- Cancer-associated fibroblasts
- mutation detection
There is convincing evidence that carcinoma-associated fibroblasts (CAF) differ phenotypically from fibroblasts associated with normal epithelium and influence tumorigenesis ( 1– 3). For example, Cunha and colleagues showed that the in vivo combination of normal human prostatic epithelial cells with CAFs led to limited tumor growth that resembled prostatic intraepithelial neoplasia, and that grafting CAFs with immortalized prostatic epithelial cells that were nontumorigenic but expressed the SV40 T-antigen resulted in the formation of malignant tumors ( 4, 5).
Because CAFs can be propagated in vitro for extended periods and still maintain their cancer-promoting phenotype ( 4, 6), it has been proposed that they might have acquired somatic genetic alterations analogous to those observed in malignant epithelium. A number of studies ( 7– 9) seemed to provide proof that the acquisition of somatic mutations in CAFs is associated with tumorigenesis. For example, frequent somatic mutations in classic tumor suppressor genes, such as PTEN and TP53, have been reported in fibroblasts associated with breast carcinomas ( 8, 10). The possibility that these mutations might be due to contaminating cancer cells was excluded because the mutations were distinct in the two cell types. CAFs have also been reported to carry a unique set of chromosomal aberrations ( 11– 14) with loss of heterozygosity (LOH) at a frequency similar to that observed in the epithelial components. For example, average marker-specific LOH frequencies of 59.7% and 28.4% have been reported in CAFs derived from BRCA1/2-related and sporadic breast carcinomas, respectively ( 7, 15). Moreover, significant associations between the CAF LOH signature and both tumor grade and lymph node metastasis have been reported in women diagnosed with sporadic breast tumors ( 16). Patocs and colleagues ( 8) reported a TP53 mutation frequency of 27% in breast CAFs and that mutation status was associated with regional nodal metastasis. In ovarian cancer, up to 63% of CAFs have been reported to show LOH at 3p21 ( 14). The same investigators subsequently assessed copy number changes at 110 loci and observed that ∼10% of these genes had undergone copy number changes in each ovarian CAF sample ( 17).
If reports of CAF somatic mutations are correct, they would provide compelling support for the hypothesis of interdependent coevolution of clonal populations of carcinoma cells and CAFs. This hypothesis is provocative as it requires that random somatic mutations generate individual fibroblast cells with an enhanced ability to promote tumorigenesis in surrounding epithelial cells. In turn, the cancer epithelial cells promote the proliferation of the fibroblast cell with the cancer-promoting mutation such that it is able to outgrow the neighboring fibroblasts and generate a dominant clonal population of CAFs. In essence then, this hypothesis proposes the simultaneous generation of two symbiotic malignancies.
Beyond the fact that in rare instances mixed malignancies are described, such as carcinosarcomas and Brenner-type ovarian tumors, there are many conceptual difficulties with the genetic coevolution model, the foremost of which is the issue of selective advantage. Although we are familiar with the concept that random somatic mutations could occasionally generate a cell with a selective advantage such that it is able to clonally expand and replace those cells without the mutation, this occurs in a cell autonomous fashion. The coevolution hypothesis necessarily invokes a model in which a mutation in a fibroblast cell provides no direct selective advantage to itself. Whereas proponents of this model would argue that the selective advantage derives from the symbiotic carcinoma cells, this would require the unlikely scenario in which only the mutant fibroblast was able to respond to the generic signals emanating from the carcinoma cells in order for that fibroblast clone to dominate the stromal population. At the very least, if fibroblasts in the vicinity of the mutant-driver fibroblast minority are responding to a cue from the neighboring epithelia, there should be evidence for increased proliferation of stromal cells in the malignant tissue over that of normal cells. Whereas such evidence exists for the rare Brenner-type tumors and carcinosarcomas, there is no such observation in the case of classic carcinoma-associated stroma. Furthermore, considering our understanding of senescence, if a mutation resulted in clonal expansion of the CAF, it would be expected to run into crisis before populating millimeters of the tissue. Even if one speculated that CAFs acquired bona fide transforming mutations, the expected outcome would be the generation of a sarcoma. Instead, CAFs appear to be histologically normal, and in culture behave in a manner similar to normal human fibroblasts showing neither anchorage-independent growth nor in vivo tumorigenicity and many begin to senesce after prolonged population doublings ( 3). 7
The theoretical difficulties of the genetic coevolution model, which seem to contravene established paradigms of tumorigenesis, have raised concerns that studies reporting frequent somatic alterations may be technically flawed. Indeed, not all studies support the existence of rampant somatic alterations in CAFs. For example, Allinen and colleagues ( 18) undertook array CGH and single nucleotide polymorphism array analysis of carcinoma cells and myofibroblasts isolated from ductal breast carcinomas. As expected, numerous alterations were identified in the epithelial components, but the myofibroblasts and other nonepithelial tumor-associated cells appeared normal. However, because the fibroblasts were isolated from whole tumor biopsies using cell type–specific markers rather than direct microdissection of tumor-juxtaposed fibroblasts, it was possible that any CAF genetic alterations might have been masked by the presence of normal fibroblasts. A recent study of CAF cultures derived from pancreatic cancer also failed to identify any evidence of somatic copy number gains or losses ( 19).
