Zhang et al. 2021 — In utero metal/trace element exposure and childhood blood pressure (Boston Birth Cohort)

Zhang and colleagues examined the individual and joint associations of in utero exposure to two heavy metals (Pb, Hg, Cd) and two essential trace elements (Se, Mn) — measured in maternal red blood cells (RBCs) collected 24–72 h after delivery as a proxy for third-trimester exposure — with child blood pressure percentile between ages 3 and 15 in a predominantly low-income, urban, racial/ethnic minority U.S. cohort (60.5% non-Hispanic Black, 19.9% Hispanic, 4.8% non-Hispanic White, 14.8% other). The headline finding is an inverse association between maternal RBC selenium and manganese and child systolic blood pressure (SBP) percentile (−6.23 and −2.62 percentile points per doubling of exposure, respectively, in adjusted linear regression); Pb, Hg, and Cd were not individually associated with child SBP. Bayesian kernel machine regression revealed a Mn–Cd interaction whereby the protective Mn–SBP association was stronger at higher Cd levels (p-interaction = 0.04), and maternal cigarette smoking during pregnancy (a major Cd exposure source) further modified the Mn–SBP relationship, with a doubling of maternal Mn associated with a 10.09-point reduction in child SBP percentile among smokers (95% CI: −18.03, −2.15) compared with a 1.49-point reduction in non-smokers (95% CI: −4.21, 1.24; p-interaction = 0.08). No metals or trace elements were associated with child diastolic blood pressure.

Key numbers

Maternal RBC concentrations and detection profile (Table 1, n = 1,194; values imputed at LOD/√2 for the small subset below LOD):

Adjusted linear-regression association of a doubling of each metal/trace element with child SBP percentile (Table 3; adjusted for maternal age at delivery, race/ethnicity, educational level, prepregnancy BMI, and maternal cigarette smoking history):

Quartile-based estimates (Table 3, Q4 vs Q1 adjusted): Se Q4 (318.0–624.0 µg/L) was −3.23 SBP percentile points (95% CI: −7.34, 0.89) compared with Q1 (152.4–248.0); Mn Q4 (48.2–109.8 µg/L) was −3.86 points (−7.95, 0.24) compared with Q1 (<LOD–28.8). Quartile estimates for Pb, Hg, and Cd showed no monotonic trend. BKMR exposure–response curves (Figure 2) were generally consistent with the linear regressions.

Pairwise interactions (BKMR and linear regression, Figure 3, Table S5): Mn × Cd on child SBP, p-interaction = 0.04. At low Mn levels there was a positive Cd–SBP association; the inverse Mn–SBP association was stronger at higher Cd. A doubling of both Mn and Cd was associated with +7.02 SBP percentile points (95% CI: −2.27, 16.31) versus +13.52 (95% CI: 1.01, 26.03) for a doubling of Cd alone. No other pairwise interactions reached the prespecified threshold (p < 0.10).

Effect modification by maternal cigarette smoking during pregnancy (Table 4; smoked n = 108 vs did not smoke n = 1,086):

(Note: the Se smokers row in Table 4 appears as +2.98 [−3.05, 9.00], identical to the Cd-smokers cell; the in-text figure for the Mn–smokers BKMR estimate is −10.09 [−18.03, −2.15], somewhat tighter than the linear-regression −14.28 cell.) Median maternal Cd was 1.41 µg/L (IQR 0.93–2.37) in mothers who smoked versus 0.65 µg/L (IQR 0.43–0.96) in mothers who did not (p < 0.001), making smoking a likely upstream driver of both Cd exposure and the Mn × Cd interaction.

Mixture-as-a-whole effects: BKMR with hierarchical variable selection assigned a group posterior inclusion probability (PIP) of 0.852 to the trace-elements group (Se, Mn) versus 0.042 to the heavy-metals group (Pb, Hg, Cd) for child SBP. Within groups, Pb had the largest conditional PIP among heavy metals and Se the largest among trace elements. The overall mixture was not associated with child SBP (Figure S7) or DBP (Figure S8). No metal or trace element was clearly associated with child DBP percentile (Table S4, Figure S5); the only possible exception was an inverse Se–DBP estimate (β = −2.75 per doubling; 95% CI: −6.90, 1.40).

Subgroup heterogeneity (Table S8): the inverse Mn–SBP association was strongest at ages 3–5 y (β = −6.29; 95% CI: −11.52, −1.06) and null at ages 10–15 y (p-interaction by age group = 0.01). The Hg–SBP and Hg–DBP associations differed by maternal race/ethnicity (p-interaction = 0.05 for SBP, 0.001 for DBP), with inverse estimates among children of non-Hispanic Black mothers and positive estimates among children of Hispanic mothers. The Mn–SBP association differed between children of Hispanic mothers (β = −4.11; 95% CI: −8.98, 0.75) and non-Hispanic Black mothers (β = 0.81; 95% CI: −1.83, 3.45; p-interaction = 0.05).

