Schizophrenia Patients Lack Normal Positive Correlation Between Age and Brain Response During Verbal Learning

Abstract

Objective

We examined whether there were differences in the relationship of age to functional brain response among individuals with schizophrenia compared those with no history of mental illness. We hypothesized that the correlation with age would be more negative among patients with schizophrenia, particularly in the frontal cortex.

Design

Cross-sectional measures of functional magnetic resonance imaging brain response were correlated with age among patients and comparison participants.

Setting

Patient participants were stable, community-dwelling outpatients who came to our university research facilities for testing.

Participants

We analyzed data from 31 patients with DSM-IV schizophrenia or schizoaffective disorder who ranged in age from 25 to 68 years and 14 healthy comparison participants who ranged in age from 21 to 70 years.

Measurements

Brain response during learning of novel word pairs compared to response during fixation was measured with functional magnetic resonance imaging in each voxel of the brain. This measure was correlated with age within each group and the correlations were compared across groups in a whole-brain analysis and also in regions of interest determined a priori and based on the whole-brain results. Other variables that correlated with age in either group were examined for potential moderating effects.

Results

The correlations between age and brain response were more positive in the healthy group than in the group of schizophrenia patients in several regions, including clusters in midline precuneus and right superior temporal gyrus identified through whole-brain analysis and in an a priori region of interest in right prefrontal cortex. Accounting for duration of illness reduced the group difference between correlations in the frontal region. Learning performance inside and outside the scanner moderated the age relationship in an opposite manner within the two groups, but accounting for performance did not reduce the group differences in age correlations.

Conclusions

Functional brain response during learning does not change significantly with age among schizophrenia patients and lacks the normal positive association with age seen in healthy individuals. There is some evidence that compensatory brain response may be possible among patients to help maintain cognitive performance in the face of normal aging.

As the world’s population ages, the number of elderly mentally ill is also expected to grow tremendously (1). Designing effective treatment and rehabilitation programs for elderly individuals with severe mental illness, such as schizophrenia, will thus become an even greater priority in the coming years (1). In order to tailor interventions for older patients, it will be necessary to understand more about the course of pathology in the disorder, particularly what happens to the brains of individuals with schizophrenia as they age. Furthermore, studies of cerebral age-related changes can help to address fundamental questions about the nature of the disorder, including whether the underlying pathophysiology of schizophrenia is static or deteriorating.

Research to date on brain changes with age in schizophrenia has focused mostly on brain structure and on the brain’s behavioral output, namely cognitive performance and psychiatric symptoms. Patients with schizophrenia exhibit characteristic structural brain abnormalities compared to healthy individuals that include ventricular enlargement and smaller gray matter volume, particularly in frontal and temporal cortex as well as in limbic regions (2). These abnormalities appear to be present early in the course of the disorder (3), but whether they progress with age is more controversial. The most rigorous examination of this issue comes from longitudinal follow-up studies, of which there are relatively few. On the whole, however, these studies suggest that there is progression over time in at least a subset of patients with schizophrenia that outpaces normal age-related declines (4, 5). Weaknesses of these studies include the relatively short duration of follow-up (i.e., 10 years or less), and a general focus on younger age ranges, such that the pattern of age-related change in brain volume among elderly patients is virtually unknown.

Middle-aged and elderly patients have been better represented in the literature focusing on cognitive changes with age. In both cross-sectional and longitudinal studies, cognitive function has been shown to be fairly stable in most patients (6-8), although a small subset of severe, institutionalized individuals may suffer accelerated cognitive aging (9). Symptom course across the life-span also has been examined, with the general result that negative symptoms tend to be stable or worsen with age while positive symptoms improve (8). Thus, against a background of accelerated aging of brain structures, there is behavioral evidence mostly for stability (negative features and cognition) or improvement (positive features). This disconnect between structural brain findings and behavioral outcome is highlighted in at least one longitudinal study of schizophrenia patients that found that individuals with the greatest increase in brain structural abnormality with age paradoxically had better functional outcomes (5).

The link between brain structure and behavioral performance is clearly an indirect one, however. Neural functioning, many aspects of which can be measured in living individuals using functional neuroimaging, likely mediates the relationship of brain volume to behavior. Examination of age-related changes in brain function should help to elucidate the complex interplay between brain structure and behavior and how these elements may or may not change with age in schizophrenia. Although several functional neuroimaging studies have compared the brain function of older patients with schizophrenia to age-matched comparison groups and found abnormalities similar to those seen in younger patients (10-15), only a small number of studies have directly examined age-related change, and limitations of these studies make it difficult to draw definitive conclusions. Table 1 summarizes the results of studies that have examined the relationship of age to brain functioning among patients with schizophrenia. This table includes studies that drew conclusions about progression of brain functioning based on the presented data, but does not encompass those that may have calculated the correlation of age with brain function without reporting on these findings in the paper’s abstract. In general, the reviewed cross-sectional studies have found evidence for negative correlations of hemispheric and regional brain blood flow and metabolism with age, and those that included a control group showed steeper negative slopes among patients in some regions including frontal and temporal cortex and the caudate. In contrast, one study found regions with more positive age-related slopes among patients (16), and the only longitudinal study observed no change in degree of hypofrontality across 18 years (17). Also, it is important to note that the largest study to date, while finding a negative correlation between age and mean hemispheric blood flow also observed a strong relationship of duration of illness to hypofrontality (18). In a follow-up study, the same authors found that the relationship of duration of illness to anterior-posterior blood flow ratio persisted despite controlling for and statistically co-varying age (19).

