Genetic mapping of APP and amyloid-β biology modulation by trisomy 21

Individuals who have Down syndrome (DS) frequently develop early onset Alzheimer’s disease (AD), a neurodegenerative condition caused by the build-up of aggregated amyloid-β and tau proteins in the brain. Amyloid-β is produced by amyloid precursor protein (APP), a gene located on chromosome 21. People who have Down syndrome have three copies of chromosome 21 and thus also an additional copy of APP; this genetic change drives the early development of Alzheimer’s disease in these individuals. Here we use a combination of next-generation mouse models of Down syndrome (Tc1, Dp3Tyb, Dp(10)2Yey and Dp(17)3Yey) and a knockin mouse model of amyloid-β accumulation (AppNL-F) to determine how chromosome 21 genes, other than APP, modulate APP/amyloid-β in the brain when in three copies. Using both male and female mice, we demonstrate that three copies of other chromosome 21 genes are sufficient to partially ameliorate amyloid-β accumulation in the brain. We go on to identify a subregion of chromosome 21 that contains the gene/genes causing this decrease in amyloid-β accumulation and investigate the role of two lead candidate genes Dyrk1a and Bace2. Thus an additional copy of chromosome 21 genes, other than APP, can modulate APP/amyloid-β in the brain under physiological conditions. This work provides critical mechanistic insight into the development of disease and an explanation for the typically later age of onset of dementia in people who have ADDS, compared to those who have familial AD caused by triplication of APP.


Introduction
DS is caused by trisomy of human chromosome 21 (Hsa21), and occurs in around 1/1000 live births in Europe (de Graaf, Buckley et al. 2021). Most individuals who have DS develop the neuropathological features of AD; amyloid-β plaques and tau neurofibrillary tangles by the age of 50 (Davidson, Robinson et al. 2018), and 80% of individuals will have developed dementia by age 65 (McCarron, McCallion et al. 2017). The high prevalence of AD in DS is in part due to the gene encoding amyloid precursor protein (APP) being located on Hsa21, thereby raising APP and amyloid-β protein levels (Glenner and Wong 1984, Cheon, Dierssen et al. 2008, Doran, Keator et al. 2017. Recent studies in preclinical systems have demonstrated that an extra copy of other genes on Hsa21 can modulate APP biology (Garcia-Cerro, Rueda et al. 2017, Wiseman, Pulford et al. 2018, Alic, Goh et al. 2020, Tosh, Rhymes et al. 2021) and thus may alter the earliest stages of AD in individuals who have DS. These extra genes may act to promote or to reduce amyloid-β accumulation, causes of AD is confounded by the different diagnostic criteria used (as necessitated by the underlying intellectual disability that occurs in people who have DS) (Benejam, Videla et al. 2020). Understanding these processes is crucial to the appropriate selection of treatments for AD-primary prevention trials in people who have DS.
Previous in vivo studies have either examined the processing of endogenous mouse APP or used APP transgenic models to address this biology, but both of these approaches have limitations (Garcia-Cerro, Rueda et al. 2017, Sasaguri, Nilsson et al. 2017, Wiseman, Pulford et al. 2018, Tosh, Rhymes et al. 2021. Mouse APP differs in sequence from the human protein. In the amyloid-β region these differences reduce both cleavage of the protein by β-secretase and the tendency of the amyloid-β generated to aggregate (Serneels, T'Syen et al. 2020), thus limiting our ability to determine how changes to biology affect accumulation of amyloid-β -a key early aspect of AD. The over-and mis-expression of APP in transgenic mouse models may cause artefactual phenotypes, masking the modulatory effect of the extra copy of Hsa21 genes and causing elevated mortality which may confound data interpretation (Saito, Matsuba et al. 2014, Sasaguri, Nilsson et al. 2017).
Here we take a combinatorial approach: assessing the effect of an additional copy of Hsa21 genes (using a series of DS mouse models (O'Doherty, Ruf et al. 2005, Yu, Li et al. 2010, Lana-Elola, Watson-Scales et al. 2016)) on the biology of endogenous mouse APP and on APP generated from a partially humanised App knock-in allele that also carries AD-causal Swedish (NL) and Iberian (F) point mutations, which does not cause elevated mortality (App NL-F ) ( Fig. 1) (Saito, Matsuba et al. 2014). These data indicate that trisomy of genes on Hsa21 reduces amyloid-β accumulation and that people who have DS are partly protected from their raised APP gene dose by the additional copy of other genes on the chromosome.
We go on to show that one of the gene or genes that cause this change in biology is located on mouse chromosome 16 between Mir802 and Zbtb21. This region contains 38 genes and we specifically test if mechanisms linked to two lead candidate genes in this region, Dyrk1a and Bace2, occur in our novel in vivo AD-DS model system.