Methods (brief)

Cohort. Boston Birth Cohort, an ongoing prospective birth cohort at the Boston Medical Center recruiting mother–infant dyads 24–72 h after delivery of a singleton live infant without major birth defects. 1,272 mothers had blood metals measured; 1,194 were retained after excluding 78 dyads missing covariates. Compared with the excluded subset (n = 307), the analytic sample was more likely to be non-Hispanic Black, less likely to be Hispanic or to have smoked, and had lower preterm and higher child BMI; RBC metal/trace-element distributions did not differ.

Exposure assessment. Maternal whole-blood samples collected 24–72 h after delivery were centrifuged, and red blood cells were stored at −80 °C in metal-free vials. RBC aliquots (0.5 mL) were transported on dry ice to the Public Health and Environmental Laboratories in Trenton, New Jersey, and measured by inductively coupled plasma–mass spectrometry (ICP-MS). All five elements were measured in the same run. Intra-assay coefficients of variation were ≤ 5.0%. Hg is reported as total Hg in RBCs (no MeHg/iHg speciation); the authors note that ~90% of RBC Hg is methylmercury in this U.S. coastal population. Maternal RBC concentrations are described as a proxy for third-trimester in utero exposure given the long red-cell lifespan (~120 d) and the biological half-lives of Pb, Hg, Cd in blood (2–4 months) and Mn (~40 d). Values below LOD were imputed as LOD/√2.

Outcome. Child blood pressure measured at the most recent annual well-child visit between ages 3 and 15 y by clinical staff using an automatic sphygmomanometer (Masimo SET) on the right brachial artery in the seated position. Child age-, sex-, and height-specific SBP and DBP percentiles were calculated using the 2017 American Academy of Pediatrics Clinical Practice Guideline (Flynn et al. 2017), which updated the 2004 Pediatric Hypertension Guideline. SBP was the primary outcome; DBP secondary. SBP z-scores and SBP ≥ 90th percentile (logistic regression) were used in sensitivity analyses.

Covariates. Multivariable models adjusted for maternal age at delivery (continuous), self-reported race/ethnicity (non-Hispanic Black, non-Hispanic White, Hispanic, others), educational level (middle school or below; high school graduate/some college; college graduate or above), prepregnancy BMI (continuous), and maternal cigarette smoking history (never smoked; quit smoking before pregnancy; smoked during pregnancy). Sensitivity analyses additionally adjusted for maternal fish intake, preeclampsia, diabetes mellitus, and child BMI; stabilized inverse-probability weights were applied to account for selection bias.

Statistical analysis. Two complementary approaches. (1) Linear regression with each metal/trace element log₂-transformed (β per doubling) and as quartiles (Q1 reference). Product terms tested pairwise metal × metal interactions on child SBP percentile (p < 0.10 threshold for effect modification, given the exploratory aim). (2) Bayesian kernel machine regression (BKMR) implemented in R package bkmr v0.2.0 (default parameters in the kmbayes function), with four parallel MCMC chains of 100,000 iterations each (50,000 burn-in) fitted via the kmbayes_parallel function in R package bkmrhat v1.0.2. Hierarchical variable selection grouped the five elements into heavy metals (Pb, Cd, Hg) versus trace elements (Mn, Se) to compute group and conditional posterior inclusion probabilities. Convergence assessed by trace plots, autocorrelation plots, density plots, and Gelman–Rubin statistics. Software: Stata 15.1 and R 3.6.3.

Strengths and limitations (per the authors). Strengths: prospective birth-cohort design, large (n = 1,194) sample, racially/ethnically diverse low-income population underrepresented in environmental health research, BKMR allowing nonlinear and interactive mixture effects, RBC measurement capturing transplacental transfer rather than the diluted plasma signal. Limitations: RBC collection within 24–72 h post-partum is a proxy for late pregnancy only and cannot resolve cumulative or trimester-specific exposure; only one BP measurement per child was used (the most recent visit), introducing measurement error; child metal levels in early life were not measured, so the in utero versus early-life mediator question cannot be addressed; total Hg in RBCs aggregates organic and inorganic forms whose BP effects may oppose each other.

Implications

Certification. This is a maternal-biomonitoring exposure study; it does not contribute occurrence data for any HMT&C product category. Its relevance to HMT&C threshold-setting is indirect: it argues that the toxicological frame for prenatal Pb, Hg, and Cd exposure should not be restricted to neurodevelopmental endpoints, but should also consider cardiovascular outcomes in childhood — a domain currently underrepresented in the HMT&C evidence base. The Mn–Cd interaction and the smoking-modified Mn–SBP association suggest that essential trace elements (which HMT&C does not currently regulate) are co-determinants of the toxic-metal health response in pregnancy, complicating any single-analyte interpretation of cord-blood or maternal-blood biomonitoring data.

