I. The document discusses developing a framework for individualizing radiation dose based on genetic and imaging data from tumor samples. Experimental and clinical studies have been conducted to understand how genetic variants affect radiation sensitivity.
II. Multi-omic databases have been created from analyzing over 1000 cancer cell lines and patient-derived xenografts. Computational and experimental pipelines were used to test over 500 genetic variants across 92 genes to identify variants that regulate survival after radiation.
III. The findings show the potential to associate genetic alterations with radiomic features to help personalize radiation treatment doses for individual patients. Further work is still needed to fully validate new genetic variants and their molecular mechanisms of radiation sensitivity.
8. To develop an information capability, you first need information…
I. >550 variants tested to date
II. Systematic unary, massively
parallel testing of variants
III. Primary screen completed
IV. Validation phase ongoing
I. 533 cell lines
II. Associations made with
>65,000 genomic features
III. Published in 2016
I. ~361 PDXs generated to date
II. Project launched in 2019.
10. CCLE is a compilation of gene
expression, chromosomal copy
number, and massively parallel
sequencing data from >1000 human
cancer cell lines.
We use established and new
algorithms to reflect omic data on
our phenotype (survival after
drug/radiation exposure) and
identify biological predictors of
response to therapy (resistance and
sensitivity).
CCLE
Nature. 2012 Mar 28;483(7391):603-7.
Multi-omic Databases
11. Nat Commun. 2016 Apr 25;7:11428.
Predicting Response in a Probabilistic Landscape
12. Genetic Determinants Of Tumor Radiation Sensitivity
*Five of the top six genes that when
mutated correlate with sensitivity to
IR have confirmed roles in the DNA
damage response: FLNA, SMG1,
TP53BP1, LRP1, and RIF1.
**WES has been made publically
available.
*
*
*
*
*
17. Residue
Number
Integral Survival
Too much noise at the gene level
Residue
Number
Integral Survival
Population AVG
Gene AVG
Gene X
Population AVG
Gene AVG
Gene Y
Stratify by Protein Domains: Enriching for Signal from the Noise
18. Genetic Variants that Regulate Survival after Irradiation
Primary Screen:
• >1600 replicates
• 508 Variants
• 92 genes
21. Variants that Confer Radiation Resistance
Are Evolutionarily Conserved & Are Under Somatic Oncogenic
Selection
22. N
TC
1
N
TC
2
K
EA
P1
K
O
1
K
EA
P1
K
O
2
0.0
0.2
0.4
0.6
FractionViable
4 Gy mCherry
Wild type
E117K
P278S
G333C
R470S
KEAP1 wild type, but not DN or l-o-f variants, restores radiation sensitivity
*
*
Dominant
Negative
Development of a Complementation Platform to
Query DN v. l-o-f Variants
KEAP1
GAPDH
NRF2
HDAC1
Cytoplasmic Nuclear
KEAP1
KEAP1 alleles
GAPDH
HDAC1
NRF2
Complementation with CRISPR-resistant cDNA
KO2 Cells
CRISPR-Cas9 targeted to an exon-intron junction
33. I. Correlations with clinical outcomes.
II. Clonal reconstructions of the primary and recurrent tumors.
III. Sex specificity (crossover).
IV. Time of day of treatment (cortisol release in mice).
V. Orthotopic versus heterotopic.
VI. Immune cell capture.
VII. microCT (radiomic feature stability and genomic
correlates—in vivo or ex vivo).
Data Collection Variables
34. The Need for a Framework for Individualizing Radiation Dose
J Thorac Oncol. 2017 Mar;12(3):510-519.
Nat Commun. 2014 Jun 3;5:4006.
https://doi.org/10.1016/S2589-7500(19)30058-5
35. We seek to augment intelligence, not replace it…
36. We seek to augment intelligence, not replace it…
This could benefit
from augmentation.
37. We seek to augment intelligence, not replace it…
38. An Image-based Framework for Individualizing Radiation Dose
https://doi.org/10.1016/S2589-7500(19)30058-5
39. An Image-based Framework for Individualizing Radiation Dose
https://doi.org/10.1016/S2589-7500(19)30058-5
40. An Image-based Framework for Individualizing Radiation Dose
https://doi.org/10.1016/S2589-7500(19)30058-5
41. An Image-based Framework for Individualizing Radiation Dose
https://doi.org/10.1016/S2589-7500(19)30058-5
42. An Image-based Framework for Individualizing Radiation Dose
https://doi.org/10.1016/S2589-7500(19)30058-5
43. An Image-based Framework for Individualizing Radiation Dose
https://doi.org/10.1016/S2589-7500(19)30058-5
44. An Image-based Framework for Individualizing Radiation Dose
https://doi.org/10.1016/S2589-7500(19)30058-5
45. An Image-based Framework for Individualizing Radiation Dose
https://doi.org/10.1016/S2589-7500(19)30058-5
49. Did the work
Titas Bera, BS
Jessica Castrillon, MS
Priyanka Gopal, MS
Aaron Petty, MS
Mishka Gidwani, MS
Semihcan Doken, MS
Alums
Eui Kui Chie, MD, PhD (SNUH)
Gwendolyn Kuzmishin, BS (CCLCM)
Kevin Rogacki, MD (Northwestern)
Craig Peacock, PhD (CWRU)
Roberto Vargas, MD (CCF)
Brian Yard, PhD (Viocera)
NIH
KL2CA188339-01
NCI R37CA222294-01
Acknowledgments
Collaborators
Drew Adams, PhD (CWRU)
Pablo Tamayo, PhD (UCSD)
Peter Hammerman (Novartis)
Ben Haibe-Kains, PhD (UoT)
Scott Bratman, PhD (UoT)
Urs Hagemann (Bayer)
Gerhard Siemeister (Bayer)
Ali Kamen, PhD (Siemens)
Bin Lou, PhD (Siemens)
Editor's Notes
CADD scores are based on diverse genomic features derived from surrounding sequence context, gene model annotations, evolutionary constraint and functional predictions. It uses a machine learning model trained on a binary distinction between simulated de novo variants and variants that have arisen and become fixed in human populations since the split between humans and chimpanzees. The differences between the de novo simulated variants and the actual observed variant that have been fixed represent the variants that have a high CADD score (Combined Annotation-Dependent Depletion).
PDX models recapitulate the genetic complexity of human tumors and are therefore predicted to represent improved models for drug and predictive biomarker development (4). These patient derived models have the promise to serve as powerful tools to enhance clinical trial success at a key step in biomarker and drug development.