ECNIS Web-based course in
Molecular Epidemiology in Cancer
Biomarker Validation
Slide B1.1
Commenti:
intraindividual variation si riduce con biomarkers per
esempio con la misurazione di POPs! (PCB ecc) perchè
sono long-term e stabili (lipofilici)
- adducts in cord blood? (Perera)
http://www.aacrmeetingabstracts.or
g/cgi/content/abstract/2005/1/512-b
Characteristics of ideal biomarkers
1.
2.
3.
4.
5.
6.
7.
Sensitive and specific
Relation with exposure
Standardized and validated
Relatively easy to perform
Non-invasive
High throughput
Inexpensive
Slide B1.3
Levels of validation:
Intra-individual variation
Inter-individual variation
and confounding
Intra-laboratory variation
Inter-laboratory variation
Validity (vs a standard) and predictive value
Time relationships
Dose-response
Ability to predict outcome
Slide B1.4
QUALITY CONTROL OF A BIOMARKER:
MEASUREMENT ERROR
Measurement error is classified as:
preanalytical (biological and sampling error)
Or
analytical (laboratory) error
Laboratory error focuses on method, instrument, reagent
or matrix effects.
Preanalytical Error
- individual genetic, environmental, behavioural and health
status-related variability (including smoking status, weight and
weight loss, physical exercise)
Example of genetic source of variation: FOLATE AND MTHFR
Health status-related: retinol or ascorbate and trauma, several
biomarkers and inflammation
- sampling error: within subject variation due to hourly, daily,
weekly, monthly … changes
To reduce Analytical Errors – Quality Control measures
Example of Quality Control Program: national cholesterol
Education Program (US)
Goals:
1. Attain analytical accuracy and precision (<3% cv)
2. Identify individual determinants of cholesterol variation
(lifestyle
factors)
3. Identify clinical determinants of variation (metabolic states,
illness)
4. Other sampling sources (fasting status, posture, serum vs.
plasma)
The NCEP guidelines have proven adequate to ensure 90%
correct classification
Overall measure of error is the
COEFFICIENT OF VARIATION = SD/MEAN x 100%
(SD in repeated measurements)
CV is ideally calculated for samples at the bottom, middle and top of
the reference concenttration range determined in healthy subjects
Log transformation is even better (Rappaport book)
OTHER EXAMPLES OF QUALITY CONTROL:
- Gunter et al (1996), international round-robin for folate involving
20 labs: CV of 27% for serum folate, and 36% for whole blood folate,
with substantial intermethod variation
- Pfeiffer et al (1999), interlaboratory comparison of homocysteine in
plasma samples (14 labs): CV=9% among labs, and 6% within labs
TWO MAJOR APPROACHES TO REDUCING
MEASUREMENT ERROR ARE:
1. TO BLIND THE ANALYST TO THE CASE-CONTROL
STATUS OF SPECIMENS
2. TO ELIMINATE SYSTEMATIC DIFFERENCES IN THE
WAY CASE AND CONTROL SPECIMENS ARE
HANDLED
Validation and relevance: some examples
Inter-centre variation (and potential confounding) for an
intermediate marker (plasma DNA amount in EPIC)
Slide B1.11
Genetic alterations in plasma DNA
* Useful when tumours not available
* Good concordance between tumour and plasma mutations
* When does tumour DNA appear in the blood?
* Can plasma DNA be used as a biomarker for genotoxic exposure?
Slide B1.12
DNA concentration sorted by EPIC number (origin)
6702
6478
6841
5960
5974
7413
5297
3687
5521
4821
4555
3505
3239
2875
2637
2357
5171
3939
7313
1600
1400
1200
1000
800
600
400
200
0
3967
DNA concentration (ng/ml)
GENAIR DNA concentration
MOC number
Utrecht
Slide B1.13
Univariate and multivariate analysis: plasma DNA amount (logarithm
transformation, dependent variable), by center, age, gender and time between
blood drawing and diagnosis (for cases only).