Table 1.

Studies Examining Relationship of Age to Brain Functioning among Patients with Schizophrenia

Citation	N	Age	Chronicity / Medication Status	Imaging Technique	Task	Analysis	Measure(s)	RESULTS: SZ has negative slope (or more negative than HC)	RESULTS: SZ has positive slope (or more positive than HC)	RESULTS: No age-related differences
Ingvar and Franzen, 1974 (40)	9 older SZ; 11 younger SZ	Old SZ: 61.3 (7.0); range = 49-71; Young SZ: 24.5 (5.2); range = 17-33	Old SZ: Chronically institutionalized; Young SZ: duration of 8 months - 15 years; All medicated (almost all typical antipsychotics)	Xenon injection	Rest vs Test (Old SZ: Object naming; Young SZ: Raven’s progressi ve matrices)	ROI (detectors)	Ratio of Left Frontal rCBF to Left postcentral rCBF	Old SZ more hypofrontal than young SZ at rest; Old SZ show fewer task-related increases than Young SZ
Mathew et al, 1988 (18)	108 SZ; 108 HC	SZ: M=31.5 (8.9) F=38.3 (12.4); HC: M=32.0 (13.1) F=38.9 (13.1)	Chronic (mean duration 11 yrs); 46 were medication withdrawn for 2 weeks, rest on mostly typicals	Xenon inhalation	Rest	ROI (detectors)	Right and left mean CBF and anterior to posterior gradient	Negative slope for age with left and right hemisphere mean CBF; negative slope for duration of illness with left and right anterior to posterior gradient
Mathew and Wilson, 1990 (19)	Long duration SZ: 27; Short duration SZ: 27; HC: 2 matched groups of 27 each	Long duration SZ: 41.8 (10.4); Short duration SZ: 31.4 (12.8); HC-long: 41.5 (11.4); HC-short: 32.4 (13.9)	Medicated inpatients, Long duration: 21.6 (8.5) years; Short duration: 2.4 (1.7) years	Xenon inhalation	Rest	ROI (detectors)	Anterior to posterior gradient	Long duration had lower anterior/posteri or gradients compared to short duration, even when age covaried or controlled for with matched comparisons
Goldstein et al, 1990 (41)	SZ: 15; Major affective: 15; HC: 15	SZ: 34.6 (11.9); Major affective: 36.9 (9.7); HC: 36.2 (12.2)	SZ: duration 2 to 28 yrs, 12/15 medicated; Major affective: duration 5 to 26 years, 14/15 medicated	Xenon inhalation	Rest	ROI (detectors)	Frontal deviation score (frontal minus posterior to sylvian sulcus / total)	Across all individuals, negative slope for frontal, posterior, whole brain, and frontal deviation in CBF
Cantor-Graae et al, 1991 (17)	SZ: 7 (of the 11 young SZ reported in Ingvar & Franzen, 1974)	Time 1: 23.0 (5.3), range = 17-32; Time 2: 41.4 (5.0), range = 34-49	Chronic: 22.9 (2.9) years duration of illness; typical antipsychotics; 5/7 had ECT	Xenon inhalation	Rest, word fluency, WCST and associate d control task	ROI (detectors)	Normalized Left Hemispher e regional CBF (prefrontal, fronto-temporal, temporal, central, parieto-temporal, occipital) and anterior to posterior gradient			Mean left hemisphere CBF and anterior-posterior gradient not significantly different after 18 years; no effect of cumulative medication dose
Siegel et al, 1994 (16)	18 SZ; 22 HC	SZ: 29.6 (7.2); range = 21-45; HC: 28.2 (7.2); range = 19-42	Never medicated; Most first episode	FDG PET	Degrade d stimulus CPT	ROI	Relative and absolute GMR, and Frontal/Occ ipital ratios	Relative GMR and frontal/occipital ratio in medial frontal cortex; Relative GMR in left dorsal hippocampus, precuneus, posterior cingulate, superior colliculus	Relative GMR in lateral temporal cortex and left lateral inferior frontal/occipital ratio
Shihabuddin et al (42)(age results reported in Buschbaum and Hazlett, 1997 (24))	SZ: 18; HC: 24	SZ: 38.5 (14.8), range = 18-65); HC: 37.0 (13.1), range = 21-52	Never medicated (7/18) or medication withdrawn for 12 days-2yrs (11/18);	FDG PET	Verbal list learning and memory task	MRI based anatomical ROIs; cluster-based approach within the caudate ROI	Relative GMR	More negative slope compared to HC in lateral and medial frontal regions, anterior temporal region, and a small region within right caudate
Schultz et al, 2002 (43)	SZ: 49	SZ: 32.6 (8.1), range = 20-51	Chronic (mean duration 10.25 (7.5) years); medication withdrawn for 3 weeks	[15O] water PET	Rest	Voxel-based regression; global CBF	Quantified blood flow	Negative slope in anterior cingulate, bilateral premotor frontal cortex (BA 8), bilateral parietal lobe (BA 40), and for whole brain CBF

