Transcript Genetics and Genomics -

Omics, Biomarkers, Personalized Medicine:
A New Era, or More of the Same?
Klaus Lindpaintner
Roche Genetics/Roche Center for Medical Genomics
Differential drug efficacy
Same symptoms
Same findings
Same disease (?)
Same Drug….
Genetic Differences
Different Effects
?
Possible Reasons:
Non-Compliance…
Drug-drug interactions…
Chance…
Or….
2
Pharmacotherapy: State-of-the-Art
Group
Incomplete/absent efficacy
AT2-antag
SSRI
ACE -I
Beta blockers
Tricycl. AD
HMGCoAR-I
Beta-2-agonists
10-25%
10-25%
10-30%
15-25%
20-50%
30-70%
40-70%
• Inter-individual differences in drug efficacy
• Significant incidence of serious adverse effects
among elderly hospitalized patients (US)


Serious
Lethal
6.7%
0.3%
2 M cases
100,000 cases
JAMA 98;279:1200
3
Pharmacogenetics and Personalized
Medicine
•
AnKnowledge
altogether
new concept?
of inter-individual differences wrt metabolism as old as
civilization: 6th century B.C. Pythagoras observes
that ingestion of fava beans is harmful to
some individuals yet innocuous to others
•
Finding the optimal treatment for every patient is as
old as medicine: differential diagnosis
•
Tailoring treatments to drug-specific test results is nothing new.
Example: antibiotics
•
•
•
Gram-positive bacteria: e.g. penicillin derivatives
Gram-negative bacteria: e.g. aminoglycosides
M. tuberculosis: isoniazid/rifampin/pyrazinamide
4
Bridging a Historical Divide
clinical diagnosis
tissue / organ physiology-pathology
molecular diagnosis
cell-biology
protein
RNA
DNA
protein
RNA
DNA
cell-biology
protein
RNA
DNA
protein
RNA
DNA
drugs
6
Pharmacogenetics, Pharmacogenomics
•
•
•
Glossary of Terms
Pharmacogenetics:
•
•
•
•
•
a concept to provide more patient/disease-specific health care*
based on the effects of inherited (or acquired) genetic variants
assessed primarily by sequence determination (or single gene expression)
one drug – many genomes (patients)
focus: patient variability
Pharmacogenomics (1):
•
•
•
•
•
a concept to provide more patient/disease-specific health care
based on the effects acquired (or inherited) genetic variants
assessed primarily by expression profiles (many mRNAs)
one drug – many genomes (patients)
focus: patient variability
Pharmacogenomics (2):
•
•
•
a tool for compound selection/drug discovery
many drugs – one genome (inbred animal/chip)
focus: compound variability
*as conceptualized by Motulsky (1957), Vogel (1959), Kalow (1962) and
endorsed in the 2003 Nuffield Council’s Report on Pharmacogenetics
7
2 Major Classes of Pharmacogenetics –
Both Resulting in Patient Stratification
•
Strictly affecting drug response – not predictive of disease risk:
“Differentiating people” (“classical” pgx: Archibald Garrod)
•
•
•
•
Related to molecular subclass of clinical diagnosis:
“Differentiating disease” (“molecular differential diagnosis”)
•
•
•
Pharmacokinetics (not only M, but also AADE)
Pharmacodynamics
Has not had much impact
Inherently linked to disease mechanism/prognosis
Likely increasing impact in indications where we begin to treat
causally – oncology, inflammatory disease
Both are conceptually rather different (and arguably the second
should not be included) but have practically the same
consequence:
Patient stratification according to novel, DNA-based parameters
8
Omeprazole response rate and CYP2C19
A/A – SLOW
100
90
A/B
B/B – FAST
response frequency (%)
80
70
60
50
40
30
20
10
0
gastric ulcer
duodenal ulcer
9
Drug metabolism
Inherited differences affect drug effects
Pharmacogenetics = molecular DD
Case Study: Herceptin®
Bimodal response:
2/3 of patients: addition of Herceptin® to chemoRx
 no benefit
1/3 of patients: addition of Herceptin® to chemoRx
 50% survival time increased by factor 1.5 (20 29 weeks)
High HER2
Low HER2
11
Xeloda® (capcitabine)
Patient stratification based on enzyme patterns
Xeloda susceptibility vs tumor TP/DPD
TP/DPD
in 24 xenografts
TS
(dThdPase/DPD)
Xeloda
P = 0.0015
100
TP
5-DFUR
DPD
5-FU
inactive
metabolites
S: susceptible
R: refractory
10
1
0.1
S R
12
Biomarkers
What’s new – and why now?
