The Role of Environmental Toxins on ALS: A Case-Control Study of Occupational Risk Factors

Abstract

Importance

Persistent environmental pollutants may represent a modifiable risk factor involved in the gene-time-environment hypothesis in amyotrophic lateral sclerosis (ALS).

Objective

To evaluate the association of occupational exposures and environmental toxins on the odds of developing ALS in Michigan, a state with historically high levels of environmental pollution.

Design

Case-control study conducted between 2011 and 2014.

Setting

Tertiary referral center/ALS referral center

Participants

ALS cases (n=156) with a diagnosis of definitive, probable, probable with laboratory support, or possible ALS by revised El Escorial criteria. Controls (n=128) were excluded if they had a diagnosis of ALS, another neurodegenerative condition, or a family history of ALS in a first- or second-degree blood relative. Additional exclusions included age less than 18 or inability to communicate in English.

Main Outcome and Measure(s)

Cases and controls completed a survey assessing occupational and residential exposures. Blood concentrations of 122 persistent environmental pollutants, including organochlorine pesticides (OCP), polychlorinated biphenyls (PCBs), and brominated flame retardants (BFRs), were measured using gas chromatography/mass spectrometry. Multivariable models with self-reported occupational exposures in various exposure time windows and environmental toxin blood concentrations were separately fit by logistic regression models. Concordance between the survey data and pollutant measurements was assessed using the nonparametric Kendall’s Tau correlation coefficient.

Results

Survey data revealed that reported pesticide exposure in the cumulative exposure windows was significantly associated with ALS (OR = 5.09, 95% CI = 1.85–14.0). Military service was also associated with ALS in two time windows. A multivariable model of measured persistent environmental pollutants in the blood, representing cumulative occupational and residential exposure, showed increased odds of ALS for 2 OCPs, 2 PCBs, and 1 BFR. There was modest concordance between survey data and the measurements of persistent environmental pollutants in blood.

Conclusions and Relevance

Persistent environmental pollutants measured in blood are significantly associated with ALS. These environmental pollutants may represent a modifiable ALS disease risk factor and should be further studied.

INTRODUCTION

Amyotrophic lateral sclerosis (ALS) is a progressive, fatal neurodegenerative disease.¹ Toxic exposures, combined with genetic susceptibility, may trigger motor neuron degeneration explained via the gene-time-environment hypothesis.²

In Michigan, geographic variations in ALS death rates may correspond to locations of environmental toxins identified by the Environmental Protection Agency’s Toxics Release Inventory³ and National Priorities List.⁴ We hypothesized that toxic exposures are more likely to be identified in ALS subjects compared to healthy controls. Organochlorine pesticides (OCPs),⁵ polychlorinated biphenyls (PCBs, electrical insulation),⁶ and polybrominated diphenyl ethers (PBDEs, brominated flame retardant (BFRs))⁷ are of particular interest as potential ALS risk factors due to known neurotoxic properties and high persistence in the environment and body.

In 2011, we initiated a case-control study to evaluate ALS environmental risk factors with the strength of combining assessments of environmental pollutants in blood with detailed exposure reporting. Importantly, the ability to validate self-reported exposures with biospecimens is necessary for large scale efforts identifying ALS risk factors. We previously showed that reported pesticide exposure is associated with ALS.⁸ This report extends and advances our previous work by incorporating blood measurements of persistent organic pollutants in conjunction with reported exposures.

METHODS

Study populations and data collection

Cases and controls were older than 18 years, and able to provide consent and communicate in English. All ALS patients meeting inclusion criteria with definitive, probable, probable with laboratory support, or possible ALS by revised El Escorial criteria⁹ were asked to participate during visits at the University of Michigan (UM) ALS Clinic; recruitment announcements were also sent via the National ALS Registry. Controls, recruited through postings and a UM clinical research volunteer database (umclinicaltrials.org), with a neurodegenerative condition, ALS or family history of ALS in a first- or second-degree blood relative were excluded. The study received Institutional Review Board approval. Controls received compensation.

