A multi-discipline phenotyping platform for late-onset Alzheimer’s disease employed on a novel, humanized APOE e 4.Trem2*R47H mouse model

Background: Late-onset Alzheimer’s Disease (AD) (LOAD) is the most common neurodegenerative 2 disease. Despite extensive efforts to understand disease progression there are currently no approved 3 disease modifying interventions to delay or reverse neurodegeneration caused by AD. Repeated failures in 4 human trials, despite promising preclinical results in amyloidogenic mouse models, highlight the need for 5 animals that better model human AD. MODEL-AD (Model organism development and evaluation for late- 6 onset AD) are identifying and integrating disease-relevant, humanized gene sequences identified from 7 public AD data repositories to create more translatable mouse models relevant to AD. 8 Methods: Strong risk factors for LOAD, APOE e 4 and Trem2*R47H , were expressed alone or in 9 combination on a congenic C57BL/6J (B6) background, in cohorts of mice established on multiple sites and 10 aged to between 4-24 months. A deep phenotyping approach was employed to assess phenotypes relative 11 to human AD. 12 Results: The LOAD1 mouse strain, expressing humanized APOE e 4 and Trem2*R47H alleles, was 13 designed to elucidate the disease state of animals expressing the two strongest genetic risk factors of 14 LOAD at endogenous levels. Robust analytical pipelines measured behavioral, transcriptomic, metabolic, 15 and neuropathological phenotypes in cross-sectional cohorts for progression of disease hallmarks at all life 16 stages. In vivo PET/MRI neuroimaging revealed regional alterations in glycolytic metabolism and vascular 17 perfusion. Transcriptional profiling by RNA-Seq of brain hemispheres identified sex and age as the main 18 sources of variation between genotypes including age-specific enrichment of AD-related processes. Similarly, age, but not genotype, was the strongest determinant of behavioral change. In the absence of 20 mouse amyloid plaque formation, many of the hallmarks of AD were not observed in this strain. However, these two alleles together form a sensitized, background strain which will serve as a platform for the characterization of additional genetic and environmental LOAD risk factors. Conclusions: Comprehensive phenotyping provided key insights into genetic and environmental effects mouse models were compared with human co-expression modules, we observed strong negative correlation between the B6. Trem2*R47H mice and immune-related human co-expression modules from multiple brain regions, and this inflammatory response is dampened in the presence of APOE e 4 in the B6. APOE4.Trem2*R47H mice. Distinct mouse models showed concordance with distinct human co- expression modules reflecting a different transcriptional response driven by the human APOE e 4 and 5 Trem2 * R47H risk variants. We also observed age dependent shift in co-expression patterns associated with LOAD pathologies. A strong negative correlation between co-expression modules associated with cell cycle and DNA repair was observed in the early-aged mouse B6. APOE4 model, whereas advanced-aged B6. APOE4 female mice showed strong positive correlation with these co-expression modules. This overlap with human late-onset co-expression signatures early in life was observed for a number of different brain regions and was absent in Trem2*R47H knock-in mice. Furthermore, aged B6. Trem2*R47H mice showed a moderate overlap with several human neuronal co-expression modules enriched for genes that play an important role in synaptic signaling and myelination. At advanced age, a strong correlation between the mouse models and immune related human co-expression modules highlights the important role of the TREM2 R47H variant in Alzheimer’s related immune processes. Our experiments predict that APOE 𝜀 4 functions through the suppression of effects brought out by expression of the Trem2*R47H allele: 455 genes by TREM2 R47H but suppressed by APOE 𝜀 4 for . Our results mirror some emerging evidence that APOE 𝜀 4 suppresses Trem2*R47H in AD risk, that there are some suggestions that APOE 𝜀 4 carriers don’t have increased AD risk with Trem2*R47H and Trem2*R47H only increases risk on APOE 𝜀 3 carriers Additionally, we observe more differentially expressed genes at middle age than at a later age supporting

