Metaplasia is familiar to pathologists but, because of its perceived modest clinical significance, it has never loomed very large as an issue in science or medicine. In fact, its clinical significance is probably greater than generally thought, as several metaplastic conditions represent steps in the development of cancer 1. However, this article is not devoted to clinical issues. It will seek to enquire about the biological basis for metaplasia and the relationship between these events and the cell type respecification associated with the recent preparation of induced pluripotential stem cells (iPS cells).
The concept of metaplasia carries with it three basic ideas: first, that a region of tissue is located in the wrong place; second, that it appeared there in postnatal life rather than in embryonic development; and third, that it actually developed in the wrong place rather than migrating there from elsewhere. Ectopic tissue arising during embryonic development is usually referred to as a heterotopia rather than a metaplasia, although there may be some overlap of mechanism involved in their formation. Misplaced tissues arising from abnormal migration during embryogenesis represent a common and important type of congenital abnormality 2, but one which is not relevant to our current concern.
In experimental biology the word ‘metaplasia’ is seldom used. The term ‘transdifferentiation’ was, quite properly, introduced to relate specifically to transformation of one differentiated cell type into another 3, 4, thus being a subset of metaplasias, which may involve tissue level transformations involving several cell types 5, 6. Unfortunately the word has been misused to describe the results of experiments in which stem cells from the bone marrow were alleged to transform into other tissue types after transplantation. Much experimentation on this matter has indicated that this occurs only to a very small extent, if at all 7, 8. This has somewhat discredited the word ‘transdifferentiation’ and led many to believe that any transformation of cell type is impossible. Such an extreme belief must be incorrect. Ultimately, if enough genes are turned on or off, then one cell type can certainly be turned into another. We know from developmental biology that it is not necessary to change the activity of hundreds of genes to alter a cell phenotype, because development is controlled by a relatively small number of genes encoding transcription factors whose activity determines developmental choices between programmes of gene expression. These critical genes are sometimes called ‘master control genes’. It is true that the correct master control genes can only very rarely be switched when the stimuli are confined to changes of cellular environment arising from transplantation, but the situation becomes very different when we consider the effect of misexpressing transcription factors in the target cells. In this paper we consider specifically the extent to which transcription factors of developmental importance can bring about cell- or tissue-type transformations.
Because of scepticism about the reality of metaplasias generally there is a need for careful criteria to prove their occurrence. The most important of these were enumerated some years ago by Eguchi and Kodama 9. First, the two cell states before and after the transition must be clearly defined, which will require both morphological and molecular characterization of the cells. Second, the cell lineage relationship between the two cell types needs to be established. Meeting these criteria is often easier using a tissue culture system than in an intact animal. Most cases of transdifferentiation do involve loss of differentiated appearance and some cell divisions, although there may be a few examples of direct transdifferentiation occurring without cell division 10. In general, metaplasias are much more likely to occur in situations where tissues or body parts are damaged and caused to regenerate, and the reasons for this will be considered below.
In human histopathology it is not unusual to find foci of a particular tissues in the wrong place 11, 12. Examples are the occurrence of bone in the soft connective tissue, or squamous patches in an epithelium that is normally glandular in histology. Metaplasias virtually always arise in tissues that have been subjected to chronic trauma, infection or abnormal hormonal stimulation, hence undergoing continuous regeneration. In some cases it is not clear from static preserved pathological specimens whether the ectopic tissue developed in situ or migrated from elsewhere. Highly unlikely to arise by migration are the glandular metaplasias, where patches of one tissue are found embedded within the epithelium of another. These can be found particularly in the gut and in the female reproductive system, perhaps because these two systems consist of a series of organs arranged as a tube, each organ being lined with a histologically different epithelium 1, 13. When a patch of metaplasia occurs, it is often composed of the tissue type normally derived from a neighbouring region in the embryo. For example, intestinal metaplasia of the stomach means the occurrence of patches of intestinal tissue within the gastric mucosa 14; and intestine and stomach develop from adjacent territories of the endoderm in the early embryo. Less obvious is the condition known as cystitis glandularis, where colonic-type tissue arises in the urinary bladder 15. This is a quite separate organ to the intestine in the adult, but is derived from neighbouring endoderm in the embryo, as the urinary bladder forms from the proximal part of the allantoic evagination of the hindgut.
