| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
Regular Articles |
From the Department of Pathology,*
Norris Cancer Center, University
of Southern California School of Medicine, Los Angeles, California; the
Departments of Biological Sciences and
Mathematics,
University of Southern
California, Los Angeles, California; the Departments of Pathology and
Medical Genetics,
Haartman Institute,
University of Helsinki, Helsinki, Finland; the Department of
Pathology,§
University of Queensland Medical
School, Herston, Australia, and the Department of Medical
Genetics,||
Haartman Institute, University of
Helsinki, Helsinki, Finland
 |
Abstract |
|---|
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionAppendixReferences |
|---|
|
Introduction |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionAppendixReferences |
|---|
The study of evolution provides a useful analogy to illustrate some of the complex differences between descriptions based on the direct examination of presumptive precursors, or molecular evolution based on genetic comparisons between current species. The fossil record depicts physical aspects of evolution and provides a classical understanding of phylogenies, which have been complemented by modern molecular approaches. Phylogenies based on the fossil record or on genetic comparisons may not agree,6 as they measure different aspects of evolution. One potential bias of the fossil record is its dependence on abundance, as rarer species may not be found. In contrast, molecular evolution can trace the divergence between species regardless of past abundance. In addition, the fossil record is replete with apparent dead ends, whereas genetic comparisons trace persistent lineages.
Similarly, multistep tumor progression may have multiple descriptions, depending on the approach and information desired. Progression models are largely based on mutation frequencies in lesions of different histological stages. By necessity this approach, similar to the analysis of fossils, is biased toward detectable clonal expansions, because one cannot physically analyze lesions that cannot be seen. The potential severity of this bias depends on (1) whether everything seen is relevant to cancer and (2) whether we see everything relevant to cancer. These issues are difficult to resolve. For example, clearly not every adenoma progresses, as the ratio of adenomas to cancers is approximately 30:1.7 However, it is uncertain whether every adenoma could progress to cancer if allowed sufficient time.
Genetic comparisons of concurrent tumors can trace their ancestors
regardless of past abundance. Mutation frequency studies and molecular
tumor phylogenies should concur if progression occurs along a single,
increasingly more "fit" and frequent lineage (Figure 1)
. In this case of sequential clonal
evolution,1
the clonal expansions are relevant to cancer
(ie along the lineage to cancer), and the cells relevant to cancer
expand (and therefore can be detected) before transformation. Physical
(detectable) emergence coincides with lineage divergence.
|
. In this case, multiple lineages
related to a single precursor diverge early and independently undergo
multistep progression either to adenoma or to cancer. Lineage
divergence may precede detectable physical emergence by a long time if
clonal expansion is contingent on mutations acquired after divergence.
Although the implications of monolineage and multilineage progression
are fundamentally different, mutation frequency studies are not
explicitly designed to trace lineages and by default yield progression
along a single lineage. Indeed, data from known multilineage
progression (tumors from unrelated individuals) can be accommodated by
a linear adenoma-cancer progression model.4
This paper traces tumor cell lineages back to their most recent common
ancestor (MRCA) through comparisons between genotypes of concurrent
tumor populations (Figure 2A)
. The
position of a MRCA provides information on when distinct precursors are
first created. As illustrated in Figure 2A
, several MRCAs can be
defined for the cells we examined. Comparisons between adjacent
adenoma-cancer genotypes estimate the time since lineage divergence
(T2). Comparisons between different parts
within the same tumor estimate the time since physical clonal
emergence. Comparisons between germline and tumor genotypes estimate
the time since initiation (T1 +
T2).
|
. With an
early branch, subpopulations with different phenotypes can be defined
as distinct lineages, because the MRCA has neither the "red" nor
the "black" phenotype. In contrast, with a late branch lineages can
be defined as related, because one subpopulation is derived from a cell
with the phenotype of the other subpopulation, which is the phenotype
of the MRCA. The new subpopulation is contingent on the first
subpopulation—"black" cannot arise without "red" (Figure 2b)
.
The trees in Figure 2B
represent a single transition. Tumor
progression, however, is a multistep process1-5
and
therefore should be constructed as a series of sequential changes.
Tumors change visible phenotypes, but may or may not change lineages
with progression. For example, progression to cancer may occur along a
single lineage (Figure 2C)
. Presumably, a new mutation in a single cell
confers upon it a selective advantage, allowing clonal expansion and
dominance over subclones that lack the mutation. Once past a
"gatekeeper" mutation,5
numerical predominance and
greater numbers of accumulated oncogenic mutations could channel
sequential selection along a single most "fit" and increasingly
frequent lineage,8,9
successively destroying and then
shifting the MRCA to the right. If a cancer arises directly from a
concurrent adenoma, adenoma-cancer lineage divergence should occur
relatively late, as a late branch of the final step (Figure 1)
.
