Cytokine Levels at Birth in Children Who Developed Acute Lymphoblastic Leukemia

Childhood acute lymphoblastic leukemia (ALL) is a malignancy of immature immune cells and there is substantial evidence that immune-system development affects the risk of this disease. The delayed infection hypothesis posits that when children are not exposed to common infections early in life, their uneducated immune systems will be more likely to react aberrantly to subsequent infections, potentially leading to the development of ALL (1). Observational studies using surrogate measures of immune-system priming early in life – such as being born via vaginal delivery (2), living with older siblings (3), having social contacts in daycare settings (4), or being vaccinated (5) – have consistently reported a subsequent reduction in childhood ALL risk, supporting the role of early immune-system education in conferring protection from potential leukemogenic-stimulating infections.

Several common ALL molecular subtypes, including those characterized by the t(12;21)(p13;q22) translocation (6) or high hyperdiploidy (a karyotype with 51 to 67 chromosomes; refs. 7–10), are initiated in utero. Studies of neonatal biomarkers suggest that altered immune-system development before birth may play a role in ALL etiology. As surrogate markers for systemic immune development in utero, our group measured neonatal levels of cytokines, the immune-communication proteins that circulate in the blood. In an analysis of archived neonatal blood spots collected at birth from 116 ALL cases and 116 controls, we showed that newborns who went on to develop childhood ALL were born with profoundly depressed levels of IL10 (11), a critical immunosuppressive cytokine that limits immune responses to infection after birth. Likewise, a study of Danish national registries assessed neonatal cytokine levels among 178 children diagnosed with B-cell ALL and 178 controls, demonstrating that children who developed ALL were born with lower levels of IL8 and higher levels of IL6 compared with controls (12). These two studies of cytokine levels at birth support the role of neonatal immune status in the etiology of childhood ALL; however, these previous studies were not sufficiently large to investigate variations in the relationship between cytokines and ALL risk by ethnicity, by leukemia subtype, by age at leukemia diagnosis, or by sex.

There is a parallel line of research that implicates prenatal chemical exposures, some of which are immunotoxic, in the etiology of childhood ALL. It is not known whether changes in cytokine levels in utero act as mediators in the relationship between prenatal exposures and childhood leukemia risk. However, there are at least four classes of chemicals, polycyclic aromatic hydrocarbons (PAH; ref. 13), polychlorinated biphenyls (PCB; ref. 14), polybrominated diphenyl ethers (PBDE; ref. 15), and pesticides (16), as well as two other exposures, cigarette smoke (17) and in-home paint use (18), that are putative prenatal risk factors for childhood ALL with the potential to induce changes in cytokine levels (19–23). For example, animal models and human studies have demonstrated that maternal smoking during pregnancy has the potential to disrupt cytokines levels at birth (24, 25). Furthermore, chemical-induced alterations in cytokine levels contribute to elevated risks of a variety of adverse health outcomes, including inflammatory conditions (26–29), autoimmune diseases (30), and cancer (31, 32).

Here we expand our initial study (11) to analyze seven cytokines in neonatal blood spots collected from more than 1,000 ALL cases and 1,000 controls. Incidence rates of childhood ALL vary by ethnicity, leukemia subtype, age, and sex, which suggests that multiple, distinct etiologies exist and in this, the largest study to investigate the relationship between neonatal cytokine profiles and the subsequent risk of childhood ALL, we were able to conduct additional analyses stratified by these factors. We also examine, for the first time, the interplay between neonatal immune status, prenatal exposures, and the risk of childhood ALL.

