INTRODUCTION

The Apolipoprotein E (ApoE) locus on chromosome 19 is the only well-documented genetic risk factor for the development of sporadic AD (Saunders et al, 1993). Among the three major isoforms of ApoE (ɛ2, ɛ3, and ɛ4), the ɛ4 allele has been reported to increase an individual's risk of developing AD, and decrease age of disease onset, in proportion to the number of ɛ4 alleles present (Corder et al, 1993; Farrer et al, 1997). The search for phenotypic correlates of ɛ4 has included neuropathological studies of the rate of β-amyloid (Aβ) deposition (Schmechel et al, 1993; Nagy et al, 1995; Norrman et al, 1995; Gomez-Isla et al, 1996), neurofibrillary tangle formation (Schmechel et al, 1993; Nagy et al, 1995; Norrman et al, 1995; Gomez-Isla et al, 1996), cholinergic markers (Poirier et al, 1995; Soininen et al, 1995), and medial temporal lobe atrophy (Lehtovirta et al, 1996b; Hashimoto et al, 2001; Basso et al, in press).

The links between the ApoE ɛ4 allele and decreased age of AD onset and increased accumulation of pathological features has prompted the hypothesis that ɛ4 may play a role in accelerating the clinical manifestations of the disease. However, that hypothesis has thus far not been supported by the majority of clinical studies, which have shown no effect of ApoE genotype on rate of cognitive decline in AD (eg Farlow et al, 1999; eg Aerssens et al, 2001; Kleiman et al, 2006). However, some investigators have reported that the presence of at least one ɛ4 allele may increase (Craft et al, 1998; Hirono et al, 2003) or even decrease (Frisoni et al, 1995; Stern et al, 1997) the rate of cognitive decline.

Investigators have also sought to identify clinical correlates of ApoE genotype by focusing on psychiatric symptoms associated with AD. Behavioral abnormalities contribute to the distress of both AD patients and caregivers (Teri, 1997) and are a significant cause of institutionalization (Bianchetti et al, 1995). Previous studies of psychiatric disturbances in AD in relation to ApoE have yielded a wide range of results. However, the most frequent positive association has been between ɛ4 and psychotic symptoms (delusions or hallucinations). Fourteen previous studies have examined psychotic symptoms, and five have reported that ɛ4 increased the risk for psychosis (Ramachandran et al, 1996; Ballard et al, 1997; Harwood et al, 1999; Weiner et al, 1999; Scarmeas et al, 2002), whereas nine have found no effect of ɛ4 (Lehtovirta et al, 1996a; Cacabelos et al, 1997; Lopez et al, 1997; Lyketsos et al, 1997; Hirono et al, 1998, 1999; Levy et al, 1999; Gabryelewicz et al, 2002; Sweet et al, 2002). Sources of variation among these studies are unclear, but may include differences in patient populations, psychosis criteria, and statistical methods.

The present study was thus designed to clarify the influence of the ApoE ɛ4 allele on the presence of psychotic symptoms in AD. We hypothesized that ɛ4+ AD patients would manifest symptoms of psychosis more frequently and severely than ɛ4– patients. We also conducted an exploratory analysis to examine the association between ɛ4 and other behavioral symptoms.

MATERIALS AND METHODS

Subjects

The study sample comprised 266 patients who met NINCDS-ADRDA criteria for probable AD (McKhann et al, 1984). Five of these patients have subsequently died and had autopsy confirming definite AD (Mirra et al, 1991). All patients enrolled in a study of phenotypic correlates of ApoE ɛ4 in AD and were initially evaluated in the Yale Alzheimer's Disease Research Unit between May 1995 and August 2003. The demographics and clinical characteristics of patients are displayed in Table 1. The racial composition of the sample was: European-American (n=259; 97.4%), African-American (n=3; 1.1%), and Hispanic (n=4; 1.5%).

