Glyphosate in house dust and risk of childhood acute lymphoblastic leukemia in California

Abstract

Background:

Residential use of pesticides has been associated with increased risk of childhood acute lymphoblastic leukemia (ALL). We evaluated determinants of glyphosate concentrations in house dust and estimated ALL risk in the California Childhood Leukemia Study (CCLS).

Methods:

The CCLS is a population-based case-control study of childhood leukemia in California. Among those <8-years (no move since diagnosis/reference date), we collected dust (2001–2007) from the room where the child spent the most time while awake and measured >40 pesticides. Three-to-eight years later, we collected a second sample from non-movers. We used Ultra-Performance Liquid Chromatography Tandem Mass Spectrometry to measure glyphosate (μg/g dust) for 181 ALL cases and 225 controls and for 45 households with a second dust sample. We used multivariable Tobit regression to evaluate determinants of glyphosate concentrations. Odds ratios (ORs) and 95% confidence intervals (CI) were calculated for ALL and quartiles of the concentration (first samples) using unconditional logistic regression. We computed the within- and between-home variance and intraclass correlation coefficient (ICC).

Results:

Glyphosate was frequently detected (cases: 98%; controls: 99%). Higher concentrations were associated with occupational pesticide exposure, nearby agricultural use, treatment for lawn weeds and bees/wasps, and sampling season. Increasing concentrations were not associated with ALL risk (adjusted OR_Q4vsQ1=0.8, CI: 0.4–1.4). We observed similar null associations for boys and girls, Hispanics and non-Hispanic whites, and among those who resided in their home since birth (76 cases/117 controls) or age two (130 cases/176 controls). The ICC was 0.32 indicating high within-home temporal variability during the years of our study.

Conclusions:

We observed higher concentrations in homes associated with expected predictors of exposure but no association with childhood ALL risk. Due to continuing use, potential exposure to young children is high. It will be important to evaluate risk in future studies with multiple dust measurements or biomarkers of exposure.

1.1. Introduction

Childhood leukemia is the most common childhood cancer worldwide and rates have increased in the United States (US) and in other developed countries over the past 20 years (Barrington-Trimis et al. 2017; Giddings et al. 2016; Roman 2017). Acute lymphoblastic leukemia (ALL), the most common histologic type in childhood, occurs at a young age with over half of cases diagnosed before five years of age (Roman 2017). Pesticide exposures have been linked to an increased risk of ALL through parental occupational exposure (Bailey et al. 2014; Wigle et al. 2009) home and garden use (Bailey et al. 2015; Chen et al. 2015; Turner et al. 2010, 2011), as well as residential proximity to agricultural applications (Park et al. 2020; Patel et al. 2020; Reynolds et al. 2005; Rull et al. 2009). However, few studies have identified specific pesticide active ingredients that may be responsible for these associations with childhood leukemia risk. Professional or homeowner use of herbicides (weed killers) have been implicated as a childhood leukemia risk factor in several studies (Ma et al. 2002; Turner et al. 2010). Herbicides are commonly applied to parks and lawns and are also the most common type of pesticide applied to agricultural fields in the United States (Atwood 2017).

Glyphosate is the active ingredient in Roundup^® herbicide formulations. It is an organophosphorus herbicide that has been used to kill broad-leaf weeds since 1974. Its primary use has been agricultural, but glyphosate is also a common active ingredient in household weed killers. Total use (agricultural and non-agricultural) increased over five-fold between 1994 and 2014 in the United States and over 12-fold globally after the development of genetically modified broad-leaf crops that were resistant to its toxic effects (Benbrook 2016). The U.S. was estimated to account for 15% of worldwide applications in 2014 (Benbrook 2016). Despite some restrictions and bans on its use in some countries, glyphosate is still the most widely used herbicide globally. The development of weed resistance to glyphosate’s herbicidal effects led to increasing application rates and more frequent applications as well as to use in combination with other herbicides (Benbrook 2016; Maggi et al. 2020). In the environment, glyphosate is metabolized, and the parent compound and metabolites are found in soils, plants, food products and animals (Alferness and Iwata 1994; El-Gendy et al. 2018; Group 2017). Glyphosate has been measured in human urine, blood, breast milk, and household dust; exposure can occur through dermal, oral, and respiratory routes (Curwin et al. 2005; Williams et al. 2000). Based on limited studies, exposure to glyphosate or glyphosate-based herbicides in humans has been shown to cause inflammation, alteration of lymphocyte functions, and effects on the immune system interactions with microorganisms (Peillex and Pelletier 2020). Based on animal data, mechanistic human data and epidemiologic studies of adult non-Hodgkin lymphoma, an International Agency for Research on Cancer (IARC) working group determined that glyphosate was probably carcinogenic to humans (IARC 2017).

