Chemogenomics Methods

1
Chemogenomics Methods
Paul Blower Phamacogenomics 725
Feb. 27, 2007
2
Definition of Chemogenomics

• Use of genomics to measure the system-wide effect
  of a compound on an intact biological system,
  either cells or whole organisms
• Also investigates the consequences of
  differential gene/protein expression on cellular
  response to compound treatment
• Combines genomics or proteomic profiling with
  chemoinformatics and statistical analysis

3
Overview

• Chemogenomic methods applied to NCI-60
  (Weinstein, NCI)
• Case study of mdr1 (Huang, Sadee)
• Yeast deletion library (Giaever, Stanford)
• Connectivity map (Lamb, MIT/Broad)

4
Conceptual Framework
5
Molecular Profiling of NCI-60
DNA
Compound screening
Protein
RNA
Weinstein, Mol. Cancer Ther., 2006 5(11) 2601-5
6
NCI Cancer Screening

• Tests compounds against 60 tumor cell lines
• Breast (5) Leukemia (6) Ovarian (7)
• CNS (6) Lung (9) Prostate (2)
• Colon (7) Melanoma (10) Kidney (8)
• Compounds tested for growth inhibition of tumor
  cell lines, determine GI50
• Since 1990, gt100,000 compounds screened

Shoemaker, Nat. Rev. Cancer, 2006, 6,
813-23 Source http//dtp.nci.nih.gov/docs/cancer/
cancer_data.html
7
NCI Gene Expression

• Compared cDNA from individual cell line (resting)
  with cDNA from a pool of 12 cell lines
• Microarray contained 9,703 DNA elements
• 3,700 named genes
• 1,900 human homologs
• 4,100 ESTs
• Selected 3748 genes passing QC and sequence
  verified
• Since 2000, NCI-60 gene expression measured on gt5
  platforms

U. Scherf, et. al., Nature Genet., 2000, 24,
23644 source http//discover.nci.nih.gov
8
Experimental Methods
Compound activity growth inhibition of tumor
cell lines

• Total protein assessed 48 hours after drug
  treatment by sulforhodamine assay
• Dose-response typically 5 points dilution series
  10-4 10-8M

9
Experimental Methods
cDNA Microarrays
mRNA from individual NCI-60 cell lines
mRNA from pool of 12 cell lines
cDNA labeled with cye5-dUTP (red) or cye3-dUTP
(green)
10
Experimental Methods
Oligonucleotide microarrays (Affymetrix)
photolabile protecting group

• 25-mer oligos synthesized by photolithography
• Probes synthesized from sequence information
  alone, no clones, PCR, etc.
• Probe set 10 (PM, MM)-oligo pairs per gene,
  selected for uniqueness
• Signal is average (PM MM) over probe set

11
AT Matrix
x

• Matrix of Pearson correlation coefficients
• Each row of A and T is normalized by mean and
  standard deviation,
• Matrices are multiplied to obtain AT
• Each entry is divided by n - 1, where n (60) is
  the number of cell lines
• Optionally, the rows and columns of the product
  matrix are arranged in cluster order

12
Gene-Compound Correlations

• Breast
• CNS
• Colon
• Leukemia
• Lung
• Melanoma
• Ovarian
• Prostate
• Renal

across NCI60 cell lines, r -0.87
13
Compound-Gene Correlations
strong correlation (r 0.88)
weak correlation (r 0.29)
Pearson correlation coefficient
14
Compound-Gene Correlations
Distribution of correlation coefficients between
4,463 compounds and 3,748 genes
15
Heat map
16
Sample Heatmap
17
Sample Heatmap
Compound-gene correlations
18
Unclustered Heatmap
Rows and columns in random order
19
Clustering

• Purpose discover natural groupings in a set of
  objects e.g., compounds, genes
• Grouping done on basis of similarity between
  pairs of objects
• No assumptions made about number of groups
  differs from classification

20
Methods of Clustering

• Hierarchical produces tree structure
• Agglomerative bottom-up
• Divisive top-down
• Non-hierarchical
• Jarvis-Patrick
• K Means

21
Agglomerative Hierarchical Clustering

• Basic Linkage method for N objects
• Start with N clusters and N x N distance matrix D
  dik, giving distance between objects i and k
• Search D for most similar pair of clusters U and
  V
• Merge U and V to form new cluster (UV)
• Update D by (a) removing the U and V rows and
  columns, and (b) adding new row and column giving
  distance to (UV)
• Repeat steps 2 4 a total of (N-1) times

22
Hierarchical Methods

• Determining distance between two clusters
• Single nearest neighbor
• Complete farthest neighbor
• Average average distance between all pair of
  two clusters

