The study participants were recruited from the Cincinnati Lead Study (CLS), a longitudinal birth cohort study designed to evaluate the effects of low to moderate environmental lead exposure. The CLS enrolled pregnant women in their first or early second trimester of pregnancy who visited four prenatal clinics within impoverished Cincinnati neighborhoods with a high concentration of older, lead-contaminated housing. Women were excluded if they were known to be addicted to drugs, were diabetic, or had any known neurologic or psychiatric malady. Newborns with birth dates between 1980 and 1985 were subsequently recruited for follow-up studies. Infants were excluded if their birth weight was < 1,500 g or if genetic or other serious medical issues were present at birth. This cohort has been described periodically as cognitive, and behavioral and social assessments were completed (^{Dietrich et al. 1985a}, ^1985b, ^1987a, ^1987b, ¹⁹⁹⁰, ¹⁹⁹¹, ^1993a, ^1993b, ²⁰⁰¹). Newborns (n = 305) were followed up quarterly through age 5 years, semiannually from 5 to 6.5 years of age, again at 10 years, and also between the ages of 15 and 17 years. CLS subjects between 19 and 24 years of age (n = 194) were consecutively recruited for this MRS study via mailings, phone calls, and flyers to meet our initial sample size goal of 150 CLS participants. This size was chosen before initiating the imaging aspect of the study, because members of the CLS cohort who were pregnant, incarcerated, or claustrophobic or had metal implanted from surgeries, embedded metal from occupational exposure (e.g., welder), or gunshot wounds would not be able have a research MRI examination. Subjects who were known to be diabetic or who had any known neurologic or psychiatric illness were also excluded. The substance use of participants was determined using a clinical urine toxicology screening immunoassay for drugs of abuse, which evaluated amphetamines, barbiturates, benzodiazepine, cocaine, cannabinoids, opiates, and phencyclidine (PCP). (The only positive findings for participants were exclusively cannabinoids.) The alcohol use within the cohort was also assessed via a self-report measure previously described (^{Dietrich et al. 2001}). Thirty-three subjects were excluded from the MRI examination because of the following: too large to fit into the MRI scanner (n = 8), failure to appear for the appointment (n = 7), pregnancy (n = 6), refusal due to claustrophobia (n = 6), nonremovable metal in their body (n = 5), and inability to give informed consent because of cognitive disability (n = 1). Two participants diagnosed with fetal alcohol syndrome as children participated in the imaging and spectroscopy session, but their data are not included in the analyses. Consistent with previous research in this cohort (e.g., ^{Wright et al. 2008}), we found no demographic or lead exposure biases associated with exclusion of these members from the analysis.

Enrollment of the participants’ mothers in the CLS permitted monitoring of blood lead levels for the participants prenatally, at birth, every 3 months to 60 months of age, and every 6 months from 60 to 78 months of age. The mean of 23 childhood blood lead assessments for each participant from 3 to 78 months of age was employed for comparison analyses. This mean childhood blood lead value has been historically used for comparisons with cognitive and behavioral outcomes (^{Dietrich et al. 1991}, ^1993b). During the first 5 years of life, at least one of the quarterly blood lead assessments exceeded 10 μg/dL for 99% of the CLS cohort. The peak lead exposure occurred between 2 and 3 years of age for the participants. By adolescence, the mean blood lead levels for the cohort were 2.8 ± 1.3 μg/dL.