). Additional cohorts, 3 investigating APOEe4 allele alone compared with littermate B6 controls, were aged to 12 months. In an effort to understand the role of risk alleles on regional glycolysis and tissue perfusion, translationally 5 relevant regional measures were acquired via 18F-FDG and 64Cu-PTSM PET/MRI and autoradiography, 6 respectively. By 12 months glycolysis was altered in key brain regions associated with sensory integration, 7 cognition, vision and motor function in B6.APOE4 and B6.APOE4.Trem2*R47H mice, when compared with 8 controls (Figure 4), and were confirmed via post-mortem autoradiography, which has a 40 folder greater 9 resolution than PET. As expected, these changes were greater in number of regions and magnitude of 10 change in female mice when compared to males (Figure 4 B,C). These changes were similarly observed 11 through time, where female mice showed significantly altered glycolysis at 4, 8 and 12 months, while male 12 mice largely showed a hypoglycolytic phenotype at 8 months, that was virtually mitigated by 12 months 13 (Supplemental Figure 6). Since these risk alleles can alter metabolic functionality and neuroinflammatory-14 driven tissue perfusion in an independent manner, we quantitatively measured changes in regional tissue in both sexes at 4 months (Supplemental Figure 7). Unlike glycolysis, these changes were largely resolved 20 by 8 months in female mice, while males continued to show regional reduction in perfusion at this same 21 age.

23
Biochemical and neuropathological effects of APOEe4 and Trem2*R47H alleles 24 Confirmation of protein expression levels in brain tissue were confirmed for alleles encoding human APOE4 25 and mouse TREM2 carrying the R47H mutation (Supplemental Figure 8). Similar to reports of R47H 26 variant-mediated reduction in Trem2 transcript levels (6), TREM2 protein levels in the brains of these 27 animals were also decreased. However, instead of a near knock-out of all TREM2 that has been reported 28 8 previously (6), levels fell by approximately 50% in Trem2*R47H animals compared to C57BL/6J 1 (Supplemental Figure 8A,B). We have previously shown APOE4 protein levels are similar to endogenous 2 mouse APOE (14) and expression of APOE4 appeared similar between male and female LOAD1 mice 3 (Supplemental Figure 8C) . Additional molecular characterization of these animals showed both age-and 4 genotype-driven differences in levels of cytokines present in the brain and blood (Supplemental Figure 9). 5 IL-6 and KC/GRO concentrations were highest in B6.APOE4.Trem2*R47H brain tissue at 8 months, while 6 blood plasma concentrations continued to increase with age in those mice (Supplemental Figure 9 B,D,E). 7 In multiple occasions a trend appeared to suggest increased cytokine concentrations in mice expressing 8 mutated allele Trem2.

9
Neuropathological features of AD were then investigated by hematoxylin and eosin (H&E, structure) and 10 luxol fast blue/cresyl violet (LFB/CV, myelin) staining but did not reveal any gross anatomical changes to 11 tissue architecture or myelin ( Figure 6A). Brain sections were also imaged via immunofluorescence and genes across all mouse models (Figure 7 A-C). We observed higher expression of human APOE gene in 25 mice carrying humanized APOEe4 (B6.APOE4 and B6.APOE4.Trem2*R47H mice), whereas mouse Apoe 26 gene was highly expressed in B6 and Trem2*R47H mice (Figure 7 A,B). As expected based on protein 27 levels (Supplemental Figure 8A,B), expression of Trem2 was significantly reduced (p< 0.05) in 9 B6.Trem2*R47H and B6.APOE4.Trem2*R47H compared to age-matched B6 (Figure 7C), an effect likely 1 caused by a novel effector splice site and truncation introduced by the R47H mutation. Furthermore, 2 expression level of Trem2 increased with age across all mouse models, but no such patterns were observed 3 in the expression levels of mouse Apoe and human APOE genes (Figure 7 A,B). In addition, there was 4 lower expression of Trem2 in B6.APOE4.Trem2*R47H compared to B6.Trem2*R47H mice at advanced 5 age (24 months), suggesting expression of Trem2 might be suppressed by APOEe4. Next, principal 6 component analysis (PCA) identified two distinct clusters corresponding to male and female samples 7 separated along the first principal component (26% of total variance), suggesting sex-specific differences 8 are profound in mice ( Figure 7D). Analysis of samples from different age groups revealed a gradient of 9 discrimination along the second principal component (14% of total variance) ( Figure 7D), implying the 10 presence of age-dependent molecular changes in the brain transcriptomes.

