- Email: [email protected]
Functional Magnetic Resonance Imaging and Working Memory in Adolescents with Gestational Cocaine Exposure
Functional Magnetic Resonance Imaging and Working Memory in Adolescents with Gestational Cocaine Exposure HALLAM HURT, MD, JOAN M. GIANNETTA, MARC KORCZYKOWSKI, MA, ANGELA HOANG, KATHY Z. TANG, LAURA BETANCOURT, PHD, NANCY L. BRODSKY, PHD, DAVID M. SHERA, SCD, MARTHA J. FARAH, PHD, AND JOHN A. DETRE, MD
Objective To assess the effect of gestational cocaine exposure on the prefrontal cortex (PFC) with functional magnetic resonance imaging (fMRI). Study design Using an n-back task, we obtained fMRI with a 3T Siemens scanner on 49 adolescents, 25 who were exposed to cocaine and 24 who were not exposed. The primary outcome was PFC activation during task performance. Five functionally derived regions of interest (ROI) were defined; in addition, 2 a priori anatomical ROIs were generated for Brodmann regions 10 and 46. Results Of the 49 adolescents who underwent imaging, data from 17 who were exposed to cocaine and 17 who were not exposed were in the final analysis. Groups had similar performance on the n-back task (P > .4), with both showing a fewer number of correct responses on the 2-back than the 1-back (P < .001), indicating increased demands on working memory with greater task difficulty. In functionally derived ROIs, imaging results showed increased activation for both groups in the 2-back versus the 1-back condition. In anatomical ROIs, both groups showed greater activation in the 2-back versus the 1-back condition, with activation in the non-exposed group proportionally greater for the left prefrontal region (P ⴝ .05). Conclusion In this sample of adolescents, participants who were exposed to cocaine and participants who were not exposed were similar in performance on an executive function task and in fMRI activation patterns during task performance. (J Pediatr 2008;152:371-7) he growing pre-clinical and clinical literature suggest that gestational cocaine exposure may affect developing monaminergic (dopamine, serotonin, and noradrenaline) systems that are widely distributed throughout the brain.1,2 These findings, in turn, have given rise to concern about injury to the prefrontal cortex in exposed children, with resultant effects on executive functioning, a diverse set of skills such as planning, problem solving, cognitive control, working memory, and reward processing.3 In this regard, investigations assessing neurobehavioral outcomes and executive function in children with gestational cocaine exposure have found effects ranging from none, to minimal, to impaired recognition memory, to altered task persistence, to self-regulatory deficits.4-13 There have been several efforts to anchor such clinical findings with neuroanatomical bases;14-16 however, no consistent patterns have been identified. We investigate the neural bases of injury through examination of patterns of brain activation by using functional magnetic resonance imaging (fMRI) during a task of executive function—in this case, non-spatial working memory as assessed by using an n-back task.17 Although our earlier behavioral assessments have failed to find an effect of cocaine exposure with both spatial and non-spatial working memory tasks,18 behavioral equivalence does not imply neural equivalence. Specifically, 2 groups may achieve the same behavioral performance by using different neural systems, with 1 group’s performance resulting from compensatory processing mechanisms that differ from the normal brain processing involved in the task.19 We therefore undertook an fMRI study to assess the location and degree of neural activation underlying working memory performance in children with and without cocaine exposure. We hypothesized that exposed and non-
T
ANOVA fMRI SES
Analysis of variance Functional magnetic resonance imaging Socioeconomic status
DLPFC ROI BOLD
Dorsolateral prefrontal cortex Region of interest Blood oxygenation level dependent
From the Department of Neonatology, The Children’s Hospital of Philadelphia, Philadelphia, Pennsylvania (H.H., J.G., N.B.); Department of Neurology and Neuroradiology, Center for Functional Neuroimaging, Hospital of the University of Pennsylvania, Philadelphia, Pennsylvania (M.K., A.H., K.T., J.D.); Center for Cognitive Neuroscience, University of Pennsylvania, Philadelphia, Pennsylvania (L.B., M.F.); and Department of Biostatistics and Epidemiology, The Children’s Hospital of Philadelphia, Philadelphia, Pennsylvania (D.S.). Supported by grants from the National Institutes of Health/National Institute on Drug Abuse (R01-DA14129), National Institutes of Health/National Center for Research Resources (M01-RR00240), and National Institute of Child Health and Human Development (MRDDRC-HD26979, R21-DA01586, and R01-HD043078). Submitted for publication Dec 13, 2006; last revision received May 25, 2007; accepted Aug 9, 2007. Reprint requests: Hallam Hurt, MD, Department of Neonatology, The Children’s Hospital of Philadelphia, 3535 Market St, Room 1509, Philadelphia, PA 19104. E-mail: [email protected]. 0022-3476/$ - see front matter Copyright © 2008 Mosby Inc. All rights reserved. 10.1016/j.jpeds.2007.08.006
371
exposed groups would perform similarly on behavioral tasks, but differ in activation patterns. Specifically, we hypothesized that the exposed group would exhibit less robust activation in anticipated areas of activation or would exhibit compensatory activation in areas not activated in the non-exposed group.
METHODS Participants Participants were selected from a cohort of exposed and non-exposed subjects who have been observed since their birth (1989-1992). Full details of enrollment have been reported earlier.20 All participants were born at a single innercity hospital to mothers of low socioeconomic status (SES). Exposed children’s mothers admitted to cocaine use in at least 2 trimesters of pregnancy. The mothers of non-exposed, or control, participants denied cocaine use, and both mothers and children had urine samples that were negative for cocaine metabolites at delivery. Mothers were excluded when they did not speak English, had a major psychiatric disorder, or used substances other than cigarettes, marijuana, or alcohol during pregnancy. Infants were excluded when they were ⬍34 weeks gestational age, had an Apgar score ⱕ5 at 5 minutes of age, or had fetal alcohol syndrome or any syndrome known to be associated with developmental delay. Some exposed subjects were also exposed to cigarettes, alcohol, and/or marijuana. One control subject was exposed to cigarettes. Subjects selected for fMRI met these criteria: they were right-handed, had no metal appliances, and were taking no medications. Because of concerns for possible confounding effects of sex and IQ, we further selected exposed and nonexposed children by sex and group quartiles of 4-year Wechsler Preschool and Primary Scale of Intelligence-Revised scores. The protocol was approved by The Children’s Hospital of Philadelphia Institutional Review Board. Informed consent was obtained from participants’ caregivers. All participants gave assent. Tasks All tasks were implemented in E-Prime software version 1.1 (Psychology Software Tools, Pittsburgh, PA), with subjects required to monitor a visually presented random sequence of letters and press a button every time a target letter was displayed on the screen. Non-target letters were all other letters viewed by subjects for which no button press was required. For the x-detection task, the target letter was an “x.” The 1-back task required that the button be pressed when any letter was repeated, one after another, with the repeated letter being the target. The 2-back task required that the subject press a button when any letter was repeated with exactly 1 intervening letter. For the 2-back task, the repeated letter was also the target. For example, in the 2-back task, for the sequence “J-M-J,” the subject would press the button when the second “J,” the target, was presented. The 1-back and 2-back tasks were administered twice to each subject, 2 runs of 1-back alternating with the x-de372
Hurt et al
tection task followed by 2 runs of 2-back alternating with the x- detection task. Thus, subjects were scanned during 4 fMRI runs. The 1-back task was used in runs 1 and 2. The 2-back task was used in runs 3 and 4. Each run consisted of 10 blocks of alternating x-detection and n-back tasks, with each block consisting of 20 trials. Within each run, there were a total of 30 target and 70 non-target trials for each task (x-detection, 1-back, or 2-back). The order of task presentation (1-back before 2-back) provided baseline behavioral measurements so that working memory, as operationalized by the 2-back, could be singled out as the cognitive process of interest. For all tasks, letters were presented on the screen individually with interstimulus intervals equal to 1500 ms and stimuli presentations equal to 500 ms.