Pesticides and other chemicals persist indoors where they are protected from degradation. House dust is a reservoir for pesticides used in and around the home and can be a useful indicator of exposure (Colt et al. 2008; Deziel et al. 2013; Deziel et al. 2015; Ward et al. 2006). In the California Childhood Leukemia Study (CCLS), household use of pesticides was associated with increased ALL risk (Ma et al. 2002). House dust concentrations of the herbicide chlorthal (also called dacthal), but not other herbicides (Metayer et al. 2013) or insecticides (Madrigal et al. 2021) were associated with increased risk of ALL. However, glyphosate was not previously measured in the CCLS due to the fact that it required a separate analytic method than the other pesticides and at the time of those analyses in the late 2000s data on carcinogenicity was limited. Here, we evaluate glyphosate concentrations in homes and risk of ALL among CCLS participants for whom we collected house dust samples.

1.2. Methods

1.2.1. Study Population

The CCLS is a population-based case-control study of childhood leukemia (<15 years) in 35 counties in the Central Valley and the San Francisco Bay area (1999 to 2015). As previously described, (Bartley et al. 2010; Chang et al. 2006; Metayer et al. 2013), cases were diagnosed with leukemia at nine pediatric clinical centers, resided in the study area, and had a parent who spoke English or Spanish. Controls were randomly selected from the California birth registry and were matched to cases by age, gender, Hispanic ethnicity, and maternal race. We conducted a second (Tier 2) interview for younger cases and controls who were <8 years old at diagnosis (similar reference date for controls) during December 1999 through January 2007 and had not moved since enrollment. The Tier 2 interviews took place from October 2001 through November 2007.

1.2.2. Tier 2 Interviews

The Tier 2 interview was conducted in the home and included information on demographics, characteristics of the residence, whether household members had occupational pesticide exposure, and pesticide use around the home and garden including lawn and garden weeds (Colt et al. 2008). Additionally, interviewers conducted an inventory of pesticides stored in the home and asked about the pests treated and timing of use (Guha et al. 2013). Residence location was determined using a global positioning system and was categorized as urban, suburban, or rural using the 2000 U.S. census block characteristics. The study protocol was approved by the institutional review boards at the University of California, Berkeley, the California Committee for Protection of Human Subjects, and the National Institutes of Health.

1.2.3. Dust Sample Collection

As described (Colt et al. 2008), we sampled the room where the child spent the most time while awake in the year prior to the diagnosis (reference date for controls) if there was a carpet at least nine square feet (0.84 square meters) that was present before the diagnosis (reference) date. Carpet dust samples were collected using a high-volume small-surface sampler (HVS3; Cascade Stack Sampling System, Venice, FL) for most study participants with additional dust collected from the used vacuum bag. In the last year of the study, only vacuum dust was collected since it was shown to be comparable to HVS3 dust for our analytes (Colt et al. 2008). A total of 324 cases and 407 controls were eligible for the Tier 2 interview and 296 cases (91%) and 333 controls (81%) participated. Of these, 17 ALL cases and 27 controls did not have carpet dust due to no eligible carpet or insufficient dust for any analysis (Deziel et al. 2014; Metayer et al. 2013; Ward et al. 2009; Ward et al. 2014). We previously measured 50 pesticides including 16 herbicides using three extraction methods with quantification by gas chromatography mass spectrometry (MS) (Colt et al. 2008; Metayer et al. 2013). We did not measure glyphosate since the analytic methods we used were not suitable. In 2019, Battelle Memorial Laboratory developed a liquid chromatography tandem MS method, and we used this new method (described below) to measure glyphosate in stored (−20 C) samples from ALL cases and controls with at least 2 grams of dust remaining (181 ALL cases [72%], 225 controls [74%]). A total of 192 samples (47%) were collected with the HVS3 and 214 (53%) were from the household vacuum; these percentages were the same for cases and controls. For cases, the time between enrollment and the dust sample collection ranged from 0.4–3.0 years (median=0.9; interquartile range [IQR] 0.7–1.3); whereas among controls it was 0.6–4.2 years (median=1.7 IQR 1.3–2.2).

In 2010, about three to eight years after the Tier 2 interviews, participants who had household vacuum dust samples remaining and had not moved from their Tier 2 home, were invited to participate in an additional (Tier 3) telephone interview with dust collection by mail from the household vacuum (Whitehead et al. 2013a; Whitehead et al. 2013b; Whitehead et al. 2014). A total of 204 of the 225 (91%) participants who were eligible participated. We selected 45 of the Tier 3 participants whose had adequate dust remaining for analyses of glyphosate in both the Tier 2 and Tier 3 samples. The median (IQR) time between repeated sample collections was 3.6 years (IQR: 3.0–4.3); the range was three to six years.

1.2.4. Laboratory Analysis and Quality Control

Dust samples were shipped on ice to the Battelle Memorial Institute (Columbus, OH) where they were sieved to remove coarse dust (>150 μm), aliquoted, and stored at −20° Celsius. Approximately 0.5 g aliquots were thawed, spiked with 25 μL of 20 μg/mL ¹³C-glyphosate, and extracted with 1% formic acid in methanol, followed by purification by C18 Solid Phase Extraction cartridges with final elution into 1 mL of methanol. We used Ultra-Performance Liquid Chromatography Tandem Mass Spectrometry in negative ion mode to quantify glyphosate. Concentrations were expressed as μg/g. All analysis was conducted by the same analyst (DK) who was blind to case-control status.