The California Childhood Leukemia Study (CCLS) is a population-based case-control study designed to identify genetic and environmental risk factors for childhood leukemia. Incident cases 0–14 years of age were ascertained from pediatric clinical centers in California from 1995 to 2015; controls, matched to cases on date of birth, sex, ethnicity, and maternal race, were selected from California birth records as described previously (33). None of the controls were diagnosed with any type of cancer in childhood. A guardian (usually the biological mother) was interviewed to ascertain information on a variety of topics including sociodemographic characteristics (mother's race and ethnicity, household annual income, and mother's age at birth), birth characteristics (delivery method, gestational age at birth, birth weight, and breastfeeding), medical histories (of the mother and child), and a series of household exposures. Because not all case-control pairs had sufficient blood for cytokine analysis, we broke the matched study design, here, to maximize the sample size and the corresponding statistical power and we included the original matching variables in our regression models. In total, 1,020 childhood ALL cases and 1,003 controls were available for cytokine analysis.

Blood spots were collected from newborns as part of statewide testing for genetic disorders conducted by the California Newborn Screening program; since 1982 the leftover blood spots have been archived. Five 14-mm blood spots were collected from each infant on filter paper by heel-stick during the first few days after birth (95% of the CCLS samples were collected within the first 72 hours, median of 27 hours). After parents were given an opportunity to opt out of any future research, two or three leftover blood spots were stored in the California Department of Public Health biobank at −20°C. We were provided with a portion of one of these archived neonatal blood spots from each participant for this analysis. All study protocols were approved by the University of California Berkeley Office for the Protection of Human Subjects and by the State of California Committee for the Protection of Human Subjects.

For each participant, two 4.7-mm punches (∼33%) of a blood spot were excised, placed together in an Eppendorf tube with 160 mL of extraction buffer (PBS, pH 7.4, 0.5% Tween-20, and a complete protease inhibitor cocktail), shaken at 600 rpm for 1 hour at room temperature, then incubated at 4°C overnight (11). Extracts were assayed in duplicate and block randomized on 96-well plates with each plate containing similar proportions of cases and controls and racial/ethnic groups. Nine cytokines, IL1, IL4, IL6, IL8, IL10, IL12p70, GM-CSF, TNFα, and VEGF, were initially measured using a Luminex bead–based assay (R&D Systems Multiplex Cytokine Assay Kit) incorporating calibration standards, quality control samples, and blanks. If the coefficient of variation between analytic duplicates was greater than 10%, the assay was repeated. The total protein content of each extract was determined with the Bradford Assay as an indicator of extraction efficiency. Cytokine measurements for each duplicate were averaged. Eight of the nine cytokines were detected in at least 99% of all blood spots. TNFα was the exception: levels of this cytokine were below the analytical limit of detection in 17% of all samples. Values below the limit of detection were replaced with half of the lowest observed value. We conducted one-way random-effects ANOVA to estimate within- and between-batch variability across 57 analytic batches. Two cytokines, IL10 and IL12p70, that were found to have intraclass correlation coefficients (ICC) in excess of 25% (i.e., large batch effects) were excluded from the statistical analysis, below. The other seven cytokines with lower between-batch variance (ICCs from 9% to 25%) underwent additional processing to account for their own (more modest) batch effects.

Because blood spots were collected over a long period of time, with births occurring between 1984 and 2012, and then archived for as long as 35 years prior to the cytokine analysis, degradation was a potential concern. As such, raw cytokine concentrations were normalized and log-transformed with the Bioconductor R package SCONE (34), which implemented and evaluated different scaling and regression-based normalization methods for removing unwanted variation while preserving differences in case status. The optimal normalization scheme accounted for the following unwanted sources of technical variation: birth year, total protein content, analytic batch, and well position. Supplementary Figure S1 shows diagnostics of the SCONE normalization using IL8 as an example. For most of the cytokines, the rank order of levels was similar before and after SCONE normalization, ie. r_s > 0.8. The two exceptions were cytokines IL10 (pre- vs. post-SCONE r_s = 0.73) and IL12p70 (r_s = 0.66), which were subsequently excluded from further statistical analysis. Supplementary Figure S1 illustrates how the SCONE protocol minimized batch effects for the seven cytokines with batch ICCs of 25% or less, for example, IL8. Any subsequent mention of cytokine levels in this paper will be referring to the levels after SCONE normalization.