21
'antigen processing and presentation', cytokine-cytokine receptor interaction', and 'complement and  In the human population, sedentary lifestyle is correlated with an increased risk of LOAD (45-47). Therefore, 4 to determine if physical activity influenced transcriptional changes in B6.APOE4.Trem2*R47H mice, a 5 running wheel was provided in the home cage of 22 month old male mice for two months. Brain tissue from 6 these animals was profiled by RNA-seq. A total of 292 DEGs (108 upregulated, 184 downregulated) were 7 identified in the running B6.APOE4.Trem2*R47H mice compared to 24 months old B6 male mice.

17
Finally, the expression of the upregulated DEGs associated with oxidative phosphorylation pathway were 18 assessed in transcriptional data from AMP-AD. Reduced expression of these running signature genes was 19 observed in AD cases compared to controls across multiple brain regions such as parahippocampal gyrus 20 (PHG) and frontal pole brain regions (FP) (Supplemental Figure 11C). This suggests that exercise induces 21 beneficial effects on health by increasing the expression of oxidative phosphorylation pathway genes that 22 are down regulated across multiple brain regions in AD patients.

25
MODEL-AD was established in response to the many shortcomings of existing mouse models of AD.

26
Aspects of human pathology have been replicated in mouse strains, most prominently the formation of beta 27 amyloid plaques via transgenic over-expression of brain-specific mutant human amyloid beta precursor 12 protein (APP), presenilin-1 (PS1), and/or microtubule associated protein tau (MAPT) bearing familial 1 Alzheimer's disease (FAD) mutations (48). Legacy preclinical models rely heavily on alleles that 2 overexpress transgenes, resulting in the removing or masking of important human-relevant biological 3 interactions. These mouse strains have been invaluable for understanding the molecular and behavioral 4 phenotypes of early-onset Alzheimer's disease (EOAD) driven by rapid and robust formation of plaques 5 and tangles in the brain and correlating hyperactivity which is a confound of many cognitive behaviors.

6
However, LOAD is ~20x more prevalent than EOAD and further implicates aging, inflammation, 7 environmental, and many more genetic risk factors in disease development. Our LOAD1 mice were healthy 8 late into life allowing a better understanding of the effect of AD risk factors in the context of aging 9 (Supplemental Figure 5). As the heterogeneity of this disease becomes more appreciated, so is the 10 importance of appropriate disease staging. Molecular targets of interest may only be available during 11 particular evolving disease stages: debris (cell fragments, plaques, tangles, etc.) accumulates over time 12 and inflammation, interruption/loss of neuronal function all also change with disease progression. In light of 13 the repeated short-comings of "fit-for-all" therapies, efforts may be better directed at targeted therapies (19, 14 49, 50). Faithfully modeling a complex, polygenic disease will be aided by the creation of platform strains 15 that carry multiple genetic risk factors to motivate with a scientific rationale rather than a grant-focused one.

