Chronic Inflammation and the Development of Malignancy In the GI Tract

Stages of Inflammation-Associated GI Neoplastic Development

In organizing our discussion of the influence of inflammation on the development of gastrointestinal epithelial cell neoplasia it is useful to assume that such influence is exerted in a discontinuous fashion reflective of two major stages. The first stage is a stealthy stage in which the cells are subjected to largely silent mutational “hits” whereas the second stage is a more obvious stage in which the initial hits are exploited to cause more frankly neoplastic changes, possibly in conjunction with new hits. In the first stage, the initial mutations do not lead to the generation of macroscopic benign and/or invasive tumors in most instances because they initiate cellular processes (such as apoptotic cell death) that, at least for a time, results in the elimination of the potentially cancerous cell. In the second stage, however, this elimination process is undermined by inflammation-associated epithelial cell signaling that, in effect, rescues the epithelial cells bearing a pro-oncogenic change from elimination and thereby renders the cell susceptible to further tumor progression. The development of Intestinal tumors during chronic inflammation de novo such those occurring in patients with inflammatory bowel disease or mice with IL-10 deficiency experience both the first and second stages of tumor development since in these instances no oncogenic factor is initially present other than the inflammation itself. In contrast, the development of intestinal tumors in mice exposed to a potentially oncogenic agent such as azoxymethane (AOM) and then to inflammation, in effect, experience only the second stage since in this case the epithelial cells have already sustained initial oncogenic changes that are now allowed to develop into frank tumors by pro-oncogenic signaling. It should be noted, however, that inflammation-dependent second stage events are occurring in and are necessary for all tumors that develop during inflammation and may in fact be operative in human tumors that are not ordinarily thought to be associated with inflammation.

1^st Stage Effects of Inflammation on Neoplastic Development

As discussed above, first stage pro-oncogenic effects of inflammation cause pre-neoplastic cellular changes that render the cell vulnerable to further neoplastic transformation. These changes are in fact similar to those that occur during sporadic epithelial cell malignancy with the possible exception of the sequence of the neoplastic mutations (e.g., Apc mutations tend to occur earlier in sporadic colorectal cancer than in inflammation-associated malignancy) and thus can ultimately lead to malignancy in the absence or inflammation or in the presence of very low grade local inflammation usually inapparent in sporadic malignancy (2, 3). In the discussion below they are characterized as frankly pro-oncogenic events, albeit ones that are influenced by the presence of chronic inflammation. However, in the two stage formulation they do not actually cause malignancy until cells are subject to further signaling during the second stage. In this sense they are also a feature of the second phase.

In some cases 1^st stage oncogenic events consist of “up-stream” factors that exert both cytokine independent and dependent effects such as factors associated with the gut microbiome that act on cells via TLR and other innate receptors to induce cytokines and growth factors. In other cases these consist of down-stream factors induced by soluble factors such as reactive oxygen or nitrogen species and eicosanoids. In either case, the net effect of these cytokine-related and unrelated factors is to induce stepwise genetic and epigenetic changes in epithelial cells that have the potential to eventuate in frank malignancy.

Cytokines implicated in inflammation-associated malignant transformation of cells are the same as those mediating the inflammation itself such as IL-1β, TNF-α and IL-6 (4). In general, these cytokines support oncogenesis by activation of well-known pro-inflammatory mechanisms involving the activation of the NF-κB and MAPK pathways and by downstream effects of the latter such as the induction of growth factors, the inhibition of apoptosis and the enhancement of ROS production (5). However, as we shall see, IL-6 and other cytokines that activate STAT3 have a particularly important role in malignant transformation of epithelial cells.

Molecular analysis of individual malignant epithelial cells has disclosed that these cells can contain as many as 100 coding region mutations including 10-20 that have known oncogenic properties (6). In addition, these cells exhibit numerous somatic rearrangements and multiple methylated genes that have also been associated with malignancy. Knowledge of the mechanisms by which chronic inflammation creates these genetic changes is only fragmentary. Generally speaking, however, they involve inflammation-associated generation of factors that cause inactivation of tumor-suppressor genes or, conversely, the activation of oncogenes. The most cogent example of tumor-suppressor gene inactivation is that involving TP53, a gene that is central to the prevention of malignant transformation as a result of its capacity to bind to sites regulating cell cycle, responses to DNA damage or oxidative stress, i.e., processes that expose cells to the possibility of malignant transformation (7). Evidence that inflammatory factors accompanying chronic inflammation can in fact lead to inactivating mutations in TP53 comes from studies showing that such mutations can be observed in chronic HCV infection and ulcerative colitis prior to the detection of neoplasia (8).