Univariate analysis:
Variable
F-value DF
p-value
Controls only (N=778)
Center
Age
Gender
11.23
5.21
0.52
<0.0001
0.023
0.47
22
1 (a)
1
Cases and controls (N=1185):
Center
Age
All deaths
and tumours
F-value DF
p-value
16.6
1.56
23
1 (a)
<0.0001
0.21
2.3
6
0.03
Slide B1.14
Biomarkers vs external/other
measurements
Cotinine and ETS (environmental tobacco smoke) from
questionnaires
Cotinine measurements. Means, SD and distribution by detectable
(greater than 0.05 ng/ml) and undetectable levels, by ETS status. Only
subjects with ETS information (N=374). Values of cotinine greater than
10 ng/ml (N=11) excluded (Vineis et l, BMJ 2004).
ETS status
Mean cotinine (N, SD)
Yes
0.55 (189, 0.96)
No
0.17 (174, 0.49)
p-value<0.0001 (Wilcoxon Rank-sum test)
ETS status
Yes
No
Cotinine
Detectable
Undetectable
89
100
37
137
OR=3.30 (95% CI 2.07, 5.23) p<0.0001
Slide B1.16
Bulky DNA adducts and dose-response relationship (Peluso
et al, AJE 2001
Slide B1.17
Comments
- The fact that adducts and other markers are related to exposure does
not imply that they are a better measure
- Biomarkers can increase biological plausibility of associations
- They can be useful for example if it is possible to show that intraindividual variability is lower with the marker than with external
exposure measurements
- They can address issues such as saturation of enzymes at high
levels of exposure (dose-response, risk assessment)
- DNA adducts are an integrated marker (over several sources of
exposure) that expresses also individual susceptibility (eg for DNA
repair), and can be predictive of cancer onset
Slide B1.18
Genotypes vs exposures
Genotyping
Comparison of four genotyping methods at the Cambridge laboratory.
The standard is represented by a panel evaluation of all results
(courtesy of A Dunning).
Method
Sensitivity
%
Specificity
%
ASO
836/864
97
753/836
90
Taqman
826/864
96
812/826
98
RsaI digest
125/173
72
103/125
82
Invader
62/92
67
45/62
73
Slide B1.20
Problems with studies on gene-environment interactions:
- low study power
- frequent false positives due to multiple testing
- functional data often missing
- early studies not confirmed by subsequent larger or better
conducted studies
- publication bias
Slide B1.21
Effects of random classification error on relative risk estimates
R=correlation coefficients between the measurement of
exposure/genotype by different assessors and a reference
standard, and the resulting observed relative risks (modified
from Hankinson et al, 1994, ref. 3).
Assessor
R
True relative risks (RRt)
1.5
2.0
2.5
Observed relative risks
1
2
3
4
0.10
0.60
0.80
0.90
1.1
1.3
1.4
1.4
1.1
1.5
1.8
1.9
1.1
1.7
2.1
2.3
Observed RR=exp (ln RRt*R)
Slide B1.22
According to estimates, the common genotyping method
Taqman has 96% sensitivity and 98% specificity, thus
allowing little error in classification.
On the contrary, sensitivity in environmental exposure
assessment is quite often lower than 70% and specificity
even lower.