23
Hierarchical Clustering Methods
Method
Distance
single
d24
d15
complete
average
24
Distance Matrix
Hypothetical set of 5 objects
1 2 3 4 5
1 2 3 4 5
25
Single Linkage Clustering
Initial distance matrix
New distance matrix
1 2 3 4 5
(35) 1 2 4
(35) 1 2 4
1 2 3 4 5
26
Single Linkage Clustering
Initial distance matrix
New distance matrix
(135) 2 4
(35) 1 2 4
(135) 2 4
(35) 1 2 4
Compute distance to cluster 1,3,5
d(135)2 mind12, d(35)2 7 d(135)4 mind14,
d(35)4 6
27
Single Linkage Dendrogram
28
Complete Linkage Clustering
Initial distance matrix
New distance matrix
1 2 3 4 5
(35) 1 2 4
(35) 1 2 4
1 2 3 4 5
Dendrogram
? ? ? ? ? 1 2
4 3 5
29
Complete Linkage Clustering
Initial distance matrix
New distance matrix
(35) (24) 1
(35) 1 2 4
(35) (24) 1
(35) 1 2 4
Dendrogram
Compute distance to cluster 2,4
d(24)(35) maxd(35)2, d(35)4 10 d(24)1
maxd12, d14 9
? ? ? ? ? 1 2
4 3 5
30
Complete Linkage Dendrogram
31
Sample Data
Random set of 20 points
32
Cluster Comparison
Set cutoff to yield 7 clusters
33
Results of Single Linkage Clustering
7 clusters one large cluster and 6 singletons
34
Results of Complete Linkage Clustering
7 clusters 5 compact clusters and 2 singletons
35
Complete vs Single Linkage

• Complete Linkage
• tends to produce clusters with equal diameters
• can be severely distorted by outliers
• Single Linkage
• imposes no constraints on the shape of clusters
• has ability to detect elongated and irregular
  clusters
• but may miss compact clusters

36
Clustering Based on Chemical Similarity

• Similarity measure calculate a numerical
  distance between any two molecules
• Characteristics of molecular descriptors
• Encode all important structural features
• Distinguish closely related analogs
• Common among diverse classes of compounds
• Easy to calculate
• Recognize chemically equivalent groups as similar

37
Molecular Descriptors (Fingerprints)
Example of predefined set of 2D structural
features
A any atom except H Z N, O, S blue
any bond
38
2D Fingerprints
A any atom except H Z N, O, S blue
any bond
Fingerprint
Compound 1
1
6
39
Chemical Similarity Measure

• Tanimoto Coefficient

a 1 bits in compound A b 1 bits in
compound B c 1 bits in both A and B
40
Example of 2D Fingerprint
Fingerprint
1
6
Compound 2
41
Calculation of Tanimoto Coefficient
a 1 bits in compound A b 1 bits in
compound B c 1 bits in both A and B
a 11
b 10
c 9
dTan 9/12 0.75
42
Case Study Multidrug Resistance in Cancer

• MDR1 (ABCB1) encodes P-glycoprotein
• ATP-dependent efflux pump
• Broad drug specificity doxorubicin, etoposide,
  paclitaxel, vincristine, bisantrene
• Tissue occurrence intestine, liver, kidney,
  placenta, blood-brain barrier

Huang, Y. et. al. Pharmacogenomics J 2005, 5,
(2), 112-25
43
ABC Transporters Involved in Chemoresistance
Common Names
Gene Symbol
Substrates
Associated disease
Neutral and cationic Organic compounds, many
anticancer drugs
Cancer chemoresistance
Pgp, MDR1
ABCB1
MRP1
ABCC1
Glutathione conjugates, organic anions, drugs
Unknown
Glutathione conjugates, organic anions, drugs
Dubin-Johnson syndrome
MRP2
ABCC2
Unknown
Glutathione conjugates, anti- Folates, bile
acids, etoposide
MRP3
ABCC3
Unknown
Nucleoside analogs, methotrexate
MRP4
ABCC4
Nucleoside analogs, cyclic nucleotides, organic
anions
MRP5
ABCC5
Unknown
MRP-6
ABCC6
Anionic cyclic pentapeptide
Pseudoxanthoma elasticum
ABCG2
Unknown
Anthracyclines, mitoxantrone
MXR, BCRP
Modified from Gottesman et al. Nature Reviews,
2002 Dean et al. 2001 Genome Research
44
Datasets Used in Study

• Microarray data
• 70-mer oligonucleotide probes
• 732 human transporter genes
• NCI 60 cancer cell lines
• Compound activities from NCI database
• Growth inhibition (-log(GI50)) of 7,466 compounds
• Tested at least twice
• Less than 50 missing data