Subjects completed a self-administered written survey, derived from Agency for Toxic Substances and Disease Registry instruments,¹⁰ assessing residential history, occupational history, military services, smoking history, and demographics. Questionnaires were distributed in person or via mail. Telephone follow-up clarified responses as needed; next of kin was contacted for patients with communication impairments. Double data entry was performed for quality assurance in a random sample of surveys. Blood samples were collected for measurements of pollutants.

Data types

Demographics included age, gender, ethnicity, education, marital status, tobacco history, and military service history. Job titles, workplace, dates of employment, and 22 exposure-related questions assessed occupational risk factors in each job (2 most recent and 2 longest-held jobs, Supplement). This report focuses on occupational exposures, from which 98 candidate covariates were derived from exposure-related questions, including job titles and workplace settings for each job. Specifically, 58 identified exposure risk factors, 20 occupational groups, and 20 industrial groups were queried for each job. Residential exposures were assessed using 54 questions for the current house and 18 for the houses where the subject was born and the 2 houses where the subject lived the longest (Supplement). Surveys with >70% missing responses were excluded from further analyses.

Blood samples measured 122 persistent neurotoxic organic pollutants (OCPs, PCBs, BFRs) utilizing NHANES and NHEXAS methods (Supplement).¹¹ These compounds had wide use and existing quality-assured measurement methods.^5–7 Due to a protocol change, some subjects provided whole blood while others provided plasma samples. Partition coefficients between plasma and whole blood for all measured compounds (eTable 1) were determined in a small study involving 21 volunteers (data not shown) and used to covert whole blood concentrations to plasma concentrations (Supplement). Compounds analyzed had a detection frequency >30% (eTable 1).

Exposure time windows

Four exposure time windows categorized past occupational exposures: Window 1 (W₁) indicates exposure occurred in at least one period over the entire job history; windows 2, 3 and 4 (W₂, W₃, W₄) respectively indicate exposures in the last 10 years, 10–30 years, and more than 30 years ago. Reference dates for cases and controls were symptom onset and survey consent date, respectively. Occupational exposure variables in each job were assigned to the appropriate time windows. An individual could be exposed in multiple time windows depending on job history.

Job-exposure matrix for occupational pesticide exposures

Lifetime cumulative occupational exposures to pesticides were derived from a job-exposure matrix developed using the participants’ reported job titles and workplace. The probability (0 ≤ P ≤ 1) and intensity of exposure (low to high: I = 0, 1, 2, 3, 4) was estimated for each occupation and industry group based on previous studies^12,13 and judgments of two experienced exposure scientists (F-CS, SB). Exposure scores equaled the product of probability and intensity and were mapped to four levels (scores = 0, 1, 2 and 3 for probability-intensity products of 0, 1–2, 3–6 and ≥7, respectively; eTable 2). The lifetime cumulative exposure was estimated by weighting exposure scores by job duration.¹⁴

Statistical analyses

Descriptive analyses, including central tendency and dispersion, characterized demographics, smoking status, and exposure to pollutants by case type. T-tests for continuous variables and chi-square tests for categorical variables examined differences between cases and controls. The normality of pollutant data was evaluated using Kolmogorov-Smirnov tests, and stratified by case type. Spearman rank-order correlations assessed correlations between pollutants within chemical groups.

Multivariable logistic regression models tested associations between ALS and reported occupational exposures. Variable selection used a three-step method. First, the 98 candidate covariates in W₁ were fitted into a logistic regression model using the stepwise method to select potential risk factors. Next, identified factors, including smoking status and military service,^15,16 were incorporated into a model with adjustments for age, gender, and education levels. Finally, the stepwise-selected risk factors with the largest p-values were sequentially excluded from the model until all remaining stepwise-selected risk factors were significant or marginally significant (p-values ≈ 0.05). The same covariates were used in each exposure time window. Hosmer-Lemeshow tests examined goodness-of-fit, with the null hypothesis representing that observed and predicted values of the dependent variable were not different.

Associations between ALS and pollutants were tested using two approaches. First, single pollutant logistic regression models provided an initial screen of the pollutant’s association. The Bonferroni correction was applied to adjust the critical value (i.e. α/number of tests = 0.05/28), which tends to increase the chance of false negative results. Tests of individual chemicals increase the likelihood of significant results by chance alone because the correction is conservative as a result of multiple comparisons¹⁷ and does not reflect effects of multiple exposures; therefore, the second approach used a multiple pollutant logistic regression model. This stepwise variable selection method accounts for multicollinearity occurring due to high correlations between pollutants. Both models adjusted for age, gender and educational level and used standardized pollutant levels (obtained by subtracting the mean and dividing by the standard deviation).

