using respiratory protection. They found a loss of adaptation from the time they were not exposed noted by decrease in both baseline FVC and immune reactions when returning to work. Endotoxin was thought to be the main culprit. In this study, reported PPE use, which included masks, N95 respirators, or cartridge respirators, was positively correlated with cross-shift FEV₁ and FVC. Wearing PPE during the shift may have protected workers from exposure to PM_2.5 endotoxin. The majority of studies observing respiratory health and particulate and endotoxin exposure either had not reported PPE use, omitted workers who used a PPE because of low numbers, or found low use [9] [15] [43] [44] . PPE has also been found to be worn incorrectly [43] . Mirabelli et al. [44] found only 11% of the participants reported wearing a mask all the time. In this study 9.7% of the dairy workers and 33.3% of the control workers reported wearing PPE more than half of their work-shift, although none of them reported wearing a mask more than half of the time during the observed shift. Among California agricultural workers, higher education was positively correlated to mask use. Smoking was negatively correlated with mask use in the same group. Ex-smokers were more likely to wear a mask compared to smokers and non-smokers [45] . PPE use is important for secondary prevention of PM. However, trying to maintain low levels of pollutants through better mitigation practices should be the primary concern.

Limitations included the need for a larger sample size so that smaller associations between lung function and PM_2.5 endotoxin exposure may be detected with better statistical power. In observational studies such as this, confounding is a concern. In this case, control workers had lower smoking prevalence and secondhand smoke exposure than dairy workers. Our analyses, however, showed that neither was a significant covariate in FEV₁ and FVC mixed linear regression models. As discussed in [28] , selection bias might be present because of the healthy worker effect. This bias was reduced by selecting similar age and ethnicities in dairy and control worker populations. Finally, the study design examined cross-shift changes in lung function and endotoxin exposure on a specific day. The health effects of exposure, however, are likely to be cumulative over months or years of exposure. It would have also been beneficial to test other possible bioaerosols, such as gram positive bacteria or molds/fungi, which may have given more insight as to workers exposure and effects on lung function. Current findings could be Type I errors and any significant association between respiratory health and exposure could have possibly arisen by chance, however, not very likely given the observed level of significance in the mixed linear regression model. Future studies, therefore should apply longitudinal designs, in which the cumulative effects of endotoxin exposure can be observed over longer periods of time.