The extant studies provide a foundation for examining the complex interplay between age-related changes in brain structure, brain function, and behavioral outcome, but there are many limitations of the literature that need to be addressed. First, with the exception of one large study (18), many of the schizophrenia samples have been small (the median sample size of reviewed studies was 18). Second, despite well-known normal age-related changes in brain function, many studies have not compared the age correlations among patients directly to those of a healthy comparison group, making it difficult to know if observed negative slopes among patients represent accelerated or normative aging. Third, a majority of the existing studies were conducted on patient samples in an era when most patients were taking typical, first-generation antipsychotic medications. Given that these agents may have different effects on brain function than second-generation medications, the relevance of these studies’ results to today’s elderly patients is called into question. Relatedly, several studies examined never-medicated or medication-withdrawn patients, samples that also may be less generalizable to the population of older, chronically-medicated patients. Fourth, many of the reviewed studies examined age relationships within a fairly young sample (average age across studies = 39.3 years), with few elderly individuals included (for those studies that reported an age range, the oldest patient was in the early 50’s for all but one study). Finally, all of the studies measured either resting blood flow or task-related glucose metabolism over a 45 minute uptake period. Cognitive challenge paradigms using ¹⁵O PET or higher-resolution functional magnetic resonance imaging arguably are more sensitive than resting perfusion or metabolic methods to differences between schizophrenia patients and healthy individuals and to normal age-related changes in brain function, yet we could find no cognitive challenge PET or fMRI studies specifically examining age-related change among patients with schizophrenia.

The aim of the present study was to use functional magnetic resonance imaging (fMRI) to examine the relationship between age and brain response during a cognitive challenge task among patients with schizophrenia and compare the relationship to that seen in healthy individuals. Although a longitudinal design would be the most powerful test of differential aging of brain function, the expense and challenges of following patients across a large number of years can be prohibitive. We chose to begin with a cross-sectional, correlational study with the hope that the results might motivate further longitudinal study and focus the hypotheses of such an investigation. We examined the relationship of age to brain response during a verbal learning task, as learning is thought to be an area of particular cognitive weakness among patients (20), and deficits in this ability are related to functional outcome (21). In addition, this task challenges frontal and temporal cortical regions that have been related to schizophrenia pathology (22) and similar tasks have revealed age-related differences in previous studies of both healthy individuals (23) and patients with schizophrenia (24). In the present study, we correlated age with brain response in a sample of 31 community-dwelling patients with schizophrenia who ranged in age from 25 to 68, most of whom had been ill for many years and were currently stably-medicated on atypical antipsychotics. We compared the age relationships of the patients to those of a group of 14 healthy individuals (age range: 21-70) in three ways: 1) identifying clusters of voxels within the brain in which there was a significant relationship of age to brain response in each group and/or a significant difference in correlation between groups, 2) comparing the correlation of age and mean activation among the patients to that of healthy individuals within regions identified in method 1 to be related to age among the healthy individuals, and 3) comparing the correlation of age and mean activation among patients to that among healthy individuals in two a priori regions of interest (ROIs) - the medial frontal and inferior frontal gyrii. We hypothesized that patients would show more negative associations between brain function and age than healthy individuals, particularly in frontal regions.

