1. CSCI2950-C Lecture 9 Cancer Genomics
October 16, 2008

2. Outline Cancer Genomes Paired-end Sequencing Rearrangements
Comparative Genomic Hybridization

3. Cell Division and Mutation
Single nucleotide
change
A major contributor to the development of cancer are somatic mutations that occur during cell division
Will focus on structural and later copy number, which is not to say that single are not as important. What is the effect of structural changes
Copy number
Structural

4. Rearrangements in Cancer
1) Change gene structure, create novel fusion genes
Gleevec targets ABL-BCR fusion
These rearrangements cause changes in gene structure…classic example is Philadelphia chrom….more subtle regulatory changes where a gene is placed in a different regulatory context. Here the myc cancer-promoting gene is activating by being put under control of an immunoglobulin gene.
2) Alter gene regulation
Burkitt’s lymphoma
IMAGE CREDIT: Gregory Schuler, NCBI, NIH, Bethesda, MD

5. Cancer Genomes Fusion gene in >50% prostate cancer patients
Classically, discovered through cytogenetic techniques like chromosome painting shown here. More complicated in solid tumors. Don’t reveal detailed architecture, genes affected. Rearrangements thought to be mostly noise. Recent identification of new fusion gene. Can we use genome sequencing?
Fusion gene in >50% prostate cancer patients
(Tomlins et al. Science 2005)

6. Shotgun Sequencing Get one or two reads from each segment
genomic segment
cut many times at random (shotgun)
No technology yet for reading a long strand of DNA in it’s entirety. Instead fragment. New tech. prodcuing shorter reads. Given that we have sequence of human genome, can detect rearrangements with less sequencing.
Get one or two reads from each segment
~500 bp
~500 bp

7. Sequencing of Cancer Genomes
etc.
What to sequence from each tumor?
Whole genome: all alterations
Specific genes: point mutations
Hybrid approach: structural rearrangements
All alternations,…if you could assemble it

8. End Sequence Profiling (ESP) C. Collins and S. Volik (2003)
Pieces of cancer genome: clones ( kb).
Cancer DNA
Sequence ends of clones (500bp).
Map end sequences to human genome.
Because of end sequencing protocol, clones have direction x y
Human DNA
Each clone corresponds to pair of end sequences (ES pair) (x,y).
Retain clones that correspond to a unique ES pair.

9. End Sequence Profiling (ESP) C. Collins and S. Volik (2003)
Pieces of cancer genome: clones ( kb).
Cancer DNA
Sequence ends of clones (500bp). L
Valid ES pairs
Lmin ≤ y – x ≤ Lmax, min (max) size of clone.
Convergent orientation.
Map end sequences to human genome.
Because of end sequencing protocol, clones have direction x y
Human DNA

10. End Sequence Profiling (ESP) C. Collins and S. Volik (2003)
Pieces of cancer genome: clones ( kb).
Cancer DNA
Sequence ends of clones (500bp). L
Map end sequences to human genome.
Because of end sequencing protocol, clones have direction. Some pairs cannot be mapped due to repeats in human genome. x y a b
Human DNA
Invalid ES pairs
Putative rearrangement in cancer
ES directions toward breakpoints (a,b):
Lmin ≤ |x-a| + |y-b| ≤ Lmax a x y b

11. ESP of Normal Cell 2D Representation All ES pairs valid. x y Human DNA
Lmin ≤ y – x ≤ Lmax
2D Representation
Each point (x,y) is ES pair.
Sometimes useful to have a 2D representation. All ES pairs near diagonal.
Genome Coordinate
Genome Coordinate

12. ESP of Tumor Cell Valid ES pairs satisfy length/direction constraints
Lmin ≤ y – x ≤ Lmax
Invalid ES pairs
indicate rearrangements
experimental errors
First show how to deal with noise….

13. Clusters and Coverage Pieces of tumor genome: clones (100-250kb).
Cancer DNA
Pieces of tumor genome: clones ( kb).
Rearrangement
Chimeric clone
Sequence ends of clones (500bp).
Primary source of noise is chimeric clones joined from random concatenation of non-contigouous portions of cancer genome. Can deal w/ noise w/ clustering. Clusters also allow us to identify breakpoints more precisely
Cluster invalid pairs
Isolated invalid pair
Map end sequences to human genome. y x
Human DNA

14. Clusters x2 y2 a b x1 y1 Clone size: Lmin  (a – x1) + (b – y1)  Lmax
Genome
coordinate
Lmax
Clustering have a nice geometric interpretation – each clone is trapezoid. Intersection allows to localized shared breakpoint.
Lmin
(a,b)
(a,b)
(x1,y1)
(x2,y2)
Genome coordinate

15. Fusion Genes Gene 1 Gene 2 Human x y a b Tumor
Can identify fusion genes by location of breakpoint.

16. Fusion Genes Intersection → probability of fusion genex y a b
Lmax
Lmin
(a,b)
Genes give rectangle in 2D representation. Overlap w/ trapezoids. Used this rep. to compute how much sequencing needed to find fusion genes.
(x1,y1)
(x2,y2)
Intersection → probability of fusion gene
Respect direction of transcription
Gene1 Gene2
Bashir, et al. (2008) PLOS Comp Biol.