Imaging Procedures Project personnel accompanied subjects to the Center for Advanced Magnetic Resonance Imaging Spectroscopy at the Hospital of the University of Pennsylvania. An initial visit introduced the subject to the MRI environment with a simulator without time constraints. The study was conducted on a 3T Siemens Trio whole body MRI scanner with a product volume head coil. Visual stimuli for the task were backprojected onto a Mylar screen with an LCD projector with computer control. Responses were recorded with a fiberoptic button box (FORP; Current Designs, Philadelphia, PA). The imaging protocol consisted of a 1-minute 3-axis localizer, a 5-minute inversion recovery prepped 3-dimensional T1-weighted anatomical scan with 1 mm isotropic resolution used for spatial normalization, and 2-dimensional multi-slice gradient-echo echoplanar imaging with relaxation time of 3 seconds, echo time of 30 msec, 3 mm isotropic resolution, and prospective motion correction (Siemens PACE) for detecting blood oxygenation level dependent (BOLD) contrast fMRI. Twelve seconds of “dummy” scans preceded the onset of the n-back tasks to allow magnetization to reach a steady state. Data Analysis Group baseline characteristics were compared with 2, t, or Mann-Whitney U test, as appropriate. Three behavioral scores were analyzed: 1) the number of correct responses to the target condition (hits), 2) the number of correct responses to the non-target condition (correct rejections), and 3) total correct, the sum of hits and correct rejections. Three separate group (exposed versus non-exposed) by gender by age analyses of variance with repeated measures (RM-ANOVA) for task were conducted, 1 for each of the 3 behavioral scores. Functional data were analyzed on Linux workstations using VoxBo (www.voxbo.org). Preprocessing included slicetiming correction, rigid body realignment to the subject’s first functional volume,21 and spatial smoothing with a 3D Gaussian filter (9 mm kernel, 3 ⫻ 3 ⫻ 3 voxels). A modified general linear model21 with a 1/f estimate of intrinsic autocorrelation22 was applied voxelwise to each individual’s fMRI data. The Journal of Pediatrics • March 2008
Figure 1. ROI masks. 1. A priori anatomical ROIs for Brodmann regions 10 and 46. A, Surface renderings of Brodmann regions 10 and 46. B, Cutaway rendering of Brodmann regions 10 and 46. 2. Functional ROIs for cingulate, right DLPFC, left DLPFC, right parietal, and left parietal. Functional ROIs were derived from the beta map of the non-exposed cohort during the 2-back condition, at a significant threshold of t ⫽ 4.16, alpha ⫽ 0.05. A, Functional ROI surface rendering. B, Functional ROI cutaway rendering.
Time series data were filtered in the frequency domain to remove the highest frequencies and lowest frequencies, up to but not including the fundamental task frequency. Data were convolved with an empirically derived estimate of hemodynamic response,23 and scan effect covariates were included to remove scan-to-scan variation. The fMRI data for 1-back and 2-back runs were analyzed both separately and in combination, and within-subject conditions included the x-detection (x), 1-back (1b), and 2-back (2b) tasks. Beta maps were constructed for these contrasts: (1b)-(x), (2b)-(x), and (2b)-(1b). Individual data were spatially normalized with Statistical Parametric Mapping 2 (www.fil.ion.ucl.ac.uk/spm) to a standard Montreal Neurologic Institute (MNI) template,24 and group maps were generated for those contrasts by using a random effects model comparing mean statistical parametric maps across subjects to zero with a t test for each voxel. Discrimination scores for the 1-back and 2-back tasks were also included as a covariate in the group analyses to remove variation caused by task performance. Discrimination scores were calculated with the equation:
共Hits ⁄ Total possible hits兲 ⫺ 共Incorrect rejections ⁄ Total possible rejections兲 . Group analyses were thresholded by cluster-constrained permutation analysis,25 with 2000 sign permutations of the individual maps to generate a null hypothesis distribution for cluster-corrected thresholds (cluster size ⫽ 30, P ⫽ .05). A priori anatomical regions of interest (ROIs) were generated for Brodmann regions 10 and 46 with the Wake Forest University PickAtlas26 in Talairach space (Figure 1), on the basis of the work of Casey et al.17 Functionally derived ROIs were also defined for cingulate, left and right dorsolat-
Figure 2. Number of correct responses to target condition in runs 1 to 4 in exposed and non-exposed groups. By repeated measures ANOVA, P ⫽ .69 for group across runs; P ⬍ .001 for runs 1 and 2 versus runs 3 and 4; P ⬍ .001 for run 3 versus runs 1, 2, and 4.
eral prefrontal cortex (DLPFC), and left and right parietal region on the basis of the work of Owen et al.27 Areas were determined by using the beta map from the (2b)-(x) condition in non-exposed subjects, the contrast of which provided the greatest amount of activation. The proportion of positive voxels within each ROI at zero threshold was determined for (1b)-(x) and (2b)-(x) by using beta values.28 The fractional ROI activation for the (2b)-(1b) condition was determined by the difference between the (2b)-(x) and (1b)-(x) condition between exposed and non-exposed groups.
RESULTS Forty-nine subjects, 25 exposed and 24 non-exposed, underwent imaging. Nine subjects (5 exposed, 4 non-exposed) were excluded because of motion artifact, data log issues, or scanner technical difficulties. Six additional subjects (3 exposed and 3 non-exposed) underwent imaging successfully in only 3 of the 4 runs and were excluded from analyses. The 34 remaining subjects underwent imaging successfully in all 4 runs, with their results forming the basis for analyses. Groups were similar, except exposed subjects were older at the time of testing and were more likely to be exposed to maternal cigarette, marijuana, and alcohol use during pregnancy (Table). The median days of exposure to cocaine for the exposed group was 117 days, which constitutes a high level of exposure and is consistent with the high rate of maternal urine test results that were positive (81%) for cocaine at delivery, also a marker for heavy maternal use.29 The 34 subjects who underwent imaging successfully were compared on the characteristics shown in the Table with the 15 excluded subjects and the remaining 71 cohort subjects who did not undergo imaging. Groups were similar in all characteristics (P ⱖ .08), except subjects who did not undergo imaging had lower full scale IQ scores than did the group of 34 subjects who underwent imaging successfully and the 15 subjects who underwent imaging but were excluded (P ⫽ .013).
Functional Magnetic Resonance Imaging and Working Memory in Adolescents with Gestational Cocaine Exposure
373
Table. Subject characteristics
Child characteristics at time of fMRI N Female gender Age at testing, years African American race Maternal characteristics at child’s birth Urine positive for cocaine at delivery Days of cocaine use in pregnancy Marijuana use in pregnancy Cigarette use in pregnancy Alcohol use in pregnancy Child characteristics at birth Birth weight ⬍10th percentile Head circumference ⬍10th percentile Apgar, 5 minutes Cranial ultrasound scanning abnormality Child examinations during childhood Any neurolgical abnormality at 6.5 years¶ ⬎1 neurological abnormality at 6.5 years¶ Full scale IQ at 4 years
Exposed
Non-exposed
P value
17 8 (47)* 14.7 ⫾ 0.8† (13.2-15.9)‡ 15 (88)
17 9 (53) 13.8 ⫾ 0.8 (12.9-15.3) 17 (100)
1 .001 .48
13/16 (81) 117§ (9-273)‡ 10/16 (62) 16 (94) 10/16 (62)
— — 0 1 (6) 0
— — ⬍.001 ⬍.001 ⬍.001
0 3 (18) 9§ 0
1 (6) 2 (12) 9 0
1 1 .82 —
3/14 (21) 1/14 (7) 84.1 ⫾ 15.1 (60-111)‡
2/16 (12) 1/16 (6) 86.3 ⫾ 7.9 (75-101)
.64 1 .6
*N (%). †Mean ⫾ SD. ‡Range. §Median. ¶Subjects examined by a developmental pediatrician.
Figure 3. Areas of activation for functional ROIs (left to right): 1-back minus x-detection; 2-back minus x-detection; and 2-back minus 1-back. N ⫽ 17 for non-exposed subjects (top), n ⫽ 17 for exposed subjects (bottom). t ⫽ 4.16.
Task Performance For the x-detection task, performance was similar across all 4 runs in both groups (data not shown). The number of correct responses to the target condition (hits), the non-target condition (correct rejections), and the sum of hits and correct rejections did not differ between the exposed and non-exposed groups, male subjects and female subjects, or across age (P ⱖ .27) with RM-ANOVA (data not shown). In comparing the performance on the 4 n-back runs for both groups, however, the number of correct responses to the target con374
Hurt et al
dition decreased between runs 1 and 2 (1-back): run 1 (exposed [Exp]): (mean (⫾) SD [ ] ⫽ range) 26.8 ⫾ 5.3 [18]; non-exposed [Non-E]: 28.1 ⫾ 2.8 [8]) and run 2 (Exp: 27.4 ⫾ 2.2 [7]; Non-E: 28.4 ⫾ 2.5 [9]) and runs 3 and 4 (2-back): (Exp: 18.8 ⫾ 5.9 [19]; Non-E: 18 ⫾ 5.9 [24]) and run 4 (Exp: 22.5 ⫾ 6.9 [21]; Non-E: 21.8 ⫾ 5.6 [19]; P ⬍ .001; Figure 2). Additionally, the number of correct responses to the target condition increased between runs 3 and 4 (P ⬍ .001). These results demonstrate the increased working memory demands that are associated with runs 3 and 4 (2-back), with a more pronounced effect observed for run 3. The improved performance between runs 3 and 4 is possibly related to a practice effect. The decrease in target response accuracy in the 2-back condition across exposed and non-exposed subjects indicates that both groups were challenged by the increasing complexity of the 2-back task. In a post hoc analysis, we reviewed the behavioral data to identify any subjects who did not achieve at least 50% total correct on the x-detection and n-back tasks, to avoid the inclusion of outliers. With this approach, no subject meeting these performance criteria was identified from either the exposed or non-exposed group. Thus, on an executive function task assessing working memory, exposed subjects performed as well as non-exposed subjects.