Slide B1.23
Genotype is stable, measured accurately (sens,
spec=90-100%), frequency of alleles is high
Environmental exposures are changing (life-course
events), often measured inaccurately, frequency may be
too low
Slide B1.24
In addition, genetic polymorphisms are investigated with
high-throughput technologies that allow researchers to
investigate hundreds of thousands of SNP at a time:
with the usual p-values this originates a large number of
false positives (see Bayesian strategy proposed by
Colhoun et al, Lancet 2003 361: 865-872)
In environmental research false negatives are an
important problem
Slide B1.25
Limitations of current biomarker studies






Some markers are not very reliable (e.g. interlaboratory
variation for adducts)
Biological meaning not always clear (e.g. mutations in
plasma DNA)
Long gap between marker development and its validation
Unknown or unsatisfactory time relationships between
exposure, marker measurement, disease
Usually only one spot biosample available (little known on
intra-individual variation)
Little known on potential confounders
Slide B1.26
Further issues in validation of biomarkers and examples
(from R Godschalk)
1. Knowledge of pharmacokinetics and relevance of
measurements in time
2. Insights in reasons for inter- and intra-individual variation
Measured variation = Inter + Intra + variation in assay
3. Surrogate vs. Target tissue
WBC vs. lung tissue in smokers
4. Comparison with other Biomarkers (“gold standard”)
for example 1-OH-Pyrene in urine
Slide B1.27
Exposure to cig smoke  lung DNA damage
and inflammation
Cig. smoke
(Particles)
ROS
Neutrophils
Cig. smoke (PAH)
Activation
CYP450s
MPO
Detoxication
GSTs
NATs
Activation
PAH-DNA adducts
H2O2
DNA repair
NER
BER
MPO +Cl
HOCl
DNA
damage/mutagenesis
Cancer
Slide B1.28
Characteristics of ideal biomarkers
Immunocytochemical staining of PAH-DNA adducts in Mouth
Brushes
1.
2.
3.
4.
5.
6.
7.
Sensitive and specific
Relation with exposure
Standardized and validated
Relatively easy to perform
Non-invasive
High throughput
Inexpensive
+
+/+
+
Slide B1.29
The use of Induced Sputum (IS) in smoking-related DNA adducts
analyses
Objectives
• To study the applicability of Induced Sputum (IS) as source of
lung derived cells
• To establish correlation between DNA adduct levels in IS derived
cells and smoking intensity
• To compare DNA adduct levels in IS with PBL
Department of Health Risk Analysis and Toxicology,
Maastricht University, Maastricht, The Netherlands
Slide B1.30
Intra-individual DNA adduct analysis
Intra-individual variation in IS is higher than in PBL
When quantitating the adduct levels in Induced Sputum of certain individuals
considerable variation could be observed. We could not find a reason for that
since smoking habits and dietary conditions were kept similar over time as much
as possible. There is also some variation in PBL but to a lower extent.
Slide B1.31
Characteristics of ideal biomarkers
Postlabeling of DNA adducts in Induced Sputum
1.
2.
3.
4.
5.
6.
7.
Sensitive and specific
Relation with exposure
Standardized and validated
Relatively easy to perform
Non-invasive
High throughput
Inexpensive
+/+
+/+/+/-
Slide B1.32
Stability of DNA adducts in PBL of smokers
21 weeks stopped smoking. Decay curve exponentially. BG in nonsmokers.
Godschalk et al. CEBP 1998
Slide B1.33
Saturation in DNA adduct levels at high exposure levels
A) DNA adduct levels
B) efficiency of DNA adduct
formation
in (○) smoking and (•) nonsmoking aluminium workers
exposed to PAH
1 foundry
2 electrolysis
3 bake oven
4 anode factory
5 pot-relining department
Van Schooten et al. Mut Res 1997
Slide B1.34
Correlation between surrogate and target tissue
HOWEVER,
* Not all studies find such
a relationship.
* DNA adduct levels in
PBL/ WBC of low exposed
subjects are often below
or near the limit of
detection
Wiencke et al (1999) J Natl Cancer Inst;91(7):614-9.
Slide B1.35
Differences between tissues
The example of cigarette smoke exposure
16
Adducts per 108 nucleotides
14
12
10
8
6
4
2
DL = detection limit
0
Lung
BAL
IS
Monocytes
Lympho- Granulocytes
cytes
WBC
Slide B1.36
Overall Conclusions

Type of tissue and choice of biomarker depends on research
goal

Most biomarkers AND surrogate tissues still need further
validation.
 Laboratory validation
 ‘Field’ validation

Non-invasive and high throughput methodologies are required
for Molecular Epidemiology studies
Slide B1.37
The end