Plasma amyloid-β ratios in autosomal dominant Alzheimer’s disease: the influence of genotype

Abstract In vitro studies of autosomal dominant Alzheimer’s disease implicate longer amyloid-β peptides in disease pathogenesis; however, less is known about the behaviour of these mutations in vivo. In this cross-sectional cohort study, we used liquid chromatography-tandem mass spectrometry to analyse 66 plasma samples from individuals who were at risk of inheriting a mutation or were symptomatic. We tested for differences in amyloid-β (Aβ)42:38, Aβ42:40 and Aβ38:40 ratios between presenilin 1 (PSEN1) and amyloid precursor protein (APP) carriers. We examined the relationship between plasma and in vitro models of amyloid-β processing and tested for associations with parental age at onset. Thirty-nine participants were mutation carriers (28 PSEN1 and 11 APP). Age- and sex-adjusted models showed marked differences in plasma amyloid-β between genotypes: higher Aβ42:38 in PSEN1 versus APP (P < 0.001) and non-carriers (P < 0.001); higher Aβ38:40 in APP versus PSEN1 (P < 0.001) and non-carriers (P < 0.001); while Aβ42:40 was higher in both mutation groups compared to non-carriers (both P < 0.001). Amyloid-β profiles were reasonably consistent in plasma and cell lines. Within the PSEN1 group, models demonstrated associations between Aβ42:38, Aβ42:40 and Aβ38:40 ratios and parental age at onset. In vivo differences in amyloid-β processing between PSEN1 and APP carriers provide insights into disease pathophysiology, which can inform therapy development.


Introduction
Understanding Alzheimer's disease pathogenesis is critical to realizing disease-modifying treatments. Autosomal dominant Alzheimer's disease (ADAD), caused by mutations in presenilin 1/2 (PSEN1/2) or amyloid precursor protein (APP), is a valuable model for characterizing the molecular drivers of Alzheimer's disease. 1 PSEN1, the catalytic subunit of c-secretase, sequentially cuts APP: initial endopeptidase cleavage generates an amyloid-b (Ab) peptide, either Ab49 (major product) or Ab48 (minor product). 2 Subsequent proteolysis largely occurs down two pathways: Ab49 4 46 4 43 4 40 or Ab48 4 45 4 42 4 38. 3 As Ab49 is the predominant endopeptidase cleavage product, normal APP processing largely leads to Ab40 formation. 2 Pathogenic ADAD mutations alter APP processing resulting in more and/or longer, aggregation prone, amyloid-b peptides, which accelerate cerebral amyloid accumulation leading to typical symptom onset in a person's thirties to fifties. 4,5 Both APP and PSEN1/2 mutations increase production of longer (e.g. Ab42) relative to shorter (e.g. Ab40) peptides. 5 However, there are intriguing inter-mutation differences in amyloid-b profiles. PSEN1 mutant lines produce increased Ab42:38 ratios reflecting impaired c-secretase processivity. 5,6 In contrast, APP mutations at the c-secretase cleavage site increase Ab38:40 ratios, consistent with preferential processing down the Ab48 pathway. 6 So far, studies examining the influence of ADAD genotypes on amyloid-b ratios in vivo have been lacking.
Increasingly sensitive mass spectrometry-based assays now make it possible to measure concentrations of different amyloid-b moieties in plasma. 7 Therefore, we aimed to analyse plasma amyloid-b levels in an ADAD cohort, explore influences of genotype and clinical stage, and examine relationships between ratios and both parental age at onset (AAO) and estimated years to/from symptom onset (EYO), while also assessing consistency with in vitro models of amyloid-b processing.

Study design and participants
We recruited 66 participants from the longitudinal ADAD study at University College London (UCL); details have been described previously. 1 Samples were collected from August 2012 to July 2019 and concomitantly a semi-structured health questionnaire and Clinical Dementia Rating (CDR) scale were completed. 8 EYO was calculated by subtracting parental AAO from the participant's age. Participants were defined as symptomatic if global CDR was 40. ADAD mutation status, determined using Sanger sequencing, was provided only to statisticians, ensuring blinding of participants and clinicians. The study had local Research Ethics Committee approval and written informed consent was obtained from all participants or a consultee.