Methods

Participants

Data from thirty-one patients with schizophrenia were included in the analyses presented in this manuscript. Patients were participants in one of two studies that involved the same fMRI paradigm and similar clinical and cognitive assessments. In these studies, a total of 50 patients were scanned; 6 had unusable data due to technical problems and 13 were excluded on the basis of excessive motion (defined as greater than 1/3 of timepoints identified as motion outliers, and/or high overall motion as determined by the degree of motion correction needed across the scan). The remaining patients had DSM-IV diagnoses of schizophrenia (n=18; 12 paranoid, 2 disorganized, 1 undifferentiated, 1 residual, and 2 data missing) or schizoaffective disorder (n=13) and ranged in age from 25 to 68 years old, with a mean (SD) age of 45.4 (11.1) years. The sample was composed predominantly of men (25 men, 6 women) with a mean (SD) education of 12.7 (2.3) years. Patients were chronically ill with a mean (SD) duration of illness of 21.8 (11.9) years, a mean (SD) score of 12.2 (4.7) on the Positive and Negative Symptom Scale (PANSS; (25)) Positive subscale, 14.0 (5.6) on the Negative subscale, and 24.3 (6.5) on the General subscale. All but one patient (who was taking trifluoperazine) were taking second-generation atypical antipsychotics: risperidone (n=13), olanzapine (n=10), quetiapine (n=2), ziprasidone (n=2), risperidone + olanzapine (n=2), and risperidone + quetiapine (n=1). The patients exhibited mild levels of impairment on several clinical neuropsychological tests, with mean (SD) T-scores of 37.6 (12.4) on Hopkins Verbal Learning Test (26) immediate memory and 34.2 (14.4) on delayed memory, and 40.2 (13.6) on Wisconsin Card Sorting Test (27) perseverative errors. All patients were right-handed, free of current substance use or dependence for the past 6 months, and had no history of neurological disorder or head injury with loss of consciousness greater than 30 minutes. The patients excluded for motion were not significantly different from the included patients on age (48.6 (10.1), t(42) = -0.89, p = 0.38), education (12.8 (1.5), t(42) = -0.08, p = 0.93), sex distribution (10 men, 3 women, χ² (1) = 0.08, p=0.78), diagnostic distribution (13 schizophrenia, 3 schizoaffective, χ² (1) = 1.41, p=0.23) or clinical symptoms (PANSS positive: 13.8 (5.8), t(40) = -0.89, p = 0.38; PANSS negative: 12.9 (7.9), t(40) = 0.50, p = 0.62; PANSS General: 27.6 (6.4), t(40) = -1.5, p =0.14). There was, however, a trend for the excluded patients to have a longer duration of illness (29.3 (13.6), t (42) = -1.8, p = 0.07), and those with large amounts of motion had significantly lower learning scores on both a clinical measure of list learning (Hopkins immediate memory: 28.2 (6.0), t(41) = 2.6, p = 0.01) and a trend toward poorer recall for the word pairs learned during scanning (cued recall accuracy proportion correct: 0.44(0.17), t(42) = 1.7, p=0.09), suggesting that the analyzed sample may not be representative of patients with the greatest levels of cognitive impairment.

Fifteen healthy individuals with no self-reported history of psychiatric illness also were scanned. One was excluded due to excessive motion as defined above. The 14 remaining comparison participants were right-handed and had no history of substance abuse, substance dependence, neurological disorder, or head injury with loss of consciousness greater than 30 minutes. The comparison group ranged in age from 21 to 70 years and did not differ significantly from the patient group in age (44.0 (15.2), t(43) = -0.351, p = 0.73) or sex distribution (10 men, 4 women; χ² (1) = 0.47, p=0.49). The comparison group was significantly more educated than the schizophrenia group (14.07 (2.1) years; Mann-Whitney U (45) = 121, p = 0.015).

All participants gave informed consent to the protocol, which was approved by the UCSD Institutional Review Board. Patients engaged in an enhanced consent procedure with computerized slide presentation of consent information, and capacity to consent was assessed with the MacArthur Competency Assessment Tool for Clinical Research (28). Data from 17 of the patients and all 14 healthy individuals have been presented in a previous report comparing brain response of the two groups (22), and data from 11 of the same patients were also reported in a manuscript examining the relationship of informed consent understanding to brain response (29).

Scanning Procedure

Details of the cognitive challenge task and functional magnetic resonance scanning protocol have been previously published (22). Briefly, the blood oxygen level dependent (BOLD) response in each 4mm³ voxel of the brain was measured with echoplanar imaging during a verbal paired-associates learning task. Brain response was measured during blocks of intentional learning of pairs of concrete, associated nouns and during blocks of fixation trials. To ensure orientation to the stimuli during scanning, participants were required to press a button to indicate which of the two words was capitalized. After scanning, a cued recall test was used to measure the degree of word pair learning. High-resolution anatomical images also were collected to allow for localization of the BOLD responses.

Data Analysis

Individual maps of the magnitude of brain response during novel word pair learning compared to fixation were made as previously described. These maps were then spatially blurred with a 8mm full width at half maximum Gaussian filter and transformed into standard atlas space (30). Three approaches to comparing the relationship of age to brain response between groups were undertaken. First, in a whole-brain, voxel-based approach, the correlation of age with brain response in each voxel was calculated for both groups. In addition, the fit coefficient for the interaction of age and group was also calculated to create a map of differential correlations between patients and healthy individuals. These three maps were then subjected to thresholding and clustering such that only those regions in which the correlation was significant at the p=0.025 level (schizophrenia r ≥ 0.40, t(30) ≥ 2.364, healthy r ≥ 0.59, t(13) ≥ 2.561, age × group r≥ 0.33, t(44) ≥ 2.327) in all voxels within a cluster volume ≥ 4736 mm³ (74 voxels) were considered to be age-related. This cluster-threshold combination was shown by Monte Carlo simulation to protect a whole-brain probability of false positives of p < 0.05. In the second approach, masks were created from the clusters identified as age-related in the healthy comparison group. The mean brain response of each participant within each “normal age-related region of interest” was calculated and then correlated with age separately for each group. The magnitude of the comparison group and patient group correlations were compared with Fisher’s z-tests. Finally, in the third approach, we calculated the mean brain response of each individual in two bilateral anatomical regions of interest (Figure 1). The medial frontal gyrus and inferior frontal gyrus were selected based on their known involvement in learning tasks and known deficits in brain response among patients with schizophrenia. In addition, these regions have been shown to be sensitive to normal aging and related to age among patients with schizophrenia. The masks were created in standard atlas space based on the Talairach Daemon software (31). Mean brain response in each hemisphere of the medial frontal and inferior frontal gyrus regions of interest (ROIs) was correlated with age in each group. The magnitude of the comparison group and patient group correlations were compared with Fisher’s z-tests. Variables found to be related to age were examined as potential moderators of observed correlations between age and brain response. A regression model using mean-centered variables and including age, the potential moderator, and the interaction of the two as predictors was tested. Furthermore, we examined whether group differences in correlations were influenced by potential moderator variables by calculating the partial correlation of age with brain response taking the moderator into account. This partial correlation was then compared to the first-order age correlations to see if the significant difference between groups remained after partialling out the contribution of the potential moderating variable.