Alternatively, multiple related but subsequently independent lineages
may persist, therefore preserving the MRCA (Figure 2D)
. Progression
remains a multistep process, because each lineage independently
undergoes sequential mutation, to either the final adenoma or final
cancer phenotype. However, lineage divergence can occur early and may
precede the visible differentiation of cancers out of adenomas. In this
case, the adenoma and cancer lineages are distinct because they branch
early and the MRCA had neither the adenoma nor cancer phenotype. The
adenoma lineage could be "erased" back to the MRCA without changing
the cancer. It is worth noting that phenotypes and clone sizes (which
may be as few as a single cell) during early progression are unknown;
these are illustrated in gray in Figure 2C and D
.
Genotypes can be compared and traced to a MRCA. Tumors deficient in DNA mismatch repair (MMR) have greatly elevated mutation rates,5 which are most prominent in simple repeat sequences or microsatellites (MS). With mutation rates as high as 0.01 per division,10,11 noncoding MS loci in mutator phenotype (MSI+) tumors can function as "molecular tumor clocks" because they are expected to become polymorphic after relatively few divisions.12,13 MS loci mutate predominantly by a relatively predictable mechanism ("slippage" during DNA replication14,15 with small repeat unit additions or deletions,10,11 allowing linkage of lineages through time. MS loci have been used to trace the emergence of modern humans out of Africa and the divergence between humans and chimpanzees.16-18
Physically adjacent colorectal adenomas and cancers are presumptive evidence that cancers arise from adenomas.19 Comparisons of their MS genotypes allow objective analysis of this relationship. When do adjacent adenomas and cancers diverge?
|
Materials and Methods |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionAppendixReferences |
|---|
Five tumors from three patients were examined (Table 1)
. Patients I and III were from
hereditary nonpolyposis colorectal cancer (HNPCC) families and had
germline mutations in hMS2 and hMLH1, respectively. Although patient II
was only 42 years old at the time of his colorectal cancer, his
familial history did not meet the criteria for HNPCC. There were two
synchronous and adjacent adenoma-cancer pairs, and two physically
distinct metachronous adenoma-cancer pairs.
|
Individual cells cannot be readily isolated and genotyped from fixed tissues. Therefore, the essential approach isolates the DNA from specific tumor regions on a thin 4–8-µm-thick microscope slide. The isolated DNA comes from a mixture of adjacent cells of similar phenotype and is fragmented in such a way that MS loci are physically separated from each other. DNA is sampled at random from this pool, and essentially single loci are typed by polymerase chain reaction (PCR) after dilution. The process is repeated from the same pool until alleles from multiple loci are typed.
MS Typing
To simplify analysis, X-chromosomal CA-dinucleotide repeat MS loci
and male patients were used. Every allele therefore represents a single
cell because MSI+ tumors characteristically lack
aneuploidy.20
The MS distributions were determined by two
methods. For DXS556, DXS1060, DXS418, and DXS453 (Research Genetics,
Huntsville, AL) and the data of Figure 3
,
multiple small tumor regions of approximately 200–400 cells and
containing at least 70% tumor cells were isolated by selective
ultraviolet radiation fractionation12
from microscopic
tissue sections. The DNA in these dots was diluted to essentially
single alleles12
with about 20–80% of reactions yielding
PCR products, which were analyzed on 6% denaturing
polyacrylamide sequencing gels and a phosphoimager (Molecular
Dynamics, Sunnyvale, CA). PCR products were labeled with
[33P]dCTP (NEN Research Products, Boston, MA)
incorporated during 38–43 PCR cycles in 5-µl reaction volumes.
|
.
For the data of Figure 4
and Table 2
, DNA was extracted in bulk from
dissected adenoma or cancer regions and then diluted for analysis. At
least 10 alleles were amplified for each additional MS locus (list
available on request) until a mode became evident.
|
|
|
Results |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionAppendixReferences |
|---|
and Figure 3a
). The MS loci were polymorphic, and their
distributions were different between the adjacent tumors (Figure 3b)
.
The adenoma-cancer pairs were further compared at a total of 34–37
loci. These additional loci were not characterized as extensively as in
Figure 3b
. Specifically, the ability to amplify single alleles
consistently at every locus was uncertain, and fewer alleles were
genotyped (~25 alleles per locus). The modes of the loci were evident
from their distributions (see, for example, Figure 3b
) and would be
less susceptible to problems of reduced PCR sensitivity. Therefore, for
experimental simplicity, the MS distributions for all of the loci are
summarized by their modes, which are assumed to approximate their means
(see Appendix). The genotypes defined by these modes differed between
the adjacent adenoma-cancer pairs (Figure 4a)
.