Measurement of plasma amyloid-b levels
EDTA plasma samples were processed, aliquoted and frozen at -80 C according to standardized procedures and shipped frozen to the Clinical Neurochemistry Laboratory, Sahlgrenska University Hospital, for analysis blinded to participants' mutation status and diagnosis. Samples were analysed using a liquid chromatographytandem mass spectrometry method using an optimized protocol for immunoprecipitation for improved analytical sensitivity ( Supplementary Figs 1 and 2). 9 Pooled plasma samples were used to track assay performance; intra-and inter-assay coefficients of variation were 55%.

Correlation of amyloid-b ratios in plasma and in induced pluripotent stem cell neurons
A sub-study investigated the consistency of amyloid-b profiles between plasma and induced pluripotent stem cell (iPSC)-derived neurons. Amyloid-b profiles were compared based on mutation for eight iPSC lines; data from six iPSC lines ave been previously reported by Arber et al. 6 Mutations tested were APP V717I (n = 2), PSEN1 intron 4 (n = 1), Y115H (n = 1), M139V (n = 1), R278I (n = 1) and E280G (n = 2). Plasma and iPSC samples were from the same participant or, where matched plasma was unavailable, plasma from a carrier of the same mutation and, if possible, a family member. Ab42:40, Ab38:40 and Ab42:38 ratios were normalized by taking the ratio of the value for each mutation carrier to the control median for each experimental setting (n = 27 non-carriers for plasma, n = 5 iPSC controls lines from non-ADAD families) (Supplementary Table 1, ratio values).
IPSC-neuronal amyloid-b was quantified as previously reported Arber et al. 6 Briefly, iPSCs were differentiated to cortical neurons for 100 days and then 48 h-conditioned culture supernatant was centrifuged removing cell debris. Amyloid-b was analysed via electrochemiluminescence on the MSD V-Plex Ab peptide panel (6E10), according to the manufacturer's instructions.

Statistical analysis
Summary descriptive statistics were calculated by mutation type (PSEN1, APP, non-carriers) and box plots produced for Ab42:38, Ab38:40 and Ab42:40 ratios. Box plots were presented by mutation type (PSEN1 versus APP versus non-carriers), and then individually for PSEN1 and APP carriers by clinical stage (presymptomatic versus symptomatic versus non-carriers) (Fig. 1).
Amyloid-b ratios are displayed on logarithmic scales. Age-and sex-adjusted differences were estimated between mutation type for each ratio, as were differences by clinical stage for each ratio, separately for APP and PSEN1 carriers. These comparisons were made using mixed models including random intercepts for clusters comprising individuals from the same family and group, with random intercept and residual variances allowed to differ for the groups being compared. Pairwise comparisons were only carried out if a joint test provided evidence of differences. Ratios were log-transformed; estimated coefficients were back-transformed to multiplicative effects.
The relationship between parental AAO, EYO and age (EYO = age -AAO) means that it is not possible to estimate separate effects of AAO and EYO on amyloid-b ratios adjusting for age using a conventional statistical analysis: if age is held constant, then a 1year increase in AAO implies a 1-year decrease in EYO and vice versa, hence their effects are aliased. However, the aim here should be to allow for 'normal ageing' (as observed in non-carriers), and this is possible. For each combination of mutation carrier group (PSEN1 and APP) and amyloid-b ratio a separate mixed model was fitted jointly to the carrier group and the non-carrier group. Each model allowed the logarithm of the amyloid-b ratio to depend on AAO, EYO and sex (but not age) in the carrier group, and on just sex and age (estimating 'normal ageing') in the non-carrier group. Random effects were included as in the between group comparisons above. In the carrier group, the effect of AAO adjusted for EYO, sex and (non-carrier) 'normal ageing' was obtained by subtracting the 'normal ageing' effect from the AAO effect (adjusted for sex and EYO). Analogously the effect of EYO adjusted for AAO, sex and 'normal ageing' was obtained by subtracting the 'normal ageing' effect from the EYO effect (adjusted for sex and AAO) in the carrier group. For Ab42:38 in PSEN1 carriers there was evidence also to include a quadratic term for parental AAO. For each analysis, the estimated geometric mean ratio (and 95% confidence interval, CI) was plotted against parental AAO, standardizing to an equal mix of males/females, an EYO of 0 (i.e. the point of symptom onset), and adjusted for 'normal ageing' relative to age 43 (the average age of mutation carriers). Analogous plots of estimated geometric mean ratio (and 95% CI) against EYO were standardized to an equal mix of males/females, an AAO of 43 (average age of mutation carriers) and adjusted for 'normal ageing' relative to age 43.
Spearman correlation coefficients were calculated to assess the association between plasma and iPSC-neuron amyloid-b ratios.
Analyses were performed using Stata v.16.

