Mid to late‐life scores of depression in the cognitively healthy are associated with cognitive status and Alzheimer's disease pathology at death

Abstract Objectives Early diagnosis of Alzheimer's disease (AD) is essential for early interventions. Symptoms of depression could represent a prodromal stage of AD. Very early mood alterations may help to stratify those at highest risk of late‐life AD. We aim to investigate associations between baseline/longitudinal scores for depression, presence of cognitive impairment and/or AD pathology at death. Methods/Design Between 1991 and 2015, participants from The University of Manchester Longitudinal Study of Cognition in Normal Healthy Old Age underwent 10 waves of assessment using the Geriatric Depression Scale (GDS). AD pathology at death was evaluated in 106 eligible cases. Analyses aimed to examine associations between GDS scores, cognitive status and AD pathology (as measured by Braak stage, Thal phase and CERAD). Results Baseline GDS scores were significantly higher for those cognitively impaired at death than those cognitively normal. Significantly higher baseline GDS scores were found for those with greater Consortium to Establish a Registry for Alzheimer’s Disease (CERAD) scores than those with lower CERAD scores. Similarly, significantly higher baseline GDS scores were found for those with a greater Braak stage than those with lower tau burden. These correlations remained after controlling for age at death, education and APOE ε4, but were less robust. Mean longitudinal GDS scores associated with cognition but not pathology. Conclusions GDS scores collected approximately 20 years before death were associated with cognitive status and AD pathology at death. We postulate that early AD‐related pathological change produces raised GDS scores due to an overlapping neural basis with depression, and that this may be considered as an early diagnostic marker for AD.


| INTRODUCTION
Depression is a frequent co-morbidity in patients with dementia and it presents unique challenges for clinicians who assess older, often cognitively impaired, adults. 1 Large, population-based studies have identified depression as a risk factor for dementia and, in particular, Alzheimer's disease (AD). 2 Recent epidemiological studies have replicated this finding. [3][4][5][6] However, critically, few studies have had neuropathological confirmation of AD diagnosis. 7 Depression affects about 50% of patients with AD, 8 and the cognitive performance of patients with both AD and depression declines at a greater rate than of patients without depression. 9 A history of depression has been shown to increase the risk of developing dementia. 10 Early-onset depression has been associated with an increased risk of AD. 11,12 However, late-onset depression has been strongly implicated in AD, suggesting that depression is a prodrome of AD. [13][14][15] One study highlighted that differences in depressive symptoms between those who do and those who do not go on to develop AD can be seen a decade before dementia diagnosis. 16 However, it is noteworthy that other studies reported no depressive symptoms in the prodromal phase of AD. 17 The major pathological changes associated with AD are abnormal accumulation of extracellular beta-amyloid (Aβ), which is thought to be an initial event of AD, and secondary intracellular tau in the form of neurofibrillary tangles. 18 It is thought that these pathological changes start many years before the onset of measurable cognitive decline. 19 Both Aβ and tau have also been implicated in depression.
Pre-clinical AD is associated with low plasma levels of Aβ 42 when compared with levels of Aβ 40 and a low Aβ 42 :Aβ 40 ratio is associated with depression in AD patients. 20 This is exacerbated further in those individuals carrying the APOE ε4 allele. 21 Levels of total tau and phosphorylated tau have been helpful in distinguishing mild cognitive impairment from major depressive disorder (MDD) which can be clinically difficult due to overlap of symptoms. 22 In addition, a mutant R406W human tau mouse model exhibited changes in depressionrelated behaviour which involved serotonergic neurons suggesting that tau may play a role in AD and depression. 23 Conversely, a PET imaging study found no difference in amyloid binding between depressed and non-depressed individuals, 24 and other studies have also failed to associate amyloid burden with depression. 25 Although clinical studies have significantly contributed to the understanding of the relationship between depression and AD, they lack the benefit of neuropathological confirmation of the disease.
There are few studies with pathological confirmation of AD and they lack consensus in the established literature. A recent review concluded that cognitive dysfunction in MDD and AD share a common origin relating to the hippocampus and the prefrontal cortex; specifically in the breakdown of connectivity between these regions. 26 It has previously been shown that AD patients with depression have a decreased number of neurons in the locus coeruleus compared to their non-depressed counterparts. 27 In addition, those with depression at intermediate stages of AD pathology (which is typically limbic) had a worse cognitive outcome than those without depression. 28 Likewise, the presence of a past history of major depression seems to lead to an increase in Aβ plaques and tau tangles in AD patients when compared with AD patients without depression. 29 Conversely, two large cohort studies that had brain donation as an end point, concluded that depression is not associated with any dementia-related pathology. 30,31 The Yesavage Geriatric Depression Scale (GDS) 32 is a widely used tool for assessing depressive symptoms in the elderly. Scores from the GDS have been shown to be reliable and relevant in those with AD. [33][34][35] This study aims to establish whether GDS scores for depression, measured longitudinally and starting many years before death, associate with the degree of AD pathology found at death as measured by Braak stage, 36 Consortium to Establish a Registry for Alzheimer's Disease (CERAD) score, 37 and Thal phase. 38  profile of this cohort has been previously described. 40,41 Between 1991 and 2003, participants underwent four faceto-face assessments of depression using the long form of the GDS score which can be used to assess level of depressive symptoms.