Figure 1.

Location of two bilateral a priori regions of interest, shown in color overlaid onto axial slices through an average anatomical image in standard, Talairach & Tournoux atlas space. Right and left inferior frontal gyrii regions are shown in yellow and light blue, respectively; Right and left middle frontal gyrii regions are shown in red and dark blue, respectively.

All variables were tested for normality and several were found to have skewed or kurtotic distributions for one or both groups. Pearson’s correlations and Student’s t were utilized for normally-distributed variables, while Spearman’s correlations and Mann-Whitney U were employed for the others. All significance tests were two-tailed and correlations were considered significant at the p<0.05 level.

Results

Behavioral Performance

Both groups performed well on the capitalization judgment task during scanning and did not differ significantly on either accuracy (proportion correct: schizophrenia: 0.96 (0.05), healthy: 0.97 (0.04), Mann Whitney U (43) = 157.5, p = 0.23) or reaction time (time in ms: schizophrenia: 1666 (571), healthy: 1430 (439), Mann Whitney U (42) = 152, p = 0.32). Cued recall of the novel paired associates presented during scanning was equivalently poor in both groups (proportion correct: schizophrenia: 0.57 (0.22), healthy: 0.57 (0.24), t(43) = -0.129, p=0.90) and reaction time to recall the words was similar in patients and comparison individuals (time in ms: schizophrenia: 3902 (1230), healthy: 4387 (1363), Mann-Whitney U (45) = 171, p = 0.26).

Relationship of Clinical, Cognitive, and Other Demographic Variables to Age: In order to identify potential moderators of any observed relationships between age and brain function, we correlated age with behavioral performance, clinical and cognitive measures, and other demographic variables within each group. These associations are presented in Table 2. Education was not significantly related to age in either group. Recall accuracy was negatively related to age within the healthy group, but the negative relationship among the patients was not significantly different from zero. Among patients, age was negatively related to learning and to delayed recall (at a trend level) on a standard neuropsychological measure, but there was no significant relationship with age for the measure of executive function. Duration of illness was highly related to age among the patient group. Age was positively associated with negative symptoms, but not positive symptoms or general psychopathology.

Table 2.

Correlations between Age and other Demographic, Clinical and Performance Measures

	Age
	HC	SZ
Educationa	-0.14	0.27
Capitalization Judgment Accuracya	0.01	-0.29
Capitalization Judgment Reaction Timea	-0.02	-0.10
Cued Recall Accuracy	-0.52*	-0.30
Cued Recall Reaction Time	-0.50	0.28
Duration of Illness	--	0.69**
PANSS Positive	--	0.04
PANSS Negative	--	0.37*
PANSS General	--	0.17
Hopkins Verbal Learning Trials 1-3	--	-0.44*
Hopkins Verbal Learning Delayed Recall	--	-0.35
Wisconsin Card Sorting Test Perseverative Errors	--	-0.07

NOTE: HC = healthy comparison, SZ = schizophrenia

^*= p<0.05

^**= p<0.01

^aSpearman nonparametric correlations used

Whole-brain correlation analysis: Figure 2 presents the results of the voxel-wise correlation analysis between age and brain response during novel word pair learning compared to fixation. Healthy individuals showed predominantly positive relationships with age, with two significant clusters identified (circled in Figure 2). The larger cluster (5056 voxels) was in midline precuneus, with its center of mass at Talairach coordinates 4 Left, 52 Posterior, and 41 Superior, and a mean t-value = 2.9, SEM = 0.40 (corresponding to a mean correlation of r=0.65). The other cluster (4736 voxels) was centered in the right superior temporal gyrus (60 Right, 25 Posterior, and 11 Superior) and voxels here had a slightly stronger relationship to age, with a mean t-value of 3.3, SEM = 0.07 (corresponding to a mean correlation of r=0.69). Patients with schizophrenia generally showed small negative correlations with age, none of which met our cluster size / threshold combination for significance. When patients and comparison subjects were compared directly, there was a single cluster (10624 voxels) of significant interaction between age and group, located in midline precuneus (center of mass: 2 Right, 53 Posterior, and 40 Superior; mean (SEM) t = 2.9 (0.04)), in which the age correlation was more positive among healthy individuals than among patients.

Figure 2.