Animal welfare and husbandry
All experiments were undertaken in accordance with the Animals (Scientific Procedures) Act 1986 (United Kingdom), after local institutional ethical review by the Medical Research Council, University College London and in accordance with ARRIVE2 guidelines (Percie du Sert, Hurst et al. 2020). Mice were housed in individually ventilated cages (Techniplast) with grade 5, autoclaved dust-free wood bedding, paper bedding and a translucent red "mouse house" at a specific pathogen free facility. Free-access to food (Picolab Rodent Diet 20 Labdiet) and water was provided. The animal facility was maintained at a constant temperature of 19-23°C with 55 ± 10% humidity in a 12 h light/dark cycle. Pups were weaned at 21 days and moved to standardised same-sex group housing with a maximum of 5 mice per cage. (App tm2.1Tcs , MGI:5637816), Tc1 (Tc(HSA21)1TybEmcf, MGI:3814712), Dp3Tyb (Dp(16Mir802-Zbtb21)3TybEmcf, MGI:5703802), Dp(10)2Yey (Dp(10Prmt2-Pdxk)2Yey, MGI:4461400) and Dp(17)3Yey (Dp(17Abcg1-Rrp1b)3Yey, MGI:4461398). Tc1 mice were maintained by mating Tc1 females to F1 (129S8 × C57BL/6) males. All other mouse strains were maintained by backcrossing males and females to C57BL/6J mice (imported from the Jackson Laboratory). Experimental cohorts for Tc1, Dp(10)2Yey and Dp(17)3Yey studies were produced by crossing mice carrying the additional Hsa21 or Hsa21 orthologous duplications with App NL-F/+ animals in a two-generation cross to generate all required genotypes from the second generation (wild-type, Tc1/Dpx, App NL-F/NL-F , Tc1/ Dpx;App NL-F/NL-F , Fig. 1, Table 1). As both the Dp3Tyb segmental duplication and the App NL-F gene are located on mouse chromosome 16 (Mmu16), for this cross we first generated a Dp3Tyb-App NL-F recombinant Mmu16, by crossing the two lines together and then back-crossing to C57BL/6J mice to identify recombined Mmu16, carrying both genetic changes on the same chromosome. Mice with the recombined Mmu16 were then crossed with App NL-F/+ animals to generate Dp3Tyb;App NL-F/NL-F progeny. For this cross App NL-F/NL-F controls were generated from App NL-F/+ x App NL-F/+ matings, in addition to rare re-recombinations resulting in offspring without the Dp3Tyb segmental duplication but two copies of the App knock-in allele. Dp3Tyb controls were generated from Dp3Tyb x C57BL/6J matings generated at the same time as the Dp3Tyb;App NL-F/NL-F mice. Wild-type (WT) controls were taken from all three matings.
Animals were euthanized by exposure to rising carbon dioxide, followed by confirmation of death by dislocation of the neck in accordance with the Animals (Scientific Procedures) Act 1986 (United Kingdom).

Tissue preparation and western blotting
For analysis of protein abundance in the hippocampus and cortex, tissues were dissected under ice cold phosphate-buffered saline (PBS) before snap freezing. Samples were then homogenised in RIPA Buffer (150 mM sodium chloride, 50 mM Tris, 1% NP-40, 0.5% sodium deoxycholate, 0.1% sodium dodecyl sulfate) plus complete protease inhibitors (Calbiochem) by mechanical disruption. Total protein content was determined by Bradford assay or Pierce 660 nm assay (Thermo Fisher). Samples from individual animals were analysed separately and were not pooled.
Equal amounts of total brain proteins were then denatured in LDS denaturing buffer (Invitrogen) and β-mercaptoethanol, prior to separation by SDS-PAGE gel electrophoresis using precast 4-12% Bis-Tris gels (Invitrogen). Proteins were transferred to nitrocellulose or PVDF membranes prior to blocking in 5% milk/PBST (0.05% Tween 20) or 5-10% bovine serum albumin (BSA)/PBST. Primary antibodies were diluted in 1% BSA/PBST, HRP-conjugated secondary anti-rabbit, anti-mouse and anti-goat antibodies (Dako) were diluted 1:10,000 in 1% BSA/PBST. Linearity of antibody binding was confirmed by a 2-fold dilution series of cortical protein samples. Band density was analysed using Image J. Relative signal of the antibody of interest compared to the internal loading control was then calculated, and the relative signal was then normalized to the mean relative signal of control samples electrophoresed on the same gel. Means of technical replicates were calculated and used for ANOVA, such that biological replicates were used as the experimental unit.
Biochemical fractionation of mouse brain tissues for the analysis of human amyloid-β Cortical proteins were fractionated as described in Shankar et al. (2009). A half cortex was weighed on a microscale and homogenised in 4 volumes of ice-cold Tris-buffered saline (TBS) (50mM Tris-HCl pH 8.0) containing a cocktail of protease and phosphatase inhibitors (Calbiochem) using a handheld mechanical homogeniser and disposable pestles (Anachem). Samples were then transferred to 1.5 ml microfuge tubes (Beckman Coulter #357448), balanced by adding more TBS and centrifuged at 175,000 × g with an RC-M120EX ultracentrifuge (Sorvall) fitted with rotor S100AT5 at 4 °C for 30 mins. Supernatant (the Tris-fraction) was removed and stored at -80 °C. The remaining pellet was homogenised in 5 volumes of ice-cold 1% Triton X-100 (Sigma-Aldrich) in TBS (50mM Tris-HCl pH 8.0), balanced and centrifuged at 175,000 × g for 30 mins at 4 °C. The resultant supernatant (the Triton-soluble fraction) was removed and stored at -80 °C. The pellet was then re-suspended in 8 volumes (by original cortical weight) of TBS (50mM Tris-HCl pH 8.0), containing 5 M guanidine HCl and left overnight at 4 °C on a rocker to ensure full re-suspension, and subsequently stored at -80 °C. A Bradford assay or Pierce 660 nm protein assay (Thermo Fisher) was performed to determine protein concentration.

