Text S1 – Supplementary Information

Participants. The Cincinnati Lead Study (CLS), a birth cohort recruited from 1979 to 1984, enrolled women who lived in neighborhoods with historically high rates of childhood lead poisoning. Women were excluded if they were known to be addicted to drugs, diabetic, or had any known neurological or psychiatric malady. Infants were excluded if their birth weight was less than 1,500 g or if genetic or other serious medical issues were present at birth. This process netted 305 newborns who were followed-up quarterly through age 5 years, semi-annually from age 5 to age 6.5 years, at age 10, and between the ages of 15-17 (N=195). Participants, who were between the ages of 19 and 24, were recruited for this study (N=194). Thirty-three participants were excluded from the MRI examination due to the following: too large to fit into the MRI scanner (N=8), failure to appear for the appointment (N=7), pregnant (N=6), refused due to claustrophobia (N=6), non-removable metal in their body (N=5), and unable to give informed consent due to cognitive disability (N=1). Two participants diagnosed with fetal alcohol syndrome as children were imaged, but not included in the analyses. Two participants imaged were excluded from the analysis due to poor data quality, likely resulting from excessive head motion during the scan. Consistent with previous research in this cohort, we found no biases associated with exclusion of these members from the VBM structural brain analysis. For the remaining N=157 members of the CLS with data analyzed, the demographic features are displayed in Table 1 of the manuscript with no significant differences are found in the demographic factors.

Blood lead concentrations were measured in this cohort every three months for the first five years of life and every six months from 5 - 6.5 years with a mean value determined for childhood. The mean childhood blood lead concentration was used for this study as this best represents individual differences and cumulative exposure during childhood.

Imaging Acquisition. All MRI and MRS investigations were acquired on a 1.5 T Signa General Electric MR scanner (Milwaukee, WI). During the study, participants reclined in a supine position on the scanner bed and a quadrature RF coil was positioned around the participant’s head. Earplugs or headphones were provided for protection from the noise of the scanner. An axial three-dimensional, inversion recovery prepped, fast spoiled gradient echo (3D IR FSPGR) was acquired (TE 5 msec, TR 12 msec, inversion time (TI) 300 msec, field of view (FOV) = 24 cm x 19.2 cm, 1.5-mm thick with contiguous slices, matrix 256 x 192 x 124 for a resolution of 0.94 mm * 1 mm * 1.5 mm)) for VBM analyses. Precautions were taken to minimize participant motion during the MRI examination by instructing participants to remain still, packing foam padding around their heads, and securing their head and body with soft restraints.

Voxel Based Morphometry. Data Pre-processing. All automated image processing was performed using Statistical Parameter Mapping software (SPM2, Wellcome Department of Cognitive Neurology, University College London, United Kingdom) running in MATLAB version 7.1 (MathWorks, Natick, Massachusetts). A single investigator (IE) determined the anterior and posterior commissures for each imaging dataset. Images were uniformly aligned with respect to head position to provide optimal starting estimates for subsequent spatial normalization.

Analysis followed the “optimized” voxel-based morphometry strategy of Good and colleagues. Images were normalized to a standard tissue probability template and subsequently segmented in normalized space. This procedure weights normalization parameters heavily towards a given tissue and minimizes contributions from other tissue types. Any remaining extra-neuronal tissue was removed by modulation with an extracted, individualized brain mask. These normalized images and partitions were used to create study-specific templates. The original images were then segmented in native space using the study-specific templates. Any remaining extra-neuronal tissue was removed by modulation with a newly-extracted, individual brain mask. These native-space tissue partitions were then normalized to the corresponding study-specific tissue-probability maps and modulated to preserve the total amount of tissue in the images.

As part of the segmentation procedure, residual image inhomogeneities were removed by modeling smoothly varying intensity changes. [1] This procedure involves the estimation of an intensity nonuniformity field, which is then applied to yield a corrected image. This procedure has been demonstrated to markedly increase the reproducibility of SPM2 segmentation results. [2]

Spatial normalization was achieved using an initial 12-parameter affine transformation, followed by 12 nonlinear iterations using 7 x 8 x 7 discrete cosine transform basis functions. [1,3] Images were written out in 2 x 2 x 2 mm resolution. To allow for the detection of volume changes, images were modulated by the Jacobian determinant of the normalization matrix, resulting in changes that take into account global and local volume changes during spatial normalization. [4] Final images were smoothed using a Gaussian kernel with a full-width half-maximum (FWHM) of 12 mm to create a local weighted image of the surrounding pixels. This filter width, as per the matched filter theorem, determines the spatial scale at which changes are most sensitively detected, and also accounts for structural variability and possibly inexact spatial normalization.

