We have previously reported smaller hippocampal volume and deficits in short-term memory in patients with combat-related PTSD relative to comparison subjects. The purpose of this study was to compare hippocampal volume in adult survivors of childhood abuse to matched controls. Magnetic resonance imaging (MRI) was used to measure volume of the hippocampus in adult survivors of childhood abuse (N=17) and healthy subjects (N=17) matched on a case-by-case basis for age, sex, race, handedness, years of education, body size, and years of alcohol abuse. All patients met criteria for posttraumatic stress disorder (PTSD) secondary to childhood abuse. PTSD patients had a 12% smaller left hippocampal volume relative to the matched controls (p<0.05), without smaller volumes of comparison regions (amygdala, caudate and temporal lobe). The findings were significant after controlling for alcohol, age, and eduction, with multiple linear regression. These findings suggest that a decrease in left hippocampal volume is associated with abuse-related PTSD.

Childhood physical and sexual abuse is now recognized as a public health problem of enormous magnitude, with rates of sexual abuse estimated to be from 11%-62% in women and 3%-39% in men (Finkelhor, 1979). Delayed recall of childhood abuse is currently a topic of considerable controversy (Loftus et al., 1994; Williams, 1994a; Williams, 1994b). A central issue in this debate is whether the lack of recall of childhood abuse experiences is due to "normal forgetting" (Loftus et al., 1994) or to a special type of forgetting, which has been described by clinicians as amnesia or active repression (Williams, 1994a; Williams, 1994b). One possibility which has received relatively little attention to date is that stress results in alterations in memory in victims of abuse which are not seen in the non-abused general population.

Preclinical and clinical evidence in fact suggests that stress has long-term effects on memory function and brain regions involved in memory (Pitman, 1989; Bremner et al., 1995a). High levels of glucocorticoids seen during stress have been associated with a loss of neurons and a decrease in dendritic branching in the hippocampus (Uno et al., 1989; Sapolksy et al., 1990) with associated deficits in memory function (Luine et al., 1994).

Studies in human subjects with a history of exposure to traumatic stress related to being a prisoner of war in Korea (Sutker et al., 1991) as well as studies in patients with PTSD related to Vietnam combat (Bremner et al., 1993; Uddo et al., 1993; Yehuda et al., 1995) have shown deficits in short term verbal memory relative to controls. We have also found smaller right hippocampal volume as measured with magnetic resonance imaging (MRI) with associated deficits in verbal memory in Vietnam combat veterans with PTSD in comparison to matched healthy controls (Bremner et al., 1995b). Some studies have shown a relationship between childhood abuse (Lewis et al., 1979) or the stress of civil war (Saigh et al., 1995) in children and adolescents and cognitive deficits measured with IQ (Lewis et al., 1979) and academic achievment tests (Saigh et al., 1995). We have recently reported deficits in verbal short-term memory measured with the Wechsler Memory Scale in adult survivors of childhood abuse identified with the Early Trauma Inventory (Bremner et al., 1995c). The purpose of this study was to use MRI to measure hippocampal volume and comparison brain structures in adult survivors of childhood abuse and healthy controls who were matched on a case-by-case basis for factors which could affect volume of the hippocampus. We hypothesized that patients with a history of childhood abuse would have smaller volume of the hippocampus, but not of other comparison brain structures, in comparison to their matched controls.

Methods:

Subject Selection:

The patient group included 17 adult survivors of severe childhood physical and/or sexual abuse who were admitted to the inpatient and outpatient psychiatric services of the West Haven VAMC over a ten month period, who met DSMIIIR criteria for a psychiatric disorder based on the Schedule for Affective Disorders and Schizophrenia (SADS-L) or psychiatric interview, and who did not have a history of combat exposure. All subjects in this study (both patients and controls) gave written informed consent for participation.

Patients were excluded with a history of meningitis, traumatic brain injury, neurological disorder, loss of consciousness of greater than 10 minutes, HIV positive status, current alcohol or substance abuse or lifetime schizophrenia based on the SADS-L, or shrapnel or other foreign bodies which would preclude MRI scanning. Twenty-two patients met criteria for the study, and 17 completed the study. SADS-L data was available in 14/17 of the childhood abuse patients (Table 1). All of the patients met criteria for posttraumatic stress disorder (PTSD) with the stressor being childhood abuse. The three patients for whom the SADS-L was not available met criteria for PTSD based on psychiatric interview.