Biochemical preparation of mouse brain tissues for the analysis of mouse amyloid-β
A half cortex was weighed on a microscale and homogenised in 3 volumes of ice-cold TBS (50mM Tris-HCl pH 8.0) containing a cocktail of protease and phosphatase inhibitors (Calbiochem) using a handheld mechanical homogeniser and disposable pestles (Anachem) based on the method in reference (Holtta, Hansson et al. 2013). Homogenates were centrifuged at 21130 × g at 4 °C for 1 hour and the resultant supernatant was stored at -80 °C for onward analysis.

Immunohistochemistry
Half brains were immersion fixed in 10% buffered formal saline (Pioneer Research Chemicals) for a minimum of 48 hours prior to being processed to wax (Leica ASP300S tissue processor). The blocks were trimmed laterally from the midline by ~0.9-1.4 mm to give a sagittal section of the hippocampal formation. Two 4 μm sections at least 40 μm apart were analysed. The sections were pre-treated with 98% formic acid for 8 mins, followed by washing. The slides were wet loaded onto a Ventana XT for staining (Ventana Medical Systems, Tucson, AZ, USA). The protocol included the following steps: heat induced epitope retrieval (mCC1) for 30 mins in Tris Boric acid EDTA buffer (pH 9.0), superblock (8 mins) and manual application of 100 μl directly biotinylated mouse monoclonal IgG1 antibody against the N-terminus of Aβ (82E1, IBL, 0.2 μg/ml) for 8 hours. Staining was visualised using a ChromoMap DAB kit followed by counterstaining with haematoxylin. The sections were dehydrated, cleared, and mounted in DPX prior to scanning (Leica SCN400F slide scanner). All images were analysed using ImageJ and by manual plaque counting by two independent researchers.

Biochemical preparation of mouse brain tissues for mass spectrometry of amyloid-β
A half cortex was weighed on a microscale and was homogenized in 5 volumes of tris(hydroxymethyl)aminomethane (Tris)-buffered saline (TBS), pH 7.6, containing cOmplete™ Protease Inhibitor Cocktail (Roche, cat: 04693116001). For the homogenization, one 5 mm bead per sample was used in a TissueLyser (Qiagen) for 4 min at 30 Hz. After homogenization, additional TBS with protease inhibitor cocktail was added up to 1 ml and transferred to a new tube to be centrifuged at 31,000 g for 1 h at 4 °C. The pellet was resuspended in 1 ml of 70% formic acid (FA) (v/v), followed by further homogenization in the TissueLyser for 2 min at 30 Hz and subsequent sonication for 30 s. The homogenate was centrifuged again at 31,000 g for 1 h at 4 °C and the supernatant (FA fraction) was dried down in a vacuum centrifuge.
Initially, 400 μl of 70% FA (v/v) was added to the dried FA fractions, shaken for 30 min at 21°C and centrifuged at 31,000 g for 1 h at4 °C. After the removal of the supernatant, neutralization with 8 ml 0.5 M Tris was performed. Immunoprecipitation (IP) was performed as previously described with some modifications . Briefly, 50 μl of sheep anti-mouse magnetic beads (Thermo Fisher Scientific) that had previously been linked with 4 μg each of mouse monoclonal 6E10 and 4G8 antibodies (Biolegend) were added to the neutralized FA fraction. This complex was incubated overnight at 4 °C in 0.2% Triton X-100 in PBS (v/v). By using an automated magnetic-particle KingFisher ml system (Thermo Fisher Scientific), the samples were then washed with PBS Triton X-100, PBS, and 50 mM ammonium bicarbonate separately before elution in 100 μl 0.5% FA. Eluates were dried down in a vacuum centrifuge and stored at -80 °C pending mass spectrometry (MS) analysis.

Mass spectrometry
Liquid chromatography-mass spectrometry (LC-MS) was conducted in a similar manner as described previously (Portelius, Tran et al. 2007). Briefly, a nanoflow liquid chromatograph was coupled to an electrospray ionization (ESI) hybrid quadrupole-orbitrap tandem MS (Dionex Ultimate 3000 system and Q Exactive, both Thermo Fisher Scientific). Samples were reconstituted in 7 μl 8% FA/8% acetonitrile in water (v/v/v) and loaded onto an Acclaim PepMap 100 C18 trap column (length 20 mm; inner diameter 75 μm; particle size 3 μm; pore size 100 Å) for online desalting, and thereafter separated on a reversed-phase Acclaim PepMap RSLC column (length 150 mm, inner diameter 75 μm; particle size 2 μm; pore size 100 Å) (both Thermo Fisher Scientific). Mobile phases were A: 0.1% FA in water (v/v), and B: 0.1% FA/84% acetonitrile in water (v/v/v). The flow rate was 300 nl/min and a linear gradient of 3-40% B for 50 min was applied. The temperature of the column oven was 60 °C. Mass spectrometer settings were as follows: positive ion mode; mass-to-charge (m/z) interval 350-1800 m/z units; data dependent acquisition with 1 precursor ion acquisition (MS) followed by up to 5 fragment ion acquisitions (MS/MS); resolution setting 70,000 (for both MS and MS/MS); number of microscans 1 (MS and MS/MS); target values 10 6 (MS and MS/MS); maximum injection time 250 ms (MS and MS/MS); fragmentation type was higher-energy collisional dissociation fragmentation (HCD); normalised collision energy (NCE) setting 25; singly charged ions and ions with unassigned charge were excluded for MS/MS selection. Database search (including isotope and charge deconvolution) and label free quantification was performed with PEAKS Studio v8.5 (Bioinformatics Solutions Inc.) against a custom-made APP database. All suggested fragment mass spectra were evaluated manually.MS signal was normalised to starting weight of the cortex prior to data analysis.