• Availability of powerful, highly parallel new screening
methods (omics) makes looking for new biomarkers a
reasonable proposition.
• Paradigm shift(?): maturation of these basic cell and
molecular biology tools makes them newly applicable
to later-stage R&D
•
•
Opportunities: personalized medicine
Challenges: technical, scientific (clinical-epidemiological)
economical, ethical
• CAVEAT 1:
Association ≠ Causality
•
Good news and bad news
13
Caveat 2 “Responders” & “NonResponders”
Reality Check
FDA benchmark: 35% improvement/response
80
70
response (%)
60
22%
50
43%
40
31%
30
20
10
0
individual patients
14
Single Gene Disease
Environment SNPs in other genes
Mutation
intermediate
phenotype
health
outcome
Heritability: h2 ≈ 1
Deterministic … possible stigma
15
M onogenic Diseases
CCD
Common
Complex
Diseases
16
Single Gene Disease
Environment SNPs in other genes
Mutation
intermediate
intermediate
phenotype
phenotype
health
health
outcome
outcome
Common Complex Disease
Environment SNPs in other genes
SNP
intermediate
phenotype
health
outcome
Probabilistic, not deterministic - no reason for stigma.
17
Complex Common Disease:
Nature and Nurture
environment
genes
Colon,
Hemobreast ApoE4
philia
Cancer AD
CF
P53, BRCA
HD
nutrition
Stroke
MI
Diabetes
Asthma
Lung cancer
tobacco --- asbestos
MVA
GSW
P450
18
Heritability estimates in CCD
Disorder or phenotype
Preeclampsia
NIDDM
Hypertension
Osteoarthritis
Stroke
Asthma
Obesity
Depression
Other dementia
Blood pressure
BMI
Rheumatoid arthritis
Death from heart disease
Coronary heart disease
IGT
SLE
Alzheimer’s (sporadic)
Protracted/recurrent otitis media
Heritability h2
0.2-0.35
0.26-0.50
0.28-0.73
0.3-0.46
0.32
0.36-0.47
0.4-0.7
0.41-0.66
0.43
0.5
0.5-0.7
0.53-0.65
0.55
0.56
0.61
0.66
0.72
0.72
19
Heritability estimates in cancer
Malignancy
Thyroid
Endocrine glands
Breast
Testis
Cervix invasive
Melanoma
Nervous system: age <15 years
Colon
Cervix in situ
Rectum
Nervous system
Non-Hodgkin lymphoma
Leukemia: age <15 years
Lung
Kidney
Urinary bladder
Leukemia
Stomach
\
Heritability h2
0.53 (0.52–0.53)
0.28 (0.27–0.28)
0.25 (0.23–0.27)
0.25 (0.15–0.37)
0.22 (0.14–0.27)
0.21 (0.12–0.23)
0.13 (0.06–0.20)
0.13 (0.12–0.18)
0.13 (0.06–0.15)
0.12 (0.08–0.13)
0.12 (0.10–0.18)
0.10 (0.08–0.10)
0.09 (0.09–0.16)
0.08 (0.05–0.09)
0.08 (0.07–0.09)
0.07 (0.02–0.11)
0.01 (0.00–0.01)
0.01 (0.01–0.06)
Czene et al, Int J Cancer 99:260; 2002
20
Medical Progress: Evolution or
Revolution?
