Outdoor artificial light at night, air pollution, and risk of childhood acute lymphoblastic leukemia in the California Linkage Study of Early-Onset Cancers

Acute lymphoblastic leukemia (ALL) is the most common type of cancer in children (age 0–14 years); however, the etiology remains incompletely understood. Several environmental exposures have been linked to risk of childhood ALL, including air pollution. Closely related to air pollution and human development is artificial light at night (ALAN), which is believed to disrupt circadian rhythm and impact health. We sought to evaluate outdoor ALAN and air pollution on risk of childhood ALL. The California Linkage Study of Early-Onset Cancers is a large population-based case–control in California that identifies and links cancer diagnoses from the California Cancer Registry to birth records. For each case, 50 controls with the same year of birth were obtained from birth records. A total of 2,782 ALL cases and 139,100 controls were identified during 2000–2015. ALAN was assessed with the New World Atlas of Artificial Night Sky Brightness and air pollution with an ensemble-based air pollution model of particulate matter smaller than 2.5 microns (PM2.5). After adjusting for known and suspected risk factors, the highest tertile of ALAN was associated with an increased risk of ALL in Hispanic children (odds ratio [OR] = 1.15, 95% confidence interval [CI] 1.01–1.32). There also appeared to be a borderline association between PM2.5 level and risk of ALL among non-Hispanic White children (OR per 10 µg/m3 = 1.24, 95% CI 0.98–1.56). We observed elevated risk of ALL in Hispanic children residing in areas of greater ALAN. Further work is needed to understand the role of ALAN and air pollution in the etiology of childhood ALL in different racial/ethnic groups.


Methods
Study population. The California Linkage Study of Early-Onset Cancers is a statewide linkage of birth records maintained by the California Department of Public Health and cancer diagnoses reported to the California Cancer Registry. The study was performed in accordance with relevant guidelines and regulations and the protocol was approved by the institutional review boards at the University of California, Berkeley, University of Southern California, and Yale University. This study was reviewed and approved by the Vital Statistics Advisory Committee and the institutional review board of the California Health and Human Services Agency and exempt from informed consent as data used was de-identified in accordance with California Code Sect. 102430 for the protection of human subjects that is approved by the federal Department of Health and Human Services and has a general assurance pursuant to Part 46 of Title 45 of the Code of Federal Regulations.
For the current analysis, eligible cases included all children who were born in California from 2000 through 2015 and were diagnosed with incident primary ALL between 2000 to 2015. ALL was defined as a diagnosis with an International Classification of Diseases for Oncology, 3rd edition, code of 9820, 9823, 9826, 9827, 9831-9837, 9940, or 9948. There were 2,819 cases initially identified, from which we excluded missing data on maternal residential address at the time of delivery (n = 6), maternal residence outside of California (n = 4), and missing data on congenital anomality or unknown information on congenital anomality (n = 27). For each case, we randomly selected 50 controls born in the same year from birth records. The final study population consisted of 2,782 childhood ALL cases and 139,100 controls. Exposure assessment. We assessed ALAN using the New World Atlas of Artificial Night Sky Brightness 18 , which provides a 750-m gridded spatial resolution global measure of luminance at the zenith (directly overhead) in millicandela per meter squared (mcd/m 2 ). In contrast to using observations only captured by a satellite, which represents light that escapes upwards into the atmosphere, the World Atlas produces a better estimate of ground level light exposure because in includes measurements from handheld sky meters 19 . The New World Atlas was developed with observations from the National Oceanic and Atmospheric Administration Visible Infrared Imaging Radiometer Suite observations from May-December 2014. While this only provides a single 6-month average that was assigned retrospectively, the improved resolution of the sensor technology greatly reduce misclassification, and multiple studies have demonstrated stability in ALAN over time and high correlation between years in satellite estimates (R 2 0.92-0.98) [20][21][22][23][24] . ALAN was assigned to each case and control child by their geocoded maternal residential addresses at the time of delivery.