Statistical analysis and experimental design
All experiments and analyses were undertaken blind to genotype and sex. A 6-digit identification number was allocated to each animal prior to genotyping which was used to blind samples. Experimental groups for all experiments were pseudorandomised using Mendelian inheritance. Mice not carrying the correct combination of alleles (i.e. App NL-F/+ heterozygous mice) were excluded from the analysis, no other animals were excluded. Some samples were lost from the study because of technical issues during tissue processing to wax or during fractionation. Data were analysed as indicated in figure legends by univariate ANOVA with between-subject factors being genetic status (App +/+ /App NL-F/NL-F and/or wild-type/Hsa21 or wild-type/Dpx) and sex. Fractionation batch was included as a between-subject factor for analysis of amyloid-β by MSD assays. The subject means of technical replicates were calculated and used in the ANOVA for western blot and MSD assays, as the number of replicates for which data was available varied between samples. Repeat measures ANOVA was used for manual plaque counts (combining the data of two independent researchers).

Trisomy of chromosome 21 decreases accumulation of amyloid-β in the cortex of the App NL-F mouse model
Following on from our previous studies, which indicated that an additional copy of Hsa21 alters APP biology and the accumulation of amyloid-β in vivo in an APP transgenic model (Wiseman, Pulford et al. 2018, Tosh, Rhymes et al. 2021), here we determined if an additional copy of Hsa21 modulated APP/amyloid-β biology in the App NL-F knock-in mouse model. We undertook a two generation cross of the Tc1 mouse model of DS (O'Doherty, Ruf et al. 2005), which contains a freely segregating copy of Hsa21 (but not a functional additional copy of APP) (Gribble, Wiseman et al. 2013), with the App NL-F model to generate 4 genotypes of mice (wild-type, Tc1, App NL-F/NL-F and Tc1;App NL-F/NL-F ).
We quantified the number of 82E1 + amyloid-β deposits in the cortex of these 4 genotypes of mice at 8-months of age. 82E1 + deposits were not observed in wild-type or Tc1 mice consistent with our previous study (Wiseman, Pulford et al. 2018). We found a significant decrease in the number of deposits in Tc1;App NL-F/NL-F compared with App NL-F/NL-F controls (Fig. 2 A, G).
We also determined if trisomy of Hsa21 modulated the biochemical aggregation of amyloid-β 40 and amyloid-β 42 in the cortex at 8-months of age, using biochemical protein fractionation by step-wise homogenisation and ultracentrifugation in sequentially more disruptive solutions (Tris-HCl, Tris-HCl 1% Triton X-100 and finally 5 M guanidine hydrochloride). We then quantified human amyloid-β 40 and amyloid-β 42 in each fraction normalised to starting brain weight (6E10 MSD triplex assay). Samples from mice without a humanised App allele, that do not produce human amyloid-β, were used as negative controls. Amyloid-β 42 in the guanidine hydrochloride fraction was not significantly reduced in the presence of the extra copy of Hsa21 (Fig. 2B). A significant increase in Tris-soluble amyloid-β 42 was seen in the cortex of Tc1;App NL-F/NL-F compared with App NL-F/NL-F controls (Fig. 2D). However, we note that this effect is being driven by one outlier and significance is lost if this animal is excluded from the analysis moreover this analyte could not be detected in a significant proportion of the samples analysed. Thus a replication study is required to determine the validity of this result. No significant difference in Triton-soluble amyloid-β 42 abundance was observed (Fig. 2C). The amount of human amyloid-β 40 in the App NL-F/NL-F model is very low because of the Iberian mutation in the modified App allele and this analyte was below the limit of detection in the Tris-soluble fraction and did not significantly differ between genotypes in the Triton and 5 M guanidine hydrochloride fractions (Fig. 2E, F). These data indicate that an additional copy of a Hsa21 gene or genes is sufficient to reduce the deposition of amyloid-β in the cortex, this may occur via an effect on amyloid-β formation, clearance or aggregation,