Some metaplasias have a clinical significance because they predispose to development of cancer. For example, the bronchi are lined with columnar epithelium, but smokers often have patches of squamous metaplasia and it is from within these patches that lung cancer usually arises 16. Adenocarcinoma of the oesophagus usually arises in areas of Barrett's metaplasia, a condition in which the normally squamous epithelium of the lower oesophagus becomes converted to columnar type, with gastric and intestinal differentiation patterns 17. In such cases the metaplasia can be regarded as the first step in a multistep progression to cancer.
Signals and transcription factor combinations in development
The examples of metaplasia in humans somewhat resemble the phenomenon of homeotic mutations, well known in Drosophila and other insects, in which one body part becomes substituted for another 18. Some years ago I proposed a unified theory for metaplasias based on the misexpression of homeotic genes; in other words, that the tissue type changes because the combination of expressed regulatory genes changes 1, 13. There are five key elements to this theory (Figure 1):
1.Different body parts arise in embryonic development because different, specific, combinations of genes encoding transcription factors become activated. Such a combination constitutes a ‘code’ specifying the body part in question. The code is built up through a hierarchy of developmental decisions, mostly depending on the response of cells to inducing factors secreted by neighbouring tissues. The code works because it leads, directly or indirectly, to the activation of the relevant set of differentiation genes.
2.Many body parts are associated with specific tissue types, for example each of the organs derived from the endoderm of the early embryo has its own specific epithelium (eg oral, bronchial, thyroid, gastric, intestinal, hepatic).
3.Wherever a tissue is sustained by cell turnover, there are stem or progenitor cells that persist into adult life. It is assumed that these cells retain the same codes as the embryonic primordia.
4.Whenever the code is changed, whether by mutation, epigenetic switching or environmental effects such as hormone action, then the tissue type produced by those stem or progenitor cells will change. A metaplasia will arise if the code is changed to another normal code and other changes, such as dysplasias, may arise when abnormal codes are generated. The initial change need only occur in only one or a few cells. If the new tissue type has a growth advantage over the old, it can expand to become a macroscopic focus of metaplasia.
5.The probability of metaplasia increases with regeneration. This is because tissue damage means that stem cell niches need to be repopulated with new cells, giving the opportunity for a metaplastic focus to expand to visible size.
Evidence for this theory has gradually accumulated in recent years. First, the general principle of the developmental hierarchy has been validated by working out, step by step, what happens from the fertilized egg to the adult in all the main experimental model organisms 19, 20. Although the story is not yet complete, it is clear that fields of cells become subdivided in response to inducing factors (such as FGF, BMP, hh, Wnt) and that expression of transcription factors is activated or repressed accordingly. In many cases the transcription factor codes for different tissue types are now known. For example, the heart is encoded by a combination of Nkx2.5, Mef2, Gata, Tbx and Hand factors 21; the pancreas requires Pdx1 and p48 to distinguish it from surrounding endoderm 22; the thyroid requires Ttx1 ( = Nkx2.1), Ttx2 and Pax8 23; and maintenance of hepatocyte differentiation requires five transcription factors (HNF1α, FoxA2, LRH1, HNF4α and HNF6) in a mutually activating web 24.
The contention that stem cell populations retain the characteristics of their precursor embryonic rudiments has received some support in recent years. Although it is difficult to make side-by-side microarray comparisons, because tissue samples are usually of mixed composition, there are now many examples known where key transcription factors are important for both the developmental commitment of an embryonic rudiment and the stem cells derived from it. For example, Sox2 is a transcription factor determining the properties both of the embryonic neuroepithelium and of the postnatal neuronal stem cells 25, 26. Runx1 is required for both the initial formation of the haematopoietic tissue in the dorsal aorta and the maintenance of adult haematopoietic stem cells 27. Cdx genes are required both to specify the intestine in the embryo and to maintain intestinal stem cell zones in the adult 28. p63 appears when embryonic epithelia become stratified and remains critical for controlling the properties of squamous epithelia and of keratinocyte stem cells 29, 30.