Table 1 AD Subject Characteristics

All patients underwent a comprehensive evaluation by a research physician and ancillary staff, including cognitive assessment, medical history, physical and neurological examinations, serum chemistries, thyroid function studies, complete blood count, B12, folate, VDRL, urinalysis, electrocardiogram, and brain MRI or CT. Subjects were excluded for any neurological or medical disorder (other than AD) that could produce cognitive deterioration or for significant psychiatric illness (prior to the onset of cognitive impairment), alcohol, or substance abuse. Some medications received by subjects at the time of evaluation—in particular, cholinesterase inhibitors, high-dose vitamin E (400 IU daily), and psychotropic drugs—may potentially have impacted behavioral variables analyzed in this study. These treatments were assumed to be independently distributed with regard to ApoE genotype; however, this assumption was also tested statistically (See below).

Family history of AD was assessed using the Alzheimer Dementia Risk Questionnaire (Breitner and Folstein, 1984) and the Dementia Questionnaire (Silverman et al, 1986) and was considered to be positive if at least one first-degree relative met criteria for primary degenerative dementia. No cases suggestive of autosomal dominant transmission were identified. Additionally, each subject was evaluated for an approximate date of disease onset, based on careful review of his or her medical records and detailed interviews with one or more primary caregivers. The date of onset was operationally defined as the date at which the ‘earliest definite AD symptom’ appeared. All subjects (or their responsible next of kin) provided written informed consent and were studied under a protocol approved by the Yale Human Investigation Committee.

Neuropsychiatric Evaluation

All subjects were evaluated for behavioral and psychological symptoms using the Neuropsychiatric Inventory (NPI), a structured interview that assessed the frequency and severity of these symptoms (Cummings et al, 1994). Using scripted questions, a caregiver was asked whether the patient's behavior had changed after the onset of dementia and whether the altered behavior had been present during the month preceding the evaluation. This format therefore distinguished between psychiatric symptoms that may have been present before the onset of dementia, and those that emerged during the disease process. The NPI assessed the following behavioral domains: delusions, hallucinations, agitation, depression, anxiety, elation, apathy, disinhibition, irritability, aberrant motor behavior, sleep disturbances and eating disturbances. Specific follow-up questions were used to confirm the presence of symptoms that were reported positive. For each domain in which symptoms were confirmed, the caregiver was asked to score the frequency with which symptoms occurred as: occasionally (1), often (2), frequently (3) or very frequently (4). The caregiver was asked also to score the severity of disturbances as: mild (1), moderate (2) or marked (3). Domains absent of disturbances were scored as 0. The product of the frequency and severity score was determined for each positive item (range=1–12), and the sum of all item scores yielded the total NPI score. Possible scores ranged from 0 (no behavioral disturbances) to 144 (all behavioral disturbances maximally present).

Cognitive performance was assessed by the Mini Mental State Examination (MMSE) (Folstein et al, 1975). The functional capacity of subjects in ADLs was assessed using the Instrumental Activities of Daily Living (IADL) questionnaire (Lawton and Brody, 1969). The IADL questionnaire evaluated everyday functioning along eight domains, such as driving and using the telephone (Lawton and Brody, 1969). A score of 1 for a given domain indicated no impairment, with higher scores denoting greater degrees of impairment. As not all domains were valid for all subjects (eg men who never did laundry before AD onset), the IADL score was calculated as the sum of individual activity scores divided by the total possible number of points valid for that subject. The range of scores was therefore 0.27 (no impairment) to 1.00 (maximal impairment). All subject data were obtained by trained raters who were unaware of the subjects' ApoE genotypes.

Determination of ApoE Genotype

DNA was prepared from whole blood in the laboratory of JG by standard procedures. Genotypes were obtained by the polymerase chain reaction (PCR)—restriction fragment length polymorphism method (Hixson and Vernier, 1990) using a PCR procedure slightly modified from Tsai et al (1994). The PCR product was digested by HhaI (New England Biolabs) and subjected to electrophoresis in 5% MetaPhor agarose (FMC Corp., Rochland, ME). Gels were stained with ethidium bromide and DNA visualized by UV transillumination. The three alleles were scored as described by Hixson and Vernier (1990). Genotypes (8%) were repeated as a quality check, with complete concordance.