Source http//dtp.nci.nih.gov/docs/cancer/cancer
_data.html
45
Conceptual Framework
60 Cell Lines
732 Genes
7,466 Cmpds
7,466 Cmpds
60 Cell Lines
27,000 Features
Weinstein, et. al. Science, 1997 275 343-49.
46
Conceptual Framework
732 Genes
60 Cell Lines
7,466 Cmpds
60 Cell Lines
7,466 Cmpds
27,000 Features
732 Genes
SATT (Feature Gene Correlation)
27,000 Features
47
Gene-Compound Correlations

• Breast
• CNS
• Colon
• Leukemia
• Lung
• Melanoma
• Ovarian
• Prostate
• Renal

OVCAR-8
across NCI60 cell lines, r -0.87
48
Compound-MDR1 Correlations
3 IQR Outliers pos 3 neg 124
49
Clustering of NCI-60 Cell Lines
Cell lines clustered on expression patterns of
1,376 selected genes Used correlation as distance
metric (1 r) Average linkage clustering Cell
line panels
U. Scherf, et. al., Nature Genet., 2000, 24,
23644, Fig. 2.
50
MDR1-expressing Cell Lines

• NCI-60 cell lines over-expressing MDR1
• NCI/ADR-RES (unknown tissue origin)
• HCT-15 (colon)
• UO-31 (renal)
• Comparison cell line OVCAR-8 (ovarian)
• Most similar to NCI/ADR-RES comparing cDNA
  expression patterns
• Does not over-expressing MDR1

51
Structure-Based Cluster Analysis

• Cluster 7,466 compounds on structural similarity
  using Leadscope feature set with Tanimoto
  coefficient
• Complete linkage clustering gave 1,859 clusters,
  Tanimoto cutoff 0.7
• 468 singletons
• For each cluster, computed mean compound-mdr1
  correlation
• 15 classes of 5 compounds with z-score gt 3.0
• 34 classes of 5 compounds with z-score lt -3.0

52
Cluster Analysis
Large classes with negative correlations
ABCB1 (mdr1)
? GI50
Count Mean Z-score Mean Z-score
44 -0.42 -16.5 -2.22 -22.2 17 -0.36 -8.5 -1
.14 -7.2 17 -0.35 -8.4 -0.99 -6.3
? GI50 difference between LNSNCI/ADR-RES and
OVAOVCAR-8
53
Cluster Analysis
Large classes with positive correlations
ABCB1 (mdr1)
? GI50
Count Mean Z-score Mean Z-score
41 0.12 6.3 0.0 1.8 15 0.21 6.1 0.24 2.8
13 0.11 3.5 -0.12 0.2
? GI50 difference between LNSNCI/ADR-RES and
OVAOVCAR-8
54
Cluster Analysis
Two subclasses of ellipticine analogs
ABCB1 (mdr1)
? GI50
Count Mean Z-score Mean Z-score
17 -0.36 -8.5 -1.14 -7.2 13 0.11 3.5 -0.12
0.2
? GI50 difference between LNSNCI/ADR-RES and
OVAOVCAR-8
55
R-Group Analysis of Ellipticines
R1 R2 Freq
MDR1 Correl.
DGI50
mean zscore
mean zscore
? GI50 difference between LNSNCI/ADR-RES and
OVAOVCAR-8
56
Compounds Selected for Dosing
57
Experimental Validation

• Cytotoxicity of selected ellipticine analogs in
  NCI/ADR-RES was studied by
• siRNA downregulation
• inhibition of MDR1 by Cyclosporin A
• Inhibition of MDR1-mediated efflux of fluorescent
  markers

58
Ellipticine Dosing Experiments
growth inhibition
drug concentration (µM)
drug concentration (µM)

• Drug only

• 5 mM CsA

• ABCB1 siRNA

• mock siRNA

Drug only
59
Ellipticine Dosing Experiments
growth inhibition
drug concentration (µM)
drug concentration (µM)

• Drug only

• 5 mM CsA

• ABCB1 siRNA

• mock siRNA

5 µM CsA
Drug only
ABCB1 siRNA
Mock siRNA
5 µM CsA
Drug only
ABCB1 siRNA
Mock siRNA
Drug only
Drug only
60
Ellipticine Dosing Experiments
growth inhibition
drug concentration (µM)
drug concentration (µM)

• Drug only

• 5 mM CsA

• ABCB1 siRNA

• mock siRNA

Drug only
61
Summary of Dosing Experiments

• Ellipticinium analogs (NSC 155694 and 359449) are
  MDR1 substrates
• siRNA mediated down-regulation of MDR1 expression
  decreased GI50 values in ADR-RES cells by at
  least 2 fold.
• Addition of cyclosporin A, an inhitibor of MDR1,
  decreased GI50 values by at least 20 fold.
• Neutral ellipticines (NSC 86717, 69187, 338258)
  are not MDR1 substrates
• No marked effect on GI50 values in ADR-RES cells
  was observed.