Imaging There was an increase in activation from the (1-b)(x) to the (2b)-(x) condition (Figure 3)in the functionally derived ROIs for all regions defined (cingulate, left and The Journal of Pediatrics • March 2008
right DLPFC, left and right parietal) in both exposed and non-exposed groups. The ROI percent activation was greater in the non-exposed group in all regions. For the (2b)-(1b) contrast, the ROI percent activation in cingulate, left DLPFC, and right DLPFC showed greater volumes of activation for the non-exposed group compared with the exposed group. However, in both right and left parietal ROIs, the exposed group demonstrated a greater ROI percent activation than the nonexposed group. These differences did not reach a statistical significance. In the a priori anatomical ROIs, Brodmann’s regions 10 and 46, a comparison of the average percent ROI activation for non-exposed versus exposed groups showed that, although both groups increased activation from the (1-b)-(x) to the (2-b)-(x), the level of activation in the non-exposed group was proportionally greater. A 1-tailed t test for the (2b)-(1b) condition between the non-exposed group and exposed group in the left hemisphere yielded a borderline significant value of P ⫽ .05. Although it is difficult to assess the clinical significance of the differences seen in activation, especially lacking a significant group performance difference, we can nevertheless use the estimated proportions activated in both functional and anatomical ROI to compute approximate post hoc observed effect sizes. For each area of interest, we computed task (1-back versus 2-back) effect size for each group (cocaineexposed and non-exposed) and then computed the difference in group effect sizes. For example, in the cingulate, the 2-back versus 1-back effect size was 0.9 SDs for the non-exposed group and 0.66 SDs for the exposed group, both large effect sizes. The group difference in effect size was 0.25 (a small effect). Results for the right and left DLPFC, parietal area, and Brodmann’s areas were similar, reflecting a large change in activation from the 1-back to the 2-back within both groups, which is relatively easy to detect, but a smaller difference in the groups.
DISCUSSION In this study, cocaine exposed and non-exposed innercity adolescents performed similarly on a non-spatial working memory task and demonstrated similar activation patterns during fMRI. We did not find differences in the behavioral results between exposed and non-exposed participants during administration of a non-spatial working memory task. The task was an appropriate one in that there was the predicted increase in difficulty when subjects first encountered the 2-back task after having performed the more simple x-detection and 1-back tasks. During the 2-back task, the correct responses to both targets (that is, subjects were less likely to correctly identify the target and perform the button press) and non-targets (that is, subjects were more likely to press the button in the absence of the 2-back target) decreased. When runs 1 and 2 (1-back task) were compared with runs 3 and 4 (2-back task), we found significant decreases in total correct responses (Figure 2) by both exposed and non-exposed participants.
Thus, for this working memory task, performance by exposed and non-exposed participants was similar. This occurred despite evidence of heavy gestational exposure: a median level of 117 days of cocaine exposure, and 81% of maternal urine sample test results were positive at delivery. In an investigation suggesting worse neurodevelopmental outcome with the heavy exposure, heavily exposed subjects were defined as having prenatal exposure for only 61 days. Whether the similar behavioral performance was associated with similar or dissimilar patterns of brain activation was examined with fMRI. A particular point of interest, in light of subjects’ similar performance on the behavioral task, was whether tasks were accomplished by using compensatory neural mechanisms, as would be evidenced through different activation patterns. Map-wise comparisons of both exposed and non-exposed subjects confirmed activation patterns in expected regions of cingulate, DLPFC, and parietal lobe,27 demonstrating the validity and reliability of the n-back task as applied to our population. Group statistical maps in both exposed and non-exposed groups showed greater activation with increased task complexity, particularly in bilateral DLPFC regions. The most robust activation occurred in expected directions during the 2-back condition, which requires the greatest working memory demands, as evidenced by performance results. Our results are consistent with earlier literature on functional activation during n-back tasks.17,30-32 In particular, Casey,17 reporting on a small healthy control sample of 9- to 11-year-old children with the n-back working memory task, observed activation of inferior and middle frontal gyri (Brodmann’s 46 and 10). In that sample of young subjects, activation patterns also increased in expected directions with greater task complexity and also correlated with behavioral performance. Thus, the imaging and behavioral performance results of the present study are consistent with Casey et al, in which activation increased and response accuracy significantly decreased across subjects with increasing task complexity. Earlier studies indicate that increasing the numbers “back” correspondingly increases the degree of activation. Therefore, decreasing performance in response to increased working memory load produces an increase in brain activation.31,33 However, the work of Callicott et al indicates that certain load-sensitive loci within the working memory network demonstrate capacity-limited response. The beta map for the non-exposed group for the 2-back minus 1-back condition demonstrates a greater degree of activation than the exposed group, despite analogous behavioral performance (Figure 3). This would suggest that the exposed cohort experiences a memory capacity breach at a lower difficulty level than the non-exposed cohort. This effect is seen particularly in the prefrontal cortex, which is analogous to the results of Calicott et al.31 Although a similar pattern of reduced activation with normal behavioral performance is seen in patients with schizophrenia,33 various other patterns also have been documented in developmental disorders. For example, adults with autism spectrum disorder who have similar behavioral performance as control subjects show increased brain activa-
Functional Magnetic Resonance Imaging and Working Memory in Adolescents with Gestational Cocaine Exposure
375
tion in tests of executive function.34 Further, medication naïve adolescents with attention deficit hyperactivity disorder showed reduced brain activation independent of task performance during an initiation and error detection task, findings felt to confirm a relationship between behavioral impulsivity and neural abnormalities.35 Our results are also consistent with studies that examine the relationship between frontal lobe mediated information maintenance and manipulation with time, central to the nback working memory task. D’Esposito et al36 contend that the DLPFC is necessary for working memory and both maintenance and manipulation tasks activate ventral and DLPFC regions. To further explore the relationship between information maintenance and manipulation in the broader context of central executive processing in normal subjects, Ragland et al32 used an n-back task to define the brain regions that activate in distinct maintenance or manipulation conditions. During a 1-back minus 0-back task designed to isolate information maintenance, they observed inferior parietal and DLPFC activation. During a 2-back minus 0-back task designed to assess the cumulative effect of maintenance and manipulation, they reported activation in inferior parietal regions and dorsal and ventrolateral PFC regions. During a 2-back minus 1-back “manipulation only” task designed to remove the maintenance demands across conditions, broader DLPFC and anterior cingulate activations were observed. The results of both studies, with the exception of observed activations in ventrolateral PFC, are consistent with this investigation. We found increasing DLPFC activation with task difficulty, and the emergence of anterior cingulate activation in the 2-back minus x detection condition, suggesting a similar role for these regions in working memory function in our cohort. Thus, although our study shows anticipated patterns of activation, there are no differences between exposed and nonexposed participants. Possible explanations for our null results are several. It may be that differences do exist, but that the number of subjects was too small (although our study is a large study by typical fMRI standards) to detect such differences. It may well be that investigations with a larger number of subjects will detect differences between exposed and nonexposed groups. Another possibility is that an enriched environment for exposed subjects placed out of the home could ameliorate effects of cocaine exposure; in our sample, however, there was no difference in out-of-home placement in the 2 groups. Is it possible that some of our non-exposed participants were, in fact, exposed to cocaine? This is possible, but doubtful because we had negative history from mothers, negative urine test samples for both mothers and babies, and a maternal profile (cocaine-using mothers were older, had less prenatal care, had more sexually transmitted diseases, and were more likely to be polysubstance users) of cocaine-using mothers that differed from non-using mothers.20 Perhaps the more compelling question is why, given our exposed subjects’ greater exposure not only to cocaine but also alcohol, cigarettes, and marijuana, did we not detect differences between 376
Hurt et al
exposed and non-exposed groups? It may be that fMRI, dependent on the BOLD signal from blood flow, simply is not sensitive enough to detect changes that do exist, or that recently described cerebral blood flow differences in our sample between exposed and non-exposed groups influenced results.37 However, perhaps any effects of gestational exposure present early on have been remediated through the maturational processes that normally occur during childhood and adolescence,38 or there may be characteristics unique to our sample that are responsible for null results. Finally, gestational cocaine exposure simply may be less injurious than predicted. In this regard, there are a number of studies of global intellectual functioning in exposed children at young ages showing these children are not severely and irreparably damaged as suggested by some early reports.39 In our sample, as described here, so far we have not found consistent differences between exposed and non-exposed in executive function.10,18,40 In some investigations in older subjects, however, differences in executive functions such as attention and arousal have been reported. Thus, the full impact of effects of gestational cocaine exposure, if any, remains to be determined. In conclusion, with a working memory task in this sample of low socioeconomic status subjects, exposed and non-exposed groups had similar task performance and brain activation patterns. These data provide a basis for future investigations that may explore subject groups of differing SES, IQ, and ethnicity.