The nonparametric Kendall's Tau correlation coefficient, which measures agreement between rankings of two ordinal variables, examined concordance between self-reported risks and occupational pollutant concentrations. OCP concentrations were first partitioned into 5 groups, from the lowest to the highest levels, and were displayed in a matrix plot of the correlation coefficients. Because pollutant measurements integrate exposures that may occur in both occupational and residential settings, the survey questions regarding potential residential exposures that might account for OCP exposure were also tested.

Missing responses and exposure measurements underwent multiple imputation, provided the missing rate was <30%. Ten imputations for incomplete values were obtained using the Markov chain Monte Carlo method.¹⁸ Both observed and imputed datasets were tested.

Logistic and linear regression analyses and multiple imputations were performed in SAS 9.2 (SAS Institute, Cary, NC, USA). The correlation plots used the package "corrplot" in R 3.2.1.¹⁹

RESULTS

Descriptive analyses

One hundred fifty-six ALS cases and 128 controls were recruited (66 cases and controls were reported in our previous survey-only study⁸). Age, gender, educational level, smoking status and occupational risk factor information was complete for 126 cases and 118 controls. Pollutant measurements were performed on subjects who provided blood equaling 129 cases and 119 controls. Together, 101 cases and 110 controls had complete demographic and pollutant data.

Demographics are summarized in Table 1. Twenty-eight of the 122 persistent organic pollutants had detection frequencies >30% (eTable 1).

Table 1.

Demographics and smoking status for cases and controls using observed data (n = 284).

Variable	Cases			Controls			p-value*
	n	Value or Count	%	n	Value or Count	%
Age (year)	156			128			0.950
Mean ± SD		60.5 ± 11.1			60.4 ± 9.4
Median		61			60
Range		28 – 86			40 – 81
Gender	156			128			0.524
Female		60	38.5		54	42.2
Male		96	61.5		74	57.8
Distance from UM	156			128			<.001
< 20 km		45	28.9		90	70.3
>= 20 km		111	71.2		38	29.7
Education	128			119			<.001
< bachelor's degree		81	63.3		47	39.5
≥ bachelor's degree		47	36.7		72	60.5
Smoking Status	129			120			0.721
Non-smoker		60	46.5		53	44.2
Former smoker		56	43.4		51	42.5
Current smoker		13	10.1		16	13.3

^*, t tests for continuous variables, and chi-square tests for categorical variables.

UM, University of Michigan.

Occupational ALS risk factors by exposure time window

Of the 98 candidate occupational risk factor variables in each time window, a stepwise logistic regression model identified 14 risk factor variables meeting significance levels for entering and removing effects of 0.30 and 0.35, respectively. The question “Did you get material on your skin or clothing?” was excluded due to a missing rate >70%. When including smoking status and military service into the model and adjusting for age, gender, and educational levels for each window, the following 6 occupational risk factors with p-values <0.05 comprised the final model: (1) lead exposure; (2) pesticide exposure; working in (3) the U.S. armed forces; (4) health care or social assistance; (5) accommodation (i.e., lodging) or food services; and (6) public administration (Table 2). Goodness-of-fit tests for the 4 models did not indicate differences between observed and predicted values (p-values >0.05).

Table 2.

Identified occupational risk factors of ALS for four exposure time windows.