Approximately 20 dust samples were analyzed in each batch and frequency-matched case and control samples were placed within batches in a random order. Likewise, repeat dust samples were analyzed in the same batch. Quality control samples consisted of three 0.5 g aliquots from a large dust sample and duplicate aliquots of different participant samples in each batch that were spiked with 500 ng/mL glyphosate. Glyphosate recoveries (percent of spiked amount quantified) were computed by subtracting the calculated amount of glyphosate in each quality control sample from the spiked amount and dividing by the spiked amount. In the large dust sample, mean recoveries ranged from 81 to 136% across batches. Mean recoveries in the replicate samples were similar in 22 batches (range: 81 to 118%) but more variable in three batches (−26%, 8.4%, 255%) due to two low recoveries and one high recovery in one of the two spiked samples. The limit of detection based on the standard curve was 0.1 μg glyphosate/g dust; however, a few samples were quantified below this limit (lowest concentration was 0.047 μg/g) and these values were used as they were considered to be the best estimate of the concentration.

Within-batch coefficients of variation (CV) were computed using the SAS procedure proc VARCOMP. The CV between batches based on the large dust sample was 1.3 (within batch CV 11.2). For the duplicate samples (different dust sample in each batch), the within-batch CV was 33. Removing one batch with an extreme outlier (low concentration) in one of the three replicates resulted in a within batch CV of 12.5.

1.2.5. Statistical Analysis

Glyphosate concentrations were log normally distributed, so we analyzed the natural log of the concentrations. We used a single imputation method (Lubin et al. 2004) to estimate concentrations below the detection limit assuming a log-normal distribution using the LIFEREG procedure in SAS (Version 9.4; SAS Institute Inc., Cary, NC). Chi-square tests were used to test bivariate differences in characteristics comparing cases to controls and by factors potentially associated with glyphosate concentrations. We used multivariate Tobit regression to evaluate factors associated with glyphosate concentrations from bivariate analyses (p≤0.1).

Glyphosate was detected in 98% of samples and we categorized the distribution into quartiles and quintiles based on the distribution among controls. Odds ratios (ORs) and 95% confidence intervals (CIs) were calculated using unconditional logistic regression adjusting for age at diagnosis (continuous), gender, and race/ethnicity (Hispanic/Latino, non-Hispanic White, or non-Hispanic other). We tested for linear trend in the ORs by including the median concentration in each quartile or quintile as a continuous variable in the model. We evaluated potential confounders, including household income, maternal age, maternal education, season and year of dust sampling, home age (decades), type of residence (single family vs other [duplex/townhouse, apartment, mobile home]), household and professional pesticide use around the home/garden, and whether a household member had a pesticide-exposed occupation. Besides age, gender, and race/ethnicity, our final models included variables that had p-values ≤0.1 during a stepwise backward elimination process including household income, residence type, season of dust collection, and interview year. Additionally, we adjusted final models for other frequently detected (≥35%) herbicides (2,4-dichlorophenoxyacetic acid [2,4-D], chlorthal, mecoprop, simazine, trifluralin) and evaluated potential joint effects. In a prior mixture analysis (Wheeler et al. 2021) (conducted before we measured glyphosate) using grouped Weighted Quantile Sum regression, we found a significant association with ALL for the herbicides. The herbicide with the highest weight was chlorthal (dacthal), which supported our results from individual herbicide models (Metayer et al. 2013). To evaluate interaction and consistency of our findings, we stratified models by the child’s sex, Hispanic ethnicity (Hispanics/Latinos, non-Hispanic white), household income (<$75,000 compared to ≥$75,000), season of sampling (winter/spring, summer/fall) and dust sampling method (HVS3, home vacuum). We conducted sensitivity analyses by limiting analyses to cases and controls who had resided at their Tier 2 residence since birth (76 cases, 117 controls) and since they were 2 years of age (130 cases, 176 controls) since early life may be a critical time period for exposure.

We used Proc MIXED in SAS (Version 9.4; SAS Institute Inc., Cary, NC) to estimate the between home variance in glyphosate concentrations (469 homes) and within home variance of the repeated measures (45 homes). We ran models with no covariates and models that included predictors (p≤0.1) of glyphosate concentrations (household used lawn/garden pesticides or treated for bees/wasps/hornets in 12 months prior to dust collection, householder had a pesticide-exposed occupation, density of agricultural glyphosate use within 1 km of the home as reported in the California Pesticide Use Reporting database). We computed the intraclass correlation coefficients (ICC) as the ratio of the between home variance to the total variance (sum of between home and repeated measure [within home] variance).

1.3. Results

Glyphosate was detected in 98% of homes of cases and 99% of controls. Median concentrations (IQR) in μg/g dust were 1.28 (0.58–2.49) for cases and 1.26 (0.65–2.90) for controls. Demographic characteristics of cases and controls are described in Table 1. Cases were more likely to be of Hispanic ethnicity (40% vs. 28% of controls), have lower household income and were less likely to live in a single-family home (79% vs. 89% of controls). Maternal education, residential density (urban, suburban, rural), having a household member employed in a pesticide-exposed job, and density of agricultural glyphosate use within 1 km of the home did not differ significantly between cases and controls. A greater proportion of cases than controls had dust samples collected in winter and summer. Case households were less likely to have used pesticides for control of fleas/ticks on pets or for indoor insects, and to have used a professional indoor pest control service. Household use of lawn/garden pesticides and weed killers specifically, did not differ between cases and controls; however, case households were more likely to have employed a professional service to apply herbicides to their lawns (18% of cases, 8% of controls). These case control differences were similar to those reported previously in analyses of herbicides and insecticides in homes for the full Tier 2 study population (Madrigal et al. 2021; Metayer et al. 2013).

