Weight trajectories in aging humanized APOE mice with translational validity to human Alzheimer's risk population: A retrospective analysis

Abstract

Translational validity of mouse models of Alzheimer's disease (AD) is variable. Because change in weight is a well-documented precursor of AD, we investigated whether diversity of human AD risk weight phenotypes was evident in a longitudinally characterized cohort of 1,196 female and male humanized APOE (hAPOE) mice, monitored up to 28 months of age which is equivalent to 81 human years. Autoregressive Hidden Markov Model (AHMM) incorporating age, sex, and APOE genotype was employed to identify emergent weight trajectories and phenotypes. In the hAPOE-AD mouse cohort, five distinct weight trajectories emerged: three trajectories were associated with a weight loss phenotype (36% of mice, n = 426), one with weight gain (13% of mice, n = 152), and one trajectory of no change in weight (34% of mice, n = 403). The AHMM model findings were validated with post-hoc survival analyses, revealing differences in survival rates across the five identified phenotypes. Further validation was performed using body composition and plasma β-amyloid data from mice within the identified gain, loss and stable weight trajectories. Weight gain trajectory was associated with elevated plasma β-amyloid levels, higher body fat composition, lower survival rates and a greater proportion of APOE4/4 carriers. In contrast, weight loss was associated with greater proportion of hAPOE3/4 carriers, better survival rates and was predominantly male. The association between weight change and AD risk observed in humans was mirrored in the hAPOE-AD mouse model. Weight trajectories of APOE3/3 mice were equally distributed across weight gain, loss and stability. Surprisingly, despite genetic uniformity, comparable housing, diet and handling, distinct weight trajectories and divergence points emerged for subpopulations. These data are consistent with the heterogeneity observed in the human population for change in body weight during aging and highlight the importance of longitudinal phenotypic characterization of mouse aging to advance the translational validity of preclinical AD mouse models.

Fig 1. Model Organism Development and Evaluation for Late-Onset Alzheimer’s Disease (MODEL-AD) humanized APOE (hAPOE) colony.

a) an overview of the mouse colony is presented, displaying the percentages of sex and APOE genotype, along with the variables included in the model and the frequency of weight measurements. b) illustrates the mean and standard deviation of longitudinal weights of the mice colony. c) shows the mean and standard deviation of weights by sex, with different colors and line styles representing sexes.

Fig 2. Autoregressive Hidden Markov Model (AHMM) structure and output.

The schema illustrates the AHMM structure, highlighting the hidden states (H) learned by the model based on the real data observations, including age, weight, sex, and APOE genotype of our aging humanized APOE mouse colony. To account for the observed sex-dependent weight differences, where male mice weight more than female (Fig 1C), an interaction term was introduced between the variables sex and weight (dashed arrow). Based on these data, AHMM learns the transition probabilities between the hidden states enabling the identification of main weight trajectories. Relevant weight trajectories are selected by grouping together mice ending in same hidden states, allowing the study of the identified subgroups.

Fig 3. AHMM transition probability matrix and diagram.

a) The matrix A illustrates transition probabilities between the possible 10 hidden states, with key ending hidden states (B,C,F,I,J) highlighted in red. Every hidden state is included as a row and as a column, and each cell in the matrix refers to the probability of transitioning from its row’s state to its column’s state. Numerical values of matrix A are provided in S2 Table. Key ending state are defined by a probability of arriving and staying in that state greater than 0.3 (P(A(a_i,j)))>0.3). Mice ending in the same hidden states are part of a trajectory and are therefore grouped together. b) Diagram of the Markov Chain related to the transition matrix A. Nodes correspond to hidden states, and edges illustrate the transitions between them (changes of states). Nodes with black circles indicate the five most probable ending states. Arrow thickness indicates transition probabilities, with thicker arrows associated with higher probabilities of transitioning from or remaining in a hidden state. The diagram depicts starting, intermediate, and ending states, where starting states have more exiting arrows, and ending states are identified by loop arrows. In detail, key ending states (B,C,I,F, and J) are characterized by thicker loops, indicating higher probabilities of staying in those states when reached.

Fig 4. Key weight trajectories in hAPOE mouse colony.

Panel a) The five most probable weight trajectories (labeled according to their hidden ending states) were derived using Autoregressive Hidden Markov Model (AHMM). The legend includes the number of mice within each trajectory. Panels b-c) Distribution of sex and APOE genotype within each identified weight trajectory. Panel d) Percent of mice by sex and genotype for each trajectory. Panels e-i) Visualization of each weight trajectory with separately fitted y-axes to enhance the clarity of the trend for each trajectory. The trajectories were obtained by plotting the mean and standard deviation for each age point meeting the minimum requirement of 3 mice per age. Panel j) presents the percent difference of weights for each trajectory from 12 months of age to the end of each trajectory. We selected 12 months to calculate weight percent differences as mice are in a growth phase until about 12 months (Figs 1B and 4A) and mice of 10–12 months are considered middle age [37].

Fig 5. Survival curves across weight phenotypes.

Survival curves associated with each of the five weight trajectories identified (B, C, I, F, and J). Significant difference in overall survival was found (log-rank p-value = 4e-09). For each state, median survival is pointed out with dash line corresponding to a survival probability of 0.5.

Fig 6. Distributions of plasma Aβ concentrations and EchoMRI data across weight loss, gain, and stable trajectories.

Panel a-b show respectively average level of plasma Aβ40 and Aβ42 concentrations. Panel c-d shows average level of plasma Aβ40 and Aβ42 concentrations by sex. Panel e-f shows the distribution of EchoMRI body compositions. *: 1.00e-02 < p-value < = 5.00e-02; **: 1.00e-03 < p-value < = 1.00e-02; ***: 1.00e-04 < p-value < = 1.00e-03; ****: p-value < = 1.00e-04.