A variety of prenatal exposures that are putative ALL risk factors have been assessed in CCLS participants previously (see Supplementary Table S1). During interviews, participants were asked to describe: maternal smoking during pregnancy (reported for 900 cases and 1,001 controls with cytokine measurements), in-home pesticide use during pregnancy (729 cases, 995 controls), and in-home paint use during pregnancy (690 cases, 955 controls). Mothers also reported infection histories from pregnancy (319 cases, 379 controls). A subset of the blood spots analyzed here for cytokines were previously screened for chemical biomarkers – in the form of human serum albumin adducts (318 cases, 332 controls) and small-molecule features (316 cases, 316 controls). Of the thousands of chemical biomarkers detected in the blood spots, we focused here on a set of 28 albumin adducts (35) and 28 endogenous metabolites (36) that were previously associated with childhood leukemia. Finally, levels of persistent chemicals have been measured in dust samples collected from CCLS homes after birth (226 cases, 260 controls). Previous analyses have indicated that these measurements are representative of prenatal chemical contamination in the home (37–39) and that dust levels of certain herbicides, PAHs, PCBs, and PBDEs are associated with ALL risk (13–15, 40).

Missing covariate data were replaced with the median non-missing value for cases and controls, separately, as follows: mother's race/ethnicity for 116 cases (Latina) and 2 controls (non-Latina white); household annual income for 145 cases ($30k to $44k) and 32 controls ($60k to $74k); mother's age at birth for 118 cases (28.7 years) and 2 controls (29.2 years); delivery method for 148 cases and 57 controls (vaginal delivery for both); gestational age at birth for 207 cases and 30 controls (40 weeks for both); birth weight for 137 cases and 25 controls (3.46 kg for both); breastfeeding for 120 cases and 3 controls (breastfed for both); and blood spot collection time for 22 cases (27 hours after birth) and 31 controls (26 hours). Replacement of missing covariate data allowed for regression analyses encompassing the entire group of participants with cytokine measurements (1,020 cases and 1,003 controls); but we also ran regression analyses among only those participants with complete covariate data (734 cases and 843 controls).

All statistical analyses were conducted in R version 3.6. Matches between case-control pairs were broken in this analysis. Bivariate relationships between cytokine levels and participant characteristics were evaluated with non-parametric methods: Kruskal–Wallis tests for categorical variables and Spearman rank correlation coefficients for ordinal or continuous variables. Relationships between individual cytokines were examined using Spearman correlation coefficients and principal component analysis. Because levels of individual cytokines tend to be highly correlated, we used principal components analysis to produce summary measures of groups of correlated cytokines for use in subsequent regression analyses of ALL risk. Individual cytokine levels from control participants were centered and scaled prior to principal component analysis with the prcomp function in R. After examining the percent of variation explained by each of the principal components (PC), two PCs were projected back onto the case participants using linear combinations of individual cytokine levels. ORs associated with an interquartile range increment in cytokine levels (as defined by the distribution in the control group) were calculated using logistic regression models with or without covariates (i.e., sociodemographic and birth characteristics). Secondary regression analyses were conducted while stratifying by race/ethnicity, ALL subtype, age at diagnosis, birth year, and sex because the risk of childhood ALL has been found to vary among these factors. We used a mediation analysis (using mediate in R) to evaluate whether cytokines were mediating the relationship between prenatal exposures and ALL risk (see Supplementary Fig. S2).

Cytokines were related to several demographic and birth characteristics, as shown in Table 1. Cytokine levels were negatively correlated with blood collection time suggesting that concentrations decreased in the hours and days after birth. Cesarean delivery, premature gestational age, and low birth weight were also linked to lower cytokine levels. Newborns that went on to develop childhood ALL were more likely to be born prematurely and via Cesarean delivery than controls (preterm birth; ref. 41) and C-section delivery (2) have been identified as ALL risk factors previously) and less likely to have blood collected within the first 24 hours after birth.