Historic Drivers of Medical Progress
Clinical expertise
…Genetics
Differential diagnosis
Risk assessment - prevention
Classical epidemiology
More differentiated, molecular understanding of pathology and drug action
Clinical Disease Definition
Clinical Diagnosis
Molecular Disease Definition
Molecular Diagnosis
in-vitro Diagnostics
21
Consumption
Phlebotomy
Tuberculosis
Tuberculosis
Cancer
Cancer
Antibacterials
Cytostatics
Heart
Heart
Failure
Failure
ACE Inhibitors
Breast Ca
Colon Ca
Lung Ca
HER-2-negative (2/3)
HER-2-positive (1/3)
Mean survival 7 yrs
Mean survival 3 yrs
Cytostatics
Cytostatics + humMAb
22
Pharmacogenetics vs. other Markers
A useful distinction?
Normal
Pharmacogenetics
Pharmacogenomics
DNA
DNA*
DNA *
mRNA
mRNA*
mRNA*
primary
protein
primary
protein*
processed
protein,
small
molecule
response to
medicine
Pharmacogenomics
PharmacoPharmacoproteomics metabonomics
DNA
DNA
mRNA*
mRNA
mRNA
primary
protein*
primary
protein*
primary
protein*
primary
protein
processed
protein,
small
molecule*
processed
protein,
small
molecule*
processed
protein,
small
molecule*
processed
protein,
small
molecule*
processed
protein,
small
molecule*
altered
response to
medicine*
altered
response to
medicine*
altered
response to
medicine*
altered
response to
medicine*
altered
response to
medicine*
DNA
* alteration germ-line in origin – heritable
* alteration somatic – acquired (environment, life-style)
23
Pharmacogenetics and beyond:
Biomarkers
•
Key concept:
More targeted medicines (“personalized medicine”)
•
•
•
•
Based on a better understanding of inter-individual differences among
patients
•
•
•
More effective
Safer
More cost-effective (?)
Inherited (the “classical” pharmacogenetics)
Acquired (“flavors” of disease, underlying molecular heterogeneity of any
one clinical diagnosis: molecular differential diagnosis)
Paradigm: carry out specific test that point to one or another medicine
as optimal for the patient before prescribing it.
What does not matter: Nature of test (DNA, RNA, protein, other)
What does matter:
Information content
24
Biomarker tests in medical practice
Two sets of considerations
• Test performance
•
•
Analytical performance – QC and accreditation of labs
Clinical performance
•
•
•
Clinical validity – retrospective/observation studies
Clinical utility – prospective intervention trials
Note: Prior probability: critical for test performance, esp.
screens (sensitivity/specificity, PPV/NPV)
• Nature of illness
•
Serious (life-threatening) illness
Default:
”don’t withhold in error”;
If in doubt:
“treat”
•
Less serious illness
Default:
“don’t treat in error”;
If in doubt:
“don’t treat”
25
EGFR Mutants
Much ado about…?
26
EGRF-R variants
Colocation with ATP-binding domain
27
Regulators are Taking Note
28
Interpretation? Consequences?
• NEJM
•
•
•
8/9 responders
7/7 non-responders
2 of 25 untreated
•
Pre-testing will increase response rate to 100% among those
who test +
Pre-testing will result in denial of treatment to 11% of who
would responders
•
• Pao et al, MSKCC (PNAS)
+ for mutation
– for mutation
+ for mutation
•
•
•
•
7/10 responders
8/8 non-responders
4/81 NSCLC smokers
7/15 non-smoker, adeno-Ca
+
–
+
+
for mutation
for mutation
for mutation
for mutation
•
Pre-testing will result in denial of treatment to 30% of who
would be responders
29
EGF-R variants and Drug Response
• Gefitinib (IRESSA) Response in Caucasians
Prevalence of variants in Boston patients
10%
2/25
(NEJM)
• Gefitinib (IRESSA) Response in Japanese
Prevalence of variants in Japanese patients
28%
26%
(Science)
• Erlotinib (TARCEVA) Monotherapy in NSCLS
EGFR Mutratoin prevalence
Response Rate
12%
42%
30
Analytical Performance: Metrology
Aything but straight-forward
• Precision
•
•
Repeatability
under same conditions, precision in a series of measurement
in the same run; and
Reproducibility
under different conditions, which are usually specified, e.g.
day-to-day or lab-to lab
• Trueness
•
•
the closeness of agreement of an average value from a large
series of measurements with a "true value" or an accepted
reference value.
Numerical value: bias
• Accuracy –
•
•
referring to a single measurement and comprising both
random and systematic influences.