62
Analysis of Compound-Gene Correlations

• Summary
• General method for discovering associations
  between compound classes and molecular targets
• Rapidly identified compound classes with relevant
  genes using NCI dataset
• In-silico technique aids experimental design
• Can provide insights into mode of resistance and
  mechanism of action

63
Analysis of Compound-Gene Correlations

• References
• Huang, Y. et. al. Correlating gene expression
  with chemical scaffolds of cytotoxic agents
  ellipticines as substrates and inhibitors of
  MDR1. Pharmacogenomics J 2005, 5, (2), 112-25
• Huang, Y. et. al. Membrane transporters and
  channels role of the transportome in cancer
  chemosensitivity and chemoresistance. Cancer Res
  2004, 64, (12), 4294-301
• Blower, P. E. et. al. Pharmacogenomic analysis
  correlating molecular substructure classes with
  microarray gene expression data. Pharmacogenomics
  J 2002, 2, (4), 259-71
• Staunton, J. E. et. al. Chemosensitivity
  prediction by transcriptional profiling. PNAS
  2001, 98, (19), 10787-10792
• Scherf, U. et. al. A gene expression database
  for the molecular pharmacology of cancer. Nat
  Genet 2000, 24, (3), 236-44
• Weinstein, J. N. et. al. An information-intensiv
  e approach to the molecular pharmacology of
  cancer. Science 1997, 275, (5298), 343-9

64
Genomic Profiling with Yeast Library

• Heterozygous yeast deletion library
• each strain has single copy of gene
• 6,000 deletion strains
• covers 96.5 of yeast genome
• Heterozygous strain sensitive to any drug acting
  on gene product
• Haploinsufficiency profiling

Giaever, G. et. al. Proc Natl Acad Sci U S A
2004, 101, (3), 793-8
65
Drug Induced Haploinsufficiency

• Drug treatment of heterozygous vs wild-type yeast
  strains
• ALG7 (asparagine-linked glycosyl-transferase)
  known target of tunicamycin

Tunicamycin 0 mg/ml
0.5 mg/ml 2.0 mg/ml
Giaever, G. et. al. Nat Genet 1999, 21, (3),
278-83.
66
Heterozygote Deletion Construct
DNA extracted and all tags amplified in a single
PCR reaction using common primers
kanamycin gene allows selection of yeast
tranformants
unique 20-base hybridization tag strain
specific bar-code
67
Drug Sensitivity Library Profiling
no drug with drug
68
Haploinsufficiency Profiling

• Heterzygous deletion library treated with 250 mM
  methotrexate
• Fitness defect score (FD) proportional to
  likelihood of seeing experimental value
• DFR1 is known target of methotrexate
• FOL2 required for biosynthesis of folic acid in
  yeast upstream from DHF
• no human homolog of FOL1
• human homolog of YBT1 encodes methotrexate
  transporter, up-regulation causes methotrexate
  resistance
• YOR072W encodes suspected methotrexate transporter

69
Haploinsufficiency Profiling
Atorvastatin (Lipitor) concentration
0 mM 62.5 mM 125 mM 250 mM
HMG1, HMG2 yeast isozymes of drug target
3-hydroxy-3-methylglutaryl-CoA (HMG-CoA) reductase
70
Haploinsufficiency Profiling

• Summary
• Systematic method for identifying gene products
  that interact with drugs
• Simultaneous screening of all interactions
  between compounds and gene products in yeast
• Chemical genetic probes for further functional
  studies in other organisms
• References
• Giaever, G. et. al. Chemogenomic profiling
  identifying the functional interactions of small
  molecules in yeast. Proc Natl Acad Sci U S A
  2004, 101, (3), 793-8.
• Giaever, G. et. al. Functional profiling of the
  Saccharomyces cerevisiae genome. Nature 2002,
  418, (6896), 387-91.
• Giaever, G. et. al. Genomic profiling of drug
  sensitivities via induced haploinsufficiency. Nat
  Genet 1999, 21, (3), 278-83.

71
Connectivity Map

• Database of drug-gene signatures for linking
  drugs, genes and diseases
• Profiled bioactive small molecules in four NCI60
  cell lines
• mRNA expression levels were measured after drug
  treatment giving a reference signature
• rank ordered list of genes ordered by
  differential expression relative to control
• database can be searched by comparing a query
  signature of up- and down-regulated genes with
  reference signatures

Lamb, J. et. al. Science 2006, 313, 1929-35
72
Connectivity Map

• Search database by comparing a query signature
  with reference signatures

Query signature list of up- and down-regulated
genes
Reference signatures - ranked gene lists for
compounds
Output lists of high and low scoring compounds
73
Summary of Chemogenomics

• Consequences of differential gene/protein
  expression on cellular response to compound
  treatment
• Chemoresistance studies using the NCI-60
• System-wide effect of a compound on an intact
  biological system
• Genomic profiling with yeast deletion library
• Connectivity map