Comparison subjects were matched a priori on a case-by-case basis with the patients to be the same sex, race, and handedness, to be within five years of age, within two years of education, and within five years of alcohol abuse. The patient and control groups were also matched so that there would be no mean difference in height or weight. Years of alcohol abuse were measured with the Addiction Severity Index (ASI) interview (McClellan et al., 1985), using a procedure described in previous publications (Bremner et al., 1993; Bremner et al., 1995b; Bremner et al., 1995c). The rationale, procedure, and strategy for controlling for potential biases related to alcohol and substance abuse is described in detail in previous publications (Bremner et al., 1995b). Comparison subjects were excluded with a history of psychiatric disorder based on psychiatric interview as well as the other exclusion criteria outlined above for the patients. Neuropsychological testing of IQ and memory was performed using the Wechsler Adult Intelligence Scale and the Wechsler Memory Scale as described in detail in previous publications (Bremner et al., 1993; Bremner et al., 1995b;

Bremner et al., 1995c). These data have been reported in a previous publication (Bremner et al., 1995c). There were no differences between patients and controls in any of the demographic variables measured (Table 2). Childhood abuse was assessed with the Early Trauma Inventory (ETI), as described in detail in another publication (Bremner et al., 1995c). We have developed specific criteria for severe abuse based on the ETI interview to identify a sample of subjects in which there is no question that abuse has occurred. Subjects were identified as having severe abuse who were exposed to physical abuse involving being hit with an object, burned, or locked in a closet, or penetrative sexual abuse, which occurred once a month or more for at least a year, which had an extremely negative effect on the individual at the time the event occurred, and which has an extremely negative effect currently emotionally or on social or occupational functioning. A composite index of abuse severity was determined based on the ETI as previously described (Bremner et al., 1995c). Patients in this sample, as is to be expected considering the selection criteria for the study, experienced high levels of physical and sexual abuse. The range of symptoms have been reported elsewhere (Bremner et al., 1995c).

MRI Acquisition, Image Processing, and Analysis:

Magnetic resonance images (MRI) were obtained using a protocol of 3 mm contiguous slices on a 1.5 Tesla General Electric Signa device, with a spoiled GRASS (gradient recall acquisition in the steady state) sequence with TR=25 msec, TE=5 msec, NEX (number of excitations)=2, matrix 256 X 256, field of view=16 cm. Images were transferred through computer network to a Sun Sparc10 Workstation, where volumetric measurements were performed using methods previously described (Bronen & Cheung, 1991; Bremner et al., 1995b) and the ANALYZE program (Mayo Foundation, Rochester, Minn.).

Measurements of a mid-hippocampal segment were performed independently by two investigators (J.D.B. and E.V.) who were blinded to subject diagnosis using methods previously described in detail (Bremner et al., 1995b). The mid-hippocampal segment included five coronal sections (15 mm) between the superior colliculus and the bifurcation of the basilar artery, with the first slice anterior to the superior colliculus (Bronen & Cheung, 1991). Volume of mid-hippocampal body measured using this method has been shown to correlate with whole volume measurements (Kim et al., 1994). The mean of the two raters was obtained for the final value of hippocampal volume. In three cases there was a greater than 20% discrepancy in measurements between the two operators. These scans were blindly reexamined and a consensus measurement was performed. Volumetric assessments were also made of three other regions for purposes of comparison, the temporal lobe, caudate, and amygdala. Methodology and rationale for measurement of temporallobe and caudate have been described in detail previously (Bremner et al., 1995b). Volume of the amygdala was determined by measuring cross sectional area of the amygdala on the non-resliced MRI in all slices anterior to the bifurcation of the basilar artery (inclusive of the slice in which the bifurcation of the basilar artery was visualized) (Watson et al., 1992), summing cross sectional areas, and multiplying by the slice thickness. Volumes reported for temporal lobe, amygdala and caudate are from measurements performed by a single rater (J.D.B.). We controlled for differences in brain size by matching patients with a comparison group of similar height and weight (Arndt et al., 1991)

Inter-rater reliability determined for the measurements in this study using the intraclass correlation coefficient with one-way ANOVA (Bartko, 1966) (with values of the coefficient approaching one representing a high level of agreement between two raters) were excellent: left hippocampus ICC=0.61 (F=4.13; df=33,34; p<0.01), right hippocampus ICC=0.79 (F=8.54; df=33,34; p<0.01), left amygdala ICC=0.56 (F=3.56; df=33,34; p<0.01), right amygdala ICC=0.56 (F=3.55; df=33,34; p<0.01). We have previously reported excellent inter-rater reliability for hippocampus, temporal lobe and caudate (Bremner et al., 1995b).