Decreased amyloid-β accumulation in the Tc1-App NL-F/NL-F model does not occur because of a reduction of APP or CTF-β abundance
The significant increase in Tris-soluble amyloid-β 42 in the Tc1;App NL-F/NL-F cortex suggests that the decrease in amyloid-β accumulation observed is likely to be caused by a change in peptide aggregation. However, previous studies have suggested that genes on Hsa21, other than APP, can increase APP protein level in vivo and modulate the abundance of the amyloid-β precursor APP-C-terminal fragment-β (CTF-β) (Garcia-Cerro, Rueda et al. 2017, Naert, Ferre et al. 2018, Wiseman, Pulford et al. 2018). Thus, we used western blotting to determine if the additional Hsa21 genes altered APP or CTF-β abundance in our new model system. Here, we found no evidence of decreased mouse or human full-length APP (FL-APP) in the cortex of the Tc1 and Tc1;App NL-F/NL-F compared with wild-type and App NL-F/NL-F mice (Fig. 3A, D). We note that significantly less FL-APP is detected in humanised models compared with controls (using antibody A8717), this may reflect a reduction in antibody binding rather than a biological reduction in protein level.
We found a significant increase in CTF-β and a significant decrease in CTF-α in the cortex of App NL-F/NL-F mice consistent with the reported effects of the introduced mutations on APP processing (Fig. 3B, C, D) (Saito, Matsuba et al. 2014). An additional copy of Hsa21 did not significantly alter wild-type APP CTF-β or CTF-α levels in the cortex of the Tc1 compared to wild-type-mice or in the Tc1;App NL-F/NL-F compared to App NL-F/NL-F . This finding contrasted with the large increase in CTF-β in male Tc1;APP transgenic model that we previously reported (Wiseman, Pulford et al. 2018). These data suggest that the reduction in deposition of amyloid-β in the Tc1;App NL-F/NL-F model is likely mediated by an enhancement of amyloid-β clearance or an impairment of peptide aggregation rather than a decrease in APP or CTF-β abundance.

Decreased accumulation of amyloid-β is caused by an additional copy of 38 Hsa21 orthologous genes
Many DS-associated phenotypes are multigenic, caused by the combined effect of multiple Hsa21 genes acting together on one biological pathway. To understand the mechanisms underlying the decrease in amyloid-β accumulation in the Tc1;App NL-F/NL-F model further, we used a series of mouse models of DS that carry an extra copy of subregions of mouse chromosomes that are orthologous with Hsa21, to identify the combination of regions/genes responsible for the changes (Fig. 1) (Yu, Li et al. 2010, Lana-Elola, Watson-Scales et al. 2016). We used 3 mouse lines that carry similar gene content to the Tc1 mouse model. However, because of the limitations of available models we were unable to explore the Hsa21 genes closest to App on Mmu16, which are in three copies in the Tc1 model, as using recombination to generate the required combination of alleles in the available Dp9Tyb model was not feasible.
An extra copy of the genes between Mir802 and Zbtb21 (Dp3Tyb) was sufficient to decrease the accumulation of amyloid-β in the cortex, as quantified by 82E1 plaque counts (Fig. 4A). However, an extra copy of the genes between Abcg1 and Rrp1b (Dp(10)2Yey) and between Prmt2 and Pdxk (Dp(17)3Yey) was not sufficient to significantly alter the accumulation of amyloid-β in the cortex, as quantified by 82E1 plaque counts in 8-month old mice (Fig. 4B, C).
The abundance of amyloid-β 42 in the guanidine hydrochloride fraction was not significantly altered in the Dp3Tyb;App NL-F/NL-F , Dp(10)2Yey;App NL-F/NL-F or Dp(17)3Yey;App NL-F/NL-F mice compared with App NL-F/NL-F controls at 8-months of age ( Fig. 5A-C). No significant difference in the abundance of Tris-or Tritonsoluble amyloid-β 42 was observed in Dp3Tyb;App NL-F/NL-F , Dp(10)2Yey;App NL-F/NL-F or Dp(17)3Yey;App NL-F/NL-F compared to controls (Fig. 5D-I). No difference in the abundance of guanidine hydrochloride or Triton-soluble amyloid-β 40 was observed between Dp(10)2Yey;App NL-F/NL-F or Dp(17)3Yey;App NL-F/NL-F compared to controls (Fig. 6A-D). These data indicate that a gene or genes in 3-copies in the Dp3Tyb model is sufficient to decrease the deposition of amyloid-β in the brain.

