Study design

The analysis was based on epidemiological and neuropathological data from the UK Medical Research Council Cognitive Function and Ageing Study (MRC CFAS). This study has been described in detail previously ^{Reference Brayne, McCracken and Matthews17,Reference Matthews, Brayne, Lowe, McKeith, Wharton and Ince18} and details, including flow diagrams of the sampling procedures and interviews, are displayed on the study website (www.cfas.ac.uk). The population base used was primary care physician lists that included institutional settings. In each of five centres random samples of approximately 2500 people aged 64 years and over with an 82% response rate underwent a screening interview including the Mini-Mental State Examination (MMSE) ^{Reference Folstein, Folstein and McHugh19} and the Automated Geriatric Examination for Computer Assisted Taxonomy (AGECAT) ^{Reference Copeland, Dewey and Griffith-Jones20} organicity items as well as self-report of medical conditions. Detailed assessment was conducted in a subsample of approximately 20% of participants stratified on cognitive function. All those who screened positive for dementia (AGECAT organicity greater than O3 at the screen, indicating possible or likely dementia) and a random subsample of the cognitively intact cohort were included in the assessment. The remaining participants was rescreened after 2 years and a further 20% of those not previously included were added to the assessment group. This assessment group was interviewed again 6 years after baseline and the entire remaining cohort was assessed 10 years after baseline. Subsamples of the assessment group underwent additional interviews at 1, 3 and 8 years after baseline. In a sixth centre, 5200 people underwent similar assessments at baseline, 1–2 years, 3–4 years, 5–6 years, 8 years and 10 years. Data on medical conditions were also ascertained from death certificates and from retrospective informant interviews. Data from all six centres were used in the current analysis.

Post-mortem follow-up

Individuals selected for the assessment interviews were initially approached by a trained liaison officer for brain donation in each of the centres. The percentage willing to take part in the donation programme varied among the centres from 20 to 55%. When a consenting individual died, the research team was notified and the next of kin was approached to give consent for brain donation and retention. Necropsy then proceeded subject to consent. The effect of the process of recruitment for brain donation on the population-representative nature of the MRC CFAS neuropathology sample has been analysed in detail previously. In summary, 456 brain donations were received before 31 July 2005, representing 3.8% of the 11 921 participants who entered the study and died before this time. Brain donors were on average 3 years older at death than those who died without donating, and were slightly more likely to have had a manual occupation during life. There was no difference between donors and non-donors in gender, years in full-time education ^{Reference Matthews, Brayne, Lowe, McKeith, Wharton and Ince18} or cause of death. ^{Reference Savva, Wharton, Ince, Forster, Matthews and Brayne21} After adjusting for the number of assessment interviews undertaken, there was no difference in depression occurrence between those who donated their brain and those who died without donating.

Ascertainment of depression and dementia

The Geriatric Mental State (GMS) is a fully structured diagnostic interview for mental disorders in older people that has been widely used and validated. ^{Reference Copeland, Dewey and Griffith-Jones20} Diagnoses are assigned using a computer algorithm (AGECAT) that generates a 0–5 scale of likelihood. In this analysis the standard 3+ cut-off for depression was used. This seeks to define depression of clinical significance (i.e. levels of symptomatology for which a clinician would be expected to intervene) and creates a broader category than major depression according to DSM–IV, ²² encompassing both moderate and severe symptomatology.

Data for this analysis were taken from individuals whose dementia status at death was known and where neuropathological data were obtained. Dementia status was based either on the death certificate or diagnosed by any interview in the study according to the AGECAT criteria. ²³ If the interview time was more than 6 months from the death, further information was gathered using a structure retrospective interview with knowledgeable informants and a final decision was made by clinicians working with the MRC CFAS using all available interview information. At necropsy, frozen samples of brain tissue were removed for storage and the remainder of the brain was fixed for standardised assessment on paraffin-embedded tissues, in line with the protocol by the Consortium to Establish a Registry for Alzheimer's Disease (CERAD). ^{Reference Mirra, Heyman, McKeel, Sumi, Crain and Brownlee24} Version 3.1 of the neuropathology data-set has been used.