Although physical proximity has been used to infer that adenomas
are direct precursors to cancers,19
the MS analysis
suggests a more distant relationship. To further understand how quickly
adenomas and cancers diverge, several other scenarios were examined.
Different regions within the same tumor should be closely related,
because they presumably arise from the same terminal clonal expansion.
For example, the invasive and superficial portions of a cancer arising
within a surveillance interval of 6 months had very similar genotypes
(Figure 4b)
. In contrast, metachronous tumors arising at different
sites and times in the same patient should be unrelated, because they
initiate from the same germline but otherwise progress independently.
Metachronous adenoma-cancer pairs exhibited large MS differences
similar to adjacent adenoma-cancer pairs (Figure 4c)
.
The data were modeled assuming a single cell initiates tumorigenesis
through loss of MMR and later splits into two different tumor lineages
(see Appendix). Using the model, we estimate the relative numbers of
divisions between initiation, divergence, and presentation (Table 2
and
Figure 5
). The estimated intervals
between initiation and presentation of the synchronous tumor pairs were
between 1600 and 2600 divisions, or 4.4–7.1 years, assuming one
division per day (Table 1)
. As expected, unrelated metachronous
adenoma-cancer pairs from patients I and III diverged very early. In
contrast, different regions within the same tumor were closely related
and diverged later. Less than 10% of their divisions occurred after
their MRCA, illustrating a random although predictable accumulation of
MS mutations in both adenoma and cancer cells.
|
.
The adjacent pairs appeared to be related because they diverged later
than the metachronous pairs, but still evolved independently for
greater than 60% of their divisions. However, as recently observed for
some adenomas,21
the adjacent tumors may also be unrelated
because the estimated confidence intervals were large (Table 2)
. For
patient II, adenoma-cancer lineage divergence substantially preceded
adenoma or cancer intratumor divergence (Figures 4d and 5)
. |
Discussion |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionAppendixReferences |
|---|
the differences in modal genotype are plotted
for each MS locus. The larger spread evident in the comparisons between
adenoma and cancer, as opposed to the smaller spread observed with
comparisons within adenoma or within cancer, is consistent with earlier
divergence.
Although the data and their analysis are complex, they yield relatively
simple tumor trees (Figure 5)
. Some aspects of these trees are known
with certainty and therefore can be examined for internal consistency.
As expected, the unrelated metachronous adenoma-cancer pairs had very
different genotypes (correlation coefficients <0.2) and diverged
early. Different regions within the same adenoma or cancer had more
similar genotypes (correlation coefficients >0.9) and therefore had
the recent divergence expected of a clonal expansion. The estimated
divergence (160 days) between superficial and invasive portions of
Cancer-2 is within the 6-month clinical surveillance interval of
patient I.
Adjacent adenoma-cancer pairs also had very different genotypes
(correlation coefficients of 0.30 and 0.36). Therefore their lineages
diverged relatively early and evolved independently for more than 60%
of their ~1600–2600 divisions since initiation. The visible
manifestations at adenoma-cancer divergence are unknown because
lineages and not phenotypes are traced. Phenotypic differentiation or
detectable emergence of the first distinct adenoma or cancer precursors
may coincide with lineage divergence or, more likely, may be delayed
and contingent on the subsequent accumulation of mutations. For
example, the adjacent tumors of patient II, with an estimated age of
2600 divisions since loss of MMR, physically emerged recently, but
their lineages diverged earlier (Figure 5)
. Since divergence occurred
substantially before detectable emergence, it is unlikely that the two
cells created at the lineage split from their MRCA had yet acquired
their respective adenoma or cancer phenotypes.
Cancer-2 of patient I also provides insight into the period before
detectable emergence, because no physical precursor was visible 6
months before it was removed. Although evolution in this cancer may
have been compressed, with accelerated formation and subsequent
destruction of an adenoma precursor, its MS genotype was very different
from its germline. With an estimated age of 1800 divisions (Table 1)
since loss of MMR, a more likely scenario is that most mutations
(~90% if cells divide once a day) accumulated in an occult or
microscopic precursor (Figure 6)
.