16
Of the candidate risk variants identified, expression of the e4 allele of APOE and the R47H mutation in 17 TREM2 were identified as the strongest candidates for initial development of a novel LOAD mouse strain.  presented in other APOE mouse models and most importantly, human patients (14,56). Therefore, APOE4 1 formed the basis for multiple platform strains that include: B6.APOE4.Trem2*R47H (for which we have 2 provided the identifier 'LOAD1').
3 APOE binds to high-density lipoproteins to facilitate cholesterol and phospholipid transport to LDL 4 receptors. As expected, (16, 56), mice expressing the humanized APOEe4 allele showed decreased plasma 5 lipoprotein levels across all time points (Supplemental Table 3). We observed an age-dependent decrease

16
Despite the heterogeneity of AD, age is the strongest risk factor in the human population. As an aging 17 disease, monitoring and evaluating mouse models in relation to age is crucial for understanding onset and

5
In the absence of additional environmental or genetic risk factors, B6.APOE4/Trem2*R47H mice did not 6 display penetrant behavioral phenotypes beyond the expected aging-related changes but did exhibit 7 decreased survival probabilities by 24 months (Supplemental Figure 5). Very few C57BL/6J mice 8 succumbed during the 24-month aging process (<5%), whereas mortality was higher in mice expressing 9 both LOAD alleles in both males (~20%) and females (~35%). Male mice expressing either allele alone had 10 survival probabilities similar to C57BL/6J, whereas females with APOE 4 or Trem2*R47H showed a 11 mortality rate of ~20%. Therefore, it would seem that these two LOAD risk alleles show an equal and 12 additive risk when expressed together, but in females their interaction appears synergistic.

13
We investigated the molecular signatures in the brain transcriptomes of LOAD mouse models at different 14 ages in both sexes. We identified age-dependent molecular changes associated with LOAD pathologies in 15 mouse models. Introduction of the R47H mutation revealed a novel Trem2 isoform identical to primary

19
mouse models carrying R47H mutation in Trem2 gene did not exhibit any significant transcriptional changes 20 at young age, in contrast APOE4 mice exhibited significant changes only at 8 months of age. We further 21 identified significant downregulation of genes associated with oxidative phosphorylation pathway in the 12 22 months old B6.Trem2*R47H mice, suggesting that the oxidative phosphorylation could be prominent early 23 feature for the onset of neurodegeneration/inflammation process. Subsequently, multiple immune related 24 processes were disrupted in 24 months old B6.Trem2*R47H and B6.APOE4.Trem2*R47H mice, supporting 25 the profound relationship between aging, Trem2 and AD. Interestingly, at 12 months of age we did not 26 observe any significant transcriptional changes in B6.APOE4.Trem2*R47H mice compared to control mice,