Maps of whole-brain, voxel-wise analysis of the relationship of age to brain response (contrast of novel word learning to fixation). Top panel shows unthresholded maps on axial slices through the brain of the magnitude of the age correlation in patients and comparison groups with warm colors indicating positive relationships to age and cool colors indicating negative relationships. The strength of the correlation is indicated by the color bar key in the middle of the figure. The bottom panel shows the location of the cluster of significant difference in correlation (group × age interaction), overlaid onto a three-dimensionally rendered exemplar brain with axial and coronal cuts to permit viewing of the cluster.

Region of Interest analyses

Figure 3 shows the scatter plots for the relationship of age to brain response in the two clusters identified above to be areas of “normal age-related change” and in the two bilateral a priori regions in the frontal cortex. Correlation values are listed in Table 3 along with correlations and partial correlations with potential moderator variables. As expected based on the method of identifying the regions of “normal age-related change”, the relationship between age and mean brain response in the midline precuneus and superior temporal gyrus clusters was significant and positive among healthy individuals. Healthy individuals also demonstrated significant positive correlations in the right inferior and middle frontal gyrus regions of interest. None of the age correlations was significantly different from zero among schizophrenia patients in any of the ROIs examined. In three of the regions where healthy individuals showed a significant positive correlation, patients had significantly less positive correlations (midline precuneus: z = 4.18, p < 0.00001; right superior temporal gyrus: z = 2.6, p = 0.008; right inferior frontal gyrus: z = 2.23, p = 0.02).

Figure 3.

Scatterplots of the relationship of age to brain response (fit coefficient for the contrast of novel word learning to fixation) among patients with schizophrenia (black squares) and healthy individuals (gray circles) in six regions of interest. The best fitting linear trend line is indicated as solid for patients and dashed for comparison participants.

Table 3.

Correlations and Partial Correlations between Regional Brain Response and Age, Duration of Illness, Negative Symptoms, and Recall Accuracy

Brain Region	Age		Duration of Illness	PANSS Negative	HVLT Learning	Recall Accuracy		Age Controlled for Illness Duration	Age Controlled for PANSS Negative	Age Controlled for HVLT Learning	Age Controlled for Recall Accuracy
	HC	SZ	SZ	SZ	SZ	HC	SZ	SZ	SZ	SZ	HC	SZ
Midline Precuneusa	0.84**	-0.26+	-0.32	0.21	0.27	-0.31	0.26	-0.10+	-0.38+	-0.22+	0.70*	-0.20+
Right Superior Temporal Gyrusa	0.79**	0.14+	-0.08	0.28	-0.05	-0.16	-0.19	0.23+	-0.10+	0.04+	0.71**	-0.04+
Left Inferior Frontal Gyrus	0.16	0.11	-0.04	0.09	-0.05	0.41	0.20	0.21	0.10	0.11	0.44	0.19
Right Inferior Frontal Gyrus	0.65*	-0.02+	-0.22	0.24	0.14	-0.14	0.16	0.20	-0.11+	0.04+	0.66*	0.04+
Left Middle Frontal Gyrus	0.40	-0.07	-0.19	0.14	-0.02	0.01	0.37*	-0.01	-0.16	-0.19	0.40	0.004
Right Middle Frontal Gyrus	0.42*	-0.20+	-0.38	-0.04	0.11	-0.27	0.21	0.22	-0.11	-0.17	0.19	-0.07

NOTE: HC = healthy comparison, SZ = schizophrenia

^*= p<0.05

^**= p<0.01

+ = significantly different from HC correlation with age

^aSpearman nonparametric correlations used

We examined whether any of the other variables that were correlated with age might have moderated the relationship between age and brain response within either group. Within the patient group, duration of illness and age were highly correlated. The relationship between duration of illness and brain response in each of the regions was not significantly different from zero, but the direction and magnitude tended to be more negative than the correlations with age (Table 3). In a regression model testing whether age, duration of illness, and/or the interaction of the two predicted brain response, none of these predictors was significant for any of the six regions of interest. We also tested whether controlling for variance associated with duration of illness would reduce the difference between patients and controls in the magnitude of the age-brain response correlation. After partialling out duration of illness, only correlations within the two regions identified through the whole-brain approach remained significantly different between groups. The correlations within the right inferior frontal cortex ROI were no longer significantly different between groups when duration of illness was accounted for among patients (z score before partialling = 2.23, z score after partialling = 1.61).

Negative symptom severity was also positively related to age among patients, but was not significantly associated with brain response in any of the ROIs (Table 3). In addition, there was no evidence of an interaction between age and negative symptoms for brain response in any of the regions. Finally, when variance associated with negative symptoms was partialled from the age-brain response correlation, the resulting partial correlations were still significantly less positive compared to the first-order age correlations seen among healthy individuals in all three areas of group difference.

The learning score from the Hopkins Verbal Learning Test was significantly negatively correlated with age among patients, but not significantly related to brain response in any of the ROIs (Table 3). Still, there was evidence that HVLT learning score moderated the relationship of age to brain response in the left middle frontal gyrus (t = 3.04, p = 0.005) and in the midline precuneus cluster (t = 2.8, p = 0.009) among patients. Specifically, we found that poorer learners had a more negative age relationship. On the other hand, partialling out the contribution of HVLT scores from the age-brain response relationship among patients had very little effect on the magnitude of that relationship, and in all regions of group difference, the correlations remained significantly different after accounting for HVLT scores.