Therefore, Cancer-2 and early adenoma-cancer lineage divergence suggest
that a cancer lineage may not require macroscopic expansion before
transformation. Adenoma precursors may not be essential in the setting
of a mutator phenotype because a high mutation rate can compensate for
the lack of clonal expansion.3,22
Therefore, a lineage
selected for visible adenoma expansion may not coincide with the
pathway to cancer,7
as other underlying occult lineages may
have greater potential for transformation. Neoplasia becomes a
consequence of rather than an obligate substrate or "direction" for
further random mutation, because some expansions represent dead ends
with respect to the final cancer lineage.
|
Early divergence requires the prolonged persistence of multiple lineages, which implies physical protection or niches against any temporally dominant subclones. The crypt architecture and stem cell renewal of normal intestinal mucosa provide protection against neoplastic expansion.26 Recent studies of MSI+ adenomas13,27 demonstrate evidence of stem cell behavior, which would inherently limit clonal succession.
The current study reconstructs MSI+ tumor lineages to peer back in time to the first point at which a cancer branches or attains a destiny distinct from that of a concurrent adenoma. This critical event may be occult and apparent only in retrospect. Although refinement of the model is possible, the substantial genetic differences between adenoma-cancer pairs (in contrast to the intratumor similarities) would be difficult to generate unless their lineages were separated by large numbers of divisions. Multilineage progression may be more universal than currently appreciated, as lineages are seldom formally traced. Unlike the predictable visible chronology of linear multistep progression (sequential clonal expansion), this study lessens the potential bias imposed by the analysis of only visible clonal expansions and therefore emphasizes the length and complexity of steps that may not be directly observed. Without observable criteria (such as clonal frequency) to order progression, one cannot predict which lineages are destined for transformation, except in retrospect. Perhaps multilineage progression reflects a similar biological inability to select early on the most likely candidate for transformation. Instead of clonal evolution through serial stepwise selection of a single most "fit" and frequent lineage from adenoma to cancer, one effective early progression strategy creates and maintains multiple evolving candidate lineages, which are subsequently selected for terminal clonal expansion.
|
Appendix |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionAppendixReferences |
|---|
For synchronous tumors, we assume a cell alive at generation T1 - 1 splits into two cells in generation T1, one of which is the precursor to the A cell population, the other the B cell population. The A and B cell populations evolve independently of each other for a further T2 divisions:
After these T2 divisions a number of cells
are sampled from cell lineage A and from lineage B. In our experimental
approach, we are able to measure the modal repeat length at each locus.
In what follows we use these modal repeat lengths as surrogates for the
average length in each sample (see Figure 3
). The average lengths
LA and
LB of a given MS locus in lineages A
and B, respectively, are measured relative to germline. To calculate
the correlation coefficient 
between
LA and
LB, we use a stepwise model of MS
evolution.28-32
Under this model, a random number of
repeats are added at the given locus at each cell division. These
repeat numbers are independent from division to division, with common
mean µ and variance 
2. (For simplicity we assume the
same parameters hold in all divisions, although our approach can be
modified to account for different mutation mechanisms at different
stages at the expense of greater complexity.)
We denote by SA the random number of divisions
taken to trace two cells chosen at random from the A cell population
back to their MRCA. In our model, we must have 1 
SA T2. The mean of
SA is denoted by
E(SA). Under this mutation
model, it is clear that
 | (1) |
Two cells chosen at random from population A share a number of
ancestral cells (those in generations 0,1, ...,
T1, T1 + 1, ...,
T1 + T2 -
SA) and therefore share part of their
evolutionary history. After time T1 +
T2 - SA, however, the
two cells have independent mutation histories. We can use this
observation to show that the variance of
LA is given by
 | (2) |
On the other hand, if we choose a cell at random from each of lineages
A and B, they will share fewer ancestral cells, those in generations
0,1, ..., T1 - 1. After time
T1, the two cells have independent mutation
histories. Using this fact, we can show that the covariance between
LA and
LB is given by
 | (3) |
Combining Equations (2) and (3)
and simplifying shows that
 | (4) |
where
 |
We note from Equation (4)
that the correlation coefficient is
always positive, and that = 0 in the particular case where
T1 = 1 (corresponding to independent evolution
of two cell lineages for time T2). Furthermore,
is expected to be small whenever T1 is small
relative to T2.