It was in this context that we undertook a high-resolution 500K single nucleotide polymorphism array–based investigation of CAFs microdissected from fresh-frozen primary human ovarian and breast cancers ( 20). We hypothesized that the explanation for the extraordinarily high frequency of somatic alterations in CAFs might be a reflection of the inherent technical limitations of the methodologies used. To ensure consistency with previous studies, we used tissue microdissection to specifically interrogate fibroblasts that were located within 5 mm of the carcinoma interface. In addition, all microdissected areas were confirmed to contain a high proportion of fibroblast cells, thereby eliminating any possibility of contaminating normal cells masking somatic alterations. In stark contrast to previous reports, none of the breast CAF preparations showed any evidence of copy number gain or loss, or LOH on any chromosome. Similarly, all but one of the ovarian CAF preparations showed a normal genomic profile. The one ovarian case with a somatic alteration showed copy number loss of the entire chromosome 22, whereas the remainder of the genome was normal. This case underscored the fact that somatic alterations can be readily detected in CAFs using the single nucleotide polymorphism array technology and also highlighted the reality that such alterations are extremely rare. Even though the single nucleotide polymorphism array data indicated an essentially 0% frequency of copy number alterations or LOH in CAFs, we nevertheless sought to verify this by analyzing for LOH using microsatellite markers identical to those reported to show LOH frequencies of up to 63% on chromosome 3 in ovarian cancer ( 14) and up to 50% on chromosome 11 in breast cancer ( 15, 16). Again, this microsatellite analysis failed to identify any evidence of LOH in any CAF preparation. In addition to assessing for copy number alterations and LOH, we also sought to validate the remarkably high frequency of somatic mutations in TP53 that was recently reported in sporadic breast CAFs ( 8). Using identical primer sequences, we analyzed exon 4 to exon 9 of TP53, but no mutation was detected in any of the 17 CAF preparations, although we readily detected somatic mutations in the carcinoma cells at a typical frequency for breast carcinoma ( 21).
Given that our analysis closely replicated the design of the studies reporting high frequencies of genetic alterations, what could explain these widely divergent observations? One striking common trait of those studies reporting frequent clonal somatic alterations in CAFs is their use of limiting numbers of cells derived from microdissected formalin-fixed, paraffin-embedded (FFPE) tissues followed by highly multiplexed PCR-based analyses. It is well-established that reliable LOH and mutation assessment is severely compromised when using limiting quantities of DNA typical of FFPE tumors ( 22– 25). At very low template concentrations, random variance in the allelic ratios, coupled with the exponential PCR amplification process, could frequently mimic LOH. To those familiar with working with low template DNA concentrations, it is no surprise that PCR-generated artifacts as high as 30% have been reported when using <0.5 ng of 100% amplifiable DNA per reaction ( 23). These problems occur even when the DNA is of high quality, but the problem is greatly magnified when using DNA from FFPE tissue as it is of much poorer quality. One inherent problem of FFPE DNA is that it is highly fragmented (generally in the range of 100–300 bp) so that amplifiable fragments containing the larger allele of a microsatellite marker are inherently less abundant, which can simulate LOH. However, by far the most important factor is the fact that much of the DNA extracted from FFPE tissues is unamplifiable, which profoundly reduces the actual template concentration. In view of this, Sieben and colleagues ( 23) recommended the use of a minimum of 10 ng of FFPE-derived DNA per reaction for reliable PCR analysis. However, more recent investigations suggest that the ratio of amplifiable DNA versus total DNA from FFPE tissues can be as low as 1:3,600 and is typically ∼1:100 ( 22). Consequently, Farrand and colleagues recommended that a minimum of 600 pg of 100% amplifiable DNA should be used for reliable LOH analysis, and even greater amounts when using multiplex reactions ( 22). On this basis, even a conservative template level of 10 ng of FFPE DNA would typically fall well below the 600 pg threshold.