Increased DYRK1A does not lead to increased APP or amyloid-β in the Dp3Tyb model of DS
The Dp3Tyb model contains an additional copy of 38 genes, including Dyrk1a, a kinase that phosphorylates APP (Ryoo, Cho et al. 2008), increasing the abundance of the protein in vivo, contributing to raised soluble amyloid-β abundance in the Ts65Dn mouse model of DS (Garcia-Cerro, Rueda et al. 2017). We wanted to determine whether we could observe this previously reported biology in our new model system. We found that an extra copy of the Dp3Tyb region raises the abundance of DYRK1A in the cortex of 3-month-old animals, including in the context of App NL-F knock-in mutations ( Fig. 7A-B). This is consistent with numerous previous reports of dosage sensitivity of DYRK1A in the mouse throughout lifespan (Sheppard, Plattner et al. 2012, Garcia-Cerro, Rueda et al. 2017, Yin, Jin et al. 2017. We also note that consistent with previous reports in other mouse model systems, 3-copies of Dyrk1a in the Dp3Tyb;App NL-F/NL-F model was associated with an increase in cortical weight (Guedj, Pereira et al. 2012) (Fig. 10A).
However, we found no evidence of changes to the abundance of humanised-APP or mouse-APP in Dp3Tyb;App NL-F/NL-F or Dp3Tyb models compared to App NL-F/NL-F controls in the cortex at 3-months of age ( Fig. 7D-E). Similarly no change in human-CTF-β or mouse-CTFβ was observed in the Dp3Tyb model (Fig. 7F, G). We also found no change in human/mouse APP or CTF-β abundance in the Dp(10)2Yey;App NL-F/NL-F mouse (Fig. 8).
We found no evidence of changes to total mouse amyloid-β 40 or amyloid-β 42 in young Dp3Tyb compared with wild-type controls (3-months of age) or insoluble human amyloidβ 40 or amyloid-β 42 in young (3-months of age) Dp3Tyb;App NL-F/NL-F compared with App NL-F/NL-F controls ( Fig. 9, 11). These data suggest that the decreased accumulation of human amyloid-β in the Dp3Tyb model is not likely to be the result of changed abundance of APP, CTF-β or amyloid-β in the young brain and a mechanism that decreases aggregation or enhances clearance of amyloid-β may be causal, consistent with the data from the Tc1;App NL-F/NL-F model.
As we had observed significant variability in the abundance of BACE2 in the cortex, we investigated whether BACE2 protein levels within individual mice predicted the abundance of amyloid-β 1-19 , amyloid-β 1-20 and amyloid-β 1-34 in the same animals. No relationships between the level of BACE2 protein and the analytes were observed (amyloid-β 1-19 R 2 = 0.0001, amyloid-β 1-20 R 2 = 0.0079, amyloid-β 1-34 R 2 = 0.0004). These data suggest that differences in BACE2 protein abundance in the young adult mouse brain are not sufficient to cause a detectable alteration in the clearance of amyloid-β or enhanced θcleavage. One caveat is that at 3-months of age we cannot yet detect an alteration of the aggregation of human amyloid-β in the Dp3Tyb;App NL-F/NL-F , as detected by MSD assay after biochemical isolation in formic acid (Fig. 11). Thus, our data could be consistent with a BACE2 aging-dependent mechanism leading to the observed decrease in amyloid-β in the Dp3Tyb;App NL-F/NL-F model at 8-months of age, but further analysis is required to support this.

