Integration of multiple sources of evidence in clinical classification of VUS

1
Integration of multiple sources of evidence in
clinical classification of VUS

• David Goldgar
• University of Utah School of Medicine

2
What do we mean by a VUS?

• Sequence variant in a gene with a clearly
  established role in a given disease
• Usually rare in the general population (or in the
  clinically tested population)
• If pathogenic would be clinically important to
  the individual carrying the variant
• Typically missense, intronic, or in-frame
  deletions (but could include others)

3
Evidence potentially useful for classification of
VUS

• Direct
• Co-segregation of VUS with disease in pedigrees
• Powerful direct evidence but often difficult to
  get additional samples from family members.
• Co-occurrence (in trans) with deleterious
  mutations
• Only useful if homozygotes/compound heterozygotes
  are embryonically lethal
• Distribution of family history of probands
  carrying a VUS
• Indirect
• Severity of amino acid change and evolutionary
  conservation of wt residue
• Effects on protein structure (if known)
• Functional evaluation in model systems
• Other evidence relevant to cancer susceptibility
  genesLOH, pathology, expression array/CGH
  signatures, MSI

4
Genetic vs. Functional/Sequence-based Approaches

• Genetic approaches normally require multiple
  observations to be useful
• However, most VUS occur lt5 times
• Functional and sequence-based analyses can be
  done (in theory) on any variant
• Relationship between functional assay and disease
  risk typically unknown
• If valid relationships could be established, many
  more VUS could be classified

5
Select UV
Quantifiable Individual or family
data Co-occurrence Family History Co-segregation
More data
LR 1
LR 2
LR 3
Combined evidence
? (LRi)
LRgt1000 or LR lt0.01
Yes
UV classified
Validation set for functional and conservation
data
No
Incorporate evidence from conservation
and functional data using existing models
Initial model
Refine model
No
Yes
LRgt1000 or LR lt0.01
Goldgar et al. AJHG 75535-44. 2004.
6
Easton et al. AJHG 2007
7
Align-GVGD
An extension of the original Grantham Difference
to protein multiple sequence alignments. It uses
two variables, GV and GD.
Grantham Variation (GV) A quantitative measure of
the range of variation present at a position in a
protein multiple sequence alignment. GV0
position is invariant GVgt 60 non-conservative
substitution is tolerated
Grantham Deviation (GD) A quantitative measure of
the fit between a missense substitution and the
range of variation observed at its position in
the protein. GD0 substitution is within the
observed range of variation GDgt 60
substitution is non-conservatively beyond the
range of variation Website http//agvd.iarc.fr
5000
8
Analysis of rare missense substitutions Distribu
tion of risk in the GV-GD plane
?0.81
?0.66
?0.29
GD
?0.00
GV
Tavtigian et al., Human Mutation, (almost) in
press
5000
9
? Raw data normalized by Renilla luciferase
driven by a constitutive promoter. Results from
triplicate experiments in which a Gal4 DBD BRCA1
1396-1863 is co-transfected with the reporter
(shown above graph) are plotted as percent of
wild type activity.
Marcelo Carvalho Alvaro Monteiro
10
Estimation of sensitivity and specificity of
functional assays (simple approach)

• For each variant with functional data, use prior
  probability based on sequence analysis and
  log-odds from genetic data to get posterior
  probability of being pathogenic
• Sample each variant as being pathogenic or
  neutral from posterior distribution
• Calculate sensitivity and specificity etc., from
  this simulated data set
• Average over many replicates to get estimated
  sensitivity/specificity and confidence interval
• For Transcriptional Activation assay, estimates
  were 0.85 (0.67 - 1.0) for sensitivity and 0.65
  (0.58 - 0.75) for specificity

11
The Lyon Meeting on VUS4-5 February, 2008

• Organised by Sean Tavtigian at IARC
• Goal to have a highly focused knowledge transfer
  exercise representing diverse opinions
• Representatives from MMR, p16, and BRCA worlds
• Assembled expertise clinical cancer genetics,
  functional assays, sequence analysis, genetic
  epidemiogy, etc.
• International US, UK, NL, Australia, France

12
(No Transcript)
13
Series of papers to be written for Human Mutation

• Introduction to the series
• Genetic variant classification using clinical and
  epidemiological data
• In vitro and ex vivo assessment of functional
  effects of genetic variants
• Splice site alteration assessment
• Tumour characteristics as an analytic tool
• Integration - the nuts and bolts of combining
  across data types
• Locus specific databases
• Clinical utility and risk communication

14
Issues in Integration

• Transferability of results from one kind of
  mutation to others (e.g., truncating to missense)
• LOH, Pathology, Co-occurrence
• Choice of appropriate prior probability
• Independence of evidence from different sources
• Incorporating discrete types of evidence into a
  probabilistic framework
• Combining everything -
• Mixture Models via MCMC
• Cluster analysis type methods

15
How to disseminate VUS information to the
research and clinical communities

• Should research information be separate/different
  from clinical use?
• Qualitative vs. Quantitative information
• What is the appropriate place to store this
  information?
• Locus specific databases, e.g. BIC?
• clinical databases?
• Human Variome database?
• All of the above?
• How much detail of the evidence should be
  presented?

16
Issues in Transfer of Knowledge to Clinical
Practice

• What are appropriate thresholds for causality and
  neutrality respectively?
• What should be reported?
• Only those variants that have been definitively
  (by above threshold) classified?
• Should the current odds of causality?
• Intermediate discrete categories, e.g., likely
  deleterious, probably neutral?
• What if variants confer intermediate risk? Can
  the methods be adapted to estimate risks? Would
  it be useful?

17
Unified Framework for Genetic Testing (including
VUS)

• Prior probability of an affected proband being a
  carrier of a pathogenic mutation in gene X based
  on proband phenotype (including e.g., pathology,
  MSI, IHC, etc.) and family history and locus
  heterogeneity
• Could be model based
• Add result of genetic testing of proband
• wildtype or sequence variant (excluding common
  polymorphisms)
• Add variant specific information
• Sequence analysis (A-GVGD, SIFT)
• Functional/structural assay if available and
  quantifiable
• Co-segregation analysis if additional family
  members available to be tested

18
Unified Framework Translation into disease risk

• From previous information can calculate the
  posterior probability that the individual carries
  a pathogenic mutation or wildtype (or a variety
  of intermediate risks if reliably estimated)
• Then disease risk for an at-risk relative of a
  proband discovered to have variant V is
• If V P(Vpath)P(Dpath)P(Vwt)P(Dwt fam hx)
• If V- P(Vpath)P(Dpop)P(Vwt)P(Dwt famhx)
• Could be integrated into a single Web-based tool
  (including sequence, family history,
  co-segregation, family hx, environmental factors,
  etc.)

19
Acknowledgements
BRCA2 functional assays F. Couch, D. Farrugia,
M. Argawal, L. Wadum Data Preparation A.
Deffenbaugh, D. Bateman, C. Frye Myriad
Genetics Sequence Analysis S. Tavtigian, A.
Thomas, G. Byrnes BRCA1 functional assays A.
Monteiro, M. Carvalho Moffitt Cancer
Center Statistical Aspects D. Easton, D.
Thompson Cambridge E. Iversen Duke
University The BIC steering committee Grants
P50CA116201 R01CA116167 ACSRSG-040220-01-CCE
(FC) and CA92309 (AM)