The same issue of low template concentration also applies to the generation of artifactual point mutations ( 24), and explains the apparent presence of somatic TP53 mutations detected in approximately one third of CAFs derived from sporadic and familial breast cancers ( 8, 10). The highly unusual type and distribution of TP53 mutations reported by Patocs and colleagues ( 8) is consistent with PCR-generated artifacts, including a predominance of mutations not previously described, and which are predicted to be neutral for an effect on p53 activity ( 26, 27). In particular, among the BRCA1/2-related cancers, almost 50% of the sequence alterations were in the exact same nucleotide in a CCG-rich region. This mutation (Pro89Ser) has only appeared twice among the 23,000 mutations reported by the universal mutation database for p53, and has never been observed among the 3,000 familial and sporadic breast cancers ( 28). 8 Furthermore, if the TP53 mutations were genuine, one would expect to observe p53 accumulation in the majority of cases harboring mutations, and within each case, the staining should be homogeneous, consistent with the presence of clonal populations of mutant fibroblasts. In fact, Patocs and colleagues only observed p53 staining in 2 out of 45 of the TP53 mutant CAFs, and the one example shown indicates that only a small proportion of scattered cells were p53 positive.
Although it is possible to obtain accurate LOH and mutation data from FFPE-derived DNA if particular care is taken to exclude samples in which the DNA is highly fragmented and/or the yield is low, this does not seem to have been the case for those studies reporting high frequencies of genetic alterations in CAFs. Taking, as an example, the work of Charis Eng and colleagues, who have published the bulk of articles relating to LOH somatic alterations in CAFs ( 7– 10, 15, 16, 29), it is clear that the experimental design does not allow a clear empirical distinction between the null and test hypotheses, as the concern of limiting DNA template numbers has not been sufficiently addressed. In these studies, apparently pure populations of epithelial and stromal cells (located within 5 mm of the epithelial interface) were obtained from FFPE tissues by laser capture microdissection. Critically, however, none of their early publications provide any details as to the thickness of the sections, number of sections, number of cells, or quantity of DNA extracted. In the few representative figures presented in these studies, it seems that relatively few stromal cells were dissected from a single slide suggesting that perhaps only a few hundred cells were taken for each case. Even assuming that 1,000 fibroblasts were microdissected per case, this would equate to a total of only 5 ng of DNA. An unspecified proportion of this DNA was then used in up to 72 independent multiplex PCR reactions, which equates to <70 pg of DNA per reaction. Assuming that a typical FFPE-derived DNA sample will contain <1% amplifiable DNA, the actual quantity of DNA per reaction in these studies is likely to be in the order of 5 pg or less. Given that current recommendations for FFPE-derived DNA is 600 pg of 100% amplifiable DNA per reaction ( 23), the generation of artifactual LOH in these studies is not only a possibility, it is a mathematical certainty.
The problems associated with the use of DNA derived from FFPE tissues or from very small numbers of fresh cells has been well publicized, but investigators reporting frequent CAF somatic mutations have tended to dismiss these problems as being relatively trivial. Consequently, many of the studies do not seem to have undertaken any validation experiments, whereas others have described controls that are flawed, inadequately described, or irrelevant. For example, Eng and colleagues describe four controls that they argue validate the multiplex-PCR method they used in ref. 15, and in subsequent LOH studies ( 7, 9, 16). Two of these controls involved re-analyzing extracted DNA from a subset of samples using additional microsatellite markers or analyzing additional samples using the original set of microsatellite markers. However, if a low DNA input were the original cause of the artifactual LOH, these controls would simply replicate the erroneous data. A third control involved repeating five samples using real-time LOH analysis, which the authors assert to be an alternate technology to PCR-fluorescent genotyping. However, this is not an alternative technology as it is fundamentally PCR amplification of a microsatellite marker with the same inherent susceptibility to low DNA template–generated artifacts. At best, this control simply assesses the reliability of allele quantitation by fluorescence detection on a capillary sequencer, which is not in dispute. The fourth control involved LOH analysis of four fresh-frozen breast cancer samples corresponding to four archived samples. The investigators reported finding identical LOH, which would seem to validate the LOH detected in CAFs. However, it seems that the analysis only compared LOH in the FFPE-derived epithelium with LOH in epithelium derived from a nonmicrodissected whole biopsy of fresh-frozen tumor. In these circumstances, it is entirely predictable that genuine LOH identified in microdissected FFPE-derived tumor epithelium will be replicated because it would be impossible to obtain a heterozygous genotype, regardless of the quality or quantity of DNA. The obvious control in this instance would have been to assess LOH in comparable microdissected epithelial and fibroblast compartments from these matching fresh-frozen samples, but it seems that this was not done.
