Paolo Vineis Imperial College London Causal models of carcinogenesis: a historical perspective

1
Paolo Vineis Imperial College
LondonCausal models of carcinogenesis a
historical perspective

2
1. Models of carcinogenesis2. Examples
smoking, asbestos3. Role of mutation, cell
selection, epigenetics4. Causal models

3
Armitage and Doll in 1954 proposed a multistage
model based on the observation that the incidence
rate of most epithelial tumors rises with a power
of age (5-6th power). They hypothesized-
that cancer is not due to age itself but to
prolonged duration of exposure to carcinogens-
that for a life-long exposure an increase with a
power of 6 means that there are 6 stages in
carcinogenesis- for discontinued exposures the
model becomes more complex
4
I(t) r1r2 r(n-1) (t-w)n-1where r is the
transition rate from a stage to the followingt
is agew is the time mecessary to last-stage
cells to give rise to a clinically overt
cancerAs an approximationI(t)K t n-1(n-1)
refers to the transition rates
5
The relationship with age holds true for most
epithelial cancers (exponential of age 6 for
oesophagus, stomach, pancreas, bladder, rectum,
colon), but not for lung and breast (cohort
phenomena)THE BASIC IDEA IS THAT IT IS NOT AGE
BUT DURATION OF EXPOSURE
6

• EXPERIMENTS BY IVERSEN TREATMENT OF MICE WITH
  DMBA
• (CARCINOGENESIS, 1991)
• A SINGLE DOSE OF 51.2 MICROGRAMS GAVE A TUMOR
  RATE OF 40, WHILE THE SAME DOSE DIVIDED INTO 50
  DOSES OF 1 MICROGRAM GAVE A 100 RATE

7

• REPEATED EXPOSURE TO SMALL DOSES SEEMS TO BE THE
  MOST HAZARDOUS SITUATION
• OBSERVED TO EXPECTED RATIO
• ACETONE 0.04
• DMBA 51.2 MICROG 0.62
• 25.6, TWICE 1.74
• 10, 6 TIMES 2.93
• 2.6, 20 TIMES 7.04
• 1, 50 TIMES 7.95
• EXPERIMENTS WITH UV LIGHT IN MICE SHOWED THAT
  CONTINUOUS EXPOSURE AT LOW DOSES WAS MOST
  EFFECTIVE
• AN INCREASING TIME INTERVAL BETWEEN EACH DOSE MAY
  ALSO INCREASE THE RISK

8
TOBACCO SMOKINGEXAMPLE ACS COHORT (Hammond et
al, 1977)Age at start SMR NNON-SMOKERS 1.0
25 3.2 2020-24 9.7 11015-19 12.8
315lt15 15.1 101Years since cessation
(20 cigs/day)0 13.7 351lt1 29.1 331-4
12.0 335-9 7.2 2210 1.0 5
9
EXAMPLE ASBESTOS (Seidman et al, 1977)Years
since cessation increase of CI5-9 -0.210
-14 0.415-19 1.220-24 1.325-30
1.7
10
On this basis, tobacco smoking has been
considered to be both an early and a late stage
carcinogen by Doll (1978) and Day and Brown
(1980)while asbestos has been considered an
early stage carcinogen for mesothelioma (risk
never decreases after cessation, age at start is
extremely important)
11
Other models introduce different
assumptions(a) clonal expansion (Cairns 1975
Moolgavkar)(a) killing of stem cells (Cairns
2002)
12
Multistage carcinogenesis and the incidence of
colorectal cancer E. Georg Luebeck and Suresh H.
Moolgavkar PNAS November 12, 2002 vol. 99
no. 23 15095-15100 The TSCE model posits that
a malignant cell arises after two rare events in
a stem cell. After the first event, assumed to
occur with rate µ1 per cell per year, the
initiated tem cell expands clonally, giving rise
to an intermediate (initiated) lesion.
Initiated stem cells divide symmetrically with
rate alpha and die or differentiate with rate
beta. With rate µ2, however, an initiated cell
may divide asymmetrically, giving rise to a
malignant daughter cell, the progenitor of a
carcinoma. The growth of the intermediate
lesion is described mathematically by a
stochastic birth-death process

13
   
 
14
SELECTION IN CANCER usual view

• It is commonly recognized that somatic MUTATION
  (irreversible change in DNA information content)
  initiates the process of carcinogenesis
• The mutated cell(s) are selected in vivo because
  of their growth advantage, loss of contact
  inhibition, loss of apoptotic pathway(s), etc.
  This is selection after mutation, i.e. SELECTION
  FOR THE MUTANT PHENOTYPE.
• (R Albertini)

15
SELECTION FOR MUTANT PHENOTYPES IS ALSO
SELECTION FOR MUTATOR PHENOTYPES(current view)
16
CHILDREN TREATED FOR LEUKEMIA

• Treatments included multiple cytotoxic and
  genotoxic agents
• All treatments included a purine analogue, e.g.
  6-MP, 6-TG