Control families were more likely than case families to be non-Latino and white with higher annual incomes. Concentrations of IL1β, IL4, and IL8 differed by mother's race/ethnicity – with newborns of non-Latina white mothers having the lowest levels. Newborns from high-income households tended to have lower levels of IL1β, IL6, and IL8 in their blood at birth than newborns from low-income households.

Levels of individual cytokines also differed by child's sex and mother's age at birth, but the male-to-female ratio and the distribution of mothers' ages were similar between case and control groups. Breastfeeding was not related to cytokine levels or case-control status.

Figure 1 shows the relationship between individual cytokines among controls. Levels of all cytokines were correlated with each other at birth, with Spearman rank correlation coefficients of 0.10 or more and all P values < 0.05. There were two groups of cytokines – GM-CSF, IL4, and VEGF (top left) and IL1β, IL8, IL6, and TNFα (bottom right) – within which the individual cytokines were particularly highly correlated (r_s of 0.34 to 0.65).

To accommodate the correlation between cytokines, we used principal component analysis to reduce the dimensionality of the data from seven individual cytokines to two principal components (PCs). These two PCs explained 66% of the variance in cytokine levels. PC1 was a linear combination of the seven individual cytokines, loaded most heavily by IL1β, IL8, VEGF, and TNFα. PC2 was positively loaded by GM-CSF, IL4, and VEGF and negatively loaded by IL1β, IL8, IL6, and TNFα, mirroring the two groups of more highly-correlated cytokines mentioned above.

Table 2 shows ORs for childhood ALL associated with cytokine levels at birth. We ran four models for each of the seven individual cytokines and four additional models which included the two PCs together. In each instance, two of the models used the entire study population (N = 1,020 cases and 1,003 controls; modeled with or without adjustment for sociodemographic and birth characteristics) and two of the models used only those participants with complete covariate data (N = 734 cases and 843 controls; modeled with or without covariates). Higher levels of PC1 were consistently associated with increased odds of childhood ALL and these results were statistically significant in fully adjusted models. In fully adjusted models, the odds ratios associated with an interquartile range increment in PC1 were 1.14 [95% confidence interval (CI): 1.00–1.30] across the entire study population and 1.17 (95% CI, 1.02–1.36) for the subset of participants with complete covariate data. ORs from the fully adjusted models for IL1β, IL8, TNFα, VEGF – the four individual cytokines which loaded most heavily on PC 1 – were likewise statistically significant, ranging from 1.13 to 1.19. Levels of the other three individual cytokines – IL4, IL6, and GM-CSF – were not associated with ALL risk.

We conducted additional regression analyses stratifying by race/ethnicity, ALL subtype, age at diagnosis, birth year, and sex because the risk of childhood ALL has been found to vary among these factors. We report the stratified analyses with the most striking differences between subgroups in the main paper (Tables 3 and 4) and present the remainder of the findings in the supplemental materials (Supplementary Tables S2–S4). In the main analyses, replacement of missing covariate data produced odds ratios that were similar – albeit somewhat smaller than – the odds ratios among participants with complete covariate data. Therefore, we continued to use imputed covariate data in these stratified analyses to maximize sample sizes.

Table 3 shows ORs for childhood ALL associated with cytokine levels when stratifying by mother's race/ethnicity. In the two largest groups – newborns with Latina or non-Latina white mothers – there was a stark contrast in risk. An interquartile range increment in PC1 levels was associated with an OR of 1.34 (95% CI, 1.09–1.66) among children with Latina mothers and 0.98 (95% CI, 0.79–1.21) among children with non-Latina white mothers (test for heterogeneity, P = 0.04). Likewise, levels of IL1β, IL8, TNFα, and VEGF – the four individual cytokines that were associated with ALL risk in the combined regression analyses – had higher (and statistically significant) odds ratios among children with Latina mothers and lower (nonsignificant) ORs among children with non-Latina white mothers. There was one (apparently sporadic) significant association between ALL risk and level of GM-CSF among children with non-Latina Asian mothers, but the sample sizes for analyses among this group were limited.