Data Analysis:

Repeated measures analysis of variance (ANOVA) with side (left vs right) as the repeated measure was performed to compare left and right hippocampal volume (as well as left and right caudate and left and right temporal lobe) between patients and controls. Multiple linear regression was used to examine the relationship between hippocampal volume and diagnosis while controlling for other variables including alcohol abuse, education, and age. Two-tailed nonpaired t-tests were used to compare hippocampal volume in PTSD patients with and without comorbid depression and alcohol/substance abuse.

 Results:

Hippocampal Volume:

Repeated measures ANOVA with hemisphere (left vs right) as the repeated factor did not show a significant difference between patients and controls when left and right hippocampal volume were combined in a single model (Table 3). There was not a significant main effect for side (left vs right hippocampal volume) (F=2.15; df=1,32; p=0.15), although there was a significant interaction between side and diagnosis (F=4.19; df=1,32; p=0.049).

Childhood abuse patients had a 12% smaller left hippocampal volume than controls, which was statistically significant by univariate analysis (p<0.01) (Fig. 1). A 5% reduction in volume of the right hippocampus was not significant (Table 3).

Volume of Comparison Regions:

Repeated measures ANOVA with side (left vs right temporal lobe volume) as the repeated factor did not show a difference in temporal lobe volume between patients and controls (ie no main effect for diagnosis) (Table 4). There was a significant main effect for side (left vs right temporal lobe volume) (F=19.53; df=1,32; p=0.0001), but no side by diagnosis interaction, which (consistent with our previous reports as well as reports from other groups) suggests a relative increase in right temporal volume in both groups relative to left. Univariate analyses showed that the childhood abuse patients had a greater volume of the left temporal lobe, but not right temporal lobe, in comparison to the matched controls (Table 4). Repeated measures ANOVA with side (left vs right caudate volume) as the repeated factor showed no difference in caudate volume between patients and controls (ie no main effect for diagnosis) no main effect for hemisphere, or side by diagnosis interaction. Univariate analyses did not show a difference for left or right caudate between childhood abuse patients and comparison subjects when examined alone. Repeated measures ANOVA with side (left vs right amygdala volume) as the repeated factor showed no difference in amygdala volume between patients and controls (ie no main effect for diagnosis), and no main effect for hemisphere. In addition, there was no side by diagnosis interaction (F=3.66; df=1,32; p=0.06). Univariate analyses did not show a difference for left or right amygdala volume between childhood abuse patients and controls when examined alone (Table 4).

We did not find a relationship between volume of the hippocampus and volumes of the caudate and temporal lobe in either the patients or the controls. There was a significant correlation within the childhood abuse patient group between mean temporal lobe and caudate (r=0.79; df=16; p=0.0001) which was not seen within the controls. This correlation was seen on both the left and the right sides within the childhood abuse patient group.

Hippocampal Volume and Memory:

We have reported in a companion publication deficits in short-term verbal memory as measured with the Wechsler Memory Scale-Logical component, for immediate and delayed recall and percent retention, in a slightly large sample of childhood abuse patients which included all of the patients in this study (Bremner et al., 1995c). There were no deficits in visual memory or IQ. In the current study there was no correlation between scores on the Wechsler Memory Scale (verbal and visual memory components) and bilateral hippocampal, amygdala, caudate or temporal volumes in either patients or controls. Within the childhood abuse patient group (but not the controls) there was a significant correlation between performance IQ measured with the WAIS-R and mean amygdala volume (r=0.64; df=14; p=0.008) and performance IQ and mean caudate volume (r=0.55; df=14; p=0.03), which was seen for both left and right amygdala volumes and left and right caudate volumes, respectively, but not for left or right hippocampus or temporal lobe. There was no relationship between verbal IQ and volume of any brain region in the childhood abuse patient group. There was no relationship between IQ scores and volume of any brain region in the controls.