Discussion
Here we use a series of mouse models of DS to identify which combination of Hsa21 genes (other than APP) are sufficient to modulate APP/amyloid-β. Our systematic approach identified a region of Hsa21 that contains 38 genes, which is sufficient to decrease the deposition of amyloid-β in vivo at an early time-point. This region contains two lead candidate genes Bace2 and Dyrk1A.
DYRK1A is widely expressed in both the developing and adult mouse brain (Marti, Altafaj et al. 2003, Hammerle, Elizalde et al. 2008 and is found in all major cell types in both the human and mouse brain (Zhang, Chen et al. 2014, Zhang, Sloan et al. 2016. It is a primer kinase that can phosphorylate a large number of proteins, including APP, which has been suggested to increase the protein's abundance (Ryoo, Cho et al. 2008, Branca, Shaw et al. 2017. Inhibition of the kinase in a transgenic mouse model of AD decreased amyloid-β accumulation (Branca, Shaw et al. 2017, Velazquez, Meechoovet et al. 2019. Consistent with this, a decrease in APP and amyloid-β abundance is observed in a mouse model of DS with a normalised Dyrk1a gene dose (Garcia-Cerro, Rueda et al. 2017). In the Dp3Tyb model, which contains an additional copy of Dyrk1a, we observe no increase in APP abundance (endogenous mouse APP and partially humanised APP SW ) or evidence of enhanced amyloid-β accumulation. This suggests that in our knock-in model system, 3 copies of Dyrk1a are not sufficient to modulate APP protein abundance or promote amyloidβ accumulation. Notably, inhibition of DYRK1A has been suggested as both a possible therapeutic to enhance cognition in people who have DS and as a possible AD therapeutic strategy (De la Torre, De Sola et al. 2014, Yin, Jin et al. 2017, Neumann, Gourdain et al. 2018, Nguyen, Duchon et al. 2018. Our data do not support the use of DYRK1A inhibitors as a treatment strategy to decrease early amyloid-β accumulation in people who have DS. However, inhibiting the kinase may have other beneficial effects, such as to slow or prevent the formation of tau neurofibrillary tangles or to alter the response of cells within the brain to accumulating pathology. Further research is needed, in next generation AD-DS mouse models, to address these key questions. BACE2 has been reported to be expressed in a subset of astrocytes and neurons in the human brain (Holler, Webb et al. 2012). In the mouse the gene is expressed in a subset of neurons with the highest level in the CA3 and subiculum and some expression is also reported in oligodendrocytes and astrocytes lining the lateral ventricles (Voytyuk, Mueller et al. 2018). The protein may function as a θ-secretase (decreasing CTF-β), a β-secretase (increasing CTF-β), and as an amyloid-β degrading protease (decreasing amyloid-β) (Sun, He et al. 2006, Abdul-Hay, Sahara et al. 2012; which mechanism predominates in the human brain is not well understood. A previous study in which BACE2 was overexpressed in a wild-type APP over-expressing mouse model reported no evidence of altered amyloidβ 40 or amyloid-β 42 abundance in the brain (Azkona, Levannon et al. 2010). Conversely, knocking-down Bace2 in the App NL-G-F mouse model of amyloid-β accumulation has been reported to decrease CTF-β and soluble amyloid-β (Wang, Xu et al. 2019). In contrast, reducing BACE2 copy number in human organoids produced from individuals with DS or APP-duplication increased amyloid-β and triggered the formation of deposits within the model system. The authors propose this occurs because of the amyloid clearance function of BACE2, as evidenced by the raised levels of amyloid-β degradation products in trisomic compared to isogenic disomic control (Alic, Goh et al. 2020).
In the Dp3Tyb model that contains an additional copy of Bace2, we observe no significant change in CTF-β (endogenous mouse APP and partially humanised APP SW ), soluble amyloid-β (endogenous mouse and partially humanised APP SW ) or amyloid-β degradation products in the young adult brain. However, we observe a significant decrease in amyloidβ deposition in older mice (8-months of age) consistent with the role of BACE2 as an amyloid-β degrading protease (Alic, Goh et al. 2020). Further studies are required in aged mice to determine if enhanced amyloid-clearance can be detected, or if BACE2 gene-dose correction is sufficient to reverse the reduction in amyloid accumulation caused by the Dp3Tyb model. Genes other than Dyrk1a or Bace2 may be responsible for the decreased amyloid-β accumulation observed in our model systems, perhaps via enhancing amyloid-β clearance pathways. However, we previously studied the rate of extracellular clearance of human amyloid-β in an alternative model system (cross of the Tc1 with the J20 APP transgenic model) and found no evidence that the amyloid-β clearance rate was modulated by the extra copy of Hsa21 genes (Wiseman, Pulford et al. 2018). In addition, in the same model system we found no evidence of increased abundance of the key clearance enzymes neprilysin or insulin degrading enzyme in the young adult brain. Changes to intracellular clearance/ formation of amyloid-β may contribute to the alterations in accumulation of the peptide reported here. Notably DS is associated with significant alterations to endosomal biology (Botte and Potier 2020); neuronal endosomes have a key role in amyloid-β formation and astrocyte and microglia endosomes in amyloid-β clearance. Further studies focussing on cell type specific effects of extra chromosome 21 genes are required to understand this complex and important biology. DS is caused by the increase in the abundance of Hsa21 gene products; however, not all genes on the chromosome are dosage-sensitive in all contexts. For example, APP dosage sensitivity has been reported to vary over life-span (Choi, Berger et al. 2009) and BACE2 has been reported to be both dosage-sensitive (Barbiero, Benussi et al. 2003) and insensitive in the human brain (Cheon, Dierssen et al. 2008, Holler, Webb et al. 2012. In contrast, the abundance of DYRK1A is highly sensitive to gene dose in the vast majority of reported studies (Arbones, Thomazeau et al. 2019), including previously in the adult brain of the Tc1 mouse model (Sheppard, Plattner et al. 2012, Ahmed, Dhanasekaran et al. 2013, Naert, Ferre et al. 2018. Consistent with this, here we report a significant increase in DYRK1A in the adult brain of the new Dp3Tyb model. However we could not detect a significant increase in BACE2, this lack of statistical significance is likely caused by the higher intra-animal variation in abundance of this protein in the mouse brain. Similar overlapping levels of Bace2 gene expression have previously been reported at RNA level in the brain of an alternative mouse model of DS compared with euploid controls (Sultan, Piccini et al. 2007).
The relative abundance of proteins in AD-DS model systems is particularly important, as Down syndrome is caused by an inbalance in the relative abundance of gene products. Previous research using APP transgenic models, which over express APP protein, may mask the subtle but physiologically relevant effects of the 50% increase in the abundance of other Hsa21 gene products. This study addresses this limitation, here we systematically investigate the effect of additional copies of Hsa21 gene orthologues, other than App, on APP biology in a knock-in mouse model system. However, because of technical limitations we were not able to study the effect of an extra copy of genes located close to App, including the role of key genes such as Adamts1 (Kunkle, Grenier-Boley et al. 2019) and Usp25 (Zheng, Li et al. 2021). We were also unable to determine if genes located far apart on the chromosome act synergistically to cause a modulation of APP biology. Thus our data does not preclude a multigenic role for the genes in the Dp(10)2Yey or Dp (17)3Yey regions when combined with other Hsa21 genes. A further limitation is the use of AD-associated mutations in APP to drive pathology as these mutations alter the subcellular localisation and processing of APP (Sasaguri, Nilsson et al. 2017), which may modulate the effect of trisomy of Hsa21 on these processes. We have partially addressed this issue by undertaking a side-by-side comparison of mouse APP/amyloid-β and partially humanised APP/amyloid-β.
We studied two time-points in our model systems. Firstly, the young adult brain at 3 months of age, prior to the accumulation of human amyloid-β in the App NL-F in order to understand the modulation of early upstream processes by the additional copies of Hsa21 orthologues. Secondly the middle-aged adult brain at 8-months of age, which in the App NL-F mouse, models the initiation of amyloid-β accumulation (Saito, Matsuba et al. 2014). Thus in this study we focus on how additional copies of Hsa21 gene orthologues modulate the earliest stages of AD pathology; the formation and early accumulation of amyloid-β in the brain. We note that the effect of the additional genes may differ later in disease, including during the maturation of amyloid-β plaques, the formation of tau neurofibrillary tangles and the response of astrocytes, microglia and neurons to AD pathology. We studied both male and female mice, to ensure the generalisability of our findings, as sex is know to modulate APP/amyloid-β in mouse models (Jankowsky and Zheng 2017). We did not find evidence that sex modulated the effect of additional Hsa21 genes on amyloid accumulation in either the Tc1 or Dp3 models.
Our results indicate that an additional copy of genes on Hsa21 in people who have DS can modulate key AD biology. The extent of this effect is likely to differ between individuals both because of genetic variation on Hsa21 and differences in life-style factors that modulate dementia risk. This may contribute to the variation in age of onset of both pathology and dementia observed in people who have DS.