Inflammation induced inactivating mutations in TP53 occurring in relation to inflammation have been attributed to free radicals that are generated during oxidative stress. However, other mutational mechanisms may also be operative. One such mechanism is that involving activation-induced cytidine deaminase (AID), an enzyme that facilitates DNA changes necessary for class-switch differentiation in B cells (9, 10). During its normal function this enzyme deaminates cytidine and thus produces a uracil that generates a U:G mismatch; if DNA replication starts prior to recognition by the repair system this, in turn, gives rise to non-mutational C:G to T:A transitions However, if replication is delayed there may be generation of an abasic site at the U position or recognition of the U:G mismatch by enzymes that generate a mutation at the C:G site or a nearby A:T site. The possibility that AID is involved in oncogenic mutations comes from the fact that its expression is up-regulated by NF-κB and thus cellular exposure to pro-inflammatory cytokines (11). Evidence that such inflammation-associated AID expression can lead to oncogenesis comes from the fact that cells expressing constitutively active AID exhibit mutations in TP53, and deficiency in AID reduces TP53 expression during the development of colitis-associated colorectal cancers (12, 13).

Aberrant epigenetic changes that normally occur during physiologic gene activation or inactivation constitute another possible mechanism of TP53 inactivation. These epigenetic changes consist of either histone or DNA methylation events that change the accessibility of transcription factors to promoter or enhancer sites. For instance, methylation of CpG islands in gene promoters can influence transcription factor accessibility to promoter sites and thus influence the rate of transcription (14). Inflammation induced IL-1β or ROS (as well as TNF-α and IL-6) have been implicated in the induction of aberrant epigenetic changes association with malignancy via their capacity to alter the activity of DNA methyl transferases, i.e., the enzymes that mediate methylation (15). In this context, it has been hypothesized that macrophage inflammatory cytokines and oxidative stress can lead to the recruitment of such methylases to CpG islands that are not blocked by RNA polymerase II and that while such methylation is generally a rare event, it can lead to more dense methylation at promoter sites to cause gene inactivation (16).

The above discussion of mechanisms of inflammation-associated inactivation of tumor suppressor genes focused on TP53 inactivation, but in reality similar mechanisms (or indeed other mechanisms) apply to other tumor suppressor genes. Conversely, similar mechanisms of mutational change apply to activation of oncogenes. A prominent example of the latter relating particularly to epithelial cell malignancy is that involving the Kras gene, a gene encoding a membrane-bound protein that is an essential down-stream component in the signaling pathway initiated by EGFR and other receptor tyrosine kinases involved in cellular proliferation and survival (17). Whereas in normal cells Kras is activated by various extra-cellular signals, in malignant cells it is locked into an activated state that does not require such signals and thus mediates excessive cellular proliferation and/or survival. Of interest, there is evidence that such abnormal activation of Kras may be mediated by inflammation-induced expression of pro-oncogenic miRNA, i.e., small RNA species that normally regulate mRNA survival (18).

Other mutations associated with the activation of oncogenes are those leading to perturbations of the epithelial cell Wnt signaling program that is ordinarily involved in epithelial differentiation and proliferation (19, 20). In the normal operation of this program Wnt signaling may be in an “off” mode, because the cells are not being subjected to stimulation by Wnt ligands. During this period, a key Wnt signaling molecule, β-catenin, complexes with several proteins, such as Apc, GSK3β and axin and is thus targeted for phosphorylation and proteosomal degradation. However, in the presence of mutations of one of these proteins, β-catenin degradation is impaired and is thus available for the formation of complexes with TCF/Lef proteins that mediate its translocation into the nucleus; at the latter cellular site β-catenin can transactivate proto-oncogens such as c-myc and cyclin D1 that lead to the development of tumors. The most important of the Wnt mutations with respect to colon cancer are those affecting the Apc gene (adenomatous polyposis coli gene), a gene that is mutated in most cases of sporatic colorectal cancer (especially that occurring in individual with familial adenomatous polyposis) as well as in the Apc^min murine model of colon cancer (21). Related mutations in the Wnt signaling pathway leading to tumor formation include those involving the GSK3β and the axin genes (19-21).