Statistical Analysis

Subject characteristics (including demographics, disease characteristics, and concomitant therapies) were compared between ApoE ɛ4+ and ɛ4− patients using Student's t-test for continuous variables or χ2 analysis for dichotomous variables. We hypothesized that ɛ4+ patients would manifest symptoms of psychosis more frequently and more severely than ɛ4− patients. We therefore calculated a psychosis rating for each subject by summing the Hallucinations and Delusions items on the NPI. The range of possible NPI-derived psychosis (NPI-Psychosis) scores was 0 to 24. As a result of the skewed distribution of NPI-Psychosis scores, these were dichotomized with scores 1 indicating the presence of psychosis. The association between ɛ4 and psychosis was analyzed using a multiple logistic regression model, with psychosis as dependent variable and the following independent variables (in addition to ɛ4 allele carrier status): age, sex, educational attainment, and MMSE as a measure of disease severity.

We performed a number of post hoc exploratory analyses. To ascertain whether our results were dependent on a particular definition of psychosis, we repeated the multiple logistic regression analysis using two progressively more restrictive criteria that have been utilized for entry into trials of antipsychotic drug for patients with dementia: (1) score 3 on either the Delusions or Hallucinations item of the NPI, corresponding with moderate severity or frequency (Street et al, 2000); (2) NPI-Psychosis 6 (Deberdt et al, 2005). We also examined the effect of four concomitant medication classes on psychotic symptoms by simultaneously adding them as independent variables to the primary logistic regression analysis. Furthermore, to determine whether the effect of ɛ4 on psychosis related to a particular stage of disease, we repeated the multiple logistic regression analyses separately for the same three severity strata (mild: MMSE20, moderate: MMSE 12–19, severe: MMSE<12) analyzed by Harwood et al (1999).

Finally, we conducted an exploratory analysis of the possible role of ApoE ɛ4 in promoting other behavioral disturbances in AD as measured by total NPI score and all 12 subscores. The distribution of total NPI scores in our sample showed a strong positive skew, so a Mann–Whitney U-test was employed to determine if ɛ4+ and ɛ4− groups differed in total NPI score. Potential associations between the presence of ɛ4 and the presence of symptoms in each individual NPI domain were examined by logistic regression analyses, using the same independent variables as described for NPI-Psychosis. No correction for multiple comparisons was used in this exploratory analysis.

RESULTS

Subject Characteristics

Table 1 summarizes AD subject characteristics as a function of ApoE ɛ4 status with regard to demographics, disease characteristics, concomitant treatments, and neuropsychological data. ApoE ɛ4+ and ɛ4− patients did not differ significantly in age (t264=0.94, P=0.35), sex distribution (χ2=0.16, df=1, P=0.69), or educational attainment (t264=−1.73, P=0.085). With regard to disease characteristics, they also did not differ in age of onset (t264=0.82, P=0.41) or duration of symptoms (t264=0.48, P=0.63). However, ɛ4+ patients had a higher frequency of positive family history of AD (χ2=4.31, df=1, P=0.038). Finally, ɛ4+ and ɛ4− patients did not differ in terms of current use of cholinesterase inhibitors (χ2=0.13, df=1, P=0.72), antipsychotics (χ2=0.019, df=1, P=0.89), antidepressants (χ2=2.02, df=1, P=0.16), and high-dose Vitamin E (χ2=0.18, df=1, P=0.68).

Effect of ApoE ɛ4 Allele on Psychotic Symptoms

Table 2 displays the frequency of psychotic symptoms (according to different criteria) in the AD patients according to ApoE ɛ4 status. When psychosis was defined by an NPI-Psychosis score 1, the presence of ɛ4 was significantly associated with psychosis (odds ratio (OR)=1.87, 95% CI=1.07–3.29, P=0.029) in a multiple logistic regression model, adjusting for the following variables: age (OR=1.05, 95% CI=1.01–1.09, P=0.008), educational attainment (OR=0.94, 95% CI=0.86–1.03, P=0.17), sex (OR=1.37, 95% CI=0.75–2.49, P=0.31), and MMSE score (OR=0.92, 95% CI=0.88–0.97, P=0.001).