Table 1.

Demographic and residential characteristics of 181 childhood ALL cases and 225 controls, the California Childhood Leukemia Study^a, 2001–2007

Children’s and Household Characteristics	Cases n (%)	Controls n (%)	p-valueb
Sex
Male	109 (60.2)	132 (58.7)	0.75
Female	72 (39.8)	93 (41.3)
Age, years
< 1	3 (1.7)	6 (2.7)	0.63
1 to <2	23 (12.7)	26 (11.5)
2 to 5	100 (55.2)	135 (60.0)
> 5 to 7	55 (30.4)	58 (25.8)
Race/ethnicity
Hispanic	72 (39.8)	62 (27.5)	0.01
White, non-Hispanic	67 (37.0)	114 (50.7)
Other, non-Hispanic	42 (23.2)	49 (21.8)
Annual household income
<$15,000	21 (11.6)	15 (6.7)	0.06
$15,000 to $29,999	24 (13.3)	22 (9.8)
$30,000 to $44,999	27 (14.9)	26 (11.6)
$45,000 to $59,999	30 (16.6)	30 (13.3)
$60,000 to $74,999	17 (9.4)	21 (9.3)
>$75,000	62 (34.2)	111 (49.3)
Maternal education c
<High school	21 (11.7)	15 (6.7)	0.23
High school	47 (26.3)	53 (23.7)
Some college/technical school	58 (32.4)	76 (33.9)
College graduate/post-graduate	53 (29.6)	80 (35.7)
Neighborhood type
Urban	127 (70.2)	164 (72.9)	0.18
Suburban	28 (15.5)	22 (9.8)
Rural	22 (12.1)	34 (15.1)
Unknown	4 (2.2)	5 (2.2)
Residence type d
Single family home	143 (79.0)	200 (89.3)	0.004
Townhouse/apartment/mobile home	38 (21.0)	24 (10.7)
Year residence built
1980 to present	71 (39.2)	93 (41.3)	0.28
1950 to 1979	58 (32.0)	82 (36.4)
Before 1950	25 (13.8)	30 (13.3)
Unknown	27 (14.9)	20 (8.9)
Season of dust collection
Winter	46 (25.4)	34 (15.1)	0.02
Spring	40 (22.1)	74 (32.9)
Summer	57 (31.5)	64 (28.4)
Fall	38 (21.0)	53 (23.6)
Year of interview/dust collection
2001–2002	48 (26.5)	47 (20.9)	0.11
2003–2004	69 (38.1)	72 (32.0)
2005–2006	40 (22.1)	70 (31.1)
2007	24 (13.3)	36 (16.0)
Self-reported pesticide use
Pest treated by householder e
Ant/cockroach	125 (70.2)	165 (74.3)	0.36
Carpenter ants/termites	6 (3.4)	14 (6.3)	0.18
Bees/wasps/hornets	31 (17.4)	44 (19.8)	0.54
Flies/mosquitos	21 (11.8)	39 (17.6)	0.11
Fleas/ticks on pets	35 (19.7)	63 (28.4)	0.04
Other indoor insects	140 (78.6)	193 (86.9)	0.03
Outdoor insects	52 (29.2)	75 (33.8)	0.33
Lawn/garden pesticides	106 (59.5)	141 (63.5)	0.42
Weeds	76 (42.7)	113 (50.9)	0.10
Professional pest treatments f
Professional indoor	15 (9.8)	34 (17.0)	0.05
Professional outdoor	40 (26.1)	60 (30.0)	0.42
Professional lawn	28 (18.3)	17 (8.5)	0.01
Occupational pesticide exposure g
Anyone in home with pesticide job	25 (13.9)	27 (12.0)	0.57
No one in home with pesticide job	155 (86.1)	198 (88.0)
Agricultural glyphosate use with 1 km of homeh kg/m2
Density=0	101 (56.4)	143 (63.8)	0.12
0< Density < 1.3	24 (13.4)	29 (13.0)
Density 1.3–9.8	27 (15.1)	25 (11.2)
Density > 9.8	27 (15.1)	27 (12.0)

^aCases and controls from the California Childhood Leukemia Study that consented to the tier 2 visit and had adequate dust samples for measurement of glyphosate

^bChi-Square tests were used to compare characteristics among cases and controls; unknown or missing values for maternal education, neighborhood type, residence type, year the residence was built, pesticide use, and occupational pesticide exposure were excluded for testing differences between cases and controls

^c2 cases, 1 control with ‘other’ educational status were excluded

^d1 control excluded due to missing data

^e3 cases and 3 controls excluded due to missing household pesticide use information