The importance of Apc mutations in colon tumor development is recognized not only by the fact that these mutations are found in the majority of human colon cancers, but also by the fact that mice bearing these mutations on a single chromosome, i.e., mice with loss of heterozygosity (the aforementioned Apc^min/+ mice) develop spontaneous intestinal microadenomas characterized by increased expression of nuclear β-catenin, as predicted from a disturbance in the Wnt signaling pathway (22). The fact that the development of tumors in these mice is not accompanied by presence of pre-existing intestinal inflammation or even changes in numbers of infiltrating inflammatory cells would suggest that the latter is not a key factor in the development of the tumors. However, this conclusion is not justified because it does not take into consideration the fact that Apc^min mice that lack innate signaling due to MyD88 deletion exhibit dramatically reduced tumor development and experience increased survival (23). Since such deletion results in a major reduction in the production of cytokines and other immunologic factors in the tumor bearing intestine one can confidently conclude that tumor development does in fact depend on one or more factors produced by activated hematopoietic cells in the area of the tumor development, even if such factors are not immediately evident.

A related important fact is that mice with MyD88 deletion express an equal number of microadenomas as those without deletion; this indicates that the loss of cytokine production was affecting the progression of the tumor to macroadenoma rather then initial tumor development. As noted below, in further studies of Apc-related tumor formation, a key factor necessary for tumor progression may be IL-23 and/or its down-stream induction of Th17 cytokines. This again emphasizes the relation of immune responses not only to the initial phase (1^st Stage) of neoplastic development but also to the second phase (2^nd Stage) of such development.

1^st Stage Oncogenic Events Induced by a Genotoxic Agent, AOM

Whereas, as noted above, the relation of intestinal inflammation to the development of tumors has been studied with the Apc^min model it has more commonly been investigated in studies of the effect of induced inflammation on tumor development in mice administered azoxymethane (AOM), a substance that induces epithelial cells to enter a cancer-prone state that generally stops short of neoplasia (24, 25). This approach is fundamentally different than that in models relying on spontaneous tumors because in the AOM models one does not rely on pre-existing albeit limited inflammation to “awaken” the neoplastic process; instead, Stage 1 is by-passed or accelerated and one focuses on the immunologic events that occur during Stage 2.

AOM is a metabolite of the pro-carcinogen 1,2-dimethylhydrazine (DMH) and requires the down-stream formation of a mutagenic metabolite to induce DNA-reactive products (26, 27). Upon administration, AOM is hydroxylated by the cytochrome p450 isoform CYP2E1 in the liver to form methylazoxymethanol (MAM) that is then transported to the colon where it is converted by alcohol dehydrogenase to methyldiazonium ion, a substance that can alkylate macromolecules. Such alkylation results in methylation of guanine leading to formation of 6-methyl-deoxyguanosine and N7-methyl-deoxyguanosie, DNA adducts with high cytotoxic, apoptotic, mutagenic, recombinogenic, potential (28). Specific DNA damage caused by these adducts include mutations in Apc and K-ras and Ctnnb genes, as well as GSK3β mutations that effect its phosphrylation; thus AOM creates several of the mutations observed in sporadic CRC (29).

It is important to mention that AOM-induced tumors are characterized by the enhanced cytosolic and nuclear expression of β-catenin indicating that such tumors have a dysregulated Wnt signaling pathway. In the setting of AOM-induced-tumorigenesis this can be caused by missense β-catenin mutations of β-catenin phosphorylation sites that prevent β-catenin degradation. Alternatively, they can be caused by mutations affecting the “β-catenin destruction complex” that facilitates the previously discussed mechanism of β-catenin phosphorylation and polyubiquitination/degradation (28). Thus, AOM can lead to neoplasia quite similar to that in the Apc^min mouse.

The fact that AOM administration results in the generation of mutagenic products that are in fact carcinogenic but does not actually cause neoplasia in the absence of inflammation suggests the operation of mechanisms that neutralize the AOM carcinogenic effect. One such mechanism involves the activation of the acute apoptotic response to a genotoxic carcinogen (AARGC) whereby stem cells and other proliferative cells in the crypt undergo apoptosis upon acquisition of a increased “mutational load” (28, 30, 31). The apoptotic response occurring in AARGC is dependent at the functionality of TP53, since loss of just one allele of TP53 reduces the apoptotic response to carcinogen by 50% and p53^+/− mice are more highly susceptible to AOM-induced neoplasia than are wild-type mice. This combined effect is exploited in the newly described Tp53^ΔIEC animal model of invasive colon tumorigenesis with lymph node metastasis (32). The existence of this and possibly other counter-neoplastic mechanisms that come into play after AOM administration suggests that inflammation leads to cellular changes that render AARGC and other pro-apoptotic processes inoperative.