Table 2 Frequency of Psychosis in AD Patients According to ApoE ɛ4 Status

Post hoc analyses were conducted to examine more restrictive definitions of psychosis, using the same independent variables in the logistic regression model (Table 2). When psychosis was alternatively defined by a score 3 on either the Delusions or Hallucinations item of the NPI (Street et al, 2000), the presence of ɛ4 was still associated with psychosis (OR=2.10, 95% CI=1.11–3.99, P=0.020). When psychosis was instead defined by NPI-Psychosis score 6 (Deberdt et al, 2005), the presence of ɛ4 was significantly associated with psychosis (OR=2.45, 95% CI=1.21–4.98, P=0.013).

Although the four aforementioned classes of concomitant medications appeared randomly distributed with regard to ApoE ɛ4 status, we also conducted a post hoc exploratory analysis of the effect of these drug classes on psychotic symptoms by simultaneously adding them as independent variables to the logistic regression analysis. Not surprisingly, current antipsychotic use was positively associated with presence of psychotic symptoms (OR=4.47, 95% CI=1.00–19.92, P=0.049), but antidepressant and Vitamin E use were not (data not shown). Interestingly, current cholinesterase inhibitor use was negatively associated with the presence of psychotic symptoms in the logistic regression model (OR=0.51, 95% CI=0.28–0.95, P=0.03). Although this effect is best studied in randomized trials of cholinesterase inhibitors (Wynn and Cummings, 2004), our data lend further support to the notion that these drugs may favorably impact psychotic symptoms (Cummings and Kaufer, 1996; Wynn and Cummings, 2004).

To determine whether the effect of ɛ4 status on psychosis was specific to a particular stage of disease, we performed an additional post hoc analysis in which we repeated the multiple logistic regression analyses separately for the same three severity strata (mild: MMSE20, n=103; moderate: MMSE 12–19, n=121; severe: MMSE<12, n=42) analyzed by Harwood et al (1999). The results of that analysis are displayed in Figure 1. Only at the severe stage was the presence of ɛ4 significantly associated with psychosis (OR=16.61, 95% CI=2.11–130.51, P=0.008), but not at the mild (OR=1.03, 95% CI=0.39–2.76, P=0.95) or moderate (OR=1.65, 95% CI=0.74–3.68, P=0.22) stages.

Figure 1
figure 1

Frequency of psychosis by ApoE ɛ4 carrier status and dementia severity. Mini Mental State Examination (MMSE) score was used to classify AD patients as mild (MMSE20, n=103), moderate (MMSE=12–19, n=121), or severe (MMSE<12, n=42). Separate multiple logistic regression analyses were performed for each severity category, with presence of psychosis as dependent variable and the following independent variables (in addition to ɛ4 carrier status): age, sex, educational attainment, and MMSE. *Only at the severe stage was the presence of ɛ4 significantly associated with psychosis (odds ratio (OR)=16.61, 95% CI=2.11–130.51, P=0.008).

Effect of ApoE ɛ4 Allele on Other Behavioral Disturbances

Exploratory analyses were also conducted for the effect of ApoE ɛ4 on other behavioral disturbances. No differences were found between ɛ4+ and ɛ4– AD patients in total NPI score (Z=−0.302, P=0.76, Mann–Whitney U-test). With regard to individual NPI symptom subscores (Table 3), ɛ4 was significantly associated with the presence of delusions (OR=2.26, 95% CI=1.24–4.13, P=0.008) and irritability (OR=1.75, 95% CI=1.02–3.00, P=0.041). In addition, nonsignificant trends were observed for associations with the presence of hallucinations (P=0.100), aberrant motor behaviors (P=0.100) and sleep disturbances (P=0.076).

Table 3 Frequency of Neuropsychiatric Symptoms in AD Patients According to ApoE ɛ4 Status

DISCUSSION

We investigated the effect of the ApoE ɛ4 allele on the development of psychotic manifestations in AD. Using multiple logistic regression analyses, we found that the presence of ɛ4 was associated with psychotic symptoms, adjusting for age, sex, education, and MMSE score. Exploratory analyses suggested that this effect accrued specifically from patients with severe-stage AD and primarily from an association between ɛ4 and delusions. The ɛ4 allele did not appear to influence the development of most other behavioral symptoms in our sample.