Courses. Useful for the perinatal-exposure module as a case study in (a) mixture modeling using BKMR vs single-analyte regression, (b) the difference between essential trace elements (Se, Mn) and toxic heavy metals in the context of pregnancy outcomes, and (c) effect modification by maternal smoking when one of the modifier’s mechanisms is upstream metal-exposure co-occurrence (cigarette smoke as a Cd source).

App. Not applicable to ingredient-level contamination profiling (exposure measured in maternal RBC, not in foods).

Microbiome. Not addressed.

Wiki pages updated on ingest

• lead
• mercury
• cadmium
• manganese
• cardiovascular-effects
• vulnerable-populations

Verification notes

2026-05-18 merge-enhance pass (Claude, /ingest-next-manual-fetch-pdf v2):

• Title corrected. Previous title read “…in a Boston Birth Cohort” — the actual published title is “…in a U.S. Urban, Low-Income, Minority Birth Cohort” (Environ Health Perspect 129(6):067005, June 2021, p. 067005-1). The Boston Birth Cohort is the data source, not the title phrasing.
• Authors corrected. Previous page listed “Zhang C, Chen A, Chen J, Chen Z, Peddada SD, Hu H, Kumar A, Wills-Karp M, Hirschhorn JN, Wang G, Pearson C, Zuckerman B, Wang X” — this is the wrong author list entirely (no first author “Zhang C” on this paper; “Hirschhorn JN,” “Peddada SD,” “Pearson C,” and “Zuckerman B” are not authors of this paper). Actual authors per the title page: Mingyu Zhang, Tiange Liu, Guoying Wang, Jessie P. Buckley, Eliseo Guallar, Xiumei Hong, Mei-Cheng Wang, Marsha Wills-Karp, Xiaobin Wang, Noel T. Mueller. The previous author list appears to have been confabulated.
• raw_handle corrected from the legacy placeholder papers-cube to the per-skill PCMF handle PCMF_in-utero-exposure-to-heavy-metals-and-trace-elemen (truncated stem of the PDF filename per the PCMF convention exemplified by PCMF_ijshr12 in dokubo2023-nigeria-seasoning-powder.md).
• raw_path corrected. Previous path pointed to a file (...Childhood Blood Pressure and Hypertension.pdf) that does not exist in the PCMF folder. Actual file is ...Childhood Blood Pressure in a U.S. Urban, Low-Income, Minority Birth Cohort.pdf.
• raw_sha256, access_url, license added to frontmatter for provenance.
• metals: updated. Previous [Pb, Hg, Cd, Se, Mn] changed to [Pb, tHg, Cd, Se, Mn] per Part 14 speciation discipline: the paper measures total Hg in RBCs without distinguishing MeHg from iHg (“the proportion of RBC inorganic Hg is expected to be very low… ~90% of MeHg is found in the RBCs” — Discussion, p. 067005-9), so tHg is the correct abbreviation. Se and Mn are essential trace elements but are recorded in metals: because that field captures the full analyte list of the study, and the wiki has no separate trace_elements: field; Se and Mn page slugs exist in the taxonomy (metals/selenium, metals/manganese).
• matrices: expanded from [blood, maternal-rbc] to [blood, maternal-rbc, gestational-exposure] to mark the in utero framing — the exposure is a proxy for third-trimester transplacental transfer, not adult biomonitoring. gestational-exposure matches the matrix vocabulary already used in ashley-martin2022-ias-biomonitoring-pregnancy.md and other prenatal biomarker pages.
• sample_population rewritten with the correct enrollment dates (Nov 2002 – Oct 2013), follow-up window (until May 2018), race/ethnicity composition (61% non-Hispanic Black, 20% Hispanic, 5% non-Hispanic White, 15% other), and child age range (3–15 y; median 8.4, IQR 6.2–10.6). Previous text understated the cohort recruitment period and gave child age range as “3–14” instead of the actual 3.0–15.4.
• Key numbers rewritten in tabular form. Previous body table reported single mean values only and gave qualitative summaries (“Not significant”) rather than effect sizes and confidence intervals. New tables capture (a) full Table 1 distribution + LOD detection profile, (b) full Table 3 adjusted β per doubling with 95% CIs, (c) Table 4 smoking-stratified estimates, (d) Mn × Cd interaction values from Table S5 and Figure 3 — none of which were on the previous page.
• Mn–Cd interaction added (p-interaction = 0.04; BKMR Figure 3, panels 3.5 and 5.3). This is one of the two headline findings of the paper and was entirely absent from the previous page.
• Smoking-modified Mn–SBP added (β = −10.09 per Mn doubling among smokers, 95% CI: −18.03, −2.15; Figure 4, Table 4). This is the second headline finding and was also absent.
• Methods section fully rewritten. Previous methods text contained multiple invented or wrong details: “NIEHS laboratory” (actual: Public Health and Environmental Laboratories, Trenton, NJ); “Dinamap Pro” sphygmomanometer (actual: Masimo SET); “linear mixed-effects models with age-SBP trajectories as outcomes” with “Holm correction” (actual: cross-sectional linear regression on a single most-recent BP visit, plus BKMR; no Holm correction is mentioned); “Covariates included maternal age, education, race/ethnicity, smoking, BMI, gestational age, parity, and child sex” (actual covariates: maternal age, race/ethnicity, education, prepregnancy BMI, smoking history — no parity, no gestational age, no child sex in the main model); “median of 2 BP measurements (range 1-5)” (actual: one BP percentile per child, from the most recent well-child visit). The previous methods read like fabricated methodology rather than the paper’s actual design.
• “Childhood mean SBP was ~110 mmHg; DBP ~62 mmHg” removed — these specific raw-mmHg values are not reported in the paper (the paper reports percentiles, not raw mm Hg means; 188 of 1,194 children, or 15.7%, had elevated SBP defined as ≥ 90th percentile, per the Results “Descriptive Results” paragraph on p. 067005-4).
• Implications softened to remove synthesis-style claims. The previous Certification line (“the null result for in-utero Pb… does not exonerate Pb; the Boston Birth Cohort has lower Pb exposures than many international populations…”) was a cross-source comparison that belongs in synthesis pages or metal-page summaries, not in a single-source page (Part 2 firewall). Replaced with a single-paper statement about what this study does and does not contribute to certification work.
• ## Wiki pages updated on ingest updated to add the three metal pages the previous list omitted (mercury, manganese — the previous list named only lead and cadmium despite the paper covering all five elements).
• Brand firewall: no brand names anywhere in this paper or in the wiki page; it is a biomonitoring study. No Part 12 risk.
• Part 2 wiki/HMTc firewall: no HMTc threshold proposals; no consumer advisories; no cross-source synthesis claims in the page body.