Variable		Imputed			Observed
		OR	95% CI		OR	95% CI
W1 Exposure ever happened in the entire occupational history
Age	Year	1.00	0.97	1.03	1.00	0.97	1.04
Gender	Male	0.76	0.37	1.56	0.80	0.37	1.72
Educational level	≥ College	0.30**	0.17	0.54	0.28**	0.15	0.51
Smoker	Current	1.06	0.33	3.43	0.72	0.25	2.07
	Former	0.79	0.35	1.78	0.85	0.46	1.57
Ever work in the US armed forces	Yes	2.31*	1.02	5.25	2.20	0.95	5.1
Occupational exposure to lead	Yes	0.38*	0.15	0.94	0.32*	0.13	0.81
Occupational exposure to pesticides	Yes	5.09**	1.85	14	5.46**	2	14.88
Industry: health care/social assistance	Yes	0.35**	0.17	0.74	0.38*	0.18	0.81
Industry: accommodation, food services	Yes	0.23*	0.07	0.77	0.19**	0.06	0.6
Industry: public administration	Yes	0.38	0.14	1.04	0.33*	0.12	0.88
W2 Exposure ever happened in the latest 10 years
Age	Year	1.00	0.97	1.03	1.00	0.97	1.04
Gender	Male	1.00	0.52	1.94	1.13	0.55	2.32
Educational level	≥ College	0.32**	0.18	0.57	0.31**	0.17	0.56
Smoker	Current	0.92	0.3	2.82	0.65	0.23	1.79
	Former	0.83	0.37	1.84	0.83	0.45	1.53
Ever work in the US armed forces	Yes	2.00	0.93	4.31	1.76	0.79	3.89
Occupational exposure to lead	Yes	1.42	0.45	4.53	1.31	0.41	4.14
Occupational exposure to pesticides	Yes	5.50*	1.4	21.5	6.18**	1.64	23.35
Industry: health care/social assistance	Yes	0.54	0.25	1.18	0.58	0.26	1.28
Industry: accommodation, food services	Yes	0.29	0.06	1.28	0.30	0.06	1.45
Industry: public administration	Yes	1.41	0.21	9.39	1.74	0.26	11.79
W3 Exposure ever happened in 10–30 years
Age	Year	1.00	0.97	1.03	1.00	0.97	1.03
Gender	Male	0.94	0.48	1.85	0.99	0.48	2.05
Educational level	≥ College	0.33**	0.18	0.58	0.33**	0.19	0.6
Smoker	Current	0.91	0.29	2.89	0.60	0.22	1.67
	Former	0.85	0.39	1.85	0.88	0.48	1.61
Ever work in the US armed forces	Yes	2.18*	1.01	4.73	2.34*	1.04	5.28
Occupational exposure to lead	Yes	0.62	0.24	1.61	0.59	0.22	1.59
Occupational exposure to pesticides	Yes	4.85**	1.56	15	4.53*	1.44	14.23
Industry: health care/social assistance	Yes	0.63	0.31	1.32	0.64	0.29	1.41
Industry: accommodation, food services	Yes	0.23*	0.05	0.96	0.24	0.06	1.01
Industry: public administration	Yes	0.31	0.09	1.12	0.27*	0.08	0.95
W4 Exposure ever happened more than 30 years ago
Age	Year	1.01	0.98	1.04	1.01	0.98	1.05
Gender	Male	1.11	0.57	2.17	1.20	0.57	2.54
Educational level	≥ College	0.33**	0.18	0.58	0.28**	0.15	0.53
Smoker	Current	0.73	0.24	2.23	0.33	0.11	1.02
	Former	0.88	0.41	1.89	0.71	0.38	1.32
Ever work in the US armed forces	Yes	2.05	0.93	4.51	2.08	0.89	4.83
Occupational exposure to lead	Yes	0.51	0.18	1.44	0.46	0.16	1.36
Occupational exposure to pesticides	Yes	3.30*	1.08	10	2.91	0.91	9.31
Industry: health care/social assistance	Yes	0.93	0.36	2.4	0.85	0.33	2.21
Industry: accommodation, food services	Yes	0.11*	0.01	0.94	0.11*	0.01	0.92
Industry: public administration	Yes	0.46	0.15	1.41	0.43	0.14	1.34

For imputed data, n = 284 (156 cases and 128 controls) for 10 imputations; for observed data, n = 244 (126 cases and 118 controls).

OR, odds ratio; CI, confidence interval;

^*, p < 0.05;

^**, p < 0.01.