Table 4 shows ORs for ALL subtypes associated with cytokine levels. The OR for B-cell ALL associated with an interquartile range increment in PC1 (1.13) was similar to the OR for ALL overall (1.14, see Table 2); although this result no longer reached statistical significance. Further stratifying patients with B-cell ALL into the two most common cytogenetic classifications – those with ALL characterized by t(12;21)(p13;q22) translocation or by high hyperdiploidy with chromosome counts from 51 to 67 – revealed subtype-specific associations with cytokines at birth. Increasing PC1 levels were associated with an OR of 1.23 (95% CI, 1.01–1.50) among children with high hyperdiploidy and 0.96 (95% CI, 0.75–1.24) among children with t(12:21) (test for heterogeneity, P = 0.10). Likewise, levels of the four individual cytokines that were associated with ALL risk in the combined regression analyses (IL1β, IL8, TNFα, and VEGF) had higher (and marginally significant) ORs among children with high hyperdiploidy and lower (nonsignificant) ORs among children with t(12:21). None of the ORs for T-cell ALL, a rare subtype, were statistically significant.

Supplementary Table S2 shows ORs for childhood ALL associated with cytokine levels when stratifying by age at diagnosis date (or corresponding reference date for controls). ORs associated with an interquartile range increment in neonatal levels of PC1 were similar for ALL diagnosed under age 2 years (OR, 1.19), from ages 2 to 5 years (OR, 1.19), and from ages 6 to 14 years (OR, 1.12), although none of these results reached statistical significance. Levels of individual cytokines were positively associated with ALL diagnosed from ages 2 to 5 (IL6, TNFα, and VEGF) or from ages 6 to 14 (IL8) with small variations in ORs between the two groups.

Supplementary Table S3 shows odds ratios for ALL associated with cytokine levels when stratifying by birth year. ORs associated with an interquartile range increment in PC1 were highest among the most recent births (1.27; 95% CI, 0.97–1.67) from 2000 to 2012 (test for heterogeneity, P = 0.28), although this result no longer reached statistical significance. Likewise, levels of IL1β, IL8, TNFα, and VEGF – the four individual cytokines which loaded most heavily on PC 1 – had higher (and marginally significant) ORs among children born in 2000 or later and these four cytokines had ORs closer to 1 among births prior to 2000.

Supplementary Table S4 shows odds ratios for ALL associated with cytokine levels when stratifying by child's sex. ORs associated with an interquartile range increment in PC1 were of similar magnitude in boys (1.15; 95% CI, 0.98–1.36) and girls (1.16; 95% CI, 0.94–1.43), but these results no longer reached statistical significance.

Figure 2 shows the relationship between levels of cytokines at birth and levels of a variety of prenatal exposures that have been previously assessed in the study participants (13–15, 35–39, 42–45). There were 15 positive links (in red) between levels of one of the 7 cytokines at birth and levels of one of the prenatal exposures, with Spearman correlation coefficients from +0.15 to +0.30. Of these 15 strongest positive relationships, 10 involved an endogenous metabolite measured at birth in the same blood spots (most commonly an unknown metabolite with putative molecular formula of C₂₆H₄₄O₂). There were 35 negative links (in blue) between a cytokine and a prenatal exposure, with correlation coefficients ranging from -0.15 to -0.33. Of these 35 strongest negative relationships, 33 involved an endogenous metabolite [most commonly the sphingolipid, sphingomyelin (d16:1/20:4), the glycerophospholipid, phosphaditylcholine (P16:0/20:4), and two unknown molecules with m/z of 570.3401 and 696.3573]. Notably, sphingomyelin (d16:1/20:4) was negatively correlated with VEGF, IL1β, and TNFα (r_s < −0.18). There were 741 neutral links (in gray/white) between a cytokine and a prenatal exposure with correlation coefficients between −0.15 and +0.15. Of note among these neutral links were the relationships between GM-CSF and an adduct of crotonaldehyde (r_s: −0.12) and between IL6 and an adduct of cysteine (r_s: −0.14), as both of these adducts were previously associated with childhood ALL risk (35).

