Transcript SEGMENTAL VARIATION

SEGMENTAL VARIATION
(Copy Number Variants and other
gross chromosomal rearrangements)
Allen E. Bale, M.D.
Dept. of Genetics
SLIDE 0
Importance of Copy Number Variants (CNVs) and
Other Rearrangements in Health and Disease
• Constitutional (germ-line) variants in hereditary conditions
– Large and small copy number variants
– Translocations and inversions: rarely cause a phenotype but
may generate CNVs due to mis-pairing during meiosis
• Somatically acquired variants in cancer
– Duplications and deletions: amplification of oncogene; loss of
tumor suppressor
– Translocations and inversions: place oncogene under control
of an active promoter
SLIDE 1
What is the origin of structural variants?
• An area of active research
• Recurrent constitutional CNVs: Often related to illegitimate
recombination between homologous, but non-identical,
sequences
• Rare, non-recurrent, constitutional CNVs: No obvious
sequence homology at breakpoints, ?non-homologous end
joining
• Tumor CNVs: Any mechanism to create a rearrangement
that favors tumor growth, often non-homologous end joining.
SLIDE 2
Cytogenetically visible CNVs and translocations
SLIDE 3
A Really Large CNV
SLIDE 4
Somatically acquired translocation
SLIDE 5
Limitations of Cytogenetics
• Cell has to be proliferating in order to arrest chromosomes
at metaphase (when they are visible under the microscope)
• Resolution is limited (in the range of 5 Mb)
• Requires highly skilled technologists and still a lot of handson time, even with sophisticated image processing
SLIDE 6
Submicroscopic CNVs: Array CGH*
*Frequently referred to as “chromosome microarray”
SLIDE 7
Example: Submicroscopic 22q deletion
• Abnormal nose, ears, and
palate
• Also heart, parathyroid, and
thymus abnormalities
SLIDE 8
Limitations of Array CGH
• Can’t detect translocations and inversions
• Resolution still limited by number of probes on the
array—typical resolution about 100 kb
• Still a fair amount of variability in results depending on
exactly which array is used
SLIDE 9
Genome-scale sequencing to detect rearrangements
If you could sequence each chromosome as one
continuous piece of DNA, from one end to the
other with no gaps in the sequence, what
structural variants would you miss?
S L I D E 10
Genome-scale sequencing to detect rearrangements
What methods are currently in use?
•Depth-of-coverage methods
Regions that are deleted or duplicated should yield
lesser or greater numbers of reads
•Detection of breakpoints by:
–Short paired reads (like Illumina paired-end sequencing)
Are the sequences at two ends of a fragment both from the
same chromosome? Are they the right distance apart?
–Long reads (kb-scale)
Direct sequencing of breakpoints
S L I D E 11
Genome-scale sequencing to detect rearrangements
•Depth-of-coverage method
•Detection of breakpoints by short paired reads
•Detection of breakpoints by long reads
Compared with cytogenetics and array CGH, how would
the approaches above perform?
• What would be missed by depth-of-coverage
reading?
• What would be missed by detection of breakpoints?
• What problems do you foresee with these two
approaches?
S L I D E 12
Depth-of-coverage example:
Whole exome sequencing as a tool to identify
both sequence variants and CNVs
S L I D E 13
Whole exome sequencing
• Capture portions of the genome containing
exons in order to efficiently sequence
coding regions
• Not designed for CNV detection, but
potentially contains information on gene
dosage
• For any gene, the number of fragments
captured on the array and sequenced
should be proportional to the
representation in the starting material
S L I D E 14
Array CGH vs. Exome Sequencing
S L I D E 15
Does this work at all?
• Total reads on the X chromosome were counted in a series of males and
females
• Gene dosage for the X chromosome in males should be half the gene
dosage for the X chromosome in females
S L I D E 16
Does it work for single exons?
Reads counted for each exon of the OTC gene on X chromosome
Males should have one half the female dosage.
• Read number varies among exons due to different capture efficiencies
but is consistent subject to subject.
• Exons with sufficient read numbers show dosage effect.
• Performs very well for this 70 kb gene taken as a single unit.
S L I D E 17
Approach to scanning the whole genome for CNVs
• The genome was divided into 50 kb windows.
• Intervals with zero reads were removed.
• Mean number of reads and standard deviations for each
interval were calculated from 10 exome sequences.
• Depth of coverage in a single patient was compared to
average and standard deviation of depth of coverage.
• Algorithms were developed for:
– Classifying X chromosome as being deleted in males
compared with females
– Classifying X chromosome as being duplicated in females
compared with males
S L I D E 18
Chromosomal coverage with non-zero, 50 kb intervals
corresponds exactly to density of coding sequences
S L I D E 19
Test case: Female with a 338 kb duplication on 5q35
Diagram shows all loci passing initial algorithm
S L I D E 20
Filter #1: Require two adjacent intervals to both
be deleted or duplicated
S L I D E 21
Filter #2: Remove “deleted regions” that contain
heterozygous variants
S L I D E 22
Filter # 3: Remove intervals with read counts <200
S L I D E 23
Application to 7 subjects with deletions or
duplications in 500 kb to 1 Mb range
S L I D E 24
Another reduced representation method: Alu-PCR*
Mei et al. BMC Genomics 2011, 12:564
S L I D E 25
Some problems with use of reduced representation data
• Intervals with no genes are not covered in exome (important?)