Comparison to Other ApoE ɛ4 AD Psychosis Studies

As displayed in Table 4, previous studies of psychosis in AD in relation to ApoE have yielded a wide range of results. However, the most frequent positive association has been between ɛ4 and psychotic symptoms. Fifteen studies (including the present one) have examined psychotic symptoms, and six (including the present one) have reported that ɛ4 increased the risk for psychosis (Ramachandran et al, 1996; Ballard et al, 1997; Harwood et al, 1999; Weiner et al, 1999; Scarmeas et al, 2002), whereas nine have found no effect of ɛ4 (Lehtovirta et al, 1996a; Cacabelos et al, 1997; Lopez et al, 1997; Lyketsos et al, 1997; Hirono et al, 1998, 1999; Levy et al, 1999; Gabryelewicz et al, 2002; Sweet et al, 2002). Among the six positive studies only two others (Weiner et al, 1999; Scarmeas et al, 2002) have considered delusions and hallucinations separately. Our exploratory finding that ɛ4 was preferentially associated with delusions was shared by Scarmeas et al (2002) who observed that ɛ4 was associated specifically with an increased risk for delusions in a dose dependent manner, whereas ɛ4 homozygotes actually had a reduced risk for hallucinations. By contrast, Weiner et al (1999) reported marginal (P=0.05) positive associations between ApoE ɛ4 and both delusions and hallucinations. This may reflect a statistical power issue, since delusions are more prevalent among AD patients than hallucinations (Hirono et al, 1998; Bassiony and Lyketsos, 2003), including in our present sample (delusions 29%, hallucinations 15%). Two recent studies have used Cox proportional hazards analysis to determine whether ApoE ɛ4 influences time to onset of psychosis in patients followed longitudinally, but have reported discrepant results (Scarmeas et al, 2002; Sweet et al, 2002). Sources of variation among these fifteen studies are several, but particularly include differences in psychosis criteria, subject severity, and method of analyzing ApoE ɛ4 status.

Table 4 Studies of Psychotic Symptoms in AD in Relation to ApoE ɛ4 Status

Psychosis criteria

Previous studies have used a variety of methods to identify psychotic symptoms in their subjects, including general clinical examination (Lehtovirta et al, 1996a; Lyketsos et al, 1997; Hirono et al, 1999), rating scales (Ballard et al, 1997; Cacabelos et al, 1997; Hirono et al, 1998; Levy et al, 1999; Weiner et al, 1999; Gabryelewicz et al, 2002; Scarmeas et al, 2002), or a combination (Ramachandran et al, 1996; Lopez et al, 1997; Sweet et al, 2002). Those studies that have employed research rating scales have used a wide range of scales (see Table 4), making it difficult to reconcile discrepant reports. Our choice of the NPI has a number of advantages and limitations, some of which derive from its reliance on caregiver interview (Cummings et al, 1994). One advantage of the NPI is that it yields a quantitative measure of psychosis based on frequency and severity of symptoms. Although our quantitative NPI scores exhibited too much skew to analyze as continuous variables with parametric statistics, they nonetheless permitted us to consider different thresholds of severity. Virtually all studies to date have considered only presence or absence of psychosis. However, when we alternatively considered two more restrictive thresholds of clinically significant psychosis from the NPI—that have been used as criteria for entry into clinical trials of antipsychotic drugs (Street et al, 2000; Deberdt et al, 2005) — we found essentially the same results. These analyses suggest that our findings did not depend heavily on the criterion employed to classify AD subjects as psychotic.

