β-amyloid pathology and hippocampal atrophy are independently associated with memory function in cognitively healthy elderly

The independent effects of different brain pathologies on age-dependent cognitive decline are unclear. We examined this in 300 cognitively unimpaired elderly individuals from the BioFINDER study. Using cognition as outcome we studied the effects of cerebrospinal fluid biomarkers for amyloid-β (Aβ42/40), neuroinflammation (YKL-40), and neurodegeneration and tau pathology (T-tau and P-tau) as well as MRI measures of white-matter lesions, hippocampal volume (HV), and regional cortical thickness. We found that Aβ positivity and HV were independently associated with memory. Results differed depending on age, with memory being associated with HV (but not Aβ) in older participants (73.3–88.4 years), and with Aβ (but not HV) in relatively younger participants (65.2–73.2 years). This indicates that Aβ and atrophy are independent contributors to memory variability in cognitively healthy elderly and that Aβ mainly affects memory in younger elderly individuals. With advancing age, the effect of brain atrophy overshadows the effect of Aβ on memory function.

imaging. The participants underwent MRI brain scanning at 3 Tesla using a standardized protocol of sequences. The volume of white matter lesions (WML; seen as hyperintensities in T2 weighted scans) was measured using the Lesion Segmentation Tool (https://www.applied-statistics.de/lst.html). Automatic segmentation using FreeSurfer software version 5.1 (http://www.freesurfer.net) was performed to measure total intracranial volume (ICV), HV, and regional cortical thickness. The sum of left and right HV was used, but analyses were also performed with left and right HV respectively to look for laterality effects. For cortical thickness, division into frontal, temporal, parietal, and occipital lobes was done using the standard FreeSurfer parcellation 43 . Additionally, entorhinal and parahippocampal cortices were combined in one meta-region, chosen for its association with memory function 44 . Thickness measures from both hemispheres were combined and adjusted for surface area.
cognition. The Alzheimer's Disease Assessment Scale-Cognitive Subscale (ADAS-Cog) 10-word delayed recall 45 was used as a measure of memory performance. The number of correct answers was used. The Trailmaking Test A (TMT-A) 46 , Symbol Digit Modalities Test (SDMT) 47 , and A Quick Test for Cognitive Speed (AQT) 48 were used to form a composite measure of attention and executive function. The raw scores were converted into z scores based on the distribution in the current population, and, if applicable, inverted so that a higher value represented better attention/executive function. The composite was the mean of these z scores.

Statistics.
A cut-off for Aβ positivity was defined using mixture modelling in a larger sample of the BioFINDER study (n = 889 in total) consisting of a group of cognitively unimpaired subjects, including the sample included in this study and an additional 25 subjects (n = 325), as well as a group of subjects with subjective cognitive decline (SCD; n = 204), MCI (n = 276), or dementia (n = 84), using the R package "mixtools". Mixture modelling is a 2-step procedure based on an expectation maximization algorithm, which assumes that the CSF Aβ42/40 ratio is a mixed sample from two different normal distributions (in this case one with a normal Aβ deposition and one with an abnormal Aβ deposition). Mixture modelling has previously successfully been used to identify cut-offs for Aβ biomarkers 49,50 .
To compare differences between groups, the chi-square test was used for dichotomous variables and the independent samples t-test for numerical variables. Linear regression models were tested to assess the effects of different biomarkers on cognition, with and without covariates (age, sex, and years of education, and HV was also adjusted for ICV). Interaction terms were tested for biomarkers and age. To facilitate interpretation of interactions and main effects, we used z scores of continuous variables. Test of statistical mediation was performed using the causal steps approach 51 . WML volume was used after logarithmic transformation (ln), because of skewed distribution. Statistical significance was defined by p < 0.05. Correction for multiple comparisons was performed by the false discovery rate when indicated. Statistical analyses were performed with R (version 3.3) and SPSS Statistics for Mac (version 24).