Finally, it is important to mention that AOM is not the only chemical that can cause pre-neoplastic changes in target cells. For instance 7,12-dimethylbenzanthrocene (DMBA) has this effect in epidermis cells (33). Thus, the tumor accelerating effect of AOM is not unique, although it is the only agent of this type that has been extensively used in the study of tumors of gut epithelial cells.

2^nd Stage Effects of Inflammation on Neoplastic Development

Epithelial cells, having been subjected to neoplastic influences by stealthy hits characteristic of the first stage of pro-oncogenic inflammatory effects or to a pro-neoplastic chemical such as AOM that provides a similar influence in a more rapid fashion, are now cancer prone cells vulnerable to 2^nd stage pro-oncogenic events. It is important to note that this transition to the 2^nd Stage may depend to some extent on random events affecting cancer prone cells such as epithelial injury associated with infection or inflammation as well as epithelial self renewal after such injury. Interestingly, this, in turn, may be regulated by inhibitory immunologic pathways, since it has been shown that NLRP6 inflammasome activity in epithelial cells prevents AOM-induced tumor formation in mice subjected to DSS colitis (34).

In any case, extensive evidence now exists that STAT3 stands at the gateway to the signaling pathway that leads to tumor formation in cancer prone cells. This evidence consists first of the fact that in experimental models in which inflammation causes tumor formation (such as the AOM-related models) epithelial cells manifest increased STAT3 activation and that each of the cytokines such as IL-6 or IL-11 and several growth factors such as EGF, HGF and VEGF shown to influence tumor growth have in common the fact that they directly or indirectly cause STAT3 activation (35-38). Secondly, in AOM-related tumor models tumors with increased size and proliferation are seen in the presence of epithelial cells with increased STAT3 activation (i.e., cells with a gp130^Y757F mutation that prevents SOCS3 activation or with SOCS3 deficiency) (39). This finding is mirrored in the Apc^min/+ tumor model in that Apc^min/+ mice with mutated gp130 that cannot bind STAT3 and thus manifest reduced STAT3 activation, exhibit a decreased number and size of tumors as well as reduced early stage tumor formation; reciprocally, Apc^min mice with gp130 that cannot bind SOCS3 and thus manifests increased STAT3 signaling, exhibit increased tumor formation (40). Taken together, these data provide more or less definitive evidence that STAT3 activation plays a necessary and sufficient role in the 2^nd Stage of inflammation associated epithelial cell tumor formation. Additional studies showing that Apc^min/+ mice with epithelial cell-specific decreased STAT3 activity manifest decreased tumor formation indicate that the STAT3 signaling occurring in epithelial cells is the critical locus of tumor-inducing STAT3 activity (39, 40).

With respect to the down-stream oncogenic effects of STAT3 signaling, it was shown that STAT3 mutations causing decreased STAT3 activation do not necessarily impact Wnt signaling and β-catenin nuclear localization(40). This indicates that STAT3 signaling effects do not act via dysregulation of Wnt signaling and the latters pro-oncogenic effects are in reality 1^st Stage proneoplastic effects that are likely facilitated or aggravated by subsequent STAT3 signaling abnormalities. In the as yet limited studies exploring the neoplastic effects of STAT3 signaling, it has been shown that Apc^min/+ mice with epithelial cell-specific inactivation of STAT3 display decreased abundance of cell cycle inhibitors p16 and p21 and, concomitantly, decreased expression of the polycomb group protein Bim-1, a protein that represses p16 and p21 (40). Moreover, evidence was obtained that STAT3 transactivates the Bmi-1 gene. Thus, at least one way STAT3 activation leads to neoplastic transformation is by down-stream activation of a gene that ordinarily suppresses cell proliferation and tumor growth.