W, window

Chemical concentrations in blood and ALS

Fitted models using single pollutants are shown in Figure 1 (imputed data) and eTable 3 (observed data). Figure 1 depicts ORs for 28 chemicals. Using the stepwise selection with imputed data, 10 compounds were selected into the multiple-compound model, including 4 OCPs, 3 PCBs, and 3 BFRs (Figure 1 and Table 3); 7 were significantly associated with ALS (3 OCPs, 2 PCBs and 2 BFRs). Results between the single-compound models and multiple-compound model were similar, which may be explained by lower correlations between selected compounds (most correlations were <0.40, except for correlations between β-HCH and cis-Chlordane, and PCB 175 and PCB 202).

Figure 1.

Results of logistic regression models for ALS and exposure pollutants using single and multiple chemicals.

Odds ratios (●, ▲, ■) and 95% CIs (─) were examined using standardized chemical concentrations from the imputed sample (n = 284, 10 imputations), and adjusted with age, gender and educational levels.

Single model shows the results of 28 regression models for individual compounds; multiple model shows the result of one regression model for 10 selected compounds.

β-HCH, beta-hexachlorocyclohexane; PCB, polychlorinated biphenyl; BFR, brominated flame retardant; TBBPA, tetrabromobisphenol A; PBDE, polybromodiphenyl ether.

Table 3.

Results of logistic regression models for ALS using stepwise selection of exposure pollutants.

Compound	Imputed			Observed
	OR	95% CI		OR	95% CI
Pesticides
Pentachlorobenzene	2.21*	1.06	4.60	3.40*	1.26	9.15
Hexachlorobenzene	-	-	-
β-HCH	2.62	0.92	7.41	3.53	0.87	14.38
Dacthal	0.47	0.22	1.02	0.42*	0.19	0.91
trans-Chlordane	-	-	-
cis-Chlordane	5.74**	1.80	18.20	4.84**	1.46	16.02
trans-Nonachlor	-	-	-
p,p'-DDE	-	-	-
PCBs
PCB 110	-	-	-
PCB 151	0.47**	0.28	0.80	0.49*	0.27	0.89
PCB 135/144	-	-	-
PCB 118	-	-	-
PCB 132/153	-	-	-
PCB 138/163	-	-	-
PCB 175	1.81**	1.20	2.72	1.64*	1.03	2.61
PCB 174	-	-	-
PCB 202	2.11**	1.36	3.27	2.06**	1.24	3.43
PCB 180	-	-	-
PCB 170/190	-	-	-
PCB 198	-	-	-
BFRs
TBBPA	0.76	0.51	1.15	0.69	0.44	1.10
PBDE 28	-	-	-
PBDE 47	2.69**	1.49	4.85	2.78**	1.49	5.20
PBDE 66	0.54*	0.32	0.91	0.59*	0.34	1.00
PBDE 100	-	-	-
PBDE 99	-	-	-
PBDE 85	-	-	-
PBDE 154	-	-	-

The logistic regression models were adjusted with age, gender and educational levels, and used standardized concentrations; for imputed data, n = 284 (156 cases and 128 controls) for 10 imputations; for observed data, n = 210 (97 cases and 107 controls).

OR, odds ratio; CI, confidence interval;

^*, p < 0.05;

^**, p < 0.01.

β-HCH, beta-hexachlorocyclohexane; PCB, polychlorinated biphenyl; BFR, brominated flame retardant; TBBPA, tetrabromobisphenol A; PBDE, polybromodiphenyl ether; -, non-selected (18 compounds were not selected into the model under the criteria of significance levels for entering/removing effects = 0.1 and 0.2).

Concordance between survey data and pollutant exposures

As both the survey and exposure measurements showed significant associations between pesticide exposures and ALS, we considered the concordance of these independent datasets. Kendall's Tau correlation coefficients (Figure 2) showed weak agreement between OCP concentrations and survey-based cumulative occupational exposure to pesticides. The use of occupational and residential variables provided modest correlation with measurements of environmental pollutants, suggesting that pesticide exposure in the study population occurred in both occupational and residential/environmental settings and that survey data may help indicate the exposure source.

Figure 2.

The matrix plot of Kendall's Tau correlation coefficients for self-reported and exposure pollutant data of pesticide exposures.

The OCP concentrations were partitioned into 5 groups from the lowest to the highest levels; the occupational cumulative exposure to pesticides was partitioned into 4 groups from the lowest to the highest levels; the residential variables of pesticide exposures were yes/no questions.