Results
Out of the 361 participants of the cohort of cognitively unimpaired in the Swedish BioFINDER Study, 300 had complete baseline MRI and CSF analyses and were included in the present study. Demographics are shown in Table 1 and Supplementary Fig. 1 shows a histogram of the age distribution in the sample. The cut-off for Aβ positivity was defined as Aβ42/40 < 0.051 ( Supplementary Fig. 2). The proportion of amyloid positive subjects in each group used for mixture modelling is shown in Supplementary Table 1.
When including all the biomarkers that were significant (not adjusted for multiple comparisons) after controlling for demographic variables in the same model, Aβ positivity (β = −0.14; p = 0.010) and smaller HV www.nature.com/scientificreports www.nature.com/scientificreports/ (β = 0.25; p < 0.001), but not temporal or frontal cortical thickness, were independently associated with worse memory (Table 3 and Fig. 1A). In Supplementary Table 2, different linear regression models including all or subsets of these biomarkers are shown.
Associations between Aβ and brain structure. There was no association between Aβ positivity and   www.nature.com/scientificreports www.nature.com/scientificreports/ Interactions between biomarkers and age to predict cognition. A significant interaction effect between total HV and age (used as a continuous predictor) on memory was seen (p = 0.040). Secondarily, we performed an exploratory analysis with the sample divided into younger and older participants, split by the median age (73.3 years). When using age as a dichotomous predictor, similar results were seen for the interaction effect (p = 0.007). When stratifying into the two age groups, the relationship between HV and memory was not statistically significant in the younger group (p = 0.066), but in the older group there was a highly significant relationship when controlling for demographic variables (β = 0.40; p < 0.001; Figs 1B and 2A, Suppl. Table 3).
No significant interaction was detected between Aβ positivity and age on memory (p = 0.38), but when stratifying into the two age groups, the opposite from HV was seen, i.e. there was an association between Aβ positivity and worse memory in the younger group (β = −0.23; p = 0.003), but not in the older group (p = 0.38; Figs 1B and 2B, Suppl. Table 3). Based on the theoretical model of amyloid pathology preceding tau pathology in AD 52 , we tested if the association between Aβ positivity and memory was mediated by P-tau. When adding P-tau in the model in the younger group, a statistical mediation effect was seen, i.e. higher P-tau (β = −0.17; p = 0.045) but not Aβ positivity (β = −0.15; p = 0.079) was significantly associated with worse memory (Fig. 1B, Suppl. Table 4), and Aβ positivity was associated with higher P-tau (β = −0.39; p < 0.001; Fig. 1B, Suppl. Table 5) when controlling for age and sex.
No significant interactions were seen between any of the other CSF/MRI biomarkers and continuous age on memory, and no interactions with age were seen for any of the biomarkers on attention/executive function (data not shown). We also looked on interactions on memory function between Aβ positivity and sex and education respectively, as well as between HV and sex and education respectively. None of these interactions were significant (data not shown).