Numerical value: total error of measurement.
31
Biomarker tests in medical practice
Two sets of considerations
• Test performance
•
•
Analytical performance – QC and accreditation of labs
Clinical performance
•
•
•
Clinical validity – retrospective/observation studies
Clinical utility – prospective intervention trials
Note: Prior probability: critical for test performance, esp.
screens (sensitivity/specificity, PPV/NPV)
• Nature of illness
•
Serious (life-threatening) illness
Default:
”don’t withhold in error”;
If in doubt:
“treat”
•
Less serious illness
Default:
“don’t treat in error”;
If in doubt:
“don’t treat”
32
Analytical performance
The dirty (not so) little secret
• Multiple complex variables:
•
•
•
•
•
Tissue heterogeneity
Limited sample quantity and quality (FFPE)
LCDM/macro-dissection commonly necessary
PCR-pre-amplification
4 exons x 2 amplification runs each
33
Analytical performance: EGFR sequencing
Sometimes, far from it…
unambiguous wt
wt vs. mut
wt vs. mut vs. artifact
wt?
wt vs. mut?
unambiguous known mut
known mut vs. new mut vs. both?
mut?
wt vs. mut vs. artifact?
known mut vs. new mut vs. both vs. indet?
unambiguous new mut
new mut?
wt?
known mut?
new mut?
unambiguous unknown
34
EGFR mutation analysis analytical performance
The dirty (not so) little secret
• Multiple complex variables:
•
•
•
•
•
Tissue heterogeneity
Limited sample quantity and quality (FFPE)
LCDM/macro-dissection
PCR-pre-amplification
4 exons x 2 amplification runs each
•
•
•
Algorithm 1:
Algorithm 2:
Algorithm 3:
• How to deal with “drop-outs”?
• How to deal with non-replicated mutations – artifact or
quantitative manifestation of relative abundance of
mutation?
• None of current publications disclose this difficulty
• Own experience – using different “calling” algorithms:
6.1% (13 mut / 200 wt / 94 indeterminate)
7.5% (15 mut / 186 wt / 106 indeterminate)
9.9% (23 mut / 210 wt / 74 indeterminate)
35
EGFR-Mutations, Erlotinib, and Survival
The picture is more complex…
36
Biomarker tests in medical practice
Two sets of considerations
• Test performance
•
•
Analytical accuracy – QC and accreditation of labs
Clinical performance
•
•
•
Clin validity – retrospective/observation studies
Clinical utility – prospective intervention trials
Note: Prior probability: critical for test performance, esp.
screens (sensitivity/specificity, PPV/NPV)
• Nature of illness
•
Serious (life-threatening) illness
Default:
”don’t withhold in error”;
If in doubt:
“treat”
•
Less serious illness
Default:
“don’t treat in error”;
If in doubt:
“don’t treat”
37
Optimizing Sensitivity vs. Specificity
Target Product Profile Definition is Essential
sensitivity
100%
0%
0%
1-specificity
100%
Note: Sliding the ROC-cutoff value may be more difficult with (categorical) genotype
data than with other (quantitative) biomarker data
38
Biomarker performance
Up and down the ROC curve
Serious illness: don’t withhold inappropriately
+ test
- test
+ response
true positive
 false negative
- response
 false positive
true negative
Efficacy marker: High sensitivity
+ test
- test
+ adverse event
true positive
 false negative
- adverse event
 false positive
true negative
Safety marker: High specificity
Less serious illness: don’t prescribe inappropriately
+ test
- test
+ response
true positive
 false negative
- response
 false positive
true negative
Efficacy marker: High specificty
+ test
- test
+ adverse event
true positive
 false negative
- adverse event
 false positive
true negative
Safety marker: High sensitivity
39
Case-in-point: Herceptin/HerCepTest
The search for new biomarkers – and its implications
Status quo,
 66% success rate
 no potential responder denied Rx
+ response
- response
+ new BM
test
19
true +
5
false +
24
79% response among
treated Her2+/BM+
- new BM
test
1
false -
5
true -
7
16% response among