Relationship between Hippocampal Volume and Demographic and Clinical Factors:

There was no significant correlation between hippocampal volume and level of childhood physical, sexual or emotional abuse measured with the ETI in the childhood abuse patients. The strongest relationship was for volume of the left hippocampus and sexual abuse (r=-0.23; df=16; p=0.23). There was a correlation, however, between emotional abuse measured with the ETI and mean amygdala volume (r=-0.58; df=16; p=0.01) which was due to an equal contribution from left and right amygdala volume. There was no relationship between abuse severity and volume of other comparison regions in this study.

There was no difference in left hippocampal volume between PTSD patients with (N=10) and without (N=14) a lifetime history of alcohol dependence (mean=1054, SD=175 vs mean=1008, SD=104), with (N=5) and without (N=9) a lifetime history of marijuana dependence (mean=1007, SD=144 vs mean=1060, SD=167), with (N=4) and without (N=10) a lifetime history of stimulant dependence (mean=1014, SD=149 vs mean=1052, SD=165), with (N=2) and without (N=12) a lifetime history of sedate/hypnotic/anxiolytic dependence (mean=1211, SD=9.9 vs mean=1012, SD=150), with (N=4) and without (N=10) a lifetime history of opiate dependence (mean=1102, SD=128 vs mean=1016, SD=165), or with (N=8) and without (N=6) a lifetime history of cocaine dependence (mean=1003, SD=133 vs mean=1093, SD=181). There was no correlation within the childhood abuse patients between left or right hippocampal volume and years of alcohol, heroin, hallucinogen, cocaine, marijuana, or amphetamine abuse. There was no significant correlation between years of alcohol abuse and left or right hippocampal volume in the group as a whole.

There was no significant correlation between age, years of education, height or weight and left or right hippocampal volume in either patients or controls when examined separately.

Assessment of Potential Confounders Related to the Findings of Smaller Left Hippocampal Volume in Abuse-related PTSD:

This study design involved matching of patients and controls for a number of factors which may affect hippocampal volume. To assess the contribution of potential confounders to the finding of smaller hippocampal volume due to the lack of a perfect matching, we utilized multiple linear regression with factors which may affect hippocampal volume, such as age, years of education, and years of alcohol abuse, as covariates in a model examining the relationship between abuse-related PTSD diagnosis and left hippocampal volume. None of these covariates was significantly related to right hippocampal volume either individually, or when included simultaneously in the same model (p>.25 for all variables). Utilizing a backward selection procedure with a significance level to stay in the model of p<.05, all three variables were eliminated from the model, confirming their nonsignificance. When only diagnosis was in the model predicting left hippocampal volume, the effect size for the model (expressed as R2), was 0.20, and abuse-related PTSD diagnosis was significantly related to left hippocampal volume (t=2.84; p=.0077). When years of alcohol abuse was added to the model, R2 was 0.2, and abuse-related PTSD diagnosis was still related to left hippocampal volume (t=2.52; p=.017), but alcohol was not (t=.42;p=.68). With inclusion of years of education in the model, R2 was 0.2, and abuse-related PTSD diagnosis was still related to left hippocampal volume (t=2.62; p=.013), but years of education was not (t=.06; p=.95). With inclusion of age in the model, R2 was 0.22, and abuse-related PTSD diagnosis was still related to left hippocampal volume (t=2.61; p=0.013), but age was not (t=1.02; p=.31).

We also measured point estimates and confidence intervals for the difference in volume of left hippocampus between patients and controls before and after adjustment for covariates. The difference in left hippocampal volume between the two groups without adjusting for other factors was 12.0% in univariate analyses (1050 vs. 1193 mm3) (95% confidence interval (C.I.) 40-245 mm3). After adjustment for alcohol abuse the point estimate for the difference between the two groups was not materially different, at 11.4% (1187 vs 1283 mm3) (95% C.I. 39-247 mm3). After adjustment for alcohol, education, and age the estimated difference was 10.6% (95% C.I. 37-249 mm3). In summary, abuse- related PTSD diagnosis was related to small left hippocampal volume after controlling for potential confounders of alcohol, education, and age.