Data availability
Data are available on reasonable request from qualified investigators, adhering to ethical guidelines.

Results
Demographic and clinical characteristics are presented in Table 1 for the 27 non-carriers and 39 mutation carriers (28 PSEN1, 11 APP). Mutation details are provided in Supplementary Table 2.
In PSEN1 and APP carriers, models that adjusted for sex, parental AAO and 'normal ageing' did not find any significant association between either Ab42:40, Ab42:38 or Ab38:40 and EYO (Supplementary Figs 4 and 5) (P 5 0.06). However, in APP carriers there was weak evidence of an association between Ab42:40 and EYO; a 1-year increase in EYO was associated with a 0.8% decrease (95% CI: 1.6% decrease, 0.0% increase, P = 0.06) in the geometric mean of Ab42:40.
Amyloid-b ratios in plasma and iPSC-conditioned media were highly associated for both Ab42:40 (q = 0.86, P = 0.01) and Ab38:40 (q = 0.79, P = 0.02), somewhat less so for Ab42:38 (q = 0.61, P = 0.10) (Fig. 3). While we did not observe perfect agreement in the Ab42:38 ratio between plasma and iPSC lines (shown by a solid line in Fig. 3), the direction of change in this ratio, i.e. either increased or decreased when compared to controls, was largely consistent across media.