The issue of growth of metaplastic foci to macroscopic size remains poorly understood. Normally, epithelia of different tissue types coexist in a stable manner at junctions, such as the oesophago-gastric junction and the gastro-duodenal junction of the gut, or the ecto-endocervical junction in the female reproductive tract. This is true even though the rate of cell proliferation on the two sides of the junction can differ substantially. Stability is achieved because the tissues are organized into structural–proliferative units, such as the intestinal crypts, within which cells are born, differentiate and die without migrating into surrounding proliferative units 31, 32. However, overgrowth of one tissue type by another can occur in certain circumstances, for example cervical ‘erosion’ is a migration of endocervical lining onto the exocervix. The possibility of overgrowth explains the universal association of metaplasia with tissue damage. Such damage will destroy some proliferative units and so allow some new ones to form in their place (Figure 2). Where two epithelial types abut, this situation provides an opportunity for competition to occupy the space, not simply in terms of cell division rate but in terms of the relative propensity of the two tissues to generate new proliferative units. This process has not been much studied, although it is known that the number of proliferative units, such as gastric glands or intestinal crypts, does increase by budding in both normal growth and conditions of regeneration 33, 34.
Examples of transformations arising from misexpression of a single gene
B lymphocytes into macrophages
It is known that B lymphocyte differentiation requires the transcription factors E2A and EBF. These activate expression of another transcription factor, Pax5, which in turn activates many mature B lymphocyte products, including the cell surface marker CD19. B lymphocytes were isolated from mouse bone marrow or spleen by sorting for CD19 35. Then the introduction of the basic leucine zipper transcription factors C/EBPα or β led to a transformation to macrophages, with loss of CD19 and gain of the macrophage marker Mac1. Inclusion of a second transcription factor, PU.1, improved the response to nearly 100%. The mechanism is a combination of repression of Pax5 function and activation of macrophage-specific genes.
The fact that the precursor cells were genuine B cells was shown by two methods. First, a Cre–lox lineage tracing experiment was done. A line of mice expressing the Cre recombinase under control of the CD19 promoter was mated to a line in which expression of a reporter gene (EYFP) is activated where Cre is present. This provides permanent labelling of the cells that have ever activated the CD19 promoter, and showed that the macrophages had formerly been B lymphocytes. Additional evidence was provided by a natural lineage label, which is the rearrangement of antibody genes that occurs during B cell differentiation. This rearrangement had occurred in the reprogrammed macrophages produced by the experiment.
More recently, pax5 itself has been ablated using a hormone-inducible version of Cre driven by the CD19 promoter (CD19CreER x floxed pax5). This appears to convert mature B lymphocytes into immature cells. These can differentiate into T lymphocytes of various kinds, or into macrophages. Again, the origin of the new cells can be traced to B lymphocytes because they still retain the DNA rearrangements of their antibody genes. This example is interesting because it does seem to show a genuine reversal of differentiation, rather than a change in the pathway of differentiation, arising from loss of a master control gene.
Exocrine pancreas to hepatocyte
There are quite a number of examples of cell type switching in lines of differentiated cells grown in tissue culture, several of which were discussed in a previous review 6. One problem with the in vitro approach is that cells often change their properties when they have been explanted from the intact organism and cultured, so their fully differentiated status cannot be guaranteed. On the other hand, the in vitro situation offers good opportunities for careful observation of cell phenotype and cell lineage. In one system worked on in our own laboratory we were able to demonstrate a direct transdifferentiation. AR42j-B13 cells, which have many of the properties of pancreatic exocrine cells, can be converted to well-differentiated hepatocytes by treatment with dexamethasone and oncostatin M 36, 37. Using the elastase promoter to drive GFP, we showed that in transdifferentiating cells both liver proteins and GFP could be present together, indicating that the hepatocytes must once have been differentiated exocrine cells. We identified the molecular basis of the switch as the induction of the transcription factor C/EBPβ (Figure 3). Transfection of C/EBPβ or α into AR42J-B13 cells provokes transdifferentiation, whereas introduction of its dominant-negative form, liver inhibitory protein, prevents the transdifferentiation. C/EBPα and β are normally expressed in the early liver rudiment but not in the adjacent pancreatic region 38, being complementary in domain to the pancreatic transcription factors Pdx1 and p48, so they are good candidates for transcription factors distinguishing liver from pancreas during development. This is also one of the rare systems in which transdifferentiation can occur without cell division, as shown by absence of BrdU labelling from at least some of the transdifferentiated hepatocytes 36. This in vitro system provides a possible explanation for the frequently observed foci of hepatocytes that appear during regeneration following toxic damage to the exocrine pancreas 39, 40.