We also examined whether variance related to recall accuracy might moderate any association of age with brain response within each group. Among healthy individuals, we observed a significant age × recall accuracy interaction effect on brain response in left inferior frontal gyrus (t = -2.77, p = 0.20), right inferior frontal gyrus (t = -2.6, p = 0.03), left middle frontal gyrus (t = -3.0, p = 0.013), midline precuneus (t = -3.4, p =0.006), and right superior temporal gyrus (t = -3.25, p = 0.009). Figure 4 shows the nature of this interaction for the right inferior frontal gyrus by splitting the group into older and younger subgroups at the median age and plotting the relationship of recall accuracy to brain response separately for the two age groups. In this region and all the others, younger individuals showed a positive relationship between recall accuracy and brain response, while older individuals showed a negative relationship. In contrast to the strong interaction of age and recall accuracy among the healthy individuals, we observed only one region with such an interaction among patients. In the right inferior frontal gyrus (Figure 4), older patients tended to have a positive relationship between recall accuracy and brain response, while the younger patients had a negative correlation (t = 2.27, p =0.03). This was opposite in direction from the interaction observed in healthy individuals (3 way interaction of age × group × recall accuracy: t = 2.98, p = 0.005), and was observed in the absence of any main effects of either age or recall accuracy for patients in this region. We then tested whether accounting for variance due to recall accuracy in each group would diminish the observed between-group differences in age associations. Partialling recall accuracy from the age relationships in both groups had very little effect, and all three group differences remained significant.

Figure 4.

Scatterplots of the relationship between brain response to novel word pairs in the right inferior frontal gyrus (IFG) and recall accuracy for the paired-associates presented during scanning for subgroups of older (filled symbols) and younger (open symbols) participants in the healthy comparison group (circles; left panel) and schizophrenia group (squares; right panel).

Discussion

As predicted, patients with schizophrenia showed more negative (or less positive) relationships between functional brain activation and age than did a healthy comparison group. Among healthy individuals, older age was associated with greater functional brain response in midline precuneus, right superior temporal gyrus, right middle frontal gyrus, and right inferior frontal gyrus. In contrast, among patients with schizophrenia there was no relationship or a slightly negative relationship with age in these brain regions, and the magnitude of the patient correlations was significantly lower than the magnitude of correlations among healthy individuals in all but the right middle frontal gyrus. When duration of illness was accounted for, only the precuneus and temporal lobe correlations remained significantly different between the groups. Accounting for negative symptoms and Hopkins Verbal Learning Test scores among patients and for recall accuracy in both groups did not diminish any of the observed group differences. These cross-sectional results suggest that patients with schizophrenia do not show typical age-related increases in functional activation during learning tasks in several cortical regions.

Using a whole-brain, voxel-based approach, we found two posterior regions (midline precuneus and right superior temporal gyrus) that were more active in older individuals. No clusters of negative relationship with age met our cluster-threshold criteria for significance. In both of the observed clusters, younger healthy individuals had task-related deactivation and older healthy individuals had task-related activation of these regions. Task-related deactivation is thought to arise as the result of suspension of task-independent, or default mode, resting activation in these regions (32). Our findings are consistent with previous studies that have observed a positive relationship between age and brain response in regions that are part of the default mode network (33, 34). In our a priori regions of interest, brain response in the right inferior frontal and middle frontal gyrii were also positively related to age, whereas the correlations were somewhat smaller on the left. Our finding is consistent with two previous studies showing greater learning-related brain response in older compared to younger individuals in right frontal regions (35, 36), but inconsistent with other studies that found underactivation among older individuals in these regions during encoding, particularly on the left (for a review, see (37)). It has been debated whether greater brain response among healthy older individuals reflects compensation or de-differentiation (37, 38). Our finding of significant interactions of age and performance seems to be more consistent with de-differentiation, because we observed a negative relationship between brain response and performance in these regions among older adults while the relationship was positive among younger individuals. This suggests that the increased activation with age seen in this healthy sample was not helpful to performance among the older individuals.