• Intervals having close homologs elsewhere in the genome can
not be accurately evaluated.
• Because this technology is evolving rapidly, the normal standard
to which a test sample is compared needs to be a pool of recent
sequences generated by identical methods (huge FDR with nonhomogeneous samples).
S L I D E 26
What works well: Targeted analysis of candidate genes
Case 1
• 30 year old woman with personal and family history of renal cell cancer
• Scanning candidate genes for point mutations or small indels:
o
o
o
o
o
o
o
o
o
o
SDHB
SDHC
SDHD
FH
VHL
MITF
BAP1
MET
FLCN
HNF1B
Neg
Neg
Neg
Neg
Neg
Neg
Neg
Neg
Neg
Neg
S L I D E 27
IGV frame showing read counts for exons in region of FH gene
(upper panel from patient—lower panel from normal control)
S L I D E 28
Case 2
• One year old male with low muscle tone and strength and poor cardiac
function
• No family history of a similar disease
• Scanning entire exome for point mutations or small indels:
• No mutations found
S L I D E 29
IGV frame showing read counts for exons in region of DMD gene
(upper panel from patient—lower panel from normal control)
S L I D E 30
Case 3
• 25 year old woman with numbness and weakness of lower legs
• Similar or more severe features in several relatives
• Scanning of entire exome for point mutations and small deletions:
• Negative
S L I D E 31
IGV frame showing read counts for exons in region of PMP22 gene
(upper panel from patient—lower panel from normal control)
S L I D E 32
For a review of published depth-of-coverage methods for
exome or genome data see:
Klambauer, G. et. al. (2012). "cn.MOPS: mixture of
Poissons for discovering copy number variations in nextgeneration sequencing data with a low false discovery
rate." Nucleic Acids Res.
Compares several programs, none of which work really
well.
Two newer programs for exome sequencing are in your
reading list.
S L I D E 33
Paired-end methods
• Illumina HiSeq, the current industry leader in highthroughput sequencing, generates short reads from
fragments 200 to 600 bp long.
• Reading both ends of the same fragment gives you
sequences that should lie 200 to 600 bp apart
• Other methods can generate paired fragments that lie
even farther apart
S L I D E 34
Long paired-end methods
Paired end mapping—up to thousands of bp apart
From Korbel et al., 2009
S L I D E 35
Identifying Structural Mutations: Deletions &
Duplications
S L I D E 36
Identifying Structural Mutations: Inversions
S L I D E 37
Identifying Structural Mutations: Translocations
S L I D E 38
Analyzing structural variations from paired end data
• PEMer (Korbel et al., 2009): For discovery of CNVs and
inversions; could also be implemented for translocations
• Breakdancer (Chen et al., 2009): For discovery of CNVs,
inversions, and translocations
S L I D E 39
Identifying Structural Mutations with paired end sequence:
What goes wrong?
S L I D E 40
How to overcome problems with paired end detection of CNVs
Separating the wheat from the chaff
• Technical artifacts (ligation of unrelated fragments during
library preparation) may be numerous but will be random
• Artifacts related to homologous sequences (see previous
slide) will be reproducible but common to all samples
• Real structural variants will be reproducible within a
sample and not common to all samples
• How much reading depth do you need to detect the real
variants?
S L I D E 41
Toward direct sequencing of breakpoints
• Long reads
– PACbio can generate reads of 1000 bp or so
– Nanopore sequencing said to generate reads in the
10s of thousands
• Strobe sequencing with PACbio:
Normally read length is limited due to inactivation of
polymerase by laser. Short bursts of laser give sample
sequences along a stretch of DNA in the 20 kb range.
S L I D E 42
Programs for analysis of longer reads that directly
sequence breakpoints
• CREST (Wang et. al., 2011): Detects small and large
structural variants by direct sequencing of breakpoints.
• SRiC (Zhang et al., 2011): Similar to CREST
• Algorithm for strobe reads (Ritz et al., 2010)
• Several methods reviewed in Abel and Duncavage, 2014
S L I D E 43
Conclusions
• Structural variation in the genome accounts for a great
deal of human phenotypic variability including disease
• Depth-of-coverage methods can detect many CNVs but
not inversions and translocations. Variation from sample
to sample limits sensitivity and specificity.
• Whole genome sequencing, which can identify all types
of structural variants, will supersede depth-of-coverage
methods.
• Large scale and small scale duplications and repetitive
sequences remain a major obstacle.
S L I D E 44