REFERENCES 1. Mayes LC. Developing brain and in utero cocaine exposure: effects on neural ontogeny. Dev Psychopathol 1999;11:685-714. 2. Harvey JA. Cocaine effects on the developing brain: current status. Neurosci Biobehav Rev 2004;27:751-64. 3. Mayes LC, Fahy T. Prenatal drug exposure and cognitive development. In: Sternberg RJ, Grigorenko EL, editors. Environmental effects on cognitive abilities. Mahwah, New Jersey: Lawrence Erlbaum Associates; 2001. p. 189-219. 4. Struthers JM, Hansen RL. Visual recognition memory in drug-exposed infants. J Dev Behav Pediatr 1992;13:108-11. 5. Jacobson SW, Jacobson JL, Sokol RJ, Martier SS, Chiodo LM. New evidence for neurobehavioral effects of in utero cocaine exposure. J Pediatr 1996;129:581-90. 6. Mayes LC, Bornstein MH, Chawarska K, Haynes OM, Granger RH. Impaired regulation of arousal in three-month-old infants exposed prenatally to cocaine and other drugs. Dev Psychopathol 1996;8:29-42. 7. Noland JS, Singer LT, Short EJ, Minnes S, Arendt RE, Kirchner HL, et al. Prenatal drug exposure and selective attention in preschoolers. Neurotoxicol Teratol 2005;27:429-38. 8. Richardson GA, Conroy ML, Day NL. Prenatal cocaine exposure: effects on the development of school-age children. Neurotoxicol Teratol 1996;18:627-34. 9. Bandstra ES, Morrow CE, Anthony JC, Accornero VH, Fried PA. Longitudinal investigation of task persistence and sustained attention in children with prenatal cocaine exposure. Neurotoxicol Teratol 2001;23:545-59. 10. Savage J, Brodsky NL, Malmud E, Giannetta JM, Hurt H. Attentional functioning and impulse control in cocaine-exposed and control children at age ten years. J Dev Behav Pediatr 2005;26:42-7. 11. Schroder MD, Snyder PJ, Sielski I, Mayes L. Impaired performance of children exposed in utero to cocaine on a novel test of visuospatial working memory. Brain Cogn 2004;55:409-12. 12. Mayes LC, Molfese DL, Key AP, Hunter NC. Event-related potentials in cocaine-exposed children during a Stroop task. Neurotoxicol Teratol 2005;27:797-813. 13. Linares TJ, Singer LT, Kirchner HL, Short EJ, Min MO, Hussey P, et al. Mental health outcomes of cocaine-exposed children at 6 years of age. J Pediatr Psychol 2006;31:85-97. 14. Sheinkopf SJ, Lester B, Eliassen JC, Sanes JN, Hutchinson ER, Seifer R, et al. Functional MRI and prenatal cocaine exposure. Pediatric Research. San Francisco, CA:2004;3295A.
The Journal of Pediatrics • March 2008
15. Warner TD, Behnke M, Eyler FD, Padgett K, Leonard CM, Hou W, et al. Diffusion tensor imaging of frontal white matter and executive functioning in cocaineexposed children. Pediatrics 2006;118:2014-24. 16. Smith LM, Chang L, Yonekura ML, Gilbride K, Kuo J, Poland RE, et al. Brain proton magnetic resonance spectroscopy and imaging in children exposed to cocaine in utero. Pediatrics⬘ 2001;107:227-31. 17. Casey BJ, Cohen JD, Jezzard P, Turner R, Noll DC, Trainor RJ, et al. Activation of prefrontal cortex in children during a nonspatial working memory task with functional MRI. Neuroimage 1995;2:221-9. 18. Betancourt LM, Brodksy NL, Malmud E, Giannetta JG, Farah M, Hurt H, et al. Does gestational cocaine exposure affect child neurocognitive outcome? PAS 2005; 57:1652 Platform Session APS/SPR Meeting, May 16, 2005. Available at: http:// www.abstracts2view.com/pas/. 19. Bush G, Frazier JA, Rauch SL, Seidman LJ, Whalen PJ, Jenike MA, et al. Anterior cingulate cortex dysfunction in attention-deficit/hyperactivity disorder revealed by fMRI and the Counting Stroop. Biol Psychiatry 1999;45:1542-52. 20. Hurt H, Brodsky NL, Braitman LE, Giannetta J. Natal status of infants of cocaine users and control subjects: a prospective comparison. J Perinatol 1995;15:297-304. 21. Worsley KJ, Friston KJ. Analysis of fMRI time-series revisited—again. Neuroimage 1995;2:173-81. 22. Aguirre GK, Zarahn E, D’Esposito M. Empirical analyses of BOLD fMRI statistics. II. Spatially smoothed data collected under null-hypothesis and experimental conditions. Neuroimage 1997;2:199-212. 23. Aguirre GK, Zarahn E, D’Esposito M. The variability of human, BOLD hemodynamic responses. Neuroimage 1998;8:360-9. 24. Collins D. 3D model-based segmentation of individual brain structures from magnetic resonance imaging data. Montreal: McGill University; 1994. 25. Nichols TE, Holmes AP. Nonparametric permutation tests for functional neuroimaging: a primer with examples. Hum Brain Mapp 2002;15:1-25. 26. Maldjian JA, Laurienti PJ, Kraft RA, Burdette JH. An automated method for neuroanatomic and cytoarchitectonic atlas-based interrogation of fMRI data sets. Neuroimage 2003;19:1233-9. 27. Owen AM, McMillan KM, Laird AR, Bullmore E. N-back working memory paradigm: a meta-analysis of normative functional neuroimaging studies. Hum Brain Mapp 2005;25:46-59. 28. Rabin M, Narayan V, Kimberg DY, Casasanto D, Glosser G, Tracy J. Functional
MRI predicts post-surgical memory following temporal lobectomy. Brain 2004;127: 2286-98. 29. Zuckerman B, Frank DA, Hingson R, Amaro H, Levenson SM, Kayne H, et al. Effects of maternal marijuana and cocaine use on fetal growth. N Engl J Med 1989;320:762-8. 30. Goldberg TE, Weinberger DR. Effects of neuroleptic medications on the cognition of patients with schizophrenia: a review of recent studies. J Clin Psychiatry 1996;57 Suppl 9:62-5. 31. Callicott JH, Mattay VS, Bertolino A, Finn K, Coppola R, Frank JA, et al. Physiological characteristics of capacity constraints in working memory as revealed by functional MRI. Cereb Cortex 1999;9:20-6. 32. Ragland JD, Turetsky BI, Gur RC, Gunning-Dixon F, Turner T, Schroeder L, et al. Working memory for complex figures: an fMRI comparison of letter and fractal n-back tasks. Neuropsychology 2002;16:370-9. 33. Weinberger DR, Mattay V, Callicott J, Kotrla K, Santha A, van Gelderen P, et al. fMRI applications in schizophrenia research. Neuroimage 1996;4:S118-26. 34. Schmitz N, Rubia K, Daly E, Smith A, Williams S, Murphy DG. Neural correlates of executive function in autistic spectrum disorders. Biol Psychiatry 2006;59:7-16. 35. Rubia K, Smith AB, Brammer MJ, Toone B, Taylor E. Abnormal brain activation during inhibition and error detection in medication-naive adolescents with ADHD. Am J Psychiatry 2005;162:1067-75. 36. D’Esposito M, Postle BR, Ballard D, Lease J. Maintenance versus manipulation of information held in working memory: an event-related fMRI study. Brain Cogn 1999;41:66-86. 37. Rao H, Wang J, Giannetta JM, Korcykowski M, Shera D, Avants B, et al. Altered resting cerebral blood flow in adolescents with in-utero cocaine exposure revealed by perfusion functional magnetic resonance imaging. Pediatrics. In press, 2006. 38. Rubia K, Overmeyer S, Taylor E, Brammer M, Williams SC, Simmons A, et al. Functional frontalisation with age: mapping neurodevelopmental trajectories with fMRI. Neurosci Biobehav Rev 2000;24:13-9. 39. Odom-Winn D, Dunagan DE. “Crack kids” in school. What to do. How to do it. Educational Activities 1991;54:54. 40. Betancourt L, Fischer R, Giannetta J, Malmud E, Brodsky NL, Hurt H. Problemsolving ability of inner-city children with and without in utero cocaine exposure. J Dev Behav Pediatr 1999;20:418-24.