Alzheimer neuropathology

The relationship between depression and Alzheimer pathology is of interest because diagnosed depression has been suggested to be a risk factor for Alzheimer's disease ^{Reference Ownby, Crocco, Acevedo, John and Loewenstein27} and several epidemiological studies have suggested that depressive symptoms occurring in later life are associated with an increased risk of later dementia ^{Reference Geerlings, Schoevers, Beekman, Jonker, Deeg and Schmand3} or cognitive decline. ^{Reference Wilson, Mendes de Leon, Bennet, Bienias and Evans28} In some cases this may be explained by depression occurring as a preclinical manifestation of dementia. However, chronic or recurrent depressive episodes have been found to be associated with hippocampal volume reduction ^{Reference Sheline, Gado and Kraemer29,Reference O'Brien, Lloyd, McKeith, Gholkar and Ferrier30} and with an increased risk of dementia, ^{Reference Kessing and Andersen31} suggesting that depression itself is a risk factor for neurodegenerative processes. When examining possible neuropathology underlying depression it is important in older populations to distinguish those who are dementia-free from people with a depression occurring in the context of dementia. A recent study found increased Alzheimer's disease-related neuropathological changes in people with Alzheimer's disease and a background history of major depression. ^{Reference Rapp, Schnaider-Beeri, Grossman, Sano, Perl and Purohit32} However, neuropathological associations in Alzheimer's disease samples may not be comparable to associations in community samples of people with depression who are dementia-free. We therefore sought to address the question of whether similar, albeit more minor, neurodegenerative changes occur in older people with late-life depression in a large community sample.

Neuropathological studies of late-life depression are uncommon and, for obvious logistical reasons, have tended to draw data from samples of people with severe syndromes known to clinical services. The difficulty with interpreting these findings is that these samples are unlikely to be representative of their source community and may be different in many respects from healthy controls. The MRC CFAS provided a valuable opportunity to investigate this issue in a representative community population. Additional advantages were that structured diagnostic instruments had been administered on repeated occasions in life so that dementia and depression were well characterised. Neuropathological data were obtained masked to clinical diagnoses using structured and widely applied criteria. Post-mortem intervals differed marginally between the comparison groups, although they were not associated with most of the pathological indices of interest and they are unlikely to have accounted for our observations. We do not have information on brain fixation times or brain pH; however, these are also felt to be unlikely to be relevant factors for the protocols adopted by CFAS.

In our analysis, we found no association between previously ascertained depression and Alzheimer pathology. This was consistent for three different neuropathological measures (diffuse and neuritic plaques and neurofibrillary tangles) and for multiple cortical and subcortical regions examined, although some evidence was seen for a relationship with tangles in the locus ceruleus. Our results, the first (we believe) from a sample representative of a general population, are also consistent with other research in this area. For example, in a large sample of 139 Catholic priests and nuns that were followed for 3.9 years, depressive symptoms measured by the Center for Epidemiologic Studies Depression Scale (CES–D) screening instrument were not significantly associated with neuritic plaques, diffuse plaques and neurofibrillary tangles at post-mortem. ^{Reference Wilson, Schneider, Bienias, Arnold, Evans and Bennet11} A case–control study on late-life major depression also reported no evidence of significant degenerative pathology in the same neocortical areas we examined. ^{Reference Thomas, Ferrier, Kalaria, Perry, Brown and O'Brien15} Hendricksen et al ^{Reference Hendricksen, Thomas, Ferrier, Ince and O'Brien13} carried out a study of neuronal density and neuritic pathology in serotonergic neurons in dorsal raphe nuclei and found no associations with depression, whether or not comorbid Alzheimer's disease had been present. Data on neocortical neuronal loss were not available in our data-set and inferences should bear in mind this limitation.

Lewy bodies

The increase in Lewy bodies in the brain-stem nuclei has not been previously reported in depression and merits further investigation. Since depression is increased in synucleinopathies, including dementia with Lewy bodies ^{Reference McKeith, Dickson, Lowe, Emre, O'Brien and Feldman33} and Parkinson's disease, ^{Reference Cummings34} it is possible that some cases of late-life depression are related to this pathology in brain-stem nuclei that contain the monoaminergic neurons believed to be involved in the neurobiology of depression. However, a previous study found no difference in Lewy body or neurite prevalence between 12 people with late-life depression or bipolar disorder compared with 27 age-matched controls. ^{Reference Jellinger16}

Neuronal loss

Despite a lack of association between depression and Alzheimer pathology, depression was significantly associated with hippocampal neuronal loss. This could indicate a separate hippocampal pathology (for example, sclerosis) that resulted in the depression. However, it could conceivably represent an effect of depression itself, since a significant proportion of people with late-life depression will have had earlier episodes, something which we were not able to evaluate in this particular data-set and which is limited in all studies of late-life depression because of the reliance on recall. As mentioned earlier, several studies have suggested that depression may be associated with volumetric changes in brain structures, and the direction of cause and effect underlying this relationship requires clarification. In an early study of depression in dementia, associations were found with locus ceruleus and substantia nigra degeneration. ^{Reference Zubenko and Moossy35} However, this may not generalise to neuropathological correlates of depression in the absence of dementia, which was our objective here.