^f28 cases and 25 controls missing professional insecticide use information

^g1 case excluded due to missing occupational data

^h179 cases and 224 controls with home locations linked to California Pesticide Use Reporting database, glyphosate use in the prior year

Abbreviations: ALL, acute lymphoblastic leukemia; %, percentage

1.3.1. Determinants of concentrations in homes

Detection rates and concentrations of other herbicides measured in the dust samples were previously described (Metayer et al. 2013). Briefly, herbicides with the highest percent detections (cases and controls combined) were 2,4-dichlorophenoxyacetic acid (2,4-D) (95%), simazine (90%), mecoprop (also known as methylchlorophenoxypropionic acid [MCPP] (83%), trifluralin (60%), and chlorthal (dacthal) (35%). Chlorthal was found at higher concentrations in case homes; whereas, 2,4-D was higher in control homes (Metayer et al. 2013). Spearman correlations between glyphosate concentrations and these herbicides are shown in Table 2. Only 2,4-D and trifluralin were substantially correlated with glyphosate (Spearman ρ: 0.29 and 0.23, respectively). There were no significant correlations between concentrations of glyphosate and the insecticides detected in >40% of homes (Spearman ρ all <0.16; not shown).

Table 2.

Spearman correlation between herbicides most frequently detected in house dust, California Childhood Leukemia Study, 2001–2007^a

	2,4-D	Dacthal	Glyphosate	Mecoprop	Simazine	Trifluralin
2,4-D	1.0	0.07	0.29***	0.80***	0.07	0.18**
Dacthal		1.0	0.06	0.05	0.01	0.25***
Glyphosate			1.0	0.17**	0.11*	0.23***
Mecoprop				1.0	0.05	0.12*
Simazine					1.0	0.15*
Trifluralin						1.0

Abbreviations: 2,4-D, dichlorophenoxyacetic acid

^aALL cases (n=181) and controls (n=225);

^***p<0.0001,

^**p<0.001,

^*p<0.05

Glyphosate concentrations by characteristics of the study population are shown in Supplemental Table 1. Concentrations did not differ significantly by sex, age, household income, education, housing type, or year of dust sampling. Median concentrations in homes were lower for children who were non-white/non-Hispanic. Levels were lowest in urban homes, older homes, and in homes sampled in the winter and spring. Households reporting pesticide use in the last 12 months on bees/wasps/hornets (outdoor insects) and lawn/garden weeds had higher glyphosate concentrations in their homes than those with no use. There were no significant differences by professional lawn treatments. Glyphosate concentrations were higher in homes with a pesticide-exposed worker and with higher agricultural glyphosate use densities within one kilometer of the home in the prior year.

In multivariable Tobit regression models (Table 3), agricultural use within 1 km, household treatment of bees/wasps/hornets and lawn/garden weeds, having an occupational worker in the home, and the season the dust sample was taken were all significant independent predictors of concentration. We observed the strongest relationship for the highest density of agricultural use near the home, lawn/garden weed treatment in the prior 12 months, having a pesticide-exposed worker in the home, and dust sampling in summer (vs. winter); these factors were each associated with a 1.7 to 1.9-fold increase in glyphosate concentrations.

Table 3.

Relative increase in glyphosate concentrations in house-dust samples associated with pesticide use and other factors, based on a Tobit multivariable regression model^a

Predictor	Exp β (95% CI)
Agricultural use within 1 km (kg/km 2 ):
0	ref
0< to <1.3	1.15 (0.83–1.61)
1.3 to 9.8	1.45 (1.03–2.03)
> 9.8	1.91 (1.36–2.69)
p-trend	0.0003
Pest treatments in last 12 mos. (ever vs. never):
Treatments for bees/wasps/hornets	1.42 (1.07–1.89)
Treatments for lawn/garden weeds	1.72 (1.37–2.14)
Pesticide exposed occupation vs. none	1.67 (1.18–2.36)
Season sample collected (vs. winter)
Fall	1.60 (1.13–2.25)
Spring	1.20 (0.86–1.67)
Summer	1.67 (1.21–2.31)

^aCases N=177 and controls N=221 from the California Childhood Leukemia Study that consented to the tier 2 visit, had adequate dust for measurement of glyphosate and non-missing data for these covariates

1.3.2. Glyphosate and ALL risk

Quintiles and quartiles of the glyphosate concentration were not associated with ALL risk in logistic regression models adjusting for age, gender, and race/ethnicity (minimally adjusted model quintiles p-trend=0.29; Table 4). Adjustment for income, season and year of sampling, and residence type resulted in modest changes to risk estimates (fully adjusted models). Further adjustment for concentrations of chlorthal and the other herbicides that were frequently detected in homes (2, 4-D, simazine, mecoprop, trifluralin; >50% detections) did not change our risk estimates and there was no evidence of interaction with chlorthal or the other herbicides (not shown). Analyses based on quartiles of the glyphosate loadings (μg glyphosate per m²; HVS3 dust only) were similar to the results for dust concentrations (not shown). Analyses limited to those who resided in their Tier 2 home since birth (76 cases, 117 controls) or age two (130 cases, 176 controls) showed similar nonsignificant inverse associations comparing those in the highest to the lowest quartile (age, gender, race/ethnicity adjusted OR_Q4vsQ1 0.55, 95% CI 0.23–1.31_; OR_Q4vsQ1 0.72, 95% CI 0.36–1.43, respectively) and no significant trends.