Functional Magnetic Resonance Imaging and Working Memory in Adolescents with Gestational Cocaine Exposure
377
Objective To assess the effect of gestational cocaine exposure on the prefrontal cortex (PFC) with functional magnetic resonance imaging (fMRI). Study design Using an n-back task, we obtained fMRI with a 3T Siemens scanner on 49 adolescents, 25 who were exposed to cocaine and 24 who were not exposed. The primary outcome was PFC activation during task performance. Five functionally derived regions of interest (ROI) were defined; in addition, 2 a priori anatomical ROIs were generated for Brodmann regions 10 and 46. Results Of the 49 adolescents who underwent imaging, data from 17 who were exposed to cocaine and 17 who were not exposed were in the final analysis. Groups had similar performance on the n-back task (P > .4), with both showing a fewer number of correct responses on the 2-back than the 1-back (P < .001), indicating increased demands on working memory with greater task difficulty. In functionally derived ROIs, imaging results showed increased activation for both groups in the 2-back versus the 1-back condition. In anatomical ROIs, both groups showed greater activation in the 2-back versus the 1-back condition, with activation in the non-exposed group proportionally greater for the left prefrontal region (P ⴝ .05). Conclusion In this sample of adolescents, participants who were exposed to cocaine and participants who were not exposed were similar in performance on an executive function task and in fMRI activation patterns during task performance. (J Pediatr 2008;152:371-7) he growing pre-clinical and clinical literature suggest that gestational cocaine exposure may affect developing monaminergic (dopamine, serotonin, and noradrenaline) systems that are widely distributed throughout the brain.1,2 These findings, in turn, have given rise to concern about injury to the prefrontal cortex in exposed children, with resultant effects on executive functioning, a diverse set of skills such as planning, problem solving, cognitive control, working memory, and reward processing.3 In this regard, investigations assessing neurobehavioral outcomes and executive function in children with gestational cocaine exposure have found effects ranging from none, to minimal, to impaired recognition memory, to altered task persistence, to self-regulatory deficits.4-13 There have been several efforts to anchor such clinical findings with neuroanatomical bases;14-16 however, no consistent patterns have been identified. We investigate the neural bases of injury through examination of patterns of brain activation by using functional magnetic resonance imaging (fMRI) during a task of executive function—in this case, non-spatial working memory as assessed by using an n-back task.17 Although our earlier behavioral assessments have failed to find an effect of cocaine exposure with both spatial and non-spatial working memory tasks,18 behavioral equivalence does not imply neural equivalence. Specifically, 2 groups may achieve the same behavioral performance by using different neural systems, with 1 group’s performance resulting from compensatory processing mechanisms that differ from the normal brain processing involved in the task.19 We therefore undertook an fMRI study to assess the location and degree of neural activation underlying working memory performance in children with and without cocaine exposure. We hypothesized that exposed and non-
T
ANOVA fMRI SES
Analysis of variance Functional magnetic resonance imaging Socioeconomic status
DLPFC ROI BOLD
Dorsolateral prefrontal cortex Region of interest Blood oxygenation level dependent
From the Department of Neonatology, The Children’s Hospital of Philadelphia, Philadelphia, Pennsylvania (H.H., J.G., N.B.); Department of Neurology and Neuroradiology, Center for Functional Neuroimaging, Hospital of the University of Pennsylvania, Philadelphia, Pennsylvania (M.K., A.H., K.T., J.D.); Center for Cognitive Neuroscience, University of Pennsylvania, Philadelphia, Pennsylvania (L.B., M.F.); and Department of Biostatistics and Epidemiology, The Children’s Hospital of Philadelphia, Philadelphia, Pennsylvania (D.S.). Supported by grants from the National Institutes of Health/National Institute on Drug Abuse (R01-DA14129), National Institutes of Health/National Center for Research Resources (M01-RR00240), and National Institute of Child Health and Human Development (MRDDRC-HD26979, R21-DA01586, and R01-HD043078). Submitted for publication Dec 13, 2006; last revision received May 25, 2007; accepted Aug 9, 2007. Reprint requests: Hallam Hurt, MD, Department of Neonatology, The Children’s Hospital of Philadelphia, 3535 Market St, Room 1509, Philadelphia, PA 19104. E-mail: [email protected]. 0022-3476/$ - see front matter Copyright © 2008 Mosby Inc. All rights reserved. 10.1016/j.jpeds.2007.08.006
371
exposed groups would perform similarly on behavioral tasks, but differ in activation patterns. Specifically, we hypothesized that the exposed group would exhibit less robust activation in anticipated areas of activation or would exhibit compensatory activation in areas not activated in the non-exposed group.
METHODS Participants Participants were selected from a cohort of exposed and non-exposed subjects who have been observed since their birth (1989-1992). Full details of enrollment have been reported earlier.20 All participants were born at a single innercity hospital to mothers of low socioeconomic status (SES). Exposed children’s mothers admitted to cocaine use in at least 2 trimesters of pregnancy. The mothers of non-exposed, or control, participants denied cocaine use, and both mothers and children had urine samples that were negative for cocaine metabolites at delivery. Mothers were excluded when they did not speak English, had a major psychiatric disorder, or used substances other than cigarettes, marijuana, or alcohol during pregnancy. Infants were excluded when they were ⬍34 weeks gestational age, had an Apgar score ⱕ5 at 5 minutes of age, or had fetal alcohol syndrome or any syndrome known to be associated with developmental delay. Some exposed subjects were also exposed to cigarettes, alcohol, and/or marijuana. One control subject was exposed to cigarettes. Subjects selected for fMRI met these criteria: they were right-handed, had no metal appliances, and were taking no medications. Because of concerns for possible confounding effects of sex and IQ, we further selected exposed and nonexposed children by sex and group quartiles of 4-year Wechsler Preschool and Primary Scale of Intelligence-Revised scores. The protocol was approved by The Children’s Hospital of Philadelphia Institutional Review Board. Informed consent was obtained from participants’ caregivers. All participants gave assent. Tasks All tasks were implemented in E-Prime software version 1.1 (Psychology Software Tools, Pittsburgh, PA), with subjects required to monitor a visually presented random sequence of letters and press a button every time a target letter was displayed on the screen. Non-target letters were all other letters viewed by subjects for which no button press was required. For the x-detection task, the target letter was an “x.” The 1-back task required that the button be pressed when any letter was repeated, one after another, with the repeated letter being the target. The 2-back task required that the subject press a button when any letter was repeated with exactly 1 intervening letter. For the 2-back task, the repeated letter was also the target. For example, in the 2-back task, for the sequence “J-M-J,” the subject would press the button when the second “J,” the target, was presented. The 1-back and 2-back tasks were administered twice to each subject, 2 runs of 1-back alternating with the x-de372
Hurt et al
tection task followed by 2 runs of 2-back alternating with the x- detection task. Thus, subjects were scanned during 4 fMRI runs. The 1-back task was used in runs 1 and 2. The 2-back task was used in runs 3 and 4. Each run consisted of 10 blocks of alternating x-detection and n-back tasks, with each block consisting of 20 trials. Within each run, there were a total of 30 target and 70 non-target trials for each task (x-detection, 1-back, or 2-back). The order of task presentation (1-back before 2-back) provided baseline behavioral measurements so that working memory, as operationalized by the 2-back, could be singled out as the cognitive process of interest. For all tasks, letters were presented on the screen individually with interstimulus intervals equal to 1500 ms and stimuli presentations equal to 500 ms.
