Using deep learning to decipher the impact of telomerase promoter mutations on the dynamic metastatic morpholome

Abstract

Melanoma showcases a complex interplay of genetic alterations and intra- and inter-cellular morphological changes during metastatic transformation. While pivotal, the role of specific mutations in dictating these changes still needs to be fully elucidated. Telomerase promoter mutations (TERTp mutations) significantly influence melanoma's progression, invasiveness, and resistance to various emerging treatments, including chemical inhibitors, telomerase inhibitors, targeted therapy, and immunotherapies. We aim to understand the morphological and phenotypic implications of the two dominant monoallelic TERTp mutations, C228T and C250T, enriched in melanoma metastasis. We developed isogenic clonal cell lines containing the TERTp mutations and utilized dual-color expression reporters steered by the endogenous Telomerase promoter, giving us allelic resolution. This approach allowed us to monitor morpholomic variations induced by these mutations. TERTp mutation-bearing cells exhibited significant morpholome differences from their wild-type counterparts, with increased allele expression patterns, augmented wound-healing rates, and unique spatiotemporal dynamics. Notably, the C250T mutation exerted more pronounced changes in the morpholome than C228T, suggesting a differential role in metastatic potential. Our findings underscore the distinct influence of TERTp mutations on melanoma's cellular architecture and behavior. The C250T mutation may offer a unique morpholomic and systems-driven advantage for metastasis. These insights provide a foundational understanding of how a non-coding mutation in melanoma metastasis affects the system, manifesting in cellular morpholome.

Fig 1. Analysis of mined information from MetMap database for quantified metastatic potential and penetrance of cell lines harboring Telomerase Promoter Mutations stratified by C228T and C250T.

C228T n = 76, and C250T n = 13. (A) Metastatic Penetrance comparison of C228T and C250T aggregated by all organs used in the MetMap database, p = 0.0330. (B, C, D, E, F) Metastatic Penetrance comparison of C228T and C250T separated by organ Brain p = 0.0240, Lung p = 0.0326, Liver p = 0.0055, Kidney p = 0.0615, and Bone p = 0.0766. (G) Metastatic Potential comparison of C228T and C250T aggregated by all organs used in the MetMap database, p = 0.0326. (H, I, J, K, L) Metastatic Potential comparison of C228T and C250T separated by organ Brain p = 0.1555, Lung p = 0.0099, Liver p = 0.0347, Kidney p = 0.0819, and Bone p = 0.3706. All p values from the Mann-Whitney test.

Fig 2. Introduction of the TERTp mutations illicit a response in the morpholome.

(A) Schematic of the engineered genomic structure of the cell lines. (B) False-colored image montage of representative cells for Telomerase promoter mutants, C228T, C250T, and WT cell lines. Telomerase gene expression separated by allele for n = 67,000 cells. (C) The bar graph shows the difference in mean mCherry Allele expression of the mutant allele containing the control WT allele and the promoter mutation C228T and C250T, respectively. WT displayed significantly lower expression than C228T p <0.0001 and C250T <0.0001. While C228T showed significantly lower expression than C250T <0.0001 (unpaired t-tests). (D) Bar graph showing the allele expression variance for the mCherry Allele quantified using the Fano Factor calculation. (E) Scatter plot overlaying the WT, C228T, and C250T allele expression. Each data point represents a single cell. (F) The standard deviation of parallel coordinates of the top 5 discriminative features from a Random Forest trained on individual mutation status features. (G) 3D UMAP of the Average Pool 4 latent embeddings in a modified Resnet-50 model trained on phase contrast images labeled by individual mutation status. (H) Standard deviation parallel coordinates of the top 5 discriminative features from a Random Forest trained on TERTp mutations (C228T and C250T) and WT features. (I) 3D UMAP of the Average Pool 4 latent embeddings in a modified Resnet-50 model trained on phase contrast images labeled by TERTp mutations (C228T and C250T) and WT.