Table 4.

Adjusted odds ratios (ORs) and 95% confidence intervals (CIs) for childhood ALL and glyphosate concentrations in homes in the California Childhood Leukemia Study, 2001–2007

Glyphosate Concentration (μg/g)	Cases	Controls	Model 1 (minimally adjusted) OR (95% CI)a	Model 2 (fully adjusted) OR (95% CI)b
Quintiles
≤ 0.55	44	45	1.0 (ref)	1.0 (ref)
> 0.55 to 1.03	35	44	0.79 (0.43–1.45)	0.83 (0.43–1.58)
> 1.03 to 1.61	28	45	0.66 (0.35–1.25)	0.71 (0.36–1.39)
> 1.61 to 3.83	48	45	1.08 (0.59–1.95)	1.11 (0.59–2.09)
> 3.83 to 135.7	26	45	0.62 (0.32–1.20)	0.65 (0.32–1.30)
p-value for trend			0.29	0.53
Per ln ug/g increase			0.89 (0.75–1.05)	0.90 (0.75–1.08)
Quartiles
≤ 0.65	50	57	1.0 (ref)	1.0 (ref)
> 0.65 to 1.26	37	56	0.76 (0.43–1.35)	0.78 (0.43–1.43)
> 1.26 to 2.90	57	56	1.17 (0.68–2.00)	1.18 (0.66–2.10)
> 2.90 to 135.7	37	56	0.77 (0.43–1.36)	0.78 (0.43–1.44)
p-value for trend			0.46	0.47
Boys
≤ 0.65	28	31	1.0 (ref)	1.0 (ref)
> 0.65 to 1.26	23	38	0.70 (0.34–1.47)	0.65 (0.30–1.41)
> 1.26 to 2.90	36	27	1.57 (0.76–3.23)	1.50 (0.70–3.23)
> 2.90 to 135.7	22	36	0.72 (0.34–1.52)	0.73 (0.33–1.60)
p-value for trend			0.44	0.52
Girls
≤ 0.65	22	26	1.0 (ref)	1.0 (ref)
> 0.65 to 1.26	14	18	0.81 (0.31–2.09)	1.11 (0.39–3.19)
> 1.26 to 2.90	21	29	0.77 (0.34–1.77)	1.17 (0.44–3.12)
> 2.90 to 135.7	15	20	0.84 (0.34–2.12)	0.87 (0.31–2.45)
p-value for trend			0.84	0.48
Latinos/Hispanics
≤ 0.65	17	15	1.0 (ref)	1.0 (ref)
> 0.65 to 1.26	14	16	0.73 (0.27–2.00)	0.79 (0.26–2.42)
> 1.26 to 2.90	24	17	1.20 (0.47–3.05)	1.14 (0.40–3.30)
> 2.90 to 135.7	17	14	1.04 (0.38–2.83)	1.00 (0.32–3.13)
p-value for trend			0.76	0.89
Non-Hispanic Whites
≤ 0.65	16	24	1.0 (ref)	1.0 (ref)
> 0.65 to 1.26	16	27	0.87 (0.36–2.11)	0.98 (0.38–2.51)
> 1.26 to 2.90	22	29	1.12 (0.48–2.61)	1.36 (0.54–3.44)
> 2.90 to 135.7	13	34	0.56 (0.23–1.38)	0.58 (0.22–1.51)
p-value for trend			0.16	0.14

^aAdjusted for age (continuous, years), gender, child’s race/ethnicity (Hispanic, non-Hispanic White, or non-Hispanic non-White); models stratified by gender and race/ethnicity were not adjusted by that factor, respectively

^bAdditionally adjusted for income (6 categories), season of dust sampling (winter, fall, spring, summer), residence type (single family vs other [duplex/townhouse, mobile home]), interview year (2001–2002, 2003–2004, 2005–2006, 2007)

The associations between quartiles of the dust concentrations and ALL risk were similar for boys and girls (fully adjusted models p-trend=0.52 and 0.48, respectively) (Table 4). There were no associations with quartiles of glyphosate concentrations among Latinos/Hispanics and non-Hispanic whites. The highest quartile was associated with a nonsignificant inverse association among non-Hispanic whites. Associations with quartiles of the concentration did not differ by income (<$75,000, ≤$75,000), season (winter/spring, summer/fall), or dust collection method (HVS3, home vacuum) (Supplemental Table 2).

1.3.3. Temporal variability of repeated measures

Among the 45 homes with two dust samples, we observed similar glyphosate detections (98% detections for both visits) and concentrations across home visits ((first visit median (IQR) μg/g: 1.49 (1.08–3.14); second visit: 1.42 (0.45–3.04)). Estimates of the within and between home variance from the random-effects model with no covariates were 1.108 and 0.521, respectively, and the ICC was 0.320. Adjusting for variables significantly associated with glyphosate concentrations (treatment for lawn/garden weeds and pests, season, agricultural use with 1 km, someone in the home with a pesticide-exposed occupation) resulted in an ICC of 0.207; whereas, estimates of the within home and between home variances were 1.150 and 0.302, respectively.