Imaging Procedures Project personnel accompanied subjects to the Center for Advanced Magnetic Resonance Imaging Spectroscopy at the Hospital of the University of Pennsylvania. An initial visit introduced the subject to the MRI environment with a simulator without time constraints. The study was conducted on a 3T Siemens Trio whole body MRI scanner with a product volume head coil. Visual stimuli for the task were backprojected onto a Mylar screen with an LCD projector with computer control. Responses were recorded with a fiberoptic button box (FORP; Current Designs, Philadelphia, PA). The imaging protocol consisted of a 1-minute 3-axis localizer, a 5-minute inversion recovery prepped 3-dimensional T1-weighted anatomical scan with 1 mm isotropic resolution used for spatial normalization, and 2-dimensional multi-slice gradient-echo echoplanar imaging with relaxation time of 3 seconds, echo time of 30 msec, 3 mm isotropic resolution, and prospective motion correction (Siemens PACE) for detecting blood oxygenation level dependent (BOLD) contrast fMRI. Twelve seconds of “dummy” scans preceded the onset of the n-back tasks to allow magnetization to reach a steady state. Data Analysis Group baseline characteristics were compared with 2, t, or Mann-Whitney U test, as appropriate. Three behavioral scores were analyzed: 1) the number of correct responses to the target condition (hits), 2) the number of correct responses to the non-target condition (correct rejections), and 3) total correct, the sum of hits and correct rejections. Three separate group (exposed versus non-exposed) by gender by age analyses of variance with repeated measures (RM-ANOVA) for task were conducted, 1 for each of the 3 behavioral scores. Functional data were analyzed on Linux workstations using VoxBo (www.voxbo.org). Preprocessing included slicetiming correction, rigid body realignment to the subject’s first functional volume,21 and spatial smoothing with a 3D Gaussian filter (9 mm kernel, 3 ⫻ 3 ⫻ 3 voxels). A modified general linear model21 with a 1/f estimate of intrinsic autocorrelation22 was applied voxelwise to each individual’s fMRI data. The Journal of Pediatrics • March 2008
Figure 1. ROI masks. 1. A priori anatomical ROIs for Brodmann regions 10 and 46. A, Surface renderings of Brodmann regions 10 and 46. B, Cutaway rendering of Brodmann regions 10 and 46. 2. Functional ROIs for cingulate, right DLPFC, left DLPFC, right parietal, and left parietal. Functional ROIs were derived from the beta map of the non-exposed cohort during the 2-back condition, at a significant threshold of t ⫽ 4.16, alpha ⫽ 0.05. A, Functional ROI surface rendering. B, Functional ROI cutaway rendering.
Time series data were filtered in the frequency domain to remove the highest frequencies and lowest frequencies, up to but not including the fundamental task frequency. Data were convolved with an empirically derived estimate of hemodynamic response,23 and scan effect covariates were included to remove scan-to-scan variation. The fMRI data for 1-back and 2-back runs were analyzed both separately and in combination, and within-subject conditions included the x-detection (x), 1-back (1b), and 2-back (2b) tasks. Beta maps were constructed for these contrasts: (1b)-(x), (2b)-(x), and (2b)-(1b). Individual data were spatially normalized with Statistical Parametric Mapping 2 (www.fil.ion.ucl.ac.uk/spm) to a standard Montreal Neurologic Institute (MNI) template,24 and group maps were generated for those contrasts by using a random effects model comparing mean statistical parametric maps across subjects to zero with a t test for each voxel. Discrimination scores for the 1-back and 2-back tasks were also included as a covariate in the group analyses to remove variation caused by task performance. Discrimination scores were calculated with the equation:
共Hits ⁄ Total possible hits兲 ⫺ 共Incorrect rejections ⁄ Total possible rejections兲 . Group analyses were thresholded by cluster-constrained permutation analysis,25 with 2000 sign permutations of the individual maps to generate a null hypothesis distribution for cluster-corrected thresholds (cluster size ⫽ 30, P ⫽ .05). A priori anatomical regions of interest (ROIs) were generated for Brodmann regions 10 and 46 with the Wake Forest University PickAtlas26 in Talairach space (Figure 1), on the basis of the work of Casey et al.17 Functionally derived ROIs were also defined for cingulate, left and right dorsolat-
Figure 2. Number of correct responses to target condition in runs 1 to 4 in exposed and non-exposed groups. By repeated measures ANOVA, P ⫽ .69 for group across runs; P ⬍ .001 for runs 1 and 2 versus runs 3 and 4; P ⬍ .001 for run 3 versus runs 1, 2, and 4.
eral prefrontal cortex (DLPFC), and left and right parietal region on the basis of the work of Owen et al.27 Areas were determined by using the beta map from the (2b)-(x) condition in non-exposed subjects, the contrast of which provided the greatest amount of activation. The proportion of positive voxels within each ROI at zero threshold was determined for (1b)-(x) and (2b)-(x) by using beta values.28 The fractional ROI activation for the (2b)-(1b) condition was determined by the difference between the (2b)-(x) and (1b)-(x) condition between exposed and non-exposed groups.
RESULTS Forty-nine subjects, 25 exposed and 24 non-exposed, underwent imaging. Nine subjects (5 exposed, 4 non-exposed) were excluded because of motion artifact, data log issues, or scanner technical difficulties. Six additional subjects (3 exposed and 3 non-exposed) underwent imaging successfully in only 3 of the 4 runs and were excluded from analyses. The 34 remaining subjects underwent imaging successfully in all 4 runs, with their results forming the basis for analyses. Groups were similar, except exposed subjects were older at the time of testing and were more likely to be exposed to maternal cigarette, marijuana, and alcohol use during pregnancy (Table). The median days of exposure to cocaine for the exposed group was 117 days, which constitutes a high level of exposure and is consistent with the high rate of maternal urine test results that were positive (81%) for cocaine at delivery, also a marker for heavy maternal use.29 The 34 subjects who underwent imaging successfully were compared on the characteristics shown in the Table with the 15 excluded subjects and the remaining 71 cohort subjects who did not undergo imaging. Groups were similar in all characteristics (P ⱖ .08), except subjects who did not undergo imaging had lower full scale IQ scores than did the group of 34 subjects who underwent imaging successfully and the 15 subjects who underwent imaging but were excluded (P ⫽ .013).
Functional Magnetic Resonance Imaging and Working Memory in Adolescents with Gestational Cocaine Exposure
373
Table. Subject characteristics
Child characteristics at time of fMRI N Female gender Age at testing, years African American race Maternal characteristics at child’s birth Urine positive for cocaine at delivery Days of cocaine use in pregnancy Marijuana use in pregnancy Cigarette use in pregnancy Alcohol use in pregnancy Child characteristics at birth Birth weight ⬍10th percentile Head circumference ⬍10th percentile Apgar, 5 minutes Cranial ultrasound scanning abnormality Child examinations during childhood Any neurolgical abnormality at 6.5 years¶ ⬎1 neurological abnormality at 6.5 years¶ Full scale IQ at 4 years
Exposed
Non-exposed
P value
17 8 (47)* 14.7 ⫾ 0.8† (13.2-15.9)‡ 15 (88)
17 9 (53) 13.8 ⫾ 0.8 (12.9-15.3) 17 (100)
1 .001 .48
13/16 (81) 117§ (9-273)‡ 10/16 (62) 16 (94) 10/16 (62)
— — 0 1 (6) 0
— — ⬍.001 ⬍.001 ⬍.001
0 3 (18) 9§ 0
1 (6) 2 (12) 9 0
1 1 .82 —
3/14 (21) 1/14 (7) 84.1 ⫾ 15.1 (60-111)‡
2/16 (12) 1/16 (6) 86.3 ⫾ 7.9 (75-101)
.64 1 .6
*N (%). †Mean ⫾ SD. ‡Range. §Median. ¶Subjects examined by a developmental pediatrician.
Figure 3. Areas of activation for functional ROIs (left to right): 1-back minus x-detection; 2-back minus x-detection; and 2-back minus 1-back. N ⫽ 17 for non-exposed subjects (top), n ⫽ 17 for exposed subjects (bottom). t ⫽ 4.16.
Task Performance For the x-detection task, performance was similar across all 4 runs in both groups (data not shown). The number of correct responses to the target condition (hits), the non-target condition (correct rejections), and the sum of hits and correct rejections did not differ between the exposed and non-exposed groups, male subjects and female subjects, or across age (P ⱖ .27) with RM-ANOVA (data not shown). In comparing the performance on the 4 n-back runs for both groups, however, the number of correct responses to the target con374
Hurt et al
dition decreased between runs 1 and 2 (1-back): run 1 (exposed [Exp]): (mean (⫾) SD [ ] ⫽ range) 26.8 ⫾ 5.3 [18]; non-exposed [Non-E]: 28.1 ⫾ 2.8 [8]) and run 2 (Exp: 27.4 ⫾ 2.2 [7]; Non-E: 28.4 ⫾ 2.5 [9]) and runs 3 and 4 (2-back): (Exp: 18.8 ⫾ 5.9 [19]; Non-E: 18 ⫾ 5.9 [24]) and run 4 (Exp: 22.5 ⫾ 6.9 [21]; Non-E: 21.8 ⫾ 5.6 [19]; P ⬍ .001; Figure 2). Additionally, the number of correct responses to the target condition increased between runs 3 and 4 (P ⬍ .001). These results demonstrate the increased working memory demands that are associated with runs 3 and 4 (2-back), with a more pronounced effect observed for run 3. The improved performance between runs 3 and 4 is possibly related to a practice effect. The decrease in target response accuracy in the 2-back condition across exposed and non-exposed subjects indicates that both groups were challenged by the increasing complexity of the 2-back task. In a post hoc analysis, we reviewed the behavioral data to identify any subjects who did not achieve at least 50% total correct on the x-detection and n-back tasks, to avoid the inclusion of outliers. With this approach, no subject meeting these performance criteria was identified from either the exposed or non-exposed group. Thus, on an executive function task assessing working memory, exposed subjects performed as well as non-exposed subjects.