Studies linking STAT3 to down-stream pro-oncogenic effects immediately suggest that cytokines present in the inflammatory milieu are the chief purveyers of inflammation-related neoplasia. A wide variety of cytokines can induce STAT3 activation, but IL-6 is perhaps the most important activator in relation to colonic cancer since in murine experimental neoplasia induced by azoxymethane in the context of either DSS or oxazolone colitis, anti-IL-6 treatment or other methods of inhibiting IL-6 signaling is sufficient to prevent neoplasia (35, 41, 42). IL-6 is a member of the IL-6 cytokine family, a family defined by its capacity to signal via the gp130 receptor chain(43, 44). However, the ability of IL-6 and other IL-6 family members to signal through this chain is determined by a co-receptor chain that is unique to each family member and that binds to gp130 upon recognition of an IL-6 family member on the cell surface (43). Thus, the ability of various family members to induce STAT3 activation in epithelial cells depends on the ability of the latter to express the second chain dedicated to the family member. Interestingly, the co-receptor for IL-6, IL-6Rα, has been shown to be a secreted receptor that can bind to IL-6 and induce gp130 signaling upon insertion into the cell membrane as an IL-6/IL-6Rα complex. Such “trans-signaling” thus facilitates IL-6 signaling in cells, such as epithelial cells, which express relatively low amounts of membrane IL-6Rα but which secrete IL-6Rα (44, 45). The source of IL-6 in the tumor milieu is primarily myeloid cells as well as epithelial cells themselves. With respect to the latter it is possible that activated STAT3 in combination with NF-κB can transactivate IL-6, suggesting the operation of a positive feedback IL-6 loop (46). Finally, It should be noted that the prooncogenic effect of IL-6 is not necessarily limited to its ability to activate epithelial STAT3. IL-6 signaling of T cells also suppresses apoptosis and thus increases the lifespan of infiltrating cells mediating carcinogenesis (44).

IL-11, another member of the IL-6 family of cytokines, is also a potent activator of STAT3 and inducer of GI malignancy, particularly gastric malignancy. Activation of Stat3 in relation to IL-11 depends on both the production of IL-11 by cells in the gastrointestinal inflammatory milieu and the expression of the gp130 co-receptor specific for IL-11, IL-11Rα1. IL-11 production may originate from epithelial cells or hematopoietic cells manifesting STAT3 activation and initiation by one of several cytokines including IL-22; thus, once IL-11 production is initiated it may exert positive feedback on its own production via its ability to induce STAT3 activation (47). Evidence that IL-11 induces tumor formation comes from studies of mice with constitutive gp130 activation that manifest decreased gastric tumors when subjected to genetic or pharmacologic blockade of IL-11 expression (47); in addition, IL-11 deficiency has a negative effect on intestinal tumor growth in the AOM/DSS and the Apc^min/+ intestinal tumor models and this negative effect, somewhat surprisingly, is more profound than that exerted by IL-6 deficiency (47, 48). The latter may be related to the as yet unproven thesis that IL-11Rα1 rather than IL-6Rα is closely associated with neoplastic epithelial stem cells or more simply that IL-6 effects depend on the presence of soluble IL-6Rα that may not always be present in sufficient amounts.

Th17 cells have been linked to tumor cell development in the gut both in experimental systems and in humans with colonic neoplasia; in addition, in studies to be described, inhibition of the differentiation of these cells leads to abrogation of tumor growth (49). This has led to the view that IL-17 itself may also be a STAT3 activating cytokine, but there is little or no evidence that IL-17 has this property. On the other hand, Th17 cells also produce several other cytokines including GM-CSF and IL-22. Most relevant here is Th17 production of IL-22, a potent STAT3 activator by virtue of the ability of STAT3 to bind to a tyrosine-less intra-cellular domain of the IL-22R1 receptor (50). It is therefore not surprising that IL-22 has been linked to GI cancer development and that neutralization of IL-22 is more effective in preventing tumor formation than neutralization of IL-17 (51, 52). Overall, then, it is likely that Th17 cells supporting STAT3-mediated tumor formation indirectly either by IL-17 induction of IL-6 (53) or directly by induction of IL-22. In addition, it is possible that IL-17 has other effects on epithelial cells that favor its malignant transformation.

Finally, it should be noted that not all cytokines that induce STAT3 activation are not necessarily involved in the support of epithelial tumor formation. IL-10 is a good case in point since signaling by this cytokine induces sustained STAT3 activation owing to the fact that the IL-10 receptor cannot activate SOCS3 (54). In fact, IL-10 production tends to reduce tumor induction because this cytokine has anti-inflammatory effects that reduce the levels of cytokines discussed above that enhance tumor formation. The reason STAT3 activation by IL-10 does not also have pro-oncogenic effects is not at all clear. One possibility is that activation of STAT3 must occur in epithelial cells to achieve a pro-oncogenic and IL-10 activation of STAT3 is mainly a feature of hematopoietic cells rather than epithelial cells. In addition, it is possible that the intra-cellular pool of STAT3 activated in the absence of SOCS3 is more diverse by virtue of post-translational modifications and thus contains STAT3 with effects on genes that decrease production of cytokines with pro-oncogenic effects.