To evaluate whether cytokines were acting as mediators between the circulating small molecules and childhood ALL risk we conducted the mediation analysis summarized in Supplementary Table S5. The mediation analysis comprised 28 endogenous metabolites which have been previously identified (in analyses stratified by age at diagnosis) as putative risk factors for childhood ALL (36). In the mediation analysis, we evaluated 28 base models for case-control outcome, each with one endogenous metabolite as an explanatory variable and covariates for sociodemographic and birth characteristics, but excluding cytokines. Next we evaluated 28 corresponding models that added a term for cytokines into the regression (as PC1). For each endogenous metabolite we estimated the total effect on ALL and partitioned it into the direct effect of the metabolite (independent of PC1) and the mediating effect of PC1 (see Supplementary Fig. S2 for a diagram of the mediation analysis). The maximum mediating effect was 0.9% and none of the mediating effects were statistically significant. These results suggest that cytokines were not acting as mediators between the endogenous metabolites and the subsequent development of childhood ALL.

In this analysis, the largest biomarker study to investigate the relationship between neonatal cytokine profiles and the subsequent risk of childhood ALL, we found that ALL patients were born with higher levels of a group of correlated cytokines – IL1β, IL8, TNFα, and VEGF – in their blood than controls. Although the positive association between neonatal cytokine levels and childhood ALL risk was of modest magnitude, it was robust and persisted after adjustments for several potential confounders. These associations were observed among children of Latina mothers but not among children of non-Latina white mothers, and were seen most dramatically for ALL with high hyperdiploidy. We found that neonatal cytokine levels were correlated with neonatal levels of some endogenous metabolites which had been previously associated with ALL risk (36); however, there was no evidence that the cytokines were mediating the relationship between these circulating metabolites and ALL risk.

Interestingly, we observed negative correlation between the sphingolipid, sphingomyelin, and proangiogenic cytokines, IL1β, TNFα, and VEGF. Sphingolipid metabolism represents a balance between angiogenesis and cell death, with the sphingolipid ceramide (and its source sphingomyelin) tending to prevent cancer growth (46). Ceramide has been shown to have antileukemic properties (47) and sphingomyelin is known to have a role in hematologic malignancy (48), having been associated with lower risk of acute myeloid leukemia (49) and large granular lymphocyte leukemia (50), Likewise, we previously reported a lower risk of childhood ALL associated with a sphingolipid (putatively-identified as ceramide) and sphingomyelin (d16:1/20:4), an observation which we recapitulated in the current analysis (see Supplementary Table S5). Proangiogenic factors like VEGF can disrupt sphingolipid metabolism (51) and certain types of sphingolipids can prevent the metastatic outburst of proangiogenic factors like VEGF (52). Higher levels of VEGF, IL1β, and TNFα were all associated with increased risk of childhood ALL in this analysis. Although we did not find any evidence that cytokines were mediating the relationship between sphingolipids and childhood ALL risk, these relationships warrant further study.