Imaging There was an increase in activation from the (1-b)(x) to the (2b)-(x) condition (Figure 3)in the functionally derived ROIs for all regions defined (cingulate, left and The Journal of Pediatrics • March 2008
right DLPFC, left and right parietal) in both exposed and non-exposed groups. The ROI percent activation was greater in the non-exposed group in all regions. For the (2b)-(1b) contrast, the ROI percent activation in cingulate, left DLPFC, and right DLPFC showed greater volumes of activation for the non-exposed group compared with the exposed group. However, in both right and left parietal ROIs, the exposed group demonstrated a greater ROI percent activation than the nonexposed group. These differences did not reach a statistical significance. In the a priori anatomical ROIs, Brodmann’s regions 10 and 46, a comparison of the average percent ROI activation for non-exposed versus exposed groups showed that, although both groups increased activation from the (1-b)-(x) to the (2-b)-(x), the level of activation in the non-exposed group was proportionally greater. A 1-tailed t test for the (2b)-(1b) condition between the non-exposed group and exposed group in the left hemisphere yielded a borderline significant value of P ⫽ .05. Although it is difficult to assess the clinical significance of the differences seen in activation, especially lacking a significant group performance difference, we can nevertheless use the estimated proportions activated in both functional and anatomical ROI to compute approximate post hoc observed effect sizes. For each area of interest, we computed task (1-back versus 2-back) effect size for each group (cocaineexposed and non-exposed) and then computed the difference in group effect sizes. For example, in the cingulate, the 2-back versus 1-back effect size was 0.9 SDs for the non-exposed group and 0.66 SDs for the exposed group, both large effect sizes. The group difference in effect size was 0.25 (a small effect). Results for the right and left DLPFC, parietal area, and Brodmann’s areas were similar, reflecting a large change in activation from the 1-back to the 2-back within both groups, which is relatively easy to detect, but a smaller difference in the groups.
DISCUSSION In this study, cocaine exposed and non-exposed innercity adolescents performed similarly on a non-spatial working memory task and demonstrated similar activation patterns during fMRI. We did not find differences in the behavioral results between exposed and non-exposed participants during administration of a non-spatial working memory task. The task was an appropriate one in that there was the predicted increase in difficulty when subjects first encountered the 2-back task after having performed the more simple x-detection and 1-back tasks. During the 2-back task, the correct responses to both targets (that is, subjects were less likely to correctly identify the target and perform the button press) and non-targets (that is, subjects were more likely to press the button in the absence of the 2-back target) decreased. When runs 1 and 2 (1-back task) were compared with runs 3 and 4 (2-back task), we found significant decreases in total correct responses (Figure 2) by both exposed and non-exposed participants.
Thus, for this working memory task, performance by exposed and non-exposed participants was similar. This occurred despite evidence of heavy gestational exposure: a median level of 117 days of cocaine exposure, and 81% of maternal urine sample test results were positive at delivery. In an investigation suggesting worse neurodevelopmental outcome with the heavy exposure, heavily exposed subjects were defined as having prenatal exposure for only 61 days. Whether the similar behavioral performance was associated with similar or dissimilar patterns of brain activation was examined with fMRI. A particular point of interest, in light of subjects’ similar performance on the behavioral task, was whether tasks were accomplished by using compensatory neural mechanisms, as would be evidenced through different activation patterns. Map-wise comparisons of both exposed and non-exposed subjects confirmed activation patterns in expected regions of cingulate, DLPFC, and parietal lobe,27 demonstrating the validity and reliability of the n-back task as applied to our population. Group statistical maps in both exposed and non-exposed groups showed greater activation with increased task complexity, particularly in bilateral DLPFC regions. The most robust activation occurred in expected directions during the 2-back condition, which requires the greatest working memory demands, as evidenced by performance results. Our results are consistent with earlier literature on functional activation during n-back tasks.17,30-32 In particular, Casey,17 reporting on a small healthy control sample of 9- to 11-year-old children with the n-back working memory task, observed activation of inferior and middle frontal gyri (Brodmann’s 46 and 10). In that sample of young subjects, activation patterns also increased in expected directions with greater task complexity and also correlated with behavioral performance. Thus, the imaging and behavioral performance results of the present study are consistent with Casey et al, in which activation increased and response accuracy significantly decreased across subjects with increasing task complexity. Earlier studies indicate that increasing the numbers “back” correspondingly increases the degree of activation. Therefore, decreasing performance in response to increased working memory load produces an increase in brain activation.31,33 However, the work of Callicott et al indicates that certain load-sensitive loci within the working memory network demonstrate capacity-limited response. The beta map for the non-exposed group for the 2-back minus 1-back condition demonstrates a greater degree of activation than the exposed group, despite analogous behavioral performance (Figure 3). This would suggest that the exposed cohort experiences a memory capacity breach at a lower difficulty level than the non-exposed cohort. This effect is seen particularly in the prefrontal cortex, which is analogous to the results of Calicott et al.31 Although a similar pattern of reduced activation with normal behavioral performance is seen in patients with schizophrenia,33 various other patterns also have been documented in developmental disorders. For example, adults with autism spectrum disorder who have similar behavioral performance as control subjects show increased brain activa-
Functional Magnetic Resonance Imaging and Working Memory in Adolescents with Gestational Cocaine Exposure
375
tion in tests of executive function.34 Further, medication naïve adolescents with attention deficit hyperactivity disorder showed reduced brain activation independent of task performance during an initiation and error detection task, findings felt to confirm a relationship between behavioral impulsivity and neural abnormalities.35 Our results are also consistent with studies that examine the relationship between frontal lobe mediated information maintenance and manipulation with time, central to the nback working memory task. D’Esposito et al36 contend that the DLPFC is necessary for working memory and both maintenance and manipulation tasks activate ventral and DLPFC regions. To further explore the relationship between information maintenance and manipulation in the broader context of central executive processing in normal subjects, Ragland et al32 used an n-back task to define the brain regions that activate in distinct maintenance or manipulation conditions. During a 1-back minus 0-back task designed to isolate information maintenance, they observed inferior parietal and DLPFC activation. During a 2-back minus 0-back task designed to assess the cumulative effect of maintenance and manipulation, they reported activation in inferior parietal regions and dorsal and ventrolateral PFC regions. During a 2-back minus 1-back “manipulation only” task designed to remove the maintenance demands across conditions, broader DLPFC and anterior cingulate activations were observed. The results of both studies, with the exception of observed activations in ventrolateral PFC, are consistent with this investigation. We found increasing DLPFC activation with task difficulty, and the emergence of anterior cingulate activation in the 2-back minus x detection condition, suggesting a similar role for these regions in working memory function in our cohort. Thus, although our study shows anticipated patterns of activation, there are no differences between exposed and nonexposed participants. Possible explanations for our null results are several. It may be that differences do exist, but that the number of subjects was too small (although our study is a large study by typical fMRI standards) to detect such differences. It may well be that investigations with a larger number of subjects will detect differences between exposed and nonexposed groups. Another possibility is that an enriched environment for exposed subjects placed out of the home could ameliorate effects of cocaine exposure; in our sample, however, there was no difference in out-of-home placement in the 2 groups. Is it possible that some of our non-exposed participants were, in fact, exposed to cocaine? This is possible, but doubtful because we had negative history from mothers, negative urine test samples for both mothers and babies, and a maternal profile (cocaine-using mothers were older, had less prenatal care, had more sexually transmitted diseases, and were more likely to be polysubstance users) of cocaine-using mothers that differed from non-using mothers.20 Perhaps the more compelling question is why, given our exposed subjects’ greater exposure not only to cocaine but also alcohol, cigarettes, and marijuana, did we not detect differences between 376
Hurt et al
exposed and non-exposed groups? It may be that fMRI, dependent on the BOLD signal from blood flow, simply is not sensitive enough to detect changes that do exist, or that recently described cerebral blood flow differences in our sample between exposed and non-exposed groups influenced results.37 However, perhaps any effects of gestational exposure present early on have been remediated through the maturational processes that normally occur during childhood and adolescence,38 or there may be characteristics unique to our sample that are responsible for null results. Finally, gestational cocaine exposure simply may be less injurious than predicted. In this regard, there are a number of studies of global intellectual functioning in exposed children at young ages showing these children are not severely and irreparably damaged as suggested by some early reports.39 In our sample, as described here, so far we have not found consistent differences between exposed and non-exposed in executive function.10,18,40 In some investigations in older subjects, however, differences in executive functions such as attention and arousal have been reported. Thus, the full impact of effects of gestational cocaine exposure, if any, remains to be determined. In conclusion, with a working memory task in this sample of low socioeconomic status subjects, exposed and non-exposed groups had similar task performance and brain activation patterns. These data provide a basis for future investigations that may explore subject groups of differing SES, IQ, and ethnicity.