Discussion:

Adult survivors of childhood abuse with the diagnosis of posttraumatic stress disorder (PTSD) had a 12% smaller left hippocampal volume in relation to controls matched on a case-by-case basis for age, sex, race, handedness, years of education, socioeconomic status, body size, and years of alcohol abuse. Right hippocampal volume was 5% smaller in the patients than in the controls, which was not statistically significant. None of the comparison regions measured in this study were significantly smaller in the patients in comparison to controls, including volume of the left or right caudate, left or right amygdala, or left or right temporal lobe volume (minus hippocampus and amygdala). There continued to be a significant relationship between left hippocampal volume and abuse-related PTSD after statistically controlling for alcohol, education, and age, with multiple linear regression.

Accumulated evidence from animal studies has shown that stress results in a loss of hippocampal neurons and a decrease in branching of the dendrites of neurons in the hippocampus (Uno et al., 1989; Sapolsky et al., 1990). Some studies suggest that high levels of glucocorticoids, released during stress, increase the vulnerability of these neurons to excitatory amino acid-induced toxicity. Consistent with this, conditions in which there are high levels of circulating glucococorticoids, such as Cushing's Disease (Starkman et al.,1992) have been found to have alterations in the morphology of the hippocampus and other cortical structures also on MRI and CT. One possible explanation for our findings is that a surge of cortisol at the time of stress may have resulted in damage to the hippocampus in the patient group. An alternative explanation is that patients who were born with smaller hippocampal volumes may have been more vulnerable to develop psychopathology in response to childhood abuse. Consistent with this idea, a recent report has found IQ to be predictive of combat-related PTSD symptomatology, although again without premorbid IQ it is not possible to determine whether this is causal or an outcome of stress (McNally & Shin, 1995).

We carefully considered the possible effects of comorbidity with alcohol and substance abuse on hippocampal volume in the childhood abuse patients. Some studies in animals have shown a relationship between neuronal damage and alcohol, although we are not aware of any evidence of neuronal damage related to other substances, or a rationale for why exposure to these substances would be expected to be associated with hippocampal neuronal damage. We therefore elected to match for alcohol only. Our finding of a decrease in left hippocampal volume in patients with childhood abuse and PTSD persisted after controlling for alcohol and substance abuse using several statistical analysis strategies outlined above. These findings should be considered preliminary, however, due to the lack of a perfect matching for alcohol between patients and controls.

We have previously reported a statistically significant 8% smaller right hippocampal volume in patients with a history of combat-related PTSD, and a 4% smaller left hippocampal volume relative to controls, which was not statistically significant. This raises the question of why patients with abuse-related PTSD had smaller left hippocampal volume, while patients with combat-related PTSD had smaller right hippocampal volume, relative to comparison subjects. One possibility is that with very large sample sizes (for example, 100 patients and 100 controls), there would be a finding of a statistically significant smaller hippocampal volume in the patients for both left and right hippocampal volume, due to the increase in power with larger samples. Another possibility is that there is a true difference in patients with early trauma versus trauma later in life. The brain is known to turn during development, a phenomenon known as "torque." A traumatic insult early in development may arrest or delay this normal process of brain rotation. This may result have the consequence that the brain does not "turn into its proper final position", as it were, which leads to an alteration in the brain's normal assymetry. Other authors have hypothesized that alterations in neurodevelopment in patients with, for instance, schizophrenia, which has also been associated with smaller hippocampal volume relative to comparison subjects (Breier et al., 1992; Suddath et al., 1990), may result in a lack of the normal asymmetry of the brain.

We found a relationship between level of emotional abuse as measured with the ETI and volume of the amygdala in the abuse patients, and between performance IQ and volume of the amygdala and caudate in the patients, but not in the controls. The amygdala plays an important role in conditioend fear, and the caudate is involved in memory and the execution of directed tasks. We did not, however, expect these findings, and report them here for the benefit of future investigations. In contrast to studies in combat veterans (Bremner et al., 1995b) we did not find a relationship between deficits in verbal memory and decreased hippocampal volume.

Our findings may have implications for psychotherapeutic treatment of PTSD related to childhood abuse. The hippocampus is hypothesized to play a role in integrating individual aspects of memory (touch, taste, vision, feeling, hearing) at the time of recall. Dysfunction of the hippocampus may underlie the fragmentation of memory which is a common part of the clinical presentation of these patients.