Her2-/BM-
Sensitivity:
Specificity:
true+/(true+ + false-)
19/(19+1)=0.95
true-/(true- + false+)
5/5+5=.5
95% sensitive
50% (94%*) specific
+ response
- response
+ new
BM test
16
true +
2
false +
18
88% response among
treated Her2+/BM+
- new
BM test
4
false -
8
true -
12
33% response among
Her2-/BM-
Sensitivity:
Specificity:
true+/(true+ + false-)
16/(16+4)=0.8
true-/(true- + false+)
8/1+9=0.9
80% sensitive
80% (98%*) specific
30
*Specificity of combined Her2 and new BM tests
+ response
- response
+Her2
test
20
true +
10
false +
30
66% response among
treated Her2+
- Her2
test
0
false -
70
true -
70
presumed 0% response
among Her2(NB: anecdotal data)
Sensitivity:
Specificity:
true+/(true+ + false-)
20/(20+0)=1
true-/(true- + false+)
70/70+10=0.875
100% sensitive
88% specific
100
Add-on-BM scenario 1
 78% success rate
 5% of would-be responders denied Rx
Add-on-BM scenario 2
 88% success rate
 20% of would-be responders denied Rx
30
40
Not all that glitters is gold: TPMT
Thiopurine-treated patients with adverse drug reactions
Total
n
25
17
7
15
41
8
Patients without
deficient TPMT-allele
n
%
20
16
4
14
29
6
80
94
57
93
70
75
Patients with one or two
deficient TPMT-alleles
n
%
5
1
3
1
12
2
20
6
43
7
30
25
Reference
Black et al. 1998
Naughton et al. 1999
Ishioka et al. 1999
Dubinsky et al. 2000
Colombel et al. 2000
Ando et al. 2001
sensitivity
positive test predicts, but negative tests by no means excludes SAE
299 negative tests for every one positive test
41
Economic considerations
How far is segmentation of markets feasible?
“Exhaustive pharmacogenetic research efforts have
narrowed your niche market down
to Harry Finkelstein of Newburg Heights here.”
42
Emergence of sub-critically small
segments
A self-limited proposition
• Retrospectively:
Given biomedical variance, biomarker-defined
segments are unlikely to be recognizable unless they
represent a significant share of the overall patient
population.
• Prospectively:
Small segments known to exist will either not be
addressed for lack of business case, or under Orphan
Drug Guidelines
43
The Tightening Reimbursement Climate
Biomarker strategies may be essential
Strategy
No test
Chemo-Rx alone
Positive HerCep Test
Chemo-Rx and Herceptin
No test
Chemo-Rx and Herceptin
Life-months
Incr.
QALYs
Incr. Cost
UK £
Incr
Cost/QUALY
UK £
28.02
1.28
26,919
21,030
29.30
1.36
33,376
24,541
29.41
1.37
49,211
35,920
Elkin et al; J Clin Oncol 2004; 22:854 ff
($/£ conv. rate 1/1/2003, not PPP-adjusted)
NB: National Institute for Clinical Excellence’s (NICE) threshold
for approving reimbursement through NHS believed to be
~UK £ 30,000 per QUALY (quality-adjusted life year)
44
Biomarkers – likely outcome:
•
The concept applies potentially to most compounds
•
It will in fact, however, become reality only for some/few compounds…
but we will have to look at all to find the few!
•
(We will likely see more examples of “pathology-related” biomarkerbased stratification (Herceptin-paradigm) that advance efficacy; and
most likely in oncology and inflammatory/autoimmune disease)
•
Multifactorial algorithms likely to emerge, rather than simple, onevariable models – but highly complex algorithms unlikely.
•
Essential: Define Target-Product-Profile
•
Key: Modesty, Realism, robust Optimism:
we will not have perfect medicines
BUT
we will have increasingly better medicines
45
No 1-on-1 custom tailoring,
but towards a much better fit …
377/8 38
39
39½ 39¾ 40
Remember: All medical decisions/knowledge are based on
group-derived (aggregate) data analysis.
“Data” on individuals (Harry Finkelstein) are anecdotal and
(largely) medically/clinically meaningless
46
Without information,
the doctor cannot act.
With information,
he cannot but act.
HL Mencken’s Law
Every complex problem
has a simple solution.
And it is always wrong.
48