To obtain a simpler, approximate formula for , we assume that the
total number of cell divisions T1 +
T2 is large relative to
E(SA) and
E(SB). It follows that
c 
1, so that
 | (5) |
Estimation of Tumor Ages
The method employed here estimates the size of
T2 relative to T1 +
T2. To estimate the absolute size of
T1 + T2, we need to
assume something more about the mutation mechanism. For example, if
mutations arise according to the simplest symmetrical mutation model
(in which with probability P a mutation occurs, and results
in the addition or loss of a single repeat unit, each with equal chance
0.5), then µ = 0 and 2
= P. With the
possible exception of the adenoma in patient I, the data are consistent
with the assumption µ = 0. Knowledge of P then allows us
to estimate T1 + T2,
using the variances calculated in Equation (2)
. This follows because
 | (6) |
, estimates of the likely numbers of
divisions necessary to generate the observed MS mutations for each
tumor. As an example, from Table 3
the variance of patient II adenoma
is estimated to be approximately 15. Hence if P = 0.005
(a value within the range of mutation rates observed in MMR
deficient cell lines),10,11
it follows that
T1 + T2 15/0.005
= 3000 divisions.
|
For synchronous tumor pairs, we can apply the theory directly to
MS variability measured (relative to germline) at several different
loci. We use metachronous tumor pairs as controls, treating them as
though they started from the same cell. Because these pairs have
evolved independently after initiation, we would expect that the
correlation between the two mean lengths to be zero; this is reflected
in the fact that the confidence interval for should include zero.
Our model makes specific predictions for the variability that may be
used informally to assess the adequacy of the model. In particular,
Equations (1) and (2)
show that
E(LA) and
E(LB) are equal, and that
the variances var(LA) and
var(LB) are equal. Estimates of
these means and variances, obtained from the different loci and A,B
pairs, are given in Table 3
.
For the synchronous tumor pairs, formal statistical tests of equalities
of means and variances are complicated because of the weak dependence
between the different MS loci (they share part of the same cell lineage
history). Assuming approximate independence of the loci, the results in
Table 2
show no obvious contradictions with the model: the tumor pairs
have approximately the same means and variances. This is also true of
the metachronous pairs.
The correlation estimates given in Table 1
are also based on comparison
of LA and
LB over different loci, for each
pair of tumors A,B. The confidence intervals 33 for are again based
on the assumed adequacy of the approximate independence of the
(LA,
LB) pairs over different loci within
a given individual.
|
|
Acknowledgements |
|---|
|
Footnotes |
|---|
Supported by grants from the National Institutes of Health to D.S. (CA58704 and CA70858) and from the National Science Foundation to S.T. (BIR95-04393).
Accepted for publication March 10, 1999.
|
References |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionAppendixReferences |
|---|
This article has been cited by other articles:
 |
|
 |
|
  D. Frumkin, A. Wasserstrom, S. Itzkovitz, T. Stern, A. Harmelin, R. Eilam, G. Rechavi, and E. Shapiro Cell Lineage Analysis of a Mouse Tumor Cancer Res., July 15, 2008; 68(14): 5924 - 5931. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
S. E. Kern and D. Shibata The Fuzzy Math of Solid Tumor Stem Cells: A Perspective Cancer Res., October 1, 2007; 67(19): 8985 - 8988. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 R. D. Cardiff, J. P. Gregg, J. W. Miller, D. E. Axelrod, and A. D. Borowsky Histopathology as a Predictive Biomarker: Strengths and Limitations J. Nutr., October 1, 2006; 136(10): 2673S - 2675S. [Full Text] [PDF]
|
|
|
|
 |
|
 K. R. Mehta, K. Nakao, M. B. Zuraek, D. T. Ruan, E. K. Bergsland, A. P. Venook, D. H. Moore, T. A. Tokuyasu, A. N. Jain, R. S. Warren, et al. Fractional Genomic Alteration Detected by Array-Based Comparative Genomic Hybridization Independently Predicts Survival after Hepatic Resection for Metastatic Colorectal Cancer Clin. Cancer Res., March 1, 2005; 11(5): 1791 - 1797. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
M. Velicescu, J. Yu, B.-S. Herbert, J. W. Shay, E. Granada, and L. Dubeau Aneuploidy and Telomere Attrition Are Independent Determinants of Crisis in SV40-transformed Epithelial Cells Cancer Res., September 15, 2003; 63(18): 5813 - 5820. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J R Jass, H Iino, A Ruszkiewicz, D Painter, M J Solomon, D J Koorey, D Cohn, K L Furlong, M D Walsh, J Palazzo, et al. Neoplastic progression occurs through mutator pathways in hyperplastic polyposis of the colorectum Gut, July 1, 2000; 47(1): 43 - 49. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
A.-E. Biemer-Hüttmann, M. D. Walsh, M. A. McGuckin, L. A. Simms, J. Young, B. A. Leggett, and J. R. Jass Mucin Core Protein Expression in Colorectal Cancers with High Levels of Microsatellite Instability Indicates a Novel Pathway of Morphogenesis Clin. Cancer Res., May 1, 2000; 6(5): 1909 - 1916. [Abstract] [Full Text]
|
|
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
