Phylogenies Derived from Matched Transcriptome in Breast Cancer Brain Metastases Yifeng Tao 1,2 , Haoyun Lei 1,2 , Adrian V. Lee 3 , Jian Ma 1 , Russell Schwartz 1,4 1 Computational Biology Department, School of Computer Science, Carnegie Mellon University 2 Joint Carnegie Mellon-University of Pittsburgh Ph.D. Program in Computational Biology 3 Department of Pharmacology and Chemical Biology, UPMC Hillman Cancer Center, Magee-Womens Research Institute, University of Pittsburgh 4 Department of Biological Sciences, Carnegie Mellon University Yifeng Tao et al. @ ISMCO'19 Phylogenies of Breast Cancer Brain Metastases 1
Background: Cancer Progression [Uchi, R. et al., PLOS Genetics . 2016] o Cancer: mainly caused by accumulated somatic alterations o Tumor cells: heterogeneous populations/clones o Tumor phylogeny: tumor cells follow a clonal evolution o Metastasis: transfer from primary site to other sites o Cell communities vs. cell clones Yifeng Tao et al. @ ISMCO'19 Phylogenies of Breast Cancer Brain Metastases 2
Background: Breast Cancer Metastasis o Breast cancer: 2 nd common cause of death from cancer in women o Metastatic breast cancer o Causes majority of those deaths o Limited viable treatment options o Early detection is important o Mechanism of tumor progression/evolution during metastasis? Yifeng Tao et al. @ ISMCO'19 Phylogenies of Breast Cancer Brain Metastases 3
Tumor Evolution Derived from Match Bulk Transcriptome [Vareslija, D. et al., Journal of the National Cancer Institute . 2018] o Given matched primary and metastatic bulk transcriptome: o Q1: How to cope with high-dim, noisy, and uninformative transcriptome? o Q2: What model and solver to unmix/deconvolve clones? o Q3: How to infer evolutionary trajectory and perturbed pathways/functions? o Yes! We proposed a three-step pipeline. Yifeng Tao et al. @ ISMCO'19 Phylogenies of Breast Cancer Brain Metastases 4
Step 1: Mapping to Gene Modules and Cancer Pathways [Desmedt, C. et al., Clinical Cancer Research . 2008] o Q1: How to cope with high-dim, noisy, and uninformative RNA? o Gene modules o Compress high dimensional and noisy data à accurate deconvolution o Cancer pathways o Markers/probes à interpretation purpose Yifeng Tao et al. @ ISMCO'19 Phylogenies of Breast Cancer Brain Metastases 5
Step 2: Deconvolution of Cell Communities [Lee, D.D. et al., NIPS . 2000] o Q2: What model to unmix/deconvolve clones? o Matrix factorization o C: expression profiles of communities o F: fractions of communities in samples o However, it is non-convex and not trivial to solve… Yifeng Tao et al. @ ISMCO'19 Phylogenies of Breast Cancer Brain Metastases 6
Step 2: Deconvolution of Cell Communities Gradient descent by backpropagation [Rumelhart, D.E. et al., Nature . 1986] o Q2: What model and solver to unmix clones? o Neural network deconvolution (NND) o # components: trade-off of model complexity vs. sample size o Mask matrix for cross-validation in NND Yifeng Tao et al. @ ISMCO'19 Phylogenies of Breast Cancer Brain Metastases 7
Step 3: Inference of Phylogeny y 4 y 6 w 14 y 7 w 26 w 12 w 37 x 2 x 1 w 23 w 15 x 3 y 8 w 38 [Nei, M. et al., Molecular y 5 Biology and Evolution . 1987] o Q3: How to infer evolutionary trajectory and perturbed pathways? o Phylogeny skeleton built using neighbor-joining algorithm o Pathway of Steiner nodes inferred by minimizing the elastic potential energy: Yifeng Tao et al. @ ISMCO'19 Phylogenies of Breast Cancer Brain Metastases 8
Effective Pathway Representation Gene expression Pathway z MSD =-2.6 z MSD =-13.4 [Park, Y. et al., Transactions on Computational Biology and Bioinformatics . 2009] o PC1: recurrent feature between primary and metastatic samples o PC2+PC3: variability between patients o Effective in separating primary tumors from metastatic tumors Yifeng Tao et al. @ ISMCO'19 Phylogenies of Breast Cancer Brain Metastases 9
Differentially Expressed Cancer Pathways -log 10 FDR Primary Metastatic o Neurotransmitter and calcium homeostasis o ErbB2/HER2 pathway o Immune activity o Dysregulation promotes tumor growth Yifeng Tao et al. @ ISMCO'19 Phylogenies of Breast Cancer Brain Metastases 10
Landscape of Cell Communities RET Immune Apoptosis, Wnt, PI3k-Akt, Neurotransmitter and ErbB2/HER2 activity Hedgehog TGF-beta calcium homeostasis pathway o The deconvolution provides more fine-grained landscape of tumor cell communities Yifeng Tao et al. @ ISMCO'19 Phylogenies of Breast Cancer Brain Metastases 11
Phylogenies of Cell Communities +ErbB -PI3K-Akt Progression +RET +RET -PI3K-Akt -PI3K-Akt Case 1: 18/22 patients o Common temporal order of perturbed pathways during metastasis o Gained ErbB caused by early events o Expansion of minor clonal populations with lost PI3K-Akt and gained RET Yifeng Tao et al. @ ISMCO'19 Phylogenies of Breast Cancer Brain Metastases 12
Conclusion and Future Work o Conclusion o Pipeline to infer tumor evolution using matched bulk transcriptome o Common temporal order of perturbed pathways in breast cancer brain metastases o Open source code, data and supp: https://github.com/CMUSchwartzLab/BrM-Phylo o Further exploration: multiple metastatic sites Similar pipeline Yifeng Tao et al. @ ISMCO'19 Phylogenies of Breast Cancer Brain Metastases 13
Acknowledgment o Authors Prof. Russell Schwartz Prof. Jian Ma Prof. Adrian V. Lee Haoyun Lei o Fundings Yifeng Tao et al. @ ISMCO'19 Phylogenies of Breast Cancer Brain Metastases 14
Recommend
More recommend