REFERENCES 1. Mayes LC. Developing brain and in utero cocaine exposure: effects on neural ontogeny. Dev Psychopathol 1999;11:685-714. 2. Harvey JA. Cocaine effects on the developing brain: current status. Neurosci Biobehav Rev 2004;27:751-64. 3. Mayes LC, Fahy T. Prenatal drug exposure and cognitive development. In: Sternberg RJ, Grigorenko EL, editors. Environmental effects on cognitive abilities. Mahwah, New Jersey: Lawrence Erlbaum Associates; 2001. p. 189-219. 4. Struthers JM, Hansen RL. Visual recognition memory in drug-exposed infants. J Dev Behav Pediatr 1992;13:108-11. 5. Jacobson SW, Jacobson JL, Sokol RJ, Martier SS, Chiodo LM. New evidence for neurobehavioral effects of in utero cocaine exposure. J Pediatr 1996;129:581-90. 6. Mayes LC, Bornstein MH, Chawarska K, Haynes OM, Granger RH. Impaired regulation of arousal in three-month-old infants exposed prenatally to cocaine and other drugs. Dev Psychopathol 1996;8:29-42. 7. Noland JS, Singer LT, Short EJ, Minnes S, Arendt RE, Kirchner HL, et al. Prenatal drug exposure and selective attention in preschoolers. Neurotoxicol Teratol 2005;27:429-38. 8. Richardson GA, Conroy ML, Day NL. Prenatal cocaine exposure: effects on the development of school-age children. Neurotoxicol Teratol 1996;18:627-34. 9. Bandstra ES, Morrow CE, Anthony JC, Accornero VH, Fried PA. Longitudinal investigation of task persistence and sustained attention in children with prenatal cocaine exposure. Neurotoxicol Teratol 2001;23:545-59. 10. Savage J, Brodsky NL, Malmud E, Giannetta JM, Hurt H. Attentional functioning and impulse control in cocaine-exposed and control children at age ten years. J Dev Behav Pediatr 2005;26:42-7. 11. Schroder MD, Snyder PJ, Sielski I, Mayes L. Impaired performance of children exposed in utero to cocaine on a novel test of visuospatial working memory. Brain Cogn 2004;55:409-12. 12. Mayes LC, Molfese DL, Key AP, Hunter NC. Event-related potentials in cocaine-exposed children during a Stroop task. Neurotoxicol Teratol 2005;27:797-813. 13. Linares TJ, Singer LT, Kirchner HL, Short EJ, Min MO, Hussey P, et al. Mental health outcomes of cocaine-exposed children at 6 years of age. J Pediatr Psychol 2006;31:85-97. 14. Sheinkopf SJ, Lester B, Eliassen JC, Sanes JN, Hutchinson ER, Seifer R, et al. Functional MRI and prenatal cocaine exposure. Pediatric Research. San Francisco, CA:2004;3295A.
The Journal of Pediatrics • March 2008
15. Warner TD, Behnke M, Eyler FD, Padgett K, Leonard CM, Hou W, et al. Diffusion tensor imaging of frontal white matter and executive functioning in cocaineexposed children. Pediatrics 2006;118:2014-24. 16. Smith LM, Chang L, Yonekura ML, Gilbride K, Kuo J, Poland RE, et al. Brain proton magnetic resonance spectroscopy and imaging in children exposed to cocaine in utero. Pediatrics⬘ 2001;107:227-31. 17. Casey BJ, Cohen JD, Jezzard P, Turner R, Noll DC, Trainor RJ, et al. Activation of prefrontal cortex in children during a nonspatial working memory task with functional MRI. Neuroimage 1995;2:221-9. 18. Betancourt LM, Brodksy NL, Malmud E, Giannetta JG, Farah M, Hurt H, et al. Does gestational cocaine exposure affect child neurocognitive outcome? PAS 2005; 57:1652 Platform Session APS/SPR Meeting, May 16, 2005. Available at: http:// www.abstracts2view.com/pas/. 19. Bush G, Frazier JA, Rauch SL, Seidman LJ, Whalen PJ, Jenike MA, et al. Anterior cingulate cortex dysfunction in attention-deficit/hyperactivity disorder revealed by fMRI and the Counting Stroop. Biol Psychiatry 1999;45:1542-52. 20. Hurt H, Brodsky NL, Braitman LE, Giannetta J. Natal status of infants of cocaine users and control subjects: a prospective comparison. J Perinatol 1995;15:297-304. 21. Worsley KJ, Friston KJ. Analysis of fMRI time-series revisited—again. Neuroimage 1995;2:173-81. 22. Aguirre GK, Zarahn E, D’Esposito M. Empirical analyses of BOLD fMRI statistics. II. Spatially smoothed data collected under null-hypothesis and experimental conditions. Neuroimage 1997;2:199-212. 23. Aguirre GK, Zarahn E, D’Esposito M. The variability of human, BOLD hemodynamic responses. Neuroimage 1998;8:360-9. 24. Collins D. 3D model-based segmentation of individual brain structures from magnetic resonance imaging data. Montreal: McGill University; 1994. 25. Nichols TE, Holmes AP. Nonparametric permutation tests for functional neuroimaging: a primer with examples. Hum Brain Mapp 2002;15:1-25. 26. Maldjian JA, Laurienti PJ, Kraft RA, Burdette JH. An automated method for neuroanatomic and cytoarchitectonic atlas-based interrogation of fMRI data sets. Neuroimage 2003;19:1233-9. 27. Owen AM, McMillan KM, Laird AR, Bullmore E. N-back working memory paradigm: a meta-analysis of normative functional neuroimaging studies. Hum Brain Mapp 2005;25:46-59. 28. Rabin M, Narayan V, Kimberg DY, Casasanto D, Glosser G, Tracy J. Functional
MRI predicts post-surgical memory following temporal lobectomy. Brain 2004;127: 2286-98. 29. Zuckerman B, Frank DA, Hingson R, Amaro H, Levenson SM, Kayne H, et al. Effects of maternal marijuana and cocaine use on fetal growth. N Engl J Med 1989;320:762-8. 30. Goldberg TE, Weinberger DR. Effects of neuroleptic medications on the cognition of patients with schizophrenia: a review of recent studies. J Clin Psychiatry 1996;57 Suppl 9:62-5. 31. Callicott JH, Mattay VS, Bertolino A, Finn K, Coppola R, Frank JA, et al. Physiological characteristics of capacity constraints in working memory as revealed by functional MRI. Cereb Cortex 1999;9:20-6. 32. Ragland JD, Turetsky BI, Gur RC, Gunning-Dixon F, Turner T, Schroeder L, et al. Working memory for complex figures: an fMRI comparison of letter and fractal n-back tasks. Neuropsychology 2002;16:370-9. 33. Weinberger DR, Mattay V, Callicott J, Kotrla K, Santha A, van Gelderen P, et al. fMRI applications in schizophrenia research. Neuroimage 1996;4:S118-26. 34. Schmitz N, Rubia K, Daly E, Smith A, Williams S, Murphy DG. Neural correlates of executive function in autistic spectrum disorders. Biol Psychiatry 2006;59:7-16. 35. Rubia K, Smith AB, Brammer MJ, Toone B, Taylor E. Abnormal brain activation during inhibition and error detection in medication-naive adolescents with ADHD. Am J Psychiatry 2005;162:1067-75. 36. D’Esposito M, Postle BR, Ballard D, Lease J. Maintenance versus manipulation of information held in working memory: an event-related fMRI study. Brain Cogn 1999;41:66-86. 37. Rao H, Wang J, Giannetta JM, Korcykowski M, Shera D, Avants B, et al. Altered resting cerebral blood flow in adolescents with in-utero cocaine exposure revealed by perfusion functional magnetic resonance imaging. Pediatrics. In press, 2006. 38. Rubia K, Overmeyer S, Taylor E, Brammer M, Williams SC, Simmons A, et al. Functional frontalisation with age: mapping neurodevelopmental trajectories with fMRI. Neurosci Biobehav Rev 2000;24:13-9. 39. Odom-Winn D, Dunagan DE. “Crack kids” in school. What to do. How to do it. Educational Activities 1991;54:54. 40. Betancourt L, Fischer R, Giannetta J, Malmud E, Brodsky NL, Hurt H. Problemsolving ability of inner-city children with and without in utero cocaine exposure. J Dev Behav Pediatr 1999;20:418-24.
Functional Magnetic Resonance Imaging and Working Memory in Adolescents with Gestational Cocaine Exposure
377