27
suggesting that the effect of Trem2 gene is suppressed due to the presence of APOEe4. Similarly, when 28 15 mouse models were compared with human co-expression modules, we observed strong negative 1 correlation between the B6.Trem2*R47H mice and immune-related human co-expression modules from 2 multiple brain regions, and this inflammatory response is dampened in the presence of APOEe4 in the 3 B6.APOE4.Trem2*R47H mice. Distinct mouse models showed concordance with distinct human co-4 expression modules reflecting a different transcriptional response driven by the human APOEe4 and 5 Trem2*R47H risk variants. We also observed age dependent shift in co-expression patterns associated 6 with LOAD pathologies. A strong negative correlation between co-expression modules associated with cell 7 cycle and DNA repair was observed in the early-aged mouse B6.APOE4 model, whereas advanced-aged 8 B6.APOE4 female mice showed strong positive correlation with these co-expression modules. This overlap 9 with human late-onset co-expression signatures early in life was observed for a number of different brain 10 regions and was absent in Trem2*R47H knock-in mice. Furthermore, aged B6.Trem2*R47H mice showed 11 a moderate overlap with several human neuronal co-expression modules enriched for genes that play an  Trem2*R47H in AD risk, that there are some suggestions that APOE 4 carriers don't have increased AD 19 risk with Trem2*R47H and Trem2*R47H only increases risk on APOE 3 carriers (65, 66). Additionally, we 20 observe more differentially expressed genes at middle age than at a later age supporting evidence of an 21 earlier aging phenotype than C57BL/6J mice, with a realignment of transcriptomes at later timepoints (58, 22 67). We employed a weighted gene co-expression network analysis (WGCNA) used to identify modules of 23 correlated genes. Each module was tested for differential expression by strain, then compared with human 24 postmortem brain modules from the Accelerating Medicine's Partnership for AD (AMP-AD) to determine 25 the LOAD-related processes affected by each genetic risk factor (6, 68, 69). This will be a useful tool in  17 that more closely align with human disease will be incorporated into the pre-clinical testing core of MODEL-1 AD to assess the potential of prioritized compounds to treat AD.
The MODEL-AD consortium has established the LOAD1 model to study the effects of two strong 5 risk factors of LOAD, APOE 4 and Trem2*R47H. In the absence of amyloid plaque formation, changes in 6 the cellular dynamics observed in the diseased human brain were not recreated in the young or aged 7 cohorts. However, metabolic traits thought to exacerbate disease severity were replicated. APOE 4 8 expression altered cholesterol and lipid metabolism pathways. In addition to APOE 4, the R47H mutation 9 of Trem2 also caused changes in glycolysis and tissue perfusion in many separate brain regions, similar to 10 clinical observations. Transcriptional analysis of brain hemispheres highlighted alterations to disease-11 related processes and, when compared to human data sets, an indication that expression of APOE 4 and 12 Trem2*R47H yield an aged mouse model more representative of human patients than C57BL/6J control 13 mice. Subsequent data sets with risk factor additions, or exclusions, will further improve our understanding    Behavioral tests were conducted as previously reported (63) in the following order with at minimum a 1-2-2 day rest period between tests: Frailty assessment with core body temperature recording, open field test, 3 spontaneous alternation, rotarod, and wheel running activity. On each test day, subjects were transported 4 from the adjacent housing room into the procedure room, tails were labeled with a non-toxic permanent 5 marker with the assigned subject ID number, and subjects were left to acclimate undisturbed to the testing 6 environment for a minimum 60 minutes prior to testing. Between subjects, all testing arenas were sanitized 7 with 70% ethanol solution and dried prior to introducing the next subject. Lighting in the testing rooms were 8 consistent with the housing room (~500 lux) unless where specifically noted. At minimum 5 days post the 9 conclusion of behavioral testing, mice were sent for tissue harvesting.

11
Frailty assessment 12 Similar to as previously described (53), subjects were individually evaluated for the absence or presence 13 of 26 aging related characteristic traits and scored a 0, 0.5, or 1 (based on presence/absence, and severity) 14 for each assessment by a trained observer, blind to genotype/age, and included the following assessments:       Prior to data analysis and while still blinded, results were adjusted to exclude data only from mice which 10 could not be tested or which data was not available inclusive of any equipment failures, escape episodes, 11 etc. Subjects were not excluded by any mathematical determination. Data was analyzed under coded 12 genotypes (A, B, C, etc.) within sex, as one-way or two-way ANOVA as appropriate versus sex-and age-13 matched WT control. The blind was revealed at the conclusion of the data analysis for interpretation.

15
Fasted blood glucose collection and measurement 16 Fasted mice were placed into a fresh cage, free of food but with fresh water, at 6am -the beginning of the 17 light-ON cycle. Mice were fasted for 6 hours, until 12pm, at which time blood glucose levels were analyzed.

18
Prior to mouse restraint, a Contour Next EZ blood glucose monitor (Ascensia, Parsippany, NJ) was 19 calibrated with Contour glucose control solution and Contour Next test strips. While restraining the animal, 20 with a 5.0mm lancet a stab incision was made into and perpendicular to the cheek, located dorsal to the to 21 the cheek skin gland at a distance equal to the height of the eye and caudal distance equal to the length of 22 the eye. One drop of blood, approximately 10µl, was applied to a blood glucose test strip and readings were 23 recorded.

25
Fecal collection 26 Parallel with measurement of animal weight, animals were placed in a clean container on a scale. Mouse 27 weight was recorded and upon production, fecal sample was collected with forceps to prevent 28 23 contamination. Sample was placed in a pre-marked 1.5mL tube and snap-frozen immediately on dry ice.