Air pollution was assessed with a validated United States (US) national model developed by Di et al 25 . Briefly, the model combines outputs from a chemical transport model with variables characterizing land use, population density, weather patterns, and satellite derived aerosol optical depth in an ensemble of machine learning models to estimate daily PM 2.5 concentrations at a 1-km resolution. Validation of this model against air monitoring stations resulted in an R 2 of 0.802 for the Pacific region. Cases were assigned the average PM 2.5 concentration from birth until diagnosis. The exposure window for matched controls began at birth and ended when the corresponding case was diagnosed (e.g., if a case was diagnosed at 30 months, the matched controls were assigned the average PM 2.5 from birth until 30 months of age).
Other variables of interest. We abstracted characteristics from birth records, including birthweight, gestational age (22-36, 37-41, 42-44 weeks, or unknown), plurality (singleton or multiple), birth order (1st, 2nd, 3rd or higher), mode of delivery (cesarean section [C-section] or vaginal), complications during pregnancy, maternal history of miscarriage or stillbirth (yes/no), or history of previous cesarean section (no/yes/unknown). We also retrieved demographic and parental characteristics including sex, race/ethnicity, maternal age at delivery, mother's place of birth, maternal education, and paternal age at delivery. We also linked maternal residential address to 2000 Census block group data to obtain the percentage of the population in the block group living below 150% of the federal poverty level. Of these factors, Hispanic ethnicity 1 , higher birthweight 26,27 , lower birth order 28,29 , caesarean delivery 30,31 , and older parental ages 32,33 , have been linked to an increased risk of childhood ALL, and the others are potential confounders for our analysis of ALAN and air pollution in relation to ALL risk. www.nature.com/scientificreports/ Statistical methods. Spearman's correlation was performed to evaluate correlation between ALAN and PM 2.5 , with the former being the primary exposure of interest in this study and the latter being a potential confounder that has been linked to both ALAN and ALL risk in previous studies 13 . The associations between outdoor ALAN, PM 2.5 and risk of childhood ALL were assessed using logistic regression models, and odds ratios (OR) and 95% confidence intervals (95% CI) were calculated. Examining the variables listed in Table 1
Among the overall study population, we did not observe a clear association between outdoor ALAN and childhood ALL (OR 3rd tertile 1.07, 95% CI 0.98-1.17) ( Table 2). We did observe an association between PM 2.5 and ALL risk (OR 1.11, 95% CI 1.01-1.22) that was no longer significant after adjusting for other covariates (adjusted OR 1.08, 95% CI 0.96-1.20). There was increased risk of childhood ALL in male children compared to female (adjusted OR 1.17, 95% CI 1.08-1.26). Compared to non-Hispanic White children, Hispanic ethnicity was associated with increased ALL risk (adjusted OR 1.29, 95% CI 1.16-1.43) while non-Hispanic Black (adjusted OR 0.65, 95% CI 0.52-0.82) children had a lower risk. Compared to children born at a normal birthweight (3000-3499 g), low birthweight (< 2500 g) was associated with decreased ALL risk (adjusted OR <2500 g 0.77, 95% CI 0.64-0.92) while high birthweight (> 4000 g) was associated with increased ALL risk (adjusted OR >4000 g 1.15, 95% CI 1.01-1.31). Delivery by C-section was associated with a 1.16 (95% CI 1.07-1.26) increased risk of ALL. Mothers older than 35 years were more likely to have children who developed ALL compared to mothers between 25-29 years (adjusted OR 1.16, 95% CI 1.01-1.32). Children with fathers under age 25 years were less likely to have ALL compared to those with fathers between the age of 30-34 years (adjusted OR 0.82, 95% CI 0.71-0.95). Compared to mothers who completed high school, those with less than 8 years of school were less likely to have children who developed ALL (adjusted OR 0.76, 95% CI 0.65-0.89). Children of foreign-born mothers were less likely to develop ALL compared to US born mothers (adjusted OR 0.90, 95% CI 0.82-0.98).