1.4. Discussion

To our knowledge, the CCLS is the first study to evaluate residential concentrations of glyphosate and other herbicides in relation to risk of childhood ALL. We observed no significant differences in residential glyphosate concentrations between cases and controls overall, nor in any population subgroup that we evaluated including those who had lived in their interview home since birth. Glyphosate was detected in almost all homes and concentrations were generally much higher than the other herbicides we previously measured including 2,4-D, which had the second highest concentration (median 0.102 μg/g) (Deziel et al. 2015; Metayer et al. 2013). Adjustment for other herbicides that were moderately correlated with glyphosate did not change our risk estimates nor was there evidence of a joint effect with chlorthal, which was the only herbicide significantly associated with increased risk in this study population. We observed large temporal variability among households with repeated samples over a three- to six-year period, which would be expected to result in substantial misclassification of exposure based on one dust sample.

Reasons for the high variability in glyphosate concentrations over time may include the relatively short half-life of glyphosate in soil (typically <60 days) (USDA 1995) as well as changes in use of glyphosate-containing products around the home, on nearby lawns, or changes in nearby agricultural use. Furthermore, different vacuuming or cleaning practices between sampling periods could have contributed to temporal variability and we were not able to account for changes in these practices. If the vacuum dust came from different carpets and areas of the home at the two visits, this could also contribute to the variability between samples.

A pesticide exposure study of farm and non-farm families in two rural Iowa counties measured glyphosate in house dust in eleven homes (Curwin et al. 2005). Concentrations in the children’s bedrooms in nonfarm homes (geometric mean [GM]: 0.51 μg/g) were lower than CCLS homes (geometric mean: 1.30 μg/g). In contrast, the farm homes had similar levels to the CCLS (GM: 1.5 μg/g). Samples were taken in multiple rooms including the child’s bedroom and playroom, at two home visits three to five weeks apart. In spite of the short time frame between samples, the temporal variability in glyphosate concentrations within homes was larger than between homes yielding an ICC of 0.25 (computed from variance components in Table VI) (Curwin et al. 2005) that was similar to our study (unadjusted ICC=0.32). In the Iowa exposure study, glyphosate concentrations were also highly variable across rooms for both farm and non-farm homes with the highest levels found in the children’s bedroom. We were not able to evaluate within home variability since about half of our dust samples were from the household vacuums, which were used in multiple rooms, and we only sampled one room with the HVS3 vacuum. The ICC estimate for glyphosate was similar to what we observed for the herbicide simazine in a small exposure study in Fresno County California with repeat samples taken 3 to 15 months apart (Deziel et al. 2013). Concentrations of chlorthal and trifluralin, the two other herbicides we measured in these repeat samples, were less variable over time.

Ours is the largest study to date to evaluate determinants of residential exposure to glyphosate in homes with children. In our multivariable models, we found that having a pesticide-exposed worker in the home, treatment of lawn weeds and bees/wasps/hornets, higher density of agricultural use within one kilometer of the home, and season of sampling were significant predictors of glyphosate concentrations in the home. These results are somewhat similar to what we found for five other herbicides we measured in CCLS homes (Deziel et al. 2015). Households that used lawn herbicides had about two-fold higher concentrations of 2,4-D, dicamba, and mecoprop; whereas, nearby agricultural use was predictive of chlorthal (dacthal) and simazine concentrations. The reason for the positive association between glyphosate concentrations in the home and treatment of bees, wasps, and hornets is not clear; treatment of these outdoor insects was less common than lawn weed treatments. The Iowa pesticide exposure study (Curwin et al. 2005) found higher dust concentrations of glyphosate when it had been sprayed on the farm in the prior 7 days compared to farms where spraying had not occurred. However, pesticide treatments of the home lawn or garden and proximity to fields (<0.5, ≥0.5 miles) were not significant predictors of glyphosate in dust in that study. Their results were based on a small number of homes, which were located in two rural Iowa counties with high agricultural production. To our knowledge, no other studies have evaluated determinants of glyphosate in house dust.

In the same Iowa exposure study (Curwin et al. 2007a), urinary glyphosate metabolites were measured in 51 children and their parents and were correlated within families. Levels were not significantly different between farm and non-farm children, which they attributed to high household use among non-farm families. Importantly, children’s urinary levels were positively correlated with the dust concentrations, suggesting that glyphosate concentrations in dust may reflect children’s exposures (Curwin et al. 2007a). The absorbed daily dose of glyphosate was estimated based on the urinary concentrations and was similar for boys and girls and farm and non-farm children (Curwin et al. 2007b). Although urine samples were collected only about a month apart, within-child temporal variability in metabolite concentrations was high (ICC=0.27 for farm children).(Curwin et al. 2007a)