Cerebrovascular pathology

We found no direct association between measures of cerebrovascular pathology and depression in this sample. This is also consistent with previous research in clinical samples. The neuropathological case–control study reported by Thomas et al, for example, found no associations between late-life depression and cerebral microvascular disease, although associations were found with more severe large vessel atherosclerosis. ^{Reference Thomas, Ferrier, Kalaria, Perry, Brown and O'Brien15} Although there is some evidence for a role of vascular factors in late-life depression derived from associations with white matter hyperintensities, ^{Reference de Groot, de Leeuw, Oudkerk, Hofman, Jolles and Breteler36} causal pathways remain controversial since there is little evidence for lesion-specific associations in post-stroke depression ^{Reference Carson, MacHale, Allen, Lawrie, Dennis and House37} and little evidence for associations with vascular risk factors such as hypertension and diabetes in community samples. ^{Reference Stewart, Prince, Richards, Brayne and Mann38,Reference Kim, Stewart, Kim, Yang, Shin and Yoon39} However, it is possible that microstructural changes associated with late-onset depression may reflect vascular pathology that was not detectable by the methods used in our study, ^{Reference Kumar, Mintz, Bilker and Gottlieb40} for example that associated with white matter hyperintensities on magnetic resonance imaging.

Limitations

Although these data have important strengths for investigating the associations of interest, there are methodological issues that should be considered. Our primary analysis involved 25 hypothesis tests and, when adjusting for multiple comparisons using either Bonferroni or Benjamini and Hochberg procedures, the threshold for statistical significance is P<0.002. Using this highly conservative approach the association between depression and neuronal loss in substantia nigra remains statistically significant; this combined with the plausibility of our findings leads us to believe that our results are not due to chance. Depression was only ascertained at previous examinations and no attempt was made to define in detail earlier or intervening episodes, which may have led to some misclassification. Furthermore, the diagnoses made through the AGECAT algorithm are relatively broad, encompassing both moderate and severe syndromes: broader than DSM–IV major depression but narrower than the cut-offs most commonly applied to brief screening instruments. Heterogeneity with respect to other findings may therefore reflect the disorder definition. Owing to the nature of recruitment to brain donation programmes, participants had varying numbers of assessments during life, and it was not possible to consider the duration or chronicity of depression in our analysis. A sensitivity analysis using only depression diagnosed at the last assessment before death did not reveal any new associations.

Our brain donor cohort was recruited from a population-representative study of ageing, and has been well characterised with respect to the remainder of the original cohort who died but did not donate their brain. ^{Reference Matthews, Brayne, Lowe, McKeith, Wharton and Ince18} The two-phase process by which brain donors were recruited led to the sample being weighted toward those with cognitive impairment, although participants with dementia at death were not included in this analysis and it is unlikely that selection bias influenced our findings. Another potential limitation is that, although the pathological measures employed are widely used, they are crude, especially those for assessing vascular pathology, and more sensitive measures may have detected differences. Thus, although Thomas et al ^{Reference Thomas, Ferrier, Kalaria, Perry, Brown and O'Brien15} found no increase in vascular pathology in their sample, using similar assessments to ours, they did identify an increase in ischaemic white matter hyperintensities and in inflammatory markers in grey and white matter. ^{Reference Thomas, Ferrier, Kalaria, Davis and O'Brien41,Reference Thomas, O'Brien, Davis, Ballard, Barber and Kalaria42} Previous findings from this cohort have suggested that the level of Alzheimer pathology alone is not sufficient to predict the occurrence of clinical dementia. ²⁵ Therefore, even moderate or severe levels of pathology may not have been a sufficiently accurate marker of actual incipient dementia.

Implications

Despite the strong statistical association with depression in the samples where subcortical Lewy bodies or neuronal loss were detected, these were relatively rare and most cases of depression in our sample were not associated with any subcortical pathology. This suggests that subcortical pathology only accounts for a small number of cases of late-life depression in the population, and that neuropathological studies of late-life depression must be large if they are to detect these associations, as well as potentially explaining why a previous study of raphe pathology within MRC CFAS ^{Reference Syed, Chatfield, Matthews, Brayne and Esiri14} failed to detect an association. Despite the large number of brains donated to the study, the numbers with subcortical pathology are small and our findings must be replicated in further population-representative cohorts.