Among Hispanic children, there was a 15% increased risk of childhood ALL (95% CI 1.01-1.32) for those residing in the highest tertile of ALAN compared to the lowest (Fig. 1). In non-Hispanic White children, there did not appear to be an association between exposure to outdoor ALAN and ALL risk (adjusted OR 3rd tertile 0.84, 95% CI 0.66-1.06, Fig. 2). The association between PM 2.5 and ALL appeared to be stronger in non-Hispanic White children (adjusted OR 1.24, 95% CI 0.98-1.56) than Hispanic children (adjusted OR 1.05, 95% CI 0.90-1.21). Delivery by C-section was the only covariate still significantly associated with increased ALL risk in non-Hispanic White children (adjusted OR 1.28, 95% CI 1.08-1.51).
When stratified by age, the risk association of ALAN and ALL was highest for the oldest age group with evidence of an increasing trend from 1.00 (95% CI 0.84-1.19, Supplemental Table 2 and Supplemental Fig. 1) in the 0-2 years age group to 1.22 (95% CI 0.88-1.69) in the 8-14 years age group. Foreign maternal place of birth was only protective in the 0-2 years age group (adjusted OR 0-2 years 0.84, 95% CI 0.73-0.97). Compared to non-Hispanic White children, the risk among non-Hispanic Black children was lower in the 0-2 years age group (adjusted OR 0-2 years 0.60, 95% CI 0.41-0.87) and 3-7 years age group (adjusted OR 3-7 years 0.61, 95% CI 0.44-0.85) and null in the oldest age group. In contrast, the risk in Hispanic children was not significant in the 0-2 years age group (adjusted OR 0-2 years 1.17, 95% CI 0.98-1.38) but increased in successive older age groups (adjusted OR 3-7 years 1.33, 95% CI 1.14-1.54, adjusted OR 8-14 years 1.60, 95% CI 1. 16-2.22). Birthweight was also only significantly associated with ALL risk in the 0-2 years age group (adjusted OR <2500 g 0.52, 95% CI 0.37-0.74).

Discussion
In this large, population-based case-control study, exposure to outdoor ALAN was associated with increased childhood ALL risk, but only among Hispanic children. These results are similar to those reported in adult leukemia 5 . Compared to many ecological studies on the association between ALAN and cancer risk, we were able to adjust for important confounders, including birth characteristics, parental factors, markers of socioeconomic status, and air pollution. Since air pollution is highly correlated with ALAN and has been previously linked to  13 . PM 2.5 did not appear to be a confounder of ALAN in multivariable analyses, suggesting that the association between ALAN exposure and childhood ALL risk was independent of air pollution. Exposure to light prior to sleep leads to suppression of melatonin and delayed onset of sleep [34][35][36][37] . Melatonin is involved in the nuclear transcription factor kappa beta pathway 38 , which modulates levels of many immune cytokines, such as interleukin-8 and interleukin-10, that have been previously implicated in the etiology of childhood ALL 39,40 . Disruptions to sleep can also alter inflammatory response from Toll-like receptors 41 , which are involved in hematopoiesis and leukeogenesis 42 . We observed the strongest association between ALAN and ALL in the older 8-14 years age group (Supplemental Fig. 1), so it may be the continued disruption to these pathways over time that contribute to development of leukemia. Additional studies have also linked ALAN exposure to circadian disruption 43 and tumor progression through potential epigenetic pathways 44,45 . Several genes involved in circadian rhythm have been found to be differentially expressed in ALL patients as well 46 . As our study is the first to report a novel association between exposure to outdoor ALAN and childhood ALL risk, it would be helpful to further examine the role of ALAN in other epidemiological studies of childhood ALL and conduct additional mechanistic evaluations.
Prior studies on ALL have demonstrated differential risk patterns between Hispanic and non-Hispanic White children -including risk factors such as daycare attendance, infections, diet, and genetics [47][48][49][50] . Hispanic children have the highest risk of ALL among all racial/ethnic groups in California and the US 51 , supporting a possibly different etiological profile compared to non-Hispanic Whites. For instance, compared to non-Hispanic White children in our study, Hispanic children resided in areas of greater PM 2.5 and ALAN (Supplemental Table 1). Differences in genetics may also explain how the environment influences the way certain individuals experience ALAN and disruptions in circadian rhythm. Studies have reported shorter sleep durations in minority populations compared to non-Hispanic Whites 52 . It could be a combination of both environment and genetics as interactions have been observed for air pollution and other conditions such as cardiovascular disease 53 and Parkinson's disease 54 . There may also be interactions with other attributes of the built environment that we did not evaluate, such as green space, which has been shown to improve health 55 and attenuate the risks observed with air pollution exposure 56,57 , especially in communities of lower socioeconomic status 58 .