1
Container and forceps were cleaned with 70% ethanol before collecting from subsequent mice. Fecal 2 samples were stored long term at -80°C until analyzed.  Blood was collected by cardiac puncture from non-fasted, anesthetized animals (see Perfusion method) at 2 harvest prior to incision of the right atrium and subsequent perfusion. A 25-gauge EDTA-coated needle, 3 attached to a 1mL syringe, is inserted into the right atrium of the exposed heart and the plunger gently 4 pulled to slowly aspirate approximately 500mL of blood, avoiding entrapping air in the syringe to prevent 5 hemolysis. After removal of the needle from the syringe, the blood was slowly injected into a 1.5mL EDTA

21
end forceps, the two skull plates were removed to expose the brain. The brain was carefully removed from 22 the skull, weighed, and divided midsagitally, into left and right hemispheres, using a brain matrix. The right 23 hemisphere was quickly homogenized on ice and equally aliquoted into three cryotubes for metabolomic, 24 proteomic, and transcriptomic analysis. Cryotubes were immediately snap frozen on dry ice, and stored 25 long-term at -80°C. The left hemisphere was immediately placed in 5mL 4% PFA at 4°C for no less than 24 26 hours, but no longer than 30 hours. The left hemisphere was then moved from PFA solution to 10mL 15% 27 sucrose at 4°C for 24 hours, or until it sinks in the sucrose, when it was then transferred to a 30% sucrose 28 25 for 24 hours at 4°C, or until it sinks in the solution. The left hemisphere was then removed from 30% sucrose 1 solution, snap frozen on a flat mold, cut-side down, floating in 2-methylbutane solution cooled by dry ice.

2
Once frozen the left hemisphere is then placed into a cryotube and stored at -80°C until used for microtome 3 sectioning and immunohistochemistry analysis.

18
To provide high contrast grey matter images, at least two days prior to PET imaging, mice were induced 19 with 5% isoflurane (balance medical oxygen), placed on the head coil, and anesthesia maintained with 1-20 3% isoflurane for scan duration. High resolution T2-weighted (T2W) MRI images were acquired using a 3T

21
Siemens Prisma clinical MRI scanner outfitted with a dedicated 4 channel mouse head coil and bed system 22 (RapidMR, Columbus OH). Images were acquired using a SPACE3D sequence (80) using the following

9
Gels were transferred to PVDF membranes for immunoblotting and imaging using an iBlot2 dry blotting 10 system (Thermo Fisher). Membranes were blocked in 5% non-fat dry milk in 1xPBS+0.1% Tween20 for 1 11 hour prior to incubating with primary antibodies diluted in 5% non-fat dry milk in 1xPBS+0.1% Tween20 for 12 1 hour at room temperature. Membranes were washed in 1xPBS+0.1% Tween20 before incubating with 13 secondary antibodies diluted in 5% non-fat dry milk in 1xPBS+0.1% Tween20. HRP-conjugated secondary standards. Any value that was below the lowest limit of detection (LLOD) for the cytokine assay was 2 replaced with ½ LLOD of the assay for statistical analysis.

26
Subsequently, we added gene annotation of human APOE gene into mouse gene annotation file.

27
Additionally, we have also introduced annotation for novel Trem2 isoform in mouse gene annotation file 28 32 (GTF file), that is identical to primary transcript, but truncated exon2 by 119 bp from its start position.

6
A detailed description on post-mortem brain sample collection, tissue and RNA preparation, sequencing, 7 and sample QC has been provided elsewhere (12,13,93). As part of a transcriptome-wide meta-analysis                                                                   Cumulative Frailty Index Score. Animals were also measured for core body temperature and weight. All 5 alleles expressed were homozygous.   animals from cross-sectional cohorts housed to the ages indicated were assessed by spontaneous 10 alternation assay in Y-maze. Percentage of successive entries into all three arms in series compared to all 11 arm entries is provided. All alleles expressed were homozygous.