2026-05-18 audit-subagent pass (general-purpose Agent, fresh context):

• ❌ Subagent flagged the BKMR Methods sentence as having two transcription errors: package kmbayes v1.0.2 → actual bkmrhat v1.0.2, and function bkmbayes_parallel → actual kmbayes_parallel. Verified independently against PDF p. 067005-3: “kmbayes_parallel function in R package bkmrhat (version 1.0.2).” Both corrections applied. The page also previously omitted that the default-parameter kmbayes function in the bkmr package is what was used for the BKMR fits themselves; clarified accordingly.
• ⚠️ Subagent flagged the race/ethnicity composition (“61% / 20% / 5% / 15%”) as rounded versus the unrounded Table 2 values (60.5% / 19.9% / 4.8% / 14.8%). The Abstract uses the rounded form; Table 2 has the exact values. Replaced both the sample_population field and the opening narrative sentence with the unrounded values.
• ⚠️ Subagent flagged that the matrices snapshot does not enumerate maternal-rbc and gestational-exposure, requesting confirmation. Verified independently: maternal-rbc is in use on okubo2023-periconceptional-diet-blood-metals-lbw.md, and gestational-exposure is in use on ashley-martin2022-ias-biomonitoring-pregnancy.md and weyde2023-gestational-metals-cerebral-palsy.md. Both are accepted in the live wiki vocabulary; the taxonomy snapshot simply does not enumerate the matrices controlled list. False positive against the page; flagging as a snapshot-completeness issue rather than a wiki-page defect.
• Subagent confirmed the Se/Cd-smokers Table 4 duplicate (+2.98 [−3.05, 9.00] appearing in both rows) is what the published paper actually prints; the wiki page correctly flags this as a probable published-paper typo rather than a wiki transcription error. Preserved as-is with the parenthetical caveat.
• Subagent’s other Check 1 findings (all 25 Table 1 distributional values, all 10 Table 3 effect sizes + 5 quartile-range cells, all 10 Table 4 cells, the Mn × Cd interaction values, the smoking-modified Mn-SBP BKMR + linear-regression numbers, all subgroup heterogeneity values, BKMR group PIPs, sample-composition and age-range values) were verified clean against the PDF.
• Subagent verdict: REVISE. 1 ❌ applied (BKMR package/function names), 1 ⚠️ applied (race/ethnicity precision), 1 ⚠️ rejected (matrices vocab — false positive, slugs are in use elsewhere). All other findings (slug vocabulary in metals:/jurisdictions:/wikilinks, speciation tHg, Methods fidelity, Check 4 brand firewall, Check 5 Part 2 firewall) clean.

Page history

The five most recent substantive edits to this page. The full version history lives in git; when DOI minting comes online (see schema docs), each entry below will also link to a version-pinned DataCite DOI.