Induced pluripotential stem cells
In the last 2 years there has been much excitement in the stem cell field relating to the induced pluripotential stem cells (iPS cells). These were first discovered by Yamanaka 41 and the work has since been confirmed and extended by several other laboratories 42–45 (Figure 4). Yamanaka showed that it was possible to insert four genes into normal fibroblasts and to recover clones of cells closely resembling embryonic stem cells (ES cells). The four genes concerned encoded Oct4, Sox2, c-Myc and Klf4. Since then it has been shown that not all four genes are absolutely necessary and that Lin28 may be substituted for c-Myc + Klf4 46–48. The genes are introduced using retrovirus or lentivirus, of which both insert genes into the host DNA. The method works for both mouse and human fibroblasts and for some epithelial cell types as well 49. The cells have the same gene expression profile as ES cells; they have demethylated promoters for oct4 and other genes associated with pluripotent behaviour. The mouse cells will integrate into early mouse embryos and contribute to all tissues of the fetus, thus passing the acid test for pluripotency 50. Both human and mouse cells will form teratomas after transplantation into immunocompromised mice and these contain derivatives of all three germ layers 51.
It is important to note that the method for iPS cell production involves a selective step. For the mouse systems this is often achieved by the use of an ES cell-specific promoter driving an antibiotic resistance gene. For the human cells it involves picking and testing colonies that grow in hES medium, which selects against the parent cell. The approximate frequency of iPS colony formation is 1/104 cells. When an inducible lentivirus is used to activate the inserted genes, it can be shown that activity is required for a minimum of 12 days before any iPS colonies are initiated 44. Subsequently the initiation of colonies occurs roughly linearly with time. The transgenes are often silenced in the eventual iPS cell lines, especially those which can form differentiated teratomas on implantation into immunocompromised mice. This suggests that a transient presence of Oct4 and Sox2 can eventually activate a stable programme of endogenous gene expression involving activation of oct4 and sox2 and other genes required for the ES cell phenotype. In theory, the transformation should be achievable without DNA integration and many laboratories are currently experimenting with non-integrating vectors, such as adenovirus, to find out whether this is possible.
The laboratory of Jaenisch has recently shown that it is possible to obtain iPS cells from mature B lymphocytes 52. This was done by combining the basic iPS protocol with one of the respecification-inducing treatments described above. Mouse iPS cells were prepared from embryonic fibroblasts using dox-inducible lentivirus and injected into blastocyts to generate adult mouse chimeras. Spleen cells from these mice were cultured and infected with an additional virus carrying the C/EBPα gene, then treated with doxycyclin to induce the integrated lentiviral genes. Under these circumstances, iPS colonies could be made which contained DNA rearrangements characteristic of mature B lymphocytes.
Use of iPS technology to achieve other transformations
Most discussion of iPS cells has focused on the potential for replacing the use of human ES cells for research and therapeutic cell production. This is indeed an important issue. Much of the long-term objective of stem cell research is to produce cells that would be suitable for treating diseases by transplantation, such as pancreatic β cells, cardiomyocytes or dopaminergic neurons. The possibility of doing this on an individual patient basis must be brought nearer by the existence of this technology. However, the occurrence of the phenomenon also provides insight into the general possibilities for reprogramming one cell type into another, suggesting that it may not be necessary to revert all the way to an ES-like cell in order to achieved the desired outcomes.