Discussion
In this study of cognitively unimpaired elderly, we found that (1) Aβ positivity, HV, and cortical thickness (temporal and frontal) were associated with worse memory, with independent effects of Aβ and HV on memory; (2) the Aβ effect on memory could be confirmed in the younger part of the sample, while the HV effect on memory was significant in the older part of the sample only; (3) Aβ positivity was not related to atrophy; and (4) biomarkers of white matter lesions and inflammation were not associated with memory or attention/executive function when controlling for demographic covariates. Taken together, our findings indicate that Aβ pathology and brain atrophy are independent contributors to subtle memory decline in cognitively healthy elderly. Furthermore, Aβ pathology mainly influences memory in the younger part of the population, possibly through mechanisms such Figure 1. Effects of amyloid, tau, and hippocampal volume on memory function. (A) Shows the independent effects of amyloid pathology and HV on memory function, using a multivariable linear regression with ADAS-Cog delayed recall as dependent variable, and amyloid positivity, HV, and frontal (ns), and temporal cortical thickness (ns) as independent variables, controlling for age, sex, education, and total intracranial volume. (B) Shows the age-dependent effects of amyloid pathology, tau pathology, and HV on memory function. The effects of amyloid positivity and HV on memory performance were tested in the two age groups separately (see Suppl. Table 3). Two separate multivariable linear regressions were performed, in the younger group with amyloid positivity as independent variable and ADAS-Cog delayed recall as dependent variable (controlling for age, sex, and education), and in the older group with HV as independent variable and ADAS-Cog delayed recall as dependent variable (controlling for age, sex, education, and total intracranial volume). Secondarily, a simple mediation analysis was performed, analysing the associations between a) amyloid positivity and P-tau (controlling for age and sex) and b) P-tau and memory performance (controlling for amyloid pathology, age, and sex). Standardized beta coefficients are presented, ***p < 0.001, **p < 0.01, *p < 0.05. (2019) 9:11180 | https://doi.org/10.1038/s41598-019-47638-y www.nature.com/scientificreports www.nature.com/scientificreports/ as tau that do not require gross atrophy. With advancing age, the effect of brain atrophy seems to overtake the effect of Aβ on memory function.
Our findings are in agreement with previous studies where brain structure and Aβ pathology also were independently associated with memory performance in cognitively unimpaired, without an association between Aβ and atrophy 10,38 . Some studies have argued that the Aβ effect on memory is mediated by neurodegeneration 9,36 , at least to some degree 16,37 . However, the studies showing that neurodegeneration mediates the effect of Aβ on memory included patients with MCI in their analyses 9,16,36,37 , while the independent effect was seen when analysing cognitively unimpaired separately or adjusting for diagnosis as a co-variate 10,38 . One interpretation of this is that later on in the AD process, the Aβ effect on memory is in part mediated through atrophy, but in the preclinical stages of the disease, Aβ pathology affects memory performance without being associated with atrophy. Such atrophy-independent effects of Aβ could depend on early tau pathology, causing dysfunction of neurons or loss of synapses, without gross atrophy. This hypothesis is supported by the statistical mediation effect of P-tau in the present study, where Aβ no longer had a significant association with memory when including P-tau in the model (Fig. 1B, Suppl. Table 4). However, the effect of P-tau on memory was not very strong and a trend was still seen for Aβ (p = 0.079) and this mediation effect needs to be studied further.
The age dependent associations between amyloid pathology, hippocampal volume, and memory have in part been described before in cognitively unimpaired subjects, where memory function has been shown to be more vulnerable to hippocampal volume loss at older age 39,40 . This could imply that the function of other areas important for memory performance is impaired at higher age, contributing to worse memory without the need of as much hippocampal atrophy as in younger individuals. This is plausible considering age as a proxy of known and unknown processes, which can affect brain structure and function, such as TDP-43 accumulation 53 . Aβ was associated with memory in the younger but not the older participants. However, in the absence of a statistically significant interaction effect between amyloid and age on memory, the interpretation of this should be made with caution. This age difference could be explained by other pathologies being more common in the older group, which may overshadow the effect of Aβ pathology on memory.
An association with attention/executive function was seen for HV, but not for any of the cortical thickness measures. This could be due to a larger variability in the HV variable, making it easier to find an existing association. Also, there are substantial interindividual differences between cortical thickness measures, making these analyses hard to interpret in cross-sectional studies 54 .
This study has its limitations. First, as mentioned in the previous paragraph, it is a cross-sectional study, which means you cannot establish temporal changes of the variables. Second, studies have shown P-/T-tau to only exhibit moderate 55,56 or no 57 correlation with tau neuropathology, while the correlation between tau-PET (AV-1451) and tau neuropathology is stronger 58 . Therefor, using tau-PET instead of CSF P-tau in the mediation analysis may give different, and more accurate, results. Third, the memory test used only has ten levels and this in combination with the high overall cognitive performance may result in a ceiling effect. This would make it harder to find an actual association, which is a reason to interpret negative findings with some caution. (A) Shows the age-dependent associations between HV and memory. The effect of HV on memory performance was tested in the two age groups separately. Linear regression were tested with HV as independent variable and ADAS-Cog delayed recall as dependent variable, controlling for age, sex, education, and total intracranial volume. Results for the younger (blue) and older (green) participants are presented separately. (B) Shows the age-dependent associations between amyloid positivity and memory with a box-plot showing the results on ADAS-Cog, divided by age group (younger to the left, older to the right) and amyloid status (Aβ negative in blue, Aβ positive in green), unadjusted.