We planned a priori to summarize individual cytokine levels with principal components, to examine the impact of missing covariate data, and to conduct stratified analysis. As a result, we have calculated many risk estimates and the modest effect sizes that were observed would not remain statistically significant after stringent corrections for multiple comparisons. Still, we believe these secondary analyses have provided meaningful insights that may spur future research. The association between cytokines and childhood ALL risk was stronger among children of Latina mothers than among children of non-Latina white mothers, a finding which resonates with our prior work on immune development and childhood ALL. We previously found that exposure to infections after birth via daycare attendance was a strong protective factor in non-Latino white families, but not among Latino families (53). While postnatal daycare exposures appear to play a larger role in ALL etiology among non-Latino white children, our current analysis suggests that altered prenatal immune development may be a stronger risk factor for Latino children. In addition, the high hyperdiploid leukemia subtype – shown here to be more sensitive to prenatal cytokines – is more common in Latinos compared to non-Latino whites in our study population (54), further pointing towards a specific etiology enriched among Latinos. One prenatal factor that affects cytokines and appears to be more common among Latinos is neonatal cytomegalovirus infection (55), which we have previously linked to childhood ALL risk (56). Future work will continue to investigate neonatal cytomegalovirus infection as a possible mechanism of altered early immune development.

Interestingly, our analysis did not recapitulate the preliminary CCLS cytokine findings we previously reported among a smaller group of 116 ALL cases and 116 controls (11). Supplementary Table S6 summarizes some of the differences between our current study and the previous Chang and colleagues analysis. Whereas Chang and colleagues reported a profound deficiency in IL10 levels at birth in children who went on to develop ALL (and, to a lesser extent, a deficiency in IL4, IL6, and IL13 as well), we found higher levels of a distinct group of neonatal cytokines in cases than controls. Results for IL4 and IL6 were mostly null in our analysis, except when stratifying by age at diagnosis where higher IL6 levels were associated with lower risk of ALL diagnosed from ages 2 to 5 years (in agreement with Chang and colleagues). Crucially, Chang and colleagues' (11) analysis was not large enough to allow for stratification by ALL subtype or by race/ethnicity, which proved to be important factors in the current analysis.

In the group of CCLS participants who were assessed for cytokine levels in both analyses (N = 211), the IL10 concentrations from Chang and colleagues (11) were not correlated with the IL10 concentrations in the current analysis (r_s = −0.09, data not shown). Because the antibodies used in the Luminex bead–based cytokine assay provided by R&D Systems are proprietary and confidential, we are unable to verify whether the antibodies used in the current study are the same ones that were used in the prior study. In addition, the Chang and colleagues analyses used polystyrene beads, whereas the current analyses used larger magnetic Luminex beads which resulted in reduced sensitivity. This reduced sensitivity – taken together with unconfirmed external validity, high between-batch variability, and concerns about potential degradation – caused us to exclude the IL10 measurements from the current statistical analyses, making it impossible for us to confirm the Chang and colleagues' (11) findings.

The current analysis included earlier birth years and used blood spots collected over a longer period of time than the prior Chang and colleagues analysis (11). Because childhood leukemia is a relatively rare disease, we prospectively enrolled a limited number of participants each year and, as a result, the neonatal blood spots used for this analysis were collected over a period of several decades, from 1984 to 2012. The blood spots were ultimately analyzed in 2019, resulting in longer storage times (median 21 years) compared to Chang and colleagues (median 8 years). Prolonged freezer storage can cause cytokine stability issues, for example, resulting in a 75% reduction in detectability rates over the course of one previous 4-year experiment (57). To mitigate this limitation, we used SCONE (34) to remove any technical variation caused by storage time (i.e., birth year) and degradation. Moreover, we also conducted stratified analyses by birth year. We found that the relationship between PC1 and childhood ALL risk was strongest among the most recent cohort of births, suggesting that degradation may have attenuated the association among samples that were stored for longer periods of time prior to analysis. Because blood-spot storage time was comparable for case and control samples, cytokine degradation would have resulted in non-differential exposure misclassification and attenuation of risk estimates.

We did not reproduce the findings of Søegaard and colleagues who reported that higher levels of IL6 and lower levels of IL8 were associated with B-cell ALL risk in a study of Danish cancer registries (see Supplementary Table S6) (12). Other neonatal inflammatory markers from the Søegaard and colleagues analysis were not assessed in the current study. The very different racial/ethnic compositions of the two study populations make it difficult to compare findings. In addition, neonatal blood spot sampling and storage conditions vary between the countries, with collection occurring later in Denmark (median of 7 days after birth).