Fig 3. Analysis of WT and TERTp Mut Cells Using PHATE, Diffusion Pseudotime, and Grad-CAM Visualizations.

(A) PHATE map showing the distribution of wild-type (WT) cells (light blue) and TERT promoter mutant (TERTp Mut) cells (orange). The map illustrates the separation and clustering of the two cell types based on their morpholomic features. (B) Diffusion pseudotime plot indicates WT cells’ progression (blue) transitioning into TERTp Mut cells (green). The plot provides a continuous trajectory of cellular differentiation over pseudotime. (C) PAGA (Partition-based graph abstraction) network illustrating the connectivity and pseudotime trajectories between different cell clusters. Light blue nodes represent WT cells, orange nodes represent TERTp Mut cells and pie charts represent WT and TPM proportions. Thicker edges indicate more robust connectivity. (D) Representative images from the transition zone for WT and TERTp Mut cells. The images are organized by cell type, with WT cells in the top two rows and TERTp Mut cells in the bottom. These images highlight the morphological changes occurring in the transition zone. (E) Grad-CAM (Gradient-weighted Class Activation Mapping) heatmaps for transition zone images, highlighting regions of interest in the cell images that are most relevant for distinguishing between WT and TERTp Mut cells. The color scale ranges from blue (low relevance) to red (high relevance). (F) Representative images from the dense cluster zone for WT and TERTp Mut cells. These images show the cells in more stable and distinct morphological states. (G) Grad-CAM heatmaps for dense cluster zone images, showing regions of significant changes between WT and TERTp Mut cells. The heatmaps emphasize areas within the cells that exhibit notable differences in texture and intensity.

Fig 4. C250T has increased spatiotemporal monolayer migration using live-cell in vitro wound-healing assay.

n = 38 movies for each cell line. (A) Violin plot comparison of healing rates across cell lines. WT compared to C228T p <0.0001, WT compared to C250T p <0.0001, and C250T compared to C228T p <0.0001. (B) Total distance in microns is the leading edge covered throughout the movies. WT compared to C228T p = 0.0175, WT compared to C250T p = 0.0047, and C250T compared to C228T p <0.0001. Aggregate comparisons of the average of each of the 12 kymograph features for all movies across cell lines. (C) Speed of each of the 12 averaged figures. WT compared to C228T p = 0.1600, WT compared to C250T p = 0.0036, and C250T compared to C228T p = 0.0332. (D) Directionality of each of the 12 averaged figures. WT compared to C228T p = 0.0242, WT compared to C250T p <0.0001, and C250T compared to C228T p = 0.0083. (E) Graph showing the speed of individual 12 kymographs features averaged among the 38 movies with SD error bars. (F) Graph depicting p-values from multiple Mann-Whitney tests for speed features. WT compared to C228T were significantly different in features 2–8 and 10–12 with p values 0.013734, 0.000344, 0.000010, 0.024413, 0.000085, 0.000010, <0.000001 and 0.002214, 0.000359, and 0.000014 respectively. Features 1 and 9 were not significantly different, with p-values of 0.061604 and 0.550616, respectively. WT compared to C250T were significantly different in all 12 features, p <0.000001. C228T compared to C250T differed substantially in all features with corresponding p values of <0.000001, <0.000001, <0.000001, 0.000012, <0.000001, <0.000001, 0.000006, 0.000887, 0.000008, 0.000003, 0.001708, 0.018402. (G) Graph showing the directionality of individual 12 kymographs features averaged among the 38 movies with SD error bars. (H) WT compared to C228T differed significantly in directionality features 1–5 and 7 with p values <0.000001, 0.001646, 0.000027, 0.000670, 0.000093, and 0.008123 respectively. While features 6 and 8–12 were not significantly different with corresponding p values of 0.621746, 0.060135, 0.993789, 0.260206, 0.203475, and 0.048283. WT compared to C250T were significantly different in all 12 features, p <0.000001. C228T compared to C250T differed significantly in all 12 features p <0.000001. All p values from Mann-Whitney tests.