Voxel Based Morphometry. Image Analysis and Statistics

Processed images (gray, white and CSF, respectively) from the composite of individual datasets were analyzed within SPM2, employing the framework of the general linear model (GLM)[5]. A simple linear regression model was designed where the mean childhood blood lead concentration was considered the parameter of interest. Two contrasts were calculated, testing for a positive or negative correlation of volume with the parameter of interest. Significance was set at a conservative unadjusted value of p0.001, with a minimum cluster size of 700 voxels. [6,7] This model provided several regions of interest (ROI) in which the mean childhood blood lead exposure was associated with a loss of gray matter volume. No significant changes in white matter or CSF volumes were found.

Other variables considered that could potentially contribute to a change in gray matter and brain volume included age, current marijuana use (obtained from a urine drug screening collected at time of imaging), sex, birth weight, gestational age at birth, maternal IQ, maternal alcohol consumption during pregnancy, maternal marijuana use during pregnancy, maternal tobacco use during pregnancy, mean childhood Hollingshead socioeconomic status (SES) score, current SES score, and home environment (mean Home Observation for Measurement of the Environment (HOME) score measured in early childhood). Upon adding a putative confounding variable into the otherwise simple linear regression between gray matter volume and mean childhood blood lead concentration, the change of regression coefficient (beta) for lead exposure was calculated in a region specific manner to evaluate the influence of the variable. The MarsBar ROI toolbox for SPM2 was used to extract beta values from the ROIs on a voxel-by-voxel basis. [8] This testing was conducted within the gray matter regions significantly correlated with the mean childhood blood lead concentration as described above in the simple linear regression model. The putative confounding variable was kept for the subsequent final multivariate analysis if adding the variable caused 20% of the pixels within the predetermined ROI to have 10% change in the beta value. Two variables, birth weight and age at time of scanning, met criteria for inclusion in the final model. Contrast and significance testing are identical (unadjusted p0.001 and 700 voxel minimum cluster size) to that described above in the simple linear regression model. The center of mass for points of maximum correlation were converted from Montreal Neurological Institute coordinates to Talairach coordinates using a nonlinear transformation. Voxels of maximal correlation within the XYZ were plotted across individual scans using SPM2 and rendered on a template derived from a standard atlas. (LONI).[9]

References

1. Ashburner J, Friston KJ (2000) Voxel-based morphometry--the methods. Neuroimage 11: 805-821.

2. Chard DT, Parker GJ, Griffin CM, Thompson AJ, Miller DH (2002) The reproducibility and sensitivity of brain tissue volume measurements derived from an SPM-based segmentation methodology. J Magn Reson Imaging 15: 259-267.

3. Ashburner J, Friston KJ (1999) Nonlinear spatial normalization using basis functions. Hum Brain Mapp 7: 254-266.

4. Good CD, Johnsrude IS, Ashburner J, Henson RN, Friston KJ, et al. (2001) A voxel-based morphometric study of ageing in 465 normal adult human brains. Neuroimage 14: 21-36.

5. Friston KJ, Holmes AP, Worsley KJ, Poline JP, Frith CD, et al. (1995) Statistical parametric maps in functional imaging: A general linear approach. Human Brain Mapping 2: 189-210.

6. Wilke M, Kaufmann C, Grabner A, Putz B, Wetter TC, et al. (2001) Gray matter-changes and correlates of disease severity in schizophrenia: a statistical parametric mapping study. Neuroimage 13: 814-824.

7. Wilke M, Kowatch RA, DelBello MP, Mills NP, Holland SK (2004) Voxel-based morphometry in adolescents with bipolar disorder: first results. Psychiatry Res 131: 57-69.

8. Brett M, Anton J-L, Valabregue R, Poline JP (2002) Region of interest analysis using an SPM toolbox.

9. Toga AW, Thompson PM (2005) Brain atlases of normal and diseased populations. Int Rev Neurobiol 66: 1-54.

6