It is possible to interpret the iPS phenomenon along the following lines. Over-expression of Oct4 and Sox2 is sufficient to reprogramme any cell to an ES phenotype, so long as these two proteins can find their target genes in the DNA, and so long as those target gene products can, in turn, find their own targets. In other words, the genome needs to be competent for the establishment of a self-sustaining web of gene activity which determines the new phenotype. The four genes used comprise the two that actually initiate the activation of this web, and two others that presumably have the function of opening up the genome to make the action of the first two possible. The fact that the overall frequency is quite low, and that the number of colonies increases with time of exposure to the active transgenes, suggests that the relevant opening process is very inefficient. Perhaps loci are opening and closing regularly and only occasionally is the configuration right for the establishment of the new phenotype. We do not know whether 1/104 represents a few rather rare events (e.g. two events at 1/100 probability) or several relatively common ones (e.g. seven events at 1/4 probability).
However, the overall conclusion is that to achieve reprogramming you need three things:
1.The right combination of transcription factors to initiate activation of a self-sustaining web of gene activity that determines the required phenotype.
2.Treatments to open the genome in the necessary way.
3.A selective system to isolate those cells that respond appropriately to the introduced genes.
Given the correct choice of means for each of these three requirements, there seems no reason why any desired cell type respecification should not be achieved.
The transcription factor combination that would be required to specify a particular tissue or cell type can often be guessed from knowledge of normal developmental biology. Although in general many transcription factors will contribute to a specific developmental pathway, they are often coupled together in an autocatalytic manner. Furthermore, the effect of factors may be magnified by over-expression at supraphysiological levels, so it is not uncommon for over-expression of one or a few transcription factors to initiate a programme involving many others. For example, Nkx2.5 is known to be closely involved in heart development and has been shown to drive formation of larger hearts in zebrafish 53. Runx2 is required for bone development and can drive bone formation in marrow stromal cells 54, 55. Pdx1 + Ngn3 are required for pancreatic endocrine cell development and can drive their formation from ductal tissue 56. All of these examples involve starting cells that are developmentally just one step away from the final cell type, and sometimes with immature starting cells. It is more difficult to achieve reprogramming over a larger developmental distance, or with fully mature cells. The iPS method indicates that the chance of achieving successful reprogramming can be increased by including chromatin-opening agents, such as the c-Myc and Klf4 proteins.
The science of genome opening is still very rudimentary. Some substances are known that are expected to have a global non-specific opening effect. For example, trichostatin A or valproic acid will inhibit the activity of histone deacetylase enzymes. Because acetylated histones often predominate in active regions of the genome 57, treatment with these agents may on occasion produce the required genome opening 58. In a biochemically distinct mechanism, methylation of DNA at CG dimers is generally associated with repression of the genome 59, 60. Maintenance of methylation through cell division requires that new DNA strands be methylated at CGs lying opposite already methylated CGs. This process can be inhibited by drugs such as 5-azacytidine or zebularine 61, 62. These drugs have also often been noticed to cause change in cell phenotype or ectopic gene expression. Finally, there are various proteins that are believed to mobilize nucleosomes, such as Brm and Brg1 63. These might perhaps be introduced along with the active transcription factors to produce a more efficient reprogramming.
Selection can only really be done in tissue culture, so its application is limited to situations where cell transplantation is contemplated rather than those involving treatment of tissues in situ.
Although metaplasia may be a rare event in nature, understanding how it happens will enable us to design methods for causing it to happen deliberately. To do this we need both an understanding of normal developmental biology and mastery of the relevant techniques of gene transfer, tissue culture, colony selection and cell phenotyping. In principle, the task is to activate or inhibit a small number of key transcription factors in the target tissue. We can imagine two general classes of application for this technology: reversion of harmful metaplasias, and generation of useful differentiatied cells for therapeutic transplantation. The former may turn out to be more difficult to achieve, as it requires methods for reprogramming cells en masse and in situ. Production of cells for transplantation, on the other hand, may prove successful, even using inefficient reprogramming methods, so long as the required cells can be selected out and expanded sufficiently in number.
Power Point slides of the figures from this Review may be found in the supporting information.