Multimodal analysis of small size data to understand the effects of low doses of ionizing radiation on atherosclerosis
Résumé
Background: Atherosclerosis is a hardening and narrowing of arteries. It can put blood flow at risk as arteries become blocked. Mechanistic understanding of the effects of low-dose ionizing radiation (LDIR) on atherosclerosis remains incomplete. The experimental studies have shown a protective effect of LDIR on atherosclerosis in rodent models 1, 2. How early responses in different cell types that are known to be involved in atherosclerosis contribute to this effect is not clear. Challenges in understanding biological mechanisms of LDIR include small size of experimental animal groups, low level changes and data heterogeneity and multimodality. Advanced methods of multimodal data integration, including machine learning algorithms, have proved useful in various biomedical applications but have not been used much in the studies of health effects of LDIR. In this study, we report results of applying a statistical and classical machine learning methods to reveal causal mechanistic links between biological responses of different modalities observed in ApoE-/- mice exposed to low to moderate doses of gamma-radiation. Revealing complex correlations and causal links related to health conditions, such as atherosclerosis, can help advance the concept adverse outcome pathway (AOP).
Analysis: Fourteen to 16 weeks old ApoE-/- mice were exposed to a single dose of 0, 0.05, 0.5, or 1 Gy from 137Cs at a dose rate of 10.35 mGy.min-1. Samples for biological assay were collected at 24 hours, 10 days and 100 days post-irradiation. Bone marrow cells from mice sacrificed at 24 hours or 10 days post-exposure were used to grow bone marrow derived macrophages for polarization into M1 or M2 phenotypes. Cells were treated with Il-4 or INFγ for M2 or M1 polarization, respectively. Polarization was validated by measuring the mRNA levels of polarization markers using RT-qPCR. Also, cell culture supernatants of polarized macrophages were harvested and cytokine expression were determined using ELISA. In this study, we focused on a single time point of 24 after irradiation with three modalities of data. Data from flow cytometry, gene expression and ELISA assay originating from the same animal were gathered. To carry out an integral analysis this complex and heterogeneous data obtained at different levels of biological organization, we employed a two-step analysis. In the first step, we used univariate and descriptive statistics analysis (ANOVA, PCA, PLS, RGCCA) and compared it with four classical machine learning methods (SVM, random forest, tree decision and Gradient boosting). In the second step, an integration of the obtained correlations into a biologically relevant mechanistic model that can be used as a basis for an AOP to atherosclerosis after exposure to LDIR was carried out.
Results:
Our preliminary results confirm the difficulty of visualizing the effects of LDIR. The results of univariate ANOVA analysis, although could test important comparisons, were limited in its ability to reveal a comprehensive integral insight when multiple variables were used. Using PCA allowed to reveal certain hidden trends in the assembly of variables, which was an improvement when compared to ANOVA. PLS and RGCCA were both more efficient in identifying those variables that are responsive to different treatment options and mediate biological phenotypic changes. Despite the low classification rate, machine learning methods allowed to identify those features/variables that contribute to differences between groups. These different analyses helped to reconstruct in more detail the mechanistic and/or causal links between various molecular endpoints measured under various treatment (radiation dose) and phenotypic (M0, M1 or M2 macrophages) conditions. Overall, our results highlight the need for comprehensive data analyses of LDIR studies and provide an example of such analyses using a multitude of statistical methods to understand mechanisms of LDIR effects on atherosclerosis.
[1] Ebrahimian TG et al. “Chronic Exposure to External Low-Dose Gamma Radiation Induces an Increase in Anti-inflammatory and Anti-oxidative Parameters Resulting in Atherosclerotic Plaque Size Reduction in ApoE-/- Mice.” Radiat Res. 2018
[2] Rey et al. “Exposure to low to moderate doses of ionizing radiation induces a reduction of pro-inflammatory Ly6Chigh monocytes and a U-curved response of T cells in ApoE -/- mice.” Dose-Response.2021