Patients with schizophrenia did not show the same age-related associations as healthy individuals. In general, brain response during learning of new word pairs was slightly lower among older vs. younger patients, although the correlations with age were not significantly different from zero. The whole brain analysis revealed a significantly different region in midline precuneus which was confirmed in region of interest analyses, and we also observed significant differences when averaging across activation of the right superior temporal cluster identified on whole brain analysis and in an a priori region of right inferior prefrontal cortex. Our results are consistent with previous studies of resting blood flow or metabolism, in that the correlations were less positive among patients than comparison participants, and echo the finding of precuneus differences in attention-related metabolism by Siegel et al (1994). The findings are also slightly inconsistent, however, in that previous studies generally observed significant negative correlations that were, in some cases, more negative than those seen in healthy individuals. This discrepancy could be related to our use of a cognitive challenge task rather than resting measures. Also, as mentioned in the Introduction, most previous studies of age correlations with brain function have used samples that were unmedicated or on typical antipsychotics. The atypical antipsychotics used by most of the patients in the current study may have diminished the magnitude of age-related decline in brain function. Our results are perhaps most consistent with the one longitudinal study that observed no change in hypofrontality with age (17). In the current sample (as in the subset reported in Eyler et al (22)), left inferior frontal gyrus brain response was significantly lower among patients than healthy individuals and we found no significant correlation with age in this region. The largest previous cross-sectional study of the relationship of age to frontal blood flow showed that the observed negative correlation was a consequence of duration of illness rather than chronological age per se (18, 19). Similarly, our results showed that the age related differences between patients and comparison participants in right frontal regions of interest were no longer significant when duration of illness was accounted for statistically. However, within the two regions that we identified with a voxel-based, whole brain approach (midline precuneus and right superior temporal gyrus), the group differences in age-related correlations were still significant after accounting for variance due to duration of illness. This suggests that schizophrenia patients may have developmental differences unrelated to duration of illness that retard the normal age-related increases in brain response in these regions. In particular, older schizophrenia patients do not seem to show the loss of task-related deactivation found with age among healthy individuals in these default-mode regions.

In the Introduction, we pointed out a discrepancy in the literature between findings of volumetric decreases with age, cognitive stability with age, and clinical improvements and decrements with age among patients with schizophrenia. Our results may help to shed some light on why these findings are discrepant. Although we did not measure brain volumes in this study, we did assess cognitive performance on standard neuropsychological tasks and on the fMRI challenge task, clinical symptoms, and brain function among the schizophrenia group. In this sample, we found a significant negative correlation between age and learning scores on a standard neuropsychological task, but no relationship of age and a measure of executive functioning. Performance on the scanner task did not significantly relate to age, although the correlation was moderately negative. Negative symptoms were positively related to age, but positive symptoms and general psychopathology were not. Thus, there is cross-sectional evidence in this sample for worsening of negative symptoms and learning performance and stability of positive symptoms and executive function with age in this sample. As mentioned above, we could not find evidence for large correlations between age and brain function among patients, suggesting that brain function is also relatively stable. Interestingly, in one region, the right inferior frontal gyrus, we observed that the relationship between brain response and task performance depended on age, and that the direction of this relationship was opposite that found in healthy individuals. Whereas normal age-related increases among healthy participants appeared to be associated with poorer task performance (consistent with a de-differentiation view of overactivation), older patients with schizophrenia who had greater brain response performed better on the task (consistent with a compensatory explanation for overactivation). Similarly, we found that older patients with greater response in the midline precuneus and left middle frontal gyrus regions performed better on standard neuropsychological list learning test. Furthermore, in a previous study of older adults, we had similarly found that the relationship between task performance and brain response during picture learning in the hippocampus was opposite to that of comparison participants, with a negative association among healthy individuals and a positive association among patients. These findings suggest that the interplay between age, task performance, and brain response is likely to be complex and may be completely opposite to that expected based on the nature of those relationships among healthy individuals. Negative symptom severity was not related to brain response in any of the regions examined, and did not appear to interact with age in predicting brain response. The nature of the relationship between brain response, age, and negative symptoms might be better addressed using a different challenge task, perhaps one involving social cognition.

There are several aspects of our study that need to be carefully considered when interpreting the results. First, we used a cross-sectional design, so our age-related differences (or lack thereof) could be related to cohort effects, survival effects, or unmeasured moderator variables. Further longitudinal studies will be necessary to confirm the apparent group differences in the slope of age-related change observed in this study. Second, although our sample of patients was fairly large compared to other studies, there were some ways in which the sample was not representative of the larger population of patients, including a predominance of men and exclusion of some of the more cognitively impaired patients due to movement during scanning. Although we suspect that the latter issue is common among imaging studies, the generalizability of our findings to women and more impaired patients should not be assumed. Third, we did not measure the volume of the brain regions examined in this study, so we are unable to directly examine the relationship between volume, age, and activation. Fourth, a concern had been raised that age-related differences in BOLD signal may be due to age-related differences in hemodynamic properties rather than in neural functioning (39). This study did not examine this issue directly and future studies should use measures of resting perfusion to help determine whether the age-related differences were neural or vascular in origin. Finally, since ours was the first study to examine age-related changes using fMRI among schizophrenia patients, we conducted analyses in several different ways, leading to a large number of statistical comparisons, and thus increasing the chance of false positive results. The findings will hopefully generate hypotheses to be confirmed in future studies.

It is clear from this study and the existing literature that the course of brain structural and functional abnormalities in schizophrenia across the lifespan is likely to be complicated, especially in its relationship to cognition and behavior. While structural abnormalities may worsen slightly, this may not translate straightforwardly into declines in brain function with age. Rather, compensatory mechanisms may result in stability of brain response during cognitive challenge with age. This pattern may be specific to certain brain regions, however, and other areas may experience age-related or duration-of-illness-related decrements in function. Whether age-related declines in brain function can be found that explain the increase in negative symptoms with age remains an open question. Our findings give hope, however, that rehabilitation programs for older patients with schizophrenia may be able to take advantage of compensatory brain mechanisms to improve everyday functioning.