All MRI and MRS investigations were acquired on a 1.5 Tesla MR scanner (Signa LX; General Electric Medical Systems, Waukesha, WI, USA) with a quadrature head radiofrequency coil positioned around the head of the participant. An axial three-dimensional, inversion recovery prepped, fast spoiled gradient-echo high-resolution imaging sequence was acquired [echo time (TE) 5 msec, repetition time (TR) 12 msec, inversion time 300 msec; field of view = 24 cm × 19.2 cm, 1.5-mm thick with contiguous slices, 256 × 192 × 124 matrix for a resolution of 0.94 mm × 1 mm × 1.5 mm]. Single voxel MRS was acquired with the proton brain examination (PROBE) software (General Electric Medical Systems) using a point-resolved spectroscopy acquisition mode (PRESS) sequence (TE = 35 msec, TR = 2 sec with 64 averages) and with automatic shimming to achieve water line widths of < 5 Hz. A reference spectrum, acquired within the PROBE acquisition, collected 16 acquisitions of unsuppressed water signal for eddy current and baseline corrections. Data were processed using the LCModel program, a commercially available, user-independent, frequency domain spectral-fitting program that provides measures of metabolite concentrations (^{Provencher 1993}). Concentration levels of NAA, Cr, Cho, mI, and GLX were determined. The concentrations are reported in institutional units (IU), rather than millimolar levels; individual corrections for magnetic field and coil variations, as well as longitudinal (T1) and transverse (T2) relaxation effects, are not accounted for because of the excessive examination times required to determine such effects. However, metabolite concentrations were adjusted for the amount of cerebrospinal fluid (CSF) in each voxel. Specifically, the LCModel software determines the metabolite concentrations assuming a parenchymal volume as defined in the MRS acquisition, typically 2 cm per voxel side for a total cubic volume of 8 cm³. We employed a K-means segmentation algorithm using custom software for analysis of the high resolution imaging sequence so as to determine gray matter, white matter, and CSF contributions to each voxel. We then applied a correction factor based on the ratio of volume of CSF to total parenchymal volume and adjusted each metabolite concentration accordingly. LCModel also provides a measure of data quality and reliability, reporting the Cramer–Rao lower-bound bounds (^{Provencher 1993}). We required the metabolite Cramer-Rao value to be < 15% for inclusion in data analyses. For a given participant, regional spectroscopy data not meeting the Cramer–Rao lower-bound criteria were excluded from analyses. Approximately 98% of the > 1,100 sampled voxels of the cohort met criteria and were included in the analyses.

Seven single voxels, approximately 8 cc in volume, were positioned within the medial frontal gray matter, the left frontal white matter, left parietal white matter, the right temporal lobe at the superior temporal gyrus, the left basal ganglia, left cerebellar hemisphere (gray and white matter), and the cerebellar vermis, respectively (Figure 1). Although we anticipated no laterality differences, the left side of the brain was sampled for several reasons: a) neural networks associated with language function are left lateralized; b) most of the study population is right handed and thus likely to be left hemisphere dominant; c) the right hemisphere is prone to technical factors (orientation of slice-selective pulses) that produce signal artifacts (i.e., scalp lipid signal contamination) within the spectrum.

We investigated the relationship between metabolite levels obtained from MRS analyses in seven brain regions with mean childhood blood lead level initially using Pearson and Spearman correlation coefficients. Correlation values with an alpha (p < 0.05) were considered significant. The metabolites examined included NAA, Cr, Cho, mI, and GLX. Recognizing that other variables could influence the association between metabolite levels and blood lead, the impact of covariates on this relationship were also examined by analyzing the bivariate relationship of the potential covariate with mean childhood blood lead level in each of the sampled brain regions, using a linear regression. Such variables that could potentially contribute included age of participant at time of imaging, Full-Scale intelligence quotient (FSIQ), current drug use (positive only for cannabinoids), sex, birth weight, and gestational age at birth. Covariates that were independently related to the outcome in one or more region at p < 0.10 level were included in the subsequent multiple linear regression analyses. For the linear regression analyses, beta values corresponded to the change in metabolite concentration associated with a 10-μg/dL increase in the mean childhood blood lead level and the 95% confidence intervals (CIs) determined for significance.

The lack of correlative findings within the temporal lobe may arise from technical factors associated with the proton MRS acquisition. Peak broadening due to magnetic susceptibility and field inhomogeneity arises at the interface between the temporal lobe and adjacent structures, including the ear and the temporal bone, resulting in incomplete water suppression, broad line widths of the metabolites, and failure of metabolite quantification. A dedicated MRS examination optimizing the homogeneity of the temporal lobe is necessary to clarify the impact of lead exposure within the temporal lobe. A dedicated study of the hippocampus is also warranted from other reports in the literature providing experimental evidence.