Related Tau Astrogliopathy, Primary Ageing-Related Tauopathy and
Limbic-predominant Age-related TDP-43 Encephalopathy owing to the fact that they are common ageing-related pathologies. Of the 134 UMLCHA participants who had donated their brain, 106 were eligible. An overview of the pathology of eligible participants is available (Table S1).

| Statistical analyses
The cohort was divided into pathology groups according to severity of AD pathology: CERAD score: 0-A (low severity) versus B-C (high severity).
Differences between pathology groups in sex, cognitive impairment and presence of APOE ε4 allele(s) were analysed with the Chisquared test while those for age at death, and years of education were analysed with the T-test.
Baseline GDS30 scores were not normally distributed. Therefore, differences between cognitive and pathology groups for baseline GDS30 scores were analysed with the Mann-Whitney U-Test. In all cases, a p value of <0.05 was considered significant.

| Demographics
The demographics of the 106 eligible participants are shown in Table 1.
There were no significant differences in sex, age at death or level of education between cognitive status groups, CERAD, Thal phase or Braak stage groups. As expected, a significantly greater proportion of cognitively impaired individuals were present in the high severity pathology groups for CERAD score (χ 2 ¼ 27.2; p < 0.001), Thal phase (χ 2 ¼ 6.1; p ¼ 0.014) and Braak stage (χ 2 ¼ 22.6; p < 0.001).
Likewise, APOE ε4 allele(s) were more likely to be present in the high ROBINSON ET AL.

| Assessment of baseline GDS30 test scores and relationship to cognition and pathology at death.
Median baseline GDS30 score, mean age at baseline testing and mean number of years between test and death are shown in Table 2.
The mean age at baseline GDS30 test was 68.4 (�5.3) years which was 20.8 (�4.0) years before death.
Those classified as having cognitive impairment at death scored higher on baseline GDS30 than their cognitively normal counterparts (p ¼ 0.007). In addition, we found that baseline GDS30 scores were significantly higher in the severe AD pathology groups for CERAD and Braak stage when compared with the corresponding lower severity group (CERAD: p ¼ 0.015; Braak stage; p ¼ 0.018). However, this was not the case when examining Thal phase (p ¼ 0.292; Figure 1).
Regression analyses showed that after accounting for sex, level of education and presence of APOE ε4 allele(s), the association between baseline GDS30 scores and cognitive impairment (p ¼ 0.104), CERAD scores (p ¼ 0.094) and Braak stage (p ¼ 0.060) were no longer significant (Table 3).

| Assessment of longitudinal GDS test scores and relationship to cognition and pathology at death.
Binary logistic regression analyses were conducted using cognitive or pathology groups as the outcome measure and mean score from the    (Table 4).

| DISCUSSION
We have shown that scores from the baseline GDS30 test, undertaken approximately 20 years before death, associate with cognitive impairment and AD pathology at death. A general trend towards correlation remained after controlling for age at death, education

