450 likes | 650 Views
Genetics and the Genome. Structural unit of the genome is a spatial unit described by:Promoter regionEnhancer regionSplice sites/intronsCoding regions3' regions for stability/transport. The genome has a three dimensional structure. The DNA is wound approx twice around 4 dimeric histones (H2A, H
E N D
1. Analysis of Methylation Silencing at Multiple Loci from Multiple Tumor Types Karl T. Kelsey, M.D.
Andres Houseman, Sc.D.
2. Genetics and the Genome Structural unit of the genome is a spatial unit described by:
Promoter region
Enhancer region
Splice sites/introns
Coding regions
3’ regions for stability/transport
3. The genome has a three dimensional structure The DNA is wound approx twice around 4 dimeric histones (H2A, H2B, H3 and H4) with H1 as the linker between each nucleosome
The interplay of DNA sequence and histone architecture is precise and crucially important to gene expression
5. Nucleosome Architecture N-terminal tails protrude from the nucleosome
Acetylation of lysines
Methylation of lysines & arginines
Phosporylation of serines and threonines
Ubiquitination and sumolaytion of lysines
ADP-ribosylation of glutamic acids
7. EPIGENETIC Alterations Modifications in gene expression
Heritable
Stable
Potentially reversible
Compared to genetic
Mutations, deletions
Not reversible
9. Understand Mechanisms of Carcinogens Epigenetics – New mechanism in cancer
Studying Epigenetics in Cancers
Looking at exposures relationships
Epigenetic changes as biomarkers
10. Getting to Cancer Oncogenes
Normal genes
Signal for growth, angiogenesis, etc.
Get mutated or amplified in tumors ? lose control
Tumor Suppressor Genes
Normal genes
Stop growth, prevent cell cycle, etc.
Get silenced in tumors
11. Classical - Genetic Damage
12. Does this explain everything? Tumor Suppressor gene
No expression in tumor
Need to inactivate both copies of gene
Not all inactivation is explained through mutation or deletion
What about non-mutagenic carcinogens?
Alternative Method of Inactivation??
13. Gene Expression
14. Where epigenetics happen!
15. DNA Methylation The covalent addition of methyl group to 5th position of cystosine
Largely confined to CpG dinucleotides
CpG islands - regions of more than 500 bp with CG content > 55%
Islands found in promoter regions of genes
Catalyzed by DNA methyl transferase.
16. Promoter CpG Island Hypermethylation
17. Consequences of Hypermethylation
18. Is this Normal? Non-coding repetitive elements
Lines, Sines, Alu repeats
Centromeric regions
Inactive X-chromosome
Imprinted genes
Some gene promoters – in cell-type specific fashion
19. DNA Methylation in Cancer Aberrant
Occurs in promoters of tumor suppressors
Tumor specific & Clonal
Silences transcription of a gene – equivalent to mutation or deletion
Alternative “hit” to inactivation of tumor suppressor
Targeting and specificity unclear
Not a “global” phenomenon
Carcinogens driving this alteration?
20. methylation silencing in cancerIs it associated with: Genes?
Tissues?
Age?
Gender?
Carcinogen exposure?
Treatment-survival?
Can this alteration be diagnostic?
26. Questions Does carcinogen exposure induce methylation?
Are genes coordinately silenced?
What is the distribution of methylation silencing?
Are all tumors the same?
Does this cluster?
27. Methylation silencing in surgically treated cancers: Lung: N=173
Bladder: N=351
Head and Neck: N=345
Mesothelioma: N=71
29. Can one look at this data through another lens?
30. Comparison of Methylation Profile Between Different Cancers
How distinct are different tumor types with respect to methylation profile?
Are methylation profiles associated with disease type?
How accurately can methylation profile “predict” disease?
31. Data 4 tumor types (910 cases)
Bladder Cancer (350)
Head & Neck Cancer (351)
Lung Cancer (138)
Mesothelioma (71)
18 genes/markers
15 genes, 3 MINTs
33. Tests for Association Traditional chi-square test (4 x 18 table)
?2 = 695.5, 51 d.f., P<10-8
Permutation Test (based on chi-square statistic)
P<10-3, (99th %ile of permutation dist = 55.8)
Conclusion: strong association
34. Prediction
35. Jackknife Prediction Error “Leave-one-out” cross-validation
For each subject i
Delete subject i from data,
Use remaining data to “train” model
Use model to predict outcome for i
Compare to actual outcome
Summarize for all subjects
Assessment:
Misclassification rate (0-1, 0 is best)
“Kappa” statistic for concordance (0-1, 1 is best)
Entropy (smaller is better)
36. Jackknife Results Best prediction from multinomial regression, but results are somewhat difficult to interpret
Cross-tabulation easiest to interpret but worst performance
CART a compromise between prediction error and interpretability
37. CART Results
38. Analysis by Disease What can we say about patterns of methylation within a specific tumor type?
How do methylation “profiles” correlate with other data (e.g. survival)?
Types of analysis:
Latent Class
Latent Trait (Rasch Model)
39. Rasch Model Essentially a (GL) random effects model
Basic model:
Each gene has a different “baseline” frequency characterized by ßj
Each subject’s overall level of methylation is determined the value of the latent variable U.
40. Rasch Model with Survival Parametric proportional hazards model
Baseline hazard modeled as Weibull
Latent variable U is a survival covariate
Similar results using Cox model with empirical Bayes estimates of U
41. Bladder Cancer
42. Lung Cancer
43. Mesothelioma
44. Head and Neck Cancer
45. Conclusions Bladder cancer survival significantly associated with methylation
Lung cancer survival marginally associated with methylation
Methylation not significantly associated with survival for mesothelioma and H&N cancer.
Similar results using LCA
46. Acknowledgements HSPH
Carmen Marsit
Brock Christensen
Heather Nelson
Kim Kraunz
Karen Heffernan
Linqian Zhao
Louise Ryan
Dartmouth University
Margaret Karagas