(A-23) Inferring the role of AP-1 protein dimerization in melanoma differentiation state heterogeneity via multivariate modeling

Melanoma is a form of metastatic skin cancer that develops when melanocytes grow out of control. Melanoma tumors exhibit a wide range of differentiation states, characterized by their distinct gene expression programs. Heterogeneity in differentiation state and its plasticity allow melanoma cells to adapt to therapeutic treatments and eventually develop therapy resistance. Understanding the underlying mechanisms of such plasticity will help us develop strategies to overcome therapy resistance. Recent work has identified a key role for Activator protein 1 (AP-1) transcription factors in the regulation of differentiation state heterogeneity in melanoma [1]. AP-1 functions as homo- and heterodimeric complexes, which are formed by bZIP domain proteins that include JUN and FOS subfamilies. These complexes bind the genome to regulate gene expression. Although the role of specific AP-1 proteins in driving differentiation state heterogeneity is recognized, the impact of AP-1 dimerization has not been explored. In this work, we perform multivariate modeling of transcriptomics data across a diverse panel of melanoma cell lines to infer the role of AP-1 dimers in prediction of melanoma differentiation states.

Based on their characterized roles [1], we focused our analysis on the JUN subfamily of AP-1 genes (including JUN, JUNB and JUND) which both homodimerize and heterodimerize with other AP-1 factors, and FOS subfamily (including FOS, FOSB, FOSL1, FOSL2) that can only heterodimerize with other AP-1 factors. We used gene expression data for FOS and JUN subfamily genes from 53 melanoma cell lines [2] and generated 16 AP-1 dimer scores (representing all physiologically possible dimers) for each cell line by multiplying the individual transcript levels and their dimerization K_D (affinity) values derived from the literature. We then used partial least square regression (PLSR) modeling to predict the differentiation state scores (including melanocytic, transitory, neural crest-like and undifferentiated state scores) for each cell line based on its corresponding AP-1 dimer scores.

The PLSR model developed based on the AP-1 dimers revealed a reasonable performance accuracy (R²) of 59.63% and a prediction accuracy (Q²) of 50.97% using leave-one-out cross-validation (LOOCV) for 3 PLS components (Fig. 1.A). We also evaluated the accuracy of predictions for each differentiation state by computing the Pearson’s correlation between the measured differentiation scores and PLSR-predicted scores (Fig. 1.B). Through the variable importance projection (VIP) score analysis (Fig. 1.C), we determined which AP-1 dimers positively and negatively contributed to prediction of each differentiation state. We found FOS-JUNB and FOS-JUND dimers to positively contribute to the melanocytic state and negatively contribute to the undifferentiation state. The importance of FOS in melanocytic cells is consistent with results from the previous version of model built based on individual AP-1 transcript levels [1]. The new results suggest that this effect may be mediated by FOS dimerization with JUND and/or JUNB. FOSL1-JUN, FOSL2-JUN, and FOSL2-JUND, on the other hand, were found to negatively correlate with the melanocytic state but positively contribute to the neural crest-like and undifferentiated states.  

Based on multivariate modeling analysis, we infer that FOS-JUNB and FOS-JUND are associated with the melanocytic state, while FOSL1-JUN, FOSL2-JUN, and FOSL2-JUND dimers are associated with neural crest-like and undifferentiation states. These results provide preliminary evidence to support the hypothesis that specific AP-1 dimers may regulate the differentiation state and its heterogeneity in melanoma tumors.

We thank the members of the Fallahi-Sichani laboratory for their suggestions. This study was funded by NSF Award #1950374 and NIH Award R35-GM133404.

Comandante-Lou, et.al.(2022). Cell Reports, https://doi.org/10.1016/j.celrep.2022.111147

Tsoi, et.al.(2018). Cancer Cell, https://doi.org/10.1016/j.ccell.2018.03.017