There are important differences between the case and control groups in the CCLS; control families tended to have higher annual incomes than case families. To mitigate this limitation, we adjusted for sociodemographic characteristics (mother's race and ethnicity, household annual income, and mother's age at birth) in multivariable logistic regression; yet these factors did not confound the observed associations, suggesting that selection bias does not explain our findings.

Cytokine levels tend to be highly variable over time, as they are transiently expressed in response to infections, the circadian clock, and other external and internal stimuli – including cancer cells (58); a phenomenon which has the potential to create positive associations between elevated cytokine levels and cancer risk due to reverse causality and one which has led to challenges in replicating findings from previous epidemiologic studies. However, unlike most of the published literature, our study measured baseline cytokine levels at birth, prior to (or concomitant with) the first antigenic exposures in a child's life.

We posit that children who are born with altered cytokine levels are set on a trajectory towards modified immune system development early in life, which renders them at an increased risk for subsequent aberrant immune reactions that can initiate ALL. We did not find any evidence to suggest that endogenous metabolites – which were present at birth and previously identified as prenatal risk factors for ALL – were the cause of the observed immune phenotype. Future studies should examine other potential causes of altered cytokine levels, and also directly evaluate the potential connection between cytokine levels at birth and immune responses later in childhood among patients with ALL.

T.P. Whitehead reports grants from National Institute of Environmental Health Sciences of the NIH, United States Environmental Protection Agency, and Children With Cancer UK during the conduct of the study. L.S. McCoy reports grants from NIEHS and USEPA during the conduct of the study. No disclosures were reported by the other authors.

The California Department of Public Health is not responsible for the results or conclusions drawn by the authors of this publication. The content of this manuscript is solely the responsibility of the authors and does not necessarily represent the official views of the NIH or the Environmental Protection Agency.

T.P. Whitehead: Conceptualization, data curation, formal analysis, visualization, writing–original draft. J.L. Wiemels: Conceptualization, supervision, funding acquisition, methodology, writing–review and editing. M. Zhou: Formal analysis, methodology, writing–review and editing. A.Y. Kang: Project administration, writing–review and editing. L.S. McCoy: Project administration, writing–review and editing. R. Wang: Validation, methodology, writing–review and editing. B. Fitch: Validation, writing–review and editing. L.M. Petrick: Formal analysis, methodology, writing–review and editing. Y. Yano: Formal analysis, methodology, writing–review and editing. P. Imani: Methodology, writing–review and editing. S.M. Rappaport: Conceptualization, supervision, funding acquisition, writing–review and editing. G.V. Dahl: Resources, writing–review and editing. S.C. Kogan: Funding acquisition, validation, writing–review and editing. X. Ma: Conceptualization, resources, supervision, funding acquisition, writing–review and editing. C. Metayer: Conceptualization, resources, supervision, funding acquisition, writing–review and editing.

We thank the families for their participation. We also thank our clinical collaborators for their help in recruiting patients. The biospecimens used in this study were obtained from the California Biobank Program, (SIS request number 26), in accordance with Section 6555(b), 17 CCR.

Research reported in this publication was supported by the National Institute of Environmental Health Sciences of the NIH under Award Numbers P01ES018172 (PI: C. Metayer), P50ES018172 (PI: C. Metayer), R01ES009137 (PI: C. Metayer), R01CA185058 (PI: C. Metayer), P42ES0470518 (PI: M.T. Smith), P42ES004705 (PI: M.T. Smith), and R24ES028524 (PI: C. Metayer); by the United States Environmental Protection Agency under assistance agreement RD83451101 (PI: C. Metayer) and RD83615901 (PI: C. Metayer) to the University of California, Berkeley; and by Children With Cancer UK, a United Kingdom–based charity dedicated to raising money for research and providing care for children with cancer and their families.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.