Only a few studies have evaluated glyphosate exposure in children (Anderson 2022; Gillezeau et al. 2019). A recent study in Portugal found that lawn and garden use and residence within 1 km of agricultural field were predictors of higher excretion of glyphosate metabolites (Ferreira et al. 2021). In contrast, a study of children living in agricultural areas of Slovenia found few agricultural predictors of urinary levels (Stajnko et al. 2020). Glyphosate was measured in a large representative sample of children in Germany (Lemke et al. 2021) but few differences in urinary excretion levels were observed by demographic and dietary factors. A recent dietary intervention study showed that switching to an organic diet significantly reduced excretion of glyphosate and its metabolite aminomethyl phosphonic acid (AMPA) in adults and children (Fagan et al. 2020). Children had higher glyphosate exposures than adults and greater reductions in exposure after changing to an organic diet. Urinary glyphosate levels in the National Health and Examination Survey (2013–2014), a nationally representative sample of the U.S. population, were highest among children ages 6–11 compared to older children and adults (CDC 2022). Additional studies with detailed information on home and garden herbicide use, nearby agricultural use, and other factors are needed to understand the main contributors to exposure among children.

Glyphosate is a competitive inhibitor of the shikimate pathway of aromatic amino acid biosynthesis in plants and microorganisms (Benbrook 2016). The potential leukemogenic effect of glyphosate via its active parent compound, metabolites, or inactive (so-called inert) ingredients in product formulations remains unknown. Limited studies in animals and humans indicate that glyphosate and glyphosate-based herbicides increase oxidative stress, a potential pathway for carcinogenesis (IARC 2017). Perhaps more importantly for immunological diseases, glyphosate has been shown to affect the immune system through alterations in lymphocyte responses, complement cascade, and increased production of pro-inflammatory cytokines (IARC 2017; Peillex and Pelletier 2020). These toxic effects warrant studies of additional health outcomes in children besides leukemia.

Occupational exposure to glyphosate has been associated with increased risk of adult non-Hodgkin lymphoma in a few case-control studies (IARC 2017; Meloni et al. 2021) but not in the Agricultural Health Study cohort of pesticide applicators (Andreotti et al. 2018). However, risk of acute myeloid leukemia (AML) was elevated in the Agricultural Health Study (Andreotti et al. 2018). Meta-analyses of studies of childhood leukemia and household pesticide use including the CCLS study (Ma et al. 2002) show consistent positive associations with risk (Bailey et al. 2015; Turner et al. 2010; Van Maele-Fabry et al. 2017). Few studies estimated specific pesticide exposures (Turner et al. 2010). In the CCLS, self-reported lawn/garden pesticide use predicted dust concentrations of herbicides commonly used during the study period. Associations were similar for cases and controls demonstrating that recall bias was not likely to have occurred.

One study evaluated agricultural use of glyphosate and childhood leukemia. A record-linkage case-control study of childhood ALL and AML among young children in California evaluated agricultural pesticide applications within 4 km of homes. In single pesticide models, glyphosate use within 4 km (ever vs. never) was associated with a significant two-fold increased risk of ALL and an elevated risk of AML (Park et al. 2020). However, risk of ALL was increased for many pesticides in addition to glyphosate. In Bayesian hierarchical models with multiple pesticides, risks were attenuated for most pesticides, but glyphosate was not evaluated in multiple pesticide models. Limitations of this study with respect to the exposure assessment for glyphosate were the lack of information on the density of agricultural applications, parental occupational exposures, and home and garden use. Our finding of a significant relationship between the density of agricultural use near homes and glyphosate concentrations in the house dust suggests that agricultural use near homes may be an important source of exposure to children. Agricultural use during potentially critical windows of exposure (pregnancy, early childhood) will be an important area for future research.

A major strength of our study is that we conducted environmental sampling to quantify levels of glyphosate and other herbicides in the homes. Other strengths include the rapid case ascertainment, population-based selection of controls, and high participation rates for the dust sampling and interviews. Further, we had repeat samples a few years apart with which to evaluate temporal variability. Our sample size for the analyses of glyphosate was not as large as that for the prior pesticide analyses due to limited amounts of dust remaining for some participants; however, we had adequate power to detect about a two-fold increased risk of ALL comparing the highest to lowest quartile. By design, samples were taken after the child’s diagnosis, which may not accurately reflect exposure levels during the etiologically relevant time periods of pregnancy and early childhood. Furthermore, we assessed temporal variability many years after diagnosis. Due to the temporal trends in household and agricultural glyphosate use, our variance estimates may not accurately reflect within and between home variability in the etiologic relevant period before diagnosis (late 1990s to 2007). Nevertheless, the considerable temporal variability in glyphosate concentrations that we noted in the subset of participants with two samples likely resulted in misclassification of exposure. Controls had significantly higher household income than cases. However, income was not a predictor of glyphosate concentrations in homes and was not a confounder in our analyses. Moreover, analyses stratified by income showed similar associations with ALL risk, providing some assurance that selection bias was not an important factor in our results.

1.5. Conclusions

In summary, we did not find an association between glyphosate concentrations in homes and childhood ALL risk using a household dust sample as an indicator of exposure. Determinants of levels in homes included agricultural applications near the home, having a pesticide-exposed worker in the home, and household lawn weed treatments. Due to the continuing use of glyphosate in residential areas and increases in some agricultural uses, the potential for exposure to the developing fetus and young children is high. It will be important to evaluate this association in future studies with repeated dust measurements or biomarkers of exposure.