Discussion
In this study, we found increases in plasma Ab42:40 in both APP and PSEN1 carriers compared to non-carriers and marked differences in amyloid-b ratios between genotypes: Ab42:38 was higher in PSEN1 versus APP, and Ab38:40 was higher in APP versus PSEN1. Importantly, more aggressive PSEN1 mutations (those with earlier ages of onset) had higher Ab42:40 and Ab42:38 ratios-in vivo evidence of the pathogenicity of these peptide ratios.
in PSEN1 may be attributed to reduced conversion of Ab42 (substrate) to 38 (product) relative to non-carriers; in contrast, APP carriers showed near identical Ab42:38 ratios compared to noncarriers. Strikingly, increases in Ab42 relative to shorter amyloidb moieties (440) were associated with earlier disease onset in PSEN1. Importantly, there were no associations between amyloidb ratios and EYO in PSEN1 carriers, suggesting these ratios represent molecular drivers of disease as opposed to being markers of disease stage. Our in vivo results recapitulate cell-based findings of reduced efficiency of c-secretase processivity in PSEN1 6,10,11 ; inefficiency attributed to impaired enzyme-substrate stability causing premature release of longer amyloid-b peptides. 10 Parental AAO is an indicator of disease severity, with a younger AAO implying a more deleterious mutation. In PSEN1, Ab42:38 (a read-out of the efficiency of the fourth c-secretase cleavage) showed a deceleration in the rate of change as parental AAO increases. This further supports the central pathogenic role of csecretase processivity in ADAD, especially in younger onset, aggressive forms of PSEN1.
In APP, production of Ab38 relative to Ab40 was increased. This is consistent with a shift in the site of endopeptidase cleavage causing increased generation of Ab48; the precursor substrate in the Ab38 production line. Our study included APP mutations located near the c-secretase cleavage site. Previous cell-based work involving mutations around this site also demonstrated increased trafficking along the Ab48 pathway. 5,6,11 In contrast, APP duplications or mutations near the b-secretase site are associated with non-differential increases in amyloid-b production. 12 Changes in Ab38:40 were also seen in PSEN1 carriers; levels were reduced compared to both APP carriers and non-carriers. Declines in Ab38:40 may reflect mutation effects on endopeptidase cleavage and/or c-secretase processivity; changes in both processes have been described in in vitro studies of PSEN1. 6,13 Premature release of longer (4Ab43) peptides may contribute to falls in Ab38:40; both increasing amyloid-b length and pathogenic PSEN1 mutations are associated with destabilization of the enzyme-substrate complex. 10 It will be important for future research to investigate the exact molecular drivers of declines in Ab38:40 in PSEN1, especially as lower levels were associated with earlier disease onset.
We also saw inter-stage differences in APP processing; Ab42:40 was higher in symptomatic compared to presymptomatic PSEN1 carriers. The reason for this is unclear and should be treated cautiously given the small group sizes and the absence of inter-stage differences in Ab42:40 among APP carriers. However, post-symptomatic increases in plasma Ab42 have been reported in Down syndrome. 14 It is possible that downstream pathogenic consequences of ADAD, such as cerebral amyloid angiopathy, may interact with, and modify, plasma levels. Additionally, as amyloid-b is produced peripherally in organs, muscle and platelets, systemic factors may contribute to inter-stage differences. 15 Our results support the hypothesis that ADAD mutations increase in vivo production of longer amyloid-b peptides (Ab 5 42) relative to Ab40. This is consistent with cell-and blood-based studies in ADAD. 11,16 Additionally, we showed plasma amyloid-b profiles were recapitulated in iPSC-media with consistent profiles for the same mutation. There is some evidence that Ab42:40 ratios also increase in the CSF of mutation carriers far from onset; however, CSF levels then fall significantly during the two decades before symptom onset 17 ; reductions are attributed to 'trapping' of longer peptides within cerebral plaques. 18 In sporadic Alzheimer's disease CSF, as well as plasma, Ab42:40 levels also fall as cerebral amyloid plaques start to accumulate, with ratio levels remaining low thereafter. 19 In contrast, we show that plasma Ab42:40 in both APP and PSEN1 carriers was raised and did not fall below non-carriers' levels, either before or after symptom onset. Taken together, these findings suggest that plasma amyloid-b ratios in ADAD are less susceptible to the effects of sequestration.
Study limitations include the small sample size, due to the rarity of ADAD; however, we included a reasonably wide array of mutations. Second, ages at onset were estimated from parental AAO, while this offers a reasonable estimate there is variability within families and imprecision in determining AAO in a preceding, often deceased, generation. 20 Finally, future studies should measure amyloid-b moieties longer than Ab42, and also investigate interactions between central and peripheral amyloid-b production (we lacked paired CSF).
In conclusion, we demonstrate the impact of pathogenic ADAD mutation on APP processing in vivo. We show marked inter-mutation difference in Ab profiles, with relative increases in longer peptides being associated with earlier disease onset. Our findings indicate that plasma amyloid-b ratios in ADAD may be useful biomarkers of APP processing. This is especially important as we enter an era of gene silencing therapies and personalized medicine, Figure 3 Comparison of amyloid-b processing in vivo and in vitro. Scatterplot comparing amyloid-b ratios profiles in plasma and iPSCderived neurons for eight mutation carriers. One to one comparison of amyloid-b ratios normalized to the median of controls for each experimental setting (n = 27 non-carrier controls for plasma, n = 5 iPSC lines from controls who were not members of ADAD families); values 41 indicate higher ratio in a mutation carrier compared to the median of controls, whereas values 51 indicate a lower ratio in a mutation carrier compared to the median of controls. Matched samples (plasma and iPSC samples donated by the same donor) are identified with triangle symbols. Unmatched samples (plasma and iPSC samples donated by different participants who carry the same mutation, and where possible are members of the same family) are identified by square symbols. The y-axis scale is logarithmic in all panes. Spearman's q and the associated P-value are shown for each scatter plot. The line displayed on each scatterplot represents line of perfect agreement i.e. x = y.
where direct read-outs of gene function will be particularly valuable.