Conclusion
DS is a complex condition that alters multiple aspects of neurobiology and physiology.
Here we demonstrate using physiological mouse models that an additional copy of Hsa21 genes reduces the accumulation of amyloid-β within the brain; one of the earliest steps in the AD pathogenic process. Thus trisomy of Hsa21 may partially protect individuals who have DS from the accumulation of amyloid-β, resultant from their extra copy of APP.
Moreover we predict that individuals who have DS will accumulate amyloid-β more slowly than other groups who develop other genetic forms of early-onset AD, including that caused by duplication of APP. However, multiple studies of adults who have DS demonstrate that these individuals do accumulate substantial amyloid-β within their brains by mid-life, thus the effect of an additional copy of other genes on the chromosome is not sufficient to delay pathology development and AD primary prevention therapies targeting amyloid-β are likely to significantly benefit individuals who have DS (Fortea, Zaman et al. 2021). Treatments for AD-DS and other medical conditions associated with DS must take into account the differences in biology of individuals with the syndrome; to ensure that interventions do not reverse the beneficial effect of the additional genes. For the purpose of Open Access, the author has applied a CC-BY public copyright license to any Author Accepted Manuscript version arising from this submission. All authors read and approved the final manuscript. We thank Dr. Amanda Heslegrave (UCL) for assistance with this project.

AD
Alzheimer's disease

LC-MS
Liquid chromatography-mass spectrometry

Significance Statement
Trisomy of chromosome 21 is a commonly occurring genetic risk factor for earlyonset Alzheimer's disease, which has been previously attributed to people with Down syndrome having three copies of the APP gene, which is encoded on chromosome 21.
However, we have shown that an extra copy of other chromosome 21 genes modifies ADlike phenotypes independently of APP copy number (Wiseman et al. 2018, Brain;Tosh et al. 2021 Scientific Reports). Here, we use a mapping approach to narrow-down the genetic cause of the modulation of pathology; demonstrating that gene(s) on chromosome 21 decrease amyloid-β accumulation in the brain, independently of alterations to fulllength APP or C-terminal fragment abundance and that just 38 genes are sufficient to cause this. (A) The additional Hsa21 gene content in Tc1 (yellow), in the Dp(10)2Yey (orange), Dp(17)3Yey (light grey) and Dp3Tyb (red) segmental duplication models of DS as shown. The number of additional human genes in the Tc1, and the additional number of mouse genes in the Dp(10)2Yey, Dp(17)3Yey and Dp3Tyb as listed. These lines were crossed with the App NL-F mouse model of amyloid-β accumulation (blue), to generate Tc1;App NL-F (green), Dp(10)2Yey;App NL-F (brown), Dp(17)3Yey;App NL-F (dark grey), and Dp3Tyb-App NL-F (purple) offspring, homozygous for the App NL-F allele and carrying the additional gene content. The location of APP/App is indicated by the blue line, this locus is altered in the App NL-F model such that the amyloid-β sequence is humanised and the model carries the AD-causal Swedish (NL) and Iberian (F) mutations. This colour scheme is used in subsequent figures to code for each of the mouse models. (B) Schematic of the generation of experimental cohorts for the duplication models (Dp(10)2Yey, Dp(17)3Yey and Dp3Tyb) used in the studies, experimental cohorts were produced by crossing Dpx;App NL-F/+ mice with App NL-F/+ mice. observed in WT (female n = 5) or Tc1 (female n = 5, male n = 3) age matched littermate controls (data not shown). (B-F) Total cortical proteins were biochemically fractionated and amyloid abundance analysed by MSD assay. (B) No statistically significant difference in the abundance of 5 M guanidine hydrochloride soluble amyloid-β 42 was observed between Tc1;App NL-F/NL-F compared with App NL-F/NL-F controls (F(1,13) = 2.262, p = 0.156). App NL-F/NL-F female n = 15, male n = 8; Tc1;App NL-F/NL-F female n = 8, male n = 10.

Summary of Mouse Cohorts
Summary of the cohorts of mice, the cross(es) used to generate them, the age at which the cohort was sacrificed, the outcome measured and the figure in which the data are presented.   Table 2 Summary of negative control samples, which do not carry an App NL-F allele and do not produce human amyloid-β, used in MSD assays.

Cohort
Negative control samples from mice that do not carry the App NL-F allele and do not produce human amyloid-β were assayed by 6E10 MSD assay alongside samples homozygous for the App NL-F allele and which do produce human amyloid-β. The epitope for 6E10 lies within the region of sequence difference between human and mouse Aβ, the antibody is against the human sequence. In some cases as indicated all samples without an App NL-F allele were below the limit of detection for the assay and thus mean and standard deviations are not available (N/A) 1% Triton amyloid-β 42 / total brain weight Dp(10)2Tyb 3 2 1 0.007 N/A Figure 5F 1% Triton amyloid-β 42 / total brain weight WT 5 2 3 0.475 0.814 Figure 5F 1% Triton amyloid-β 42 / total brain weight Dp (17) Gnd HCl amyloid-β 40 /total brain weight Dp(10)2Tyb 3 0 3 2.255 1.796