Larger circles indicate larger correlation coefficients. There was weak agreement between OCP concentrations and survey-based cumulative occupational exposure to pesticides (coefficients = −0.11 to 0.05, p-values >0.05, depending on OCP). For residential variables, the “use of pesticides in home or yard” was positively correlated with pentachlorobenzene (coefficient = 0.19, p-value <0.01), and negatively correlated with Dacthal (coefficient = −0.18, p-value = 0.01); “storing lawn care products in garage” was positively correlated with pentachlorobenzene and trans-nonachlor (coefficients = 0.15 and 0.24, p-values = 0.02 and <0.01 respectively); and “storing pesticides or lawn care products in garage” was positively correlated with pentachlorobenzene (coefficient = 0.16, p-value = 0.02).

Sensitivity analysis in a smaller geographic region

Given the geographic variation between cases and controls, a subgroup analysis on subjects who lived <70 km from UM was performed. Fifty-three cases and 90 controls were selected to balance sample size and geographic matching. Single pollutant logistic regression models for all subjects and this subgroup showed high agreement for OCPs and PCBs (5 OCPs and 2 PCBs showed significant associations with ALS before Bonferroni correction; eTable 3). ORs for BFRs were very similar; however, the significant associations with ALS obtained using all subjects (5 BFRs with significant results before Bonferroni correction) no longer attained statistical significance, probably due to reduced sample size. Therefore, geographic variation between the cases and controls was a limited confounder.

DISCUSSION

The gene-time-environment hypothesis supports the contention that toxic exposures combined with genetic susceptibility may trigger motor neuron degeneration and ALS.² Most studies assessing ALS environmental and occupational risks, however, use surveys and are limited by recall bias and exposure misclassification.²⁰ Our study used both surveys and measurements of pollutants in blood. Notably, the survey used assesses exposure windows to identify periods of susceptibility for developing ALS to help clarify exposure-time interaction effects. Exposure to critical risk factors during such periods may have a greater effect than during other periods,²¹ but other than our previous study,⁸ windows of ALS susceptibility have not been investigated.

We report that service in the US military is statistically or marginally significant in multiple exposure windows, agreeing with reports that military personnel are high-risk populations for ALS.^22–24 While triggers remain unknown,²⁴ suspected risk factors include environmental exposures, multiple vaccinations, physical activity, and traumatic injury.^25,26 Similar to our initial report of survey data,⁸ we again found associations of ALS with reported occupational pesticide use and lower educational attainment. Unexpectedly, lead showed a statistically significant protective effect across all occupational exposure windows. Lead has received extensive attention as an ALS risk factor,²⁷ and may increase survival,²⁸ indicating this topic deserves further attention.

Associations between persistent organic pollutants and ALS are strengthened utilizing biological exposure measurements. Prior studies used a small set of exposure measurements to assess ALS risk. Our approach of testing a large number of chemicals, contrasts to studies testing single compounds. Applying multipollutant models, 7 of 10 model-selected compounds were significantly associated with ALS (3 OCPs, 2 PCBs and 2 BFRs). Results between the single-compound models and multiple-compound model were similar, and agree with studies demonstrating an increased ALS-pesticide association.^29–32 Surprisingly, two chemicals (PCB 151 and PBDE 66) showed protective effects. The positive association between PCB 151/PBDE 66 and ALS may result from correlations among co-existent chemicals and weak associations between the chemical themselves and the disease. The analysis of a large number of pollutants and other exposure metrics can increase the likelihood of such effects, as well as false associations resulting from multiple comparisons. The confidence in our results, particularly for OCPs, is increased by the agreement between pollutant and survey data.

Our approach is important because as persons are likely to be exposed to combinations of pollutants, such toxins may work in concert to activate several disease mechanisms. Similar methods in cognitive research show varying associations of disease risk.^33–38 Environmental pollutants cause biological effects that overlap with hypothesized ALS mechanisms suggesting biologic plausibility. For example, p,p’-DDE causes sustained depolarization, leading to release of neurotransmitters (including glutamate) and hyperexcitability.³⁹ PCBs, which accumulate in the brain, impair neurotransmitter reuptake (including glutamate), calcium homeostasis, and signal transduction, and accelerate cell death.^40,41 That PCBs can modify glutamate levels is consistent with proposed ALS disease mechanisms,⁴² and both PCBs⁴³ and PBDEs result in neuronal apoptosis in vitro.⁴⁴ Finally, persistent environmental pollutants alter global DNA methylation,⁴⁵ which may play a role in ALS pathogenesis.⁴⁶ Further work is needed to understand the aforementioned potential pathogenic implications in ALS.