17
(No Transcript)
18
Proliferation of Mutators in a Cell
PopulationMao EF, Lane L, Lee J Miller
JHJournal of Bacteriology (1997)Vol 179 (2)
417-422
19
(No Transcript)
20
IN HUMANS, AS IN BACTERIA, SELECTION FOR
MUTANT PHENOTYPES IS ALSO SELECTION FOR MUTATOR
PHENOTYPES(WHICH ARE PRESENT AT LOW FREQUENCIES
IN MOST INDIVIDUALS)(R. Albertini)
21
A NON-CANCER MODEL OF DARWINIAN MECHANISM
PNH Paroxysmal nocturnal hemoglobinuria
(PNH) is an acquired stem cell disorder
characterized by intravascular hemolysis, and
bone marrow failure. The characteristic
defect in PNH is the somatic mutation of the
PIG-A gene in hematopoietic cells.
22
The current hypothesis explaining the disorder
suggests that there are two components (1)
hematopoietic stem cells with the characteristic
defect are present in the marrow of many if not
all normal individuals in very small numbers (2)
some aplastogenic influence suppresses the
normal stem cells but does not suppress the
defective stem cells, thus allowing the
proportion of these cells to increase.
(darwinian interpretation) Bessler M,
Mason P, Hillmen P, Luzzatto L. Somatic mutations
and cellular selection in paroxysmal nocturnal
haemoglobinuria.Lancet 1994 Apr
16343(8903)951-3
23
A new paradigmepigenetics
24
(No Transcript)
25
Information heritable during cell division other
than the DNA sequence itself.
26
(No Transcript)
27
(No Transcript)
28
Models of causality from simple to complex
29
Necessary and sufficient causes
Necessary
Yes No
Yes Chr 21 Guillotine
Down syndr.
Cut neck Sufficient No
Mycobacterium Tobacco
Pulmonary TB Lung cancer

(R Saracci, 2005)


30
What about genetic causation? Nec
essary
Yes No Yes
two Rb XP-related
mutations cancer
Sufficient No one Rb
All others mutation
31
Insufficient Non-redundant component of an
Unnecessary Sufficient Complex (INUS)
32

Use of graphical models to disentangle complex
GEI (Vineis et al, paper in preparation)
33
What is interaction?
34
The absent minded Mr Smith A. the probability
that Mr Smith leaves the gas alight is 50, or
p(A) 0.5 (environmental exposure) B. the
probability that the alarm system does not work
is 1, or p(B) 0.01 (genetic risk) C. the
probability that a fire develops for reasons
other than those considered here (the background
risk) is 1/1,000, or p(not A and not B) p(C)
0.001
35
1. The scenario of population average (prior
probability). The probability of a fire occurring
through the causal chain involving only two
factors is p(A and B) - p(not A and not B)
(0.5 x 0.01)-0.001 0.005-0.0010.004. The
relative risk of a fire occurring through this
chain, compared to the risk of fire through some
other causal chain (C, the background risk)
is 0.005/0.0015.
36
2. A scenario of partial knowledge of individual
risk. If Mr Smith knows that he left the gas on
but he does not know if the alarm works, then the
probability of a fire is p(B given A) - p(non-A
and non-B) 0.01 - 0.001 0.009. The relative
risk for this causal chain compared to the
background risk is 0.01/0.001 10. 3. The
scenario of perfect knowledge of extrinsic and
intrinsic risk factors. If Mr Smith knows both
that he left the gas on AND that the alarm does
not work, then the probability of a fire is 1
the probability that the fire arises as a
consequence of this particular causal chain is 1
- 0.001, and the relative risk is 1/0.001 1000.
37
What really counts is the combination of
factors, and in particular the fact that some
exposures can complete an incomplete causal
chain. What makes this insight particularly
important for the problem of attributing causes
of cancer (or any other disease) is that while we
are confident that multiple factors act through
causal chains such as these, we are almost always
quite ignorant about what components make up
these chains. (Vineis and Kriebel, Enviromental
Health, 2006)
38

• My favourite approach Schaffners
  conditionalized realism
• A theory is true conditionally on
• Truth of auxiliary hypotheses (e.g. data in
  animals, molecular biology)
• Lack of valid alternative explanations
• Role of middle range theories (eg cell
  selection)
• Also Wesley Salmons idea of propagation of a
  mark contributes to seeing causal inference as
  mutual support between different layers of
  reality (molecules to populations)

39
Summary 1(a) different mathematical
models are compatible witht the evidence on
age-specific cancer incidence(b) different
biological models (e.g. involving clonal
expansion or stem cell death) are compatible with
epidemiologic evidence(c) however, it is likely
that selection of mutated clones AND of clones
with mutator phenotype is involved(d) a rapidly
expanding new paradigm based on epigenetics is
now developing
40

• Summary 2
• Different epistemological models are compatible
  with the evidence, except a naive one based on
  necessary and sufficient causes
• (a) Mackie's model of INUS
• (b) Salmon's propagation of a mark
• (c) Shaffner's multilayer model based on
  middle-range theories and conditionalized
  realism
• (d) a Bayesian approach to interaction

41
THE END