In light of the self-evident technical issues relating to mutation detection in the context of limiting and poor-quality template DNA, the immediate null hypothesis would be that no genetic aberration would be observed twice if the experiment was replicated. However, almost without exception, reported CAF mutations have been derived from a single assay. Where replicate experiments have been undertaken, the description of these experiments has been cursory and no relevant primary data has ever been presented that would allow proper peer review ( 11). A critical control that has rarely been undertaken in studies reporting frequent CAF mutations is to exactly replicate the LOH and mutations analyses in tissues with a known normal genotype. Even when such controls have been reported, it would seem that the tissues were not microdissected and/or the DNA was not in limiting amounts. Consequently, such controls would not be prone to artifacts leading to a false impression that the assays were robust ( 13, 14, 30). Again, even for these inadequate controls, no primary data has been published despite the fact that such information is crucial to discerning fact from artifact.
Studies reporting the existence of frequent clonal somatic genetic alterations in CAFs have used techniques and tissues that are highly prone to generating artifacts. On the other hand, those studies reporting the absence of somatic alterations have no identifiable technical issues. It is still possible that a rare ostensive mutant CAF could be missed in an analysis of bulk stromal genomic DNA extracts, but otherwise it is difficult to conceive how the detection of a normal genotype could be the result of an artifact. Our own study used tissues and technologies that overcome the potential weaknesses of previous investigations and analyzed CAF populations that exactly mirrored those reputed to harbor rampant somatic alterations. Our data shows that whereas somatic alterations can occur in CAFs, they are exceedingly rare and are unlikely to be responsible for the stable cancer-promoting attributes of CAFs. Notwithstanding the absence of genetic alterations in CAFs, the fact remains that substantial evidence exists that the differences between normal and cancer-associated fibroblasts are heritable, indicating that other mechanisms, such as epigenetic changes ( 18, 31– 34), are operating to maintain this phenotype.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
- Received January 22, 2009.
- Accepted June 12, 2009.
- ©2009 American Association for Cancer Research.
Many independent investigators who are savvy in evolutionary genetics believe in the coevolution of tumor epithelium and mesenchyme, and there is ample evidence of epithelial-to-mesenchymal transition, as well as mesenchymal-to-epithelial transition, as has been published. Since the findings of somatic loss-of-heterozygosity and somatic mutations in the stroma of various neoplasias and other nonmalignant processes have been independently found and published by independent groups using different tissues, different preparations, different techniques, different genes, and different polymorphic markers, as outlined in the present review (Campbell and colleagues), we will assume for these purposes, the veracity of these findings in vivo.
Allow us to, again, address several technical details which are routine in this field. To begin, the amount of DNA extracted depends on which region of the tissue is used (uniformly populated versus sparsely populated), and the latter is meticulously selected. Only tumor stroma is selected given that stroma distant from the tumor does not contain genomic alterations. This type of selection is dependent on the skill of the investigators. The laser capture microdissection procedure minimizes contamination, but also controls how much stroma-specific tissue needs to be extracted to achieve an already established (by others) yield of DNA for each compartment that will yield robust results (this number is minimally 10–40 ng/μL). Importantly, DNA concentration is crucial for all downstream procedures, genotyping included. Relatedly, for all our retroprospective studies, we routinely used a minimum of 100 ng of DNA as a standard for genome-wide genotyping (whether single nucleotide polymorphism-based or microsatellite-based). Low DNA concentrations were not taken to the next level. Microsatellite genotyping has not changed much since the study of ABI377, and is still being used in population studies because they are more informative from a genetic point of view. We acknowledge the importance of DNA quality in genotyping, sequence, etc. As a routine quality control measure, the quality of the DNA is always checked and compared with germline DNA, e.g., quality amplicons in both germline and tissue. Another routine quality control measure includes experiments in duplicates and/or triplicates from both archived and frozen samples, and positive and negative controls when doing single nucleotide polymorphism and microsatellite genotyping as well as sequencing, following the determination of concordance to eliminate genotyping/loss of heterozygosity errors. Other mutation screening procedures such as multiplexing procedures (denaturing gradient gel electrophoresis) are being used routinely in our laboratory, and by others using both germline DNA and DNA from tissues. In this procedure, DNA concentrations that are relatively low have been previously replicated in DNA derived from formalin-fixed, paraffin-embedded tissues. For example, laser capture microdissection–derived single endometrial PTEN protein null glands yielded DNA with replicable somatic PTEN mutations. By comparison, those who are experienced in archeological genetics successfully use far smaller quantities of much poorer quality DNA to yield valid results.