17. Results: Fusion Gene in Breast Cancer BCAS3-BCAS4
Probability of Fusion = 1
Note: More precise sizing information available for some clones
Bashir, et al. (2008) In Press.

18. ESP Data Coverage of human genome: ≈ 0.34 for MCF7, BT474
Sample
End Sequenced
Uniquely Mapped
Invalid ES Pairs
MCF7
19831
12143
491
BT474
9850
7547
186
SKBR3
9267
6950
187
Breast
9401
6540
164
7623
5381
113
Prostate
5013
3296 96
Ovary
5570
3714 87
Brain
4198
3051 67
Breast Cancer
Cell Lines
Tumors
Fusion gene from MCF7 b.c. cell line. Have also applied techniqe to several other cell lines and tumor samples. Still do not have complete coverage of genome, but new sequencing technologies should make this more straightforward. Even w/ low coverage, still very interesting patterns in data.
Coverage of human genome:
≈ 0.34 for MCF7, BT474
Raphael, et al. (2008)

19. Candidate Fusion Genesx y a b
Gene1
Gene2
Probability
Fusion
MCF7
ASTN2
PTPRG
1.000
BCAS4
BCAS3
KCND3
PPM1E
NTNG1
BCAS1
0.996
BT474
NCOA2
ZNF704
ZFP64
PHACTR3
0.632
SKBR3
PTPRT
PHF20
KCNQ3
RIMS2
0.933
Confirmed by clone sequencing 3 9
97kb
PTPRG
ASTN2
Sequenced Clone

20. Breakpoint Detection
Detect a rearrangement breakpoint when clone includes breakpoint.
Cancer Genome
breakpoint ζ
Normal Genome xC yC

21. Lander-Waterman Statistics
Given: N clones of length L from a genome of size G P(ζ covered by clone) = 1 – (1 – L/G)N ≈1 – e-c, where c = N L / G is coverage P(breakpoint ζ detected) ≈1 – e-c c
P(detection) 1
0.632 2
0.864 4
0.982 8
0.999

22. Cancer Genome Organization
Can we do more and reconstruct this picture. Not just identify single rearrangements?
What are detailed organization of cancer genomes?
What sequence of rearrangements produce these architectures?

23. ESP Genome Reconstruction Problem
Human genome
(known) A B C D E
Unknown sequence of
rearrangements
Tumor genome
(unknown)
Map ES pairs to
human genome.
Reconstruct tumor
genome
Problem we call ESP G.R. Problem. x2 y2 x3 x4 y1 x5 y5 y4 y3 x1
Location of ES pairs
in human genome.
(known)

24. ESP Genome Reconstruction Problem
Human genome
(known) A B C D E
Unknown sequence of
rearrangements
Tumor genome
(unknown) -C -D E A B
Map ES pairs to
human genome.
Reconstruct tumor
genome x2 y2 x3 x4 y1 x5 y5 y4 y3 x1
Location of ES pairs
in human genome.
(known)

25. ESP Plot 2D Representation of ESP Data Each point is ES pair.
(x3,y3)
(x4,y4) D
(x2,y2)
2D Representation of ESP Data
Each point is ES pair.
Can we reconstruct the tumor genome from the positions of the ES pairs?
Human
(x1,y1) C B A A B C D E
Human

26. ESP Plot → Tumor Genome E E D -D Human C -C B B A A A B C D E Human
Reconstructed
Tumor Genome -C -D E A B

27. Real data noisy and incomplete!
Valid ES pairs
satisfy length/direction
constraints
Lmin ≤ y – x ≤ Lmax
Invalid ES pairs
indicate rearrangements
experimental errors
Clustering can help us with the noise. Injecting a bit of extra biological knowledge also helps noise and addresses incomplete data. Namely the rearrangements in genomes results from specific processes that cause particular rearrangements.

28. Computational Approach
Use known genome rearrangement mechanisms s A t C -B B
inversion
Human
Tumor s A t -B -C B D C
translocation
Appeal to parsimony.
Find simplest explanation for ESP data, given these mechanisms.
Motivation: Genome rearrangements studies in evolution/phylogeny.