Because survey data and blood measurements both implicated pesticide exposure in ALS, we examined the concordance between survey responses addressing cumulative occupational exposure of pesticides and actual chemical measurements of OCPs. The fact that we found only a weak concordance is not unsurprising. First, pollutant measurements reflect not only occupational exposures but also residential exposures and diet. Second, one-time collected pollutants may not represent lifetime cumulative occupational exposures due to temporal variation, the finite biological half-lives of the chemicals, and sampling variation. Third, job exposure matrix-based estimates along with self-reported exposures vary in validity and reliability, despite widespread use in occupational case-control studies.⁴⁷ Fourth, many currently used pesticides are short-lived organophosphate compounds (not measured in the current study) and while exposure to these compounds was likely included in responses to the occupational survey, these short-lived compounds were not measured in the current study. Fifth, and finally, individuals may also falsely believe they experienced pesticide exposures.

The lack of concordance between survey data and actual measurements of blood pollutants supports the use of direct pollutant assessments when possible, particularly for compounds that have long half-lives, and thereby reflect both current and historical exposures. For example, organochlorine pesticides have environmental half-lives that potentially reach hundreds of years for DDT and DDE,⁵ and in humans, over 7 years for p,p'-DDE⁴⁸ and β-HCH.^49,50 While most OCPs have been banned or restricted in the US, exposure still occurs, primarily through diet. As the relative contribution of diet on environmental exposures increases, agreement between occupational survey and pollutant data will likely even further decrease.

Given the case-control study design, selection and recall biases are possibilities in our study, and varied validity and reliability of self-reported data and single biological measurements may occur. Seventy percent of controls lived <20 km from UM due to the recruitment strategy, although a sensitivity analysis of a smaller geographic region supported the association of pollutants with ALS. Controls had higher educational levels, which contrasts to National ALS Registry data showing a higher rate of ALS in more educated persons.⁵¹ While other ALS studies do have control groups with higher educational attainment,²⁹ the use of a primarily internet-based recruitment strategy for controls may have biased the group towards higher socioeconomic levels. Every model adjusted for education to control for potential confounding. Additionally, conditional logistic regression models on a frequency matched dataset, matching cases with controls for age, gender, and education did not show changes in effect associations (eTables 8–10). Next, self-reported data typically involve recall bias that can lead to exposure misclassification and affect risk estimates.⁵² Due to a protocol change, the current pollutant measurements also used both plasma and whole blood samples, although method- and chemical-specific partition coefficients were measured and used to determine unified measures with very high validity (data not shown). Lastly, pollutant measurements weight more recent exposures; however, most chemicals measured in this study have long half-lives, and thus are a good indicator of historical exposure.^5–7

In conclusion, our findings identify classes of pollutants that increase the likelihood of ALS and therefore are modifiable disease risk factors. Relative to the previous literature, the present study includes (1) a thorough survey assessment of occupational exposure risk factors over time, (2) neurotoxic occupational pollutant measurements, and (3) concordance testing between the survey and pollutant concentrations. We report that military service increases ALS risk throughout an individual’s lifetime, and survey data reveal an association with pesticide exposure and ALS. Measured blood pollutants, particularly OCPs, increased the odds of ALS across all reported time exposure windows, although there was a low concordance between survey reports of pesticide exposure and actual blood measurements. These results highlight differences in reported versus measured exposures and underscore the need to understand how survey responses relate to actual exposures; this understanding is essential in many applications, particularly large scale efforts to identify ALS risk factors such as the National ALS Registry. Finally, as environmental factors that affect the susceptibility, triggering, and progression of ALS remain largely unknown, we contend future studies are needed to evaluate longitudinal trends in exposure measurements, assess newer and non-persistent chemicals, consider pathogenic mechanisms, and assess phenotypic variations.⁵³