F I G U R E 1 Boxplots comparing baseline Geriatric Depression Scale (GDS)30 scores between cognitive (Panel A) and
Alzheimer's disease pathology groups (Panels B, C and D). The boxes represent the interquartile (IQ) range which contains the middle 50% of the records. The whiskers represent the highest and lowest values which are no greater than 1.5 times the IQ range. The line across the boxes indicates the median. Differences between cognitive and pathology groups for baseline GDS30 scores were analysed with the Mann-Whitney U-Test -717 level and presence of APOE ε4 allele(s). Although the difference in GDS30 score between the cognitive groups and also between the AD pathology groups is subtle and remains sub-clinical for depression as measured by GDS, it nonetheless suggests that, if undertaken 20 years before death, the GDS is able to differentiate between, and could be predictive of, those who go on to develop AD pathology and associated cognitive impairment from those who will not.
Some studies have reported that cognitive test scores (usually of memory) are able to predict clinical dementia approximately 4-18 years before onset of symptoms. [59][60][61][62] Our findings show that a simple and widely used test for depression can also discriminate between those who become significantly cognitively impaired from those whose cognition remains relatively intact; and can do this many years before death.
A drawback to studies which are limited to clinical measures of dementia is the lack of neuropathological confirmation of disease type. Cognitive impairment is a 'loose' term encompassing many types of dementing illness. Here, instead, we ensured that our study group included only those with AD pathology and those with pathological findings that would be considered normal for the age of the subject. This gave us the opportunity to examine AD pathology alone as cases with possible concomitant pathology were excluded.
It is widely thought that the pathological processes underlying AD occur many years before any cognitive change that may lead to an individual seeking medical intervention. 63 Many studies have shown that cognitive performance scores, obtained 1-6 years before death, can predict ultimate levels of AD pathology. [64][65][66] However, most recently our group has shown that these findings can be extrapolated to approximately 20 years before death. 67 The findings described in the present study, again, show that scores of cognitive performance, in this case GDS30 baseline scores, collected 20 years prior to death, associate with AD pathology at post-mortem.
Neuroimaging and post-mortem studies have attempted to elucidate the underlying pathophysiology of depression. Those with MDD show abnormal brain activity in frontal and temporal cortices, insula and cerebellum. 68 In addition, those with MDD have moderate volume reduction in the hippocampus and striatum 69 and a reduction of glia cell density in the prefrontal cortices and amygdalae. 70,71 However, the most striking findings are in the cingulate cortex where volume reduction has been shown early in MDD. 72 Although the present study does not include individuals with known MDD, it is highly pertinent that the areas affected in MDD overlap somewhat with those affected relatively early on in AD by Aβ and tau; specifically the hippocampus, temporal cortex and amygdala. Moreover, the presence of depressive symptoms, not severe enough to be considered MDD, could result from neurofibrillary damage to serotonergic neurones in areas considered to be the first affected in AD, such as the dorsal raphe and locus coeruleus. [73][74][75] It has previously been shown that damage to serotonergic neurones can lead to depressive symptoms in animal models. 23 Thus, it would not be unreasonable to suggest that very early degenerative changes in these brain stem nuclei in humans could be responsible for generating mild depressive symptoms years before the onset of clinical AD. It is arguable therefore that the differences we report here in baseline GDS30 scores could reflect very early, subtle AD pathological changes, rather than the onset of a mild depressive disorder per se. This argument is bolstered by the fact that although the GDS30 baseline scores suggest mild depressive symptoms, they remain in the 'normal' range of the GDS protocol and do not indicate a diagnosis of depression.
Changes in neurogenesis could be another underlying mechanism explaining the very early differences in GDS scores between AD pathology groups. Both rodent models of AD and human AD postmortem tissue have shown decreased neurogenesis, especially within the ventral dentate gyrus of the hippocampus (and area of the brain implicated in both AD and MDD), while residual low-level neurogenesis is likely to be still ongoing. 76,77 Interestingly, neurogenesis has also been implicated in the pathology of MDD and is argued to be one of the ways in which anti-depressant medications exert their effect. 78 One hypothesis could therefore be that early AD pathology induces localised disruption of neurogenesis in the dentate gyrus.
This would prevent new cell proliferation and network integration, thereby causing early depressive symptoms and later, mild cognitive impairment, and finally, dementia.
It is worthy of note that only the baseline scores for GDS30 differentiated between CERAD and Braak stage pathology groups.

-
In conclusion, GDS30 scores collected approximately 20 years before death were associated with cognitive status at death and AD pathology, but mean longitudinal GDS scores were only associated with cognition and not with pathology. We postulate that the very early pathological changes associated with AD lead to higher scores on the GDS30 due to the overlap between brain regions implicated in depression and those compromised in very early AD. We suggest that a test battery combining assessments of cognition and symptoms of depression may well identify those individuals who will go on to develop AD, many years before they would have otherwise been diagnosed using the usual clinical assessments. However, for this to be robust, factors such as age at death, education level and APOE genotype would have to be considered.

CONFLICTS OF INTEREST
The authors have declared that they have no conflict of interest to disclose.

AUTHOR CONTRIBUTIONS
All authors have read the manuscript and have agreed to be listed as authors. Andrew C Robinson devised and designed the study, performed data/statistical analysis and wrote the paper. Federico

ETHICS COMMITTEE APPROVAL
The study was approved by Manchester Brain Bank Management

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author upon reasonable request.