Subject severity

Given the well-documented increase in psychotic symptoms with AD severity (Lopez et al, 1997; Hirono et al, 1998; Harwood et al, 1999; Gabryelewicz et al, 2002), variations in subject disease stage may represent an important difference among studies. Harwood et al (1999), observed that the elevated risk for psychosis was present specifically at the severe stage (MMSE<12) of cognitive impairment among AD patients carrying the ɛ4 allele. When we stratified our sample using the same cutpoints as Harwood et al, we also found that the ɛ4 effect was statistically significant only for the severe-stage patients (Figure 1). Consequently an association between ApoE ɛ4 and psychotic symptoms may not be detected by studies that include only (eg Levy et al, 1999) or predominantly (eg Hirono et al, 1999) mild-to-moderate stage patients. Although several studies do not provide clear description of the severity range of their subject samples, two other positive studies either included severe-stage patients (Weiner et al, 1999) or followed patients longitudinally into the severe stage (Scarmeas et al, 2002). Discrepant results are not accounted for entirely by differences in patient severity, however, as one positive study was restricted to mild-to-moderate stage patients (Ramachandran et al, 1996), and two negative studies enrolled large numbers of severe-stage patients (Cacabelos et al, 1997; Gabryelewicz et al, 2002).

ApoE ɛ4 carrier status vs dose

Although we chose to evaluate ApoE ɛ4 carrier status dichotomously, several other studies have examined the number or ‘dose’ of ɛ4 alleles in association with psychotic symptoms, including two of the positive studies (Weiner et al, 1999; Scarmeas et al, 2002). Among the six positive studies only Scarmeas et al, 2002 reported a full dose effect of ɛ4 (2ɛ4>1ɛ4>0ɛ4) on psychotic symptoms in AD. If we alternatively entered ɛ4 allele number (instead of carrier status) in our multiple logistic regression model, we would still find that ɛ4 number was significantly associated with psychosis (OR=1.60, 95% CI=1.07–2.42, P=0.024). However, we would not observe a full dose effect, as the 1ɛ4 and 2ɛ4 groups would not differ in the presence of psychotic symptoms (Table 2).

An important additional limitation of the present study—and of ApoE psychosis studies broadly—pertains to selection bias, with regard to ethnicity, disease severity, and particularly, the range of psychotic symptoms included. Our subjects tended to be recruited from those participating in other research protocols, some of which specifically selected against antipsychotic drug use (thus the low rate of antipsychotic drug use=5%). The overall prevalence of psychosis in our study (33%) is comparable to other similar studies buy likely underestimates the true population prevalence. Population-based studies that include institutionalized patients would be valuable to confirm our results across the full spectrum of dementia severity.

Interpretation of an Increased Risk of Psychosis in ApoE ɛ4+ AD

The mechanism by which ApoE ɛ4 may increase the risk for psychotic symptoms in AD is unclear. Although the ɛ4 allele has been shown to promote the neuropathological features of AD, including β-amyloid (Aβ) deposition (Schmechel et al, 1993; Nagy et al, 1995; Norrman et al, 1995; Gomez-Isla et al, 1996) and neurofibrillary tangle formation (Schmechel et al, 1993; Nagy et al, 1995; Norrman et al, 1995; Gomez-Isla et al, 1996), attempts to relate these features to psychotic symptoms in AD have yielded conflicting results (Zubenko et al, 1991; Förstl et al, 1994; Mukaetova-Ladinska et al, 1995; Sweet et al, 2000). Additionally, ApoE ɛ4 has been linked with more profound cholinergic loss in the frontal cortex (Soininen et al, 1995) and medial temporal lobe (Poirier et al, 1995), and acetylcholine levels have in turn been implicated in psychotic disturbances in AD (Cummings and Kaufer, 1996). Neuroimaging studies may shed additional light on the association between ApoE ɛ4 and psychotic manifestations of AD. One in vivo SPECT study has suggested that delusions in AD may be associated with hypoperfusion in the temporal lobes (Starkstein et al, 1994), and some (Lee et al, 2003) but not all (van Dyck et al, 1998) functional imaging studies have shown that AD patients who carry the ɛ4 allele have reduced temporal lobe function. The structural MRI literature is more unified in showing greater medial temporal lobe atrophy in association with the ɛ4 allele in AD (Lehtovirta et al, 1996b; Hashimoto et al, 2001; Basso et al, in press).

In conclusion, AD patients who carry the ApoE ɛ4 allele are at greater risk than noncarriers for developing psychotic symptoms, particularly delusions. This effect appears to be associated specifically with the severe stages of the disease. Although an association has been made between the ApoE ɛ4 genotype and psychosis in AD, more research must be conducted to examine the pathological links between the gene and the symptom.