Strengths of our study include the large, population-based sample that we were able to assemble by linking statewide cancer diagnosis information to the statewide birth records in California, the most populous state in Table 1. Distribution of outdoor artificial light at night, air pollution, and birth characteristics in the California Linkage Study of Early-Onset Cancers. * Percentages may not add up to 100 due to rounding.   www.nature.com/scientificreports/ the US. The California Cancer Registry is a part of the Surveillance, Epidemiology, and End Results program and provides high quality and comprehensive ascertainment of cancer cases diagnosed in the state. Linkage to the birth records provides information on many important covariates encompassing known/suspected risk factors for childhood ALL and other potential confounders for our analysis of ALAN and air pollution. Notably, no cases or controls had to be contacted for participation in the study, minimizing selection bias. In addition, we derived high resolution estimates of ALAN and PM 2.5 based on maternal residential address at the time of delivery, which was documented in birth records prior to the diagnosis of childhood ALL, therefore reducing recall bias.
A primary limitation of this study is the lack of data on residential history. While we do have additional address information for cases at time of ALL diagnosis, the corresponding information for the reference date (age at which the matched case was diagnosed) is not available for our controls. Residential mobility appears to be greater in early childhood, though most moves tend to be to similar neighborhoods 59,60 . In addition, we have no reason to believe that residential mobility, in regards to ALAN or PM 2.5 exposure, would differ systematically between cases and controls. Another limitation is that the measure of outdoor ALAN that we used was based on observations from 2014 and consisted of an estimate of luminance at the zenith. It did not characterize light at various wavelengths, which may be an important distinction. Animal studies show that shorter, blue spectrum wavelength light (450-500 nm) was more detrimental to sleep 8,61 , which could be an issue as cities are moving away from amber lights towards more shorter wavelength LED (light emitting diode) streetlights 44 20 . Another important constraint is that we do not have any data on indoor exposure to light at night, which could be influenced by curtains or other light blocking accessories, type of windows and glasses, and type of indoor light sources. A comparison of outdoor ALAN and personal exposure in children in the Netherlands only found a weak correlation of 0.31 62 . However, disruptions to levels of melatonin have been shown to persist hours after exposure to light stimuli 34,35 . The primary exposure of interest in this study is outdoor ALAN, not indoor ALAN or all ALAN. Hopefully, our work can inspire investigators of future studies to expand the scope of exposure assessment to include all types of ALAN.
With regard to air pollution, we recognize that a limitation of our study is the inability to evaluate the composition of particulate matter, e.g., air pollution from farming may contain pesticides as opposed to carbon and brake dust from road traffic. Elucidation of the components of air pollution is needed to better understand the constituents behind the risk association between air pollution and childhood ALL. A recent review found levels of outdoor PM 2.5 to be correlated with indoor PM 2.5 , though it varied depending upon the composition 63 . Lastly, despite our ability to adjust for many known/suspected risk factors of childhood ALL, residual confounding due to imprecise measurements or unmeasured factors (e.g., exposure to smoking) remains a possibility.
In summary, exposure to outdoor ALAN was associated with an increased risk of ALL among Hispanic children. Future studies of ALAN and childhood cancers will require additional assessment of indoor ALAN and the relationship between ALAN and sleep. Inclusion of genetics may identify potential gene-environment interactions that cause individuals to differentially respond to their environment. This is especially important in Hispanic children, a subgroup of the population in which childhood ALL rates have been increasing in recent years 1 .

Data availability
The datasets generated during and/or analyzed during the current study are not publicly available due to data sharing policies set by the state of California. We welcome questions from other investigators or requests for additional analyses that are pertinent to the data presented in this manuscript, and potential data sharing when permitted by the California Health and Human Services Agency Committee for the Protection of Human Subjects.