29. ESP Sorting Problem Given: ES pairs (x1, y1), …, (xn, yn) Find:
G = [0,M], unichromosomal genome.
Inversion (Reversal) s,t A B C G x1 y1 x2 y2
x, if x < s or x > t,
t – (x – s), otherwise. s t 
s,t(x) = A -B C
G’ = G x1 y1 x2 y2 s t
Given: ES pairs (x1, y1), …, (xn, yn)
Find:
Minimum number of reversals s1,t1, …, sn, tn such that if  = s1,t1… sn, tn,
then ( x1,  y1 ), …, ( xn,  yn) are valid ES pairs.
Complexity of this problem unknown.

30. Sparse Data Assumptions
Each cluster results from single inversion or translocation.
human y1 x2 x3 y3 y2 x1 y1 x2 x3 y3 y2 x1
tumor
2. Each clone contains at most one breakpoint.
tumor

31. ESP Genome Reconstruction: Discrete Approximation
Human
Remove isolated invalid pairs (x,y)
We can get a discrete approximation of this problem as follows. First…
Human

32. ESP Genome Reconstruction: Discrete Approximation
Human
Remove isolated invalid pairs (x,y)
Define segments from clusters
Human

33. ESP Genome Reconstruction: Discrete Approximation
Human
Remove isolated invalid pairs (x,y)
Define segments from clusters
ES Orientations define links between segment ends
Human

34. ESP Genome Reconstruction: Discrete Approximation
(x2, y2)
(x3, y3) t
(x1, y1) s
Human
Remove isolated invalid pairs (x,y)
Define segments from clusters
ES Orientations define links between segment ends
Human

35. ESP Graph Edges: Human genome segments ES pairs2 3 5 1 4 2 3 5 1 4
Edges:
Human genome
segments
ES pairs
Paths in graph are tumor genome architectures.
Tumor
Genome
( )
Human
( )
Minimal sequence* of
translocations and
inversions 2 3 5 1 4
*Hannenhalli-Pevzner theory

36. Sorting Permutations by Reversals
(Sankoff et al.1990)
 = 12…n signed permutation
Reversal (i,j) [inversion]
1…i-1 -j ... -i j+1…n
Problem: Given , find a sequence of reversals 1, …, t with such that:
 . 1 . 2 … t = (1, 2, …, n) and t is minimal.
Solution: Analysis of breakpoint graph ← ESP graph
Polynomial time algorithms
O(n4) : Hannenhalli and Pevzner, O(n2) : Kaplan, Shamir, Tarjan, 1997.
O(n) [distance t] : Bader, Moret, and Yan, O(n3) : Bergeron, 2001.

37. Sorting Permutations  1 -3 -4 2 5 -3 -2 4 5 1 1 2 3 4 5

38. Breakpoint Graph Black edges: adjacent elements of   1 -3 -4 2 5
start 1 -3 -4 2 5
end
Gray edges:
adjacent elements of i = 1 2 3 4 5
start
end
Key parameter:
Black-gray cycles

39. Breakpoint Graph ESP Graph → Tumor Permutation and Breakpoint Graph
Black edges:
adjacent elements of  
start 1 -3 -4 2 5
end
start -3 -2 4 5 1
end
Gray edges:
adjacent elements of i = 1 2 3 4 5
start
end
Key parameter:
Black-gray cycles
ESP Graph → Tumor Permutation and Breakpoint Graph
Theorem: Minimum number of reversals to transform  to identity permutation i is:
d() ≥ n+1 - c()
where c() = number of gray-black cycles.

40. Breakpoint Graph ESP Graph → Tumor Permutation and Breakpoint Graph
Black edges:
adjacent elements of  
start 1 -3 -4 2 5
end
start -3 -2 4 5 1
end
Gray edges:
adjacent elements of i = 1 2 3 4 5
start
end
ESP Graph → Tumor Permutation and Breakpoint Graph
Theorem: Minimum number of reversals to transform  to identity permutation i is:
d() = n+1 - c() + h() + f()
where c() = number of gray-black cycles.

41. Multichromosomal Sorting
Concatenate chromosomes
Translocations modeled by reversals in concatenate
Minimal sequence in polynomial time (Hannenhalli & Pevzner 1996, Tesler 2003, Ozery-Flato and Shamir, 2003.) A1 A2 A1 B2
translocation B1 B2 B1 A2
concatenation
concatenation
reversal A1 A2
-B2
-B1 A1 B2
-A2
-B1

42. MCF7 Breast Cancer Cell Line
Applied this technique to MCF7.

43. MCF7 Breast Cancer Cell Line
Can be many minimal sequences. Some events are early in many minimal sequences. Left out an important event: duplications.
Sequence
Human chromosomes
MCF7 chromosomes
5 inversions
15 translocations
Raphael, et al. (2003) Bioinformatics

44. What about duplications?
Clusters were biased to a very small region of the genome. These regions are duplicated – can be measured as such independently. Take a duplication centric view
11240 ES pairs
10453 valid (black)
737 invalid
489 isolated (red)
248 form 70 clusters (blue)
33/70 clusters
Total length: 31Mb