The Microbiome and Induction of Epithelial Neoplasia

In recent years it has become increasingly apparent the a major driver of the inflammation-associated cell signaling in both the first and second stage of malignancy induction discussed above, but especially in the second stage signaling is related to innate stimulation initiated by the microbiome. The concept that the microbiome of the gut influences the development of intestinal neoplasia derives strong experimental support from the observation that in a number of different mouse models of inflammation also associated with the development of neoplasia, the institution of a germ-free state prevents both the inflammation and the neoplasia. This was first noted in DSS colitis (55), but later in a variety of colitides associated with cytokine abnormalities such as IL-10 deficient mice and mice that lack both RAG2 and T-bet and thus produce excessive amounts of TNF-α (TRUC mice) (56-58). It could be argued that in this situation, the gut microflora are not acting alone, but are instead taking advantage of (and acting in conjunction with) an abnormal mucosal immune system. However, as already noted, Apc^Min/+ mice exhibit decreased tumor formation in the small intestine (albeit not the colon) when reared in a germ-free environment even though these mice do not manifest gut inflammation (59).

As summarized by Yang and Pei, particular bacteria appear to have a special propensity to induce neoplasia in mice such as C.rodentium or H. hepaticus but usually if not always in mice with gut inflammation or a cancer-associated genetic abnormality (56). A notable example of this is Bacteroides fragilis, a commensal organism that induces tumor formation in Apc^min mice when it is capable of producing an enterotoxin (36). In this case, the carcinogenesis could be traced to the ability of the enterotoxin to induce T cells to produce IL-17 that, in turn, cause production of IL-6 and epithelial cell expression of STAT3. Another, somewhat less clear example is mice lacking IL-10 as well as epithelial cell PTEN, the latter a phosphatase that regulates susceptibility to TLR stimulation (60). These mice develop a more severe colitis than that observed in mice with IL-10 deficiency alone that is accompanied by the appearance of colonic neoplasms. Interestingly, these manifestations are ameliorated if the mice are crossed with TLR-deficient mice, suggesting that they have increased TLR responsivity that leads to pro-oncogenic immune responses. More relevant to the present discussion is the fact that a Bacteroides organism (B. acidifacians, not B. fragilis) is increased in the fecal microflora of these mice so it is possible that this organism is also involved in signaling that causes neoplasia, but this remains to be proven.

The ability of particular organisms to induce tumor formation, noted above, has recently obtained increased salience with the appearance of evidence that carcinogenesis occurring in IL-10-deficient mice may be due to organisms that have special tumor-inducing properties. In the relevant studies it was shown that germ-free IL-10-deficient mice administered AOM and then mono-associated with either adherent-invasive Group B2 E. coli developed invasive adenocarcinoma whereas those mono-associated with Enterococcus faecalis merely developed benign tumors (61). Since the inflammations induced by infection with both organisms were similar in intensity and cytokine profile, it was apparent that the difference in the tumor response was due to the infecting organism. These findings thus led to a BLAST search of the E. coli genome that revealed that the enteroinvasive E. coli causing adenocarcinoma but not the E. faecalis harbored a polyketide synthases (pks) pathogenicity island that encodes enzymes capable of synthesizing Colibactin, an as yet poorly described peptide-polyketide hybrid previously shown to cause DNA damage and genomic instability and gene mutations in mammalian cells (62). Subsequent studies with E.coli with pks deletion revealed that without the pks island the E.coli had reduced capacity to induce DNA damage in epithelial cells in vitro and, more importantly, attenuation of the ability to induce invasive carcinoma in AOM-treated IL-10-deficient mice. In further control studies it was found that the E.coli bearing the pks island does not induce tumors in WT mice and does not induce invasive adenocarcinoma in IL-10-deficient mice not treated with AOM. This indicates that E.coli bearing the pks island is not carcinogenic in the absence of immunodeficiency and accompanying inflammation and in the absence of AOM treatment that induces initial epithelial dysplasia. In other words, the Colibactin produced by the pks island acts to move already cancer-prone (1^st Stage) cells to cells with a higher level of carcinogenicity (2^nd Stage cells). In follow-up of these finding the same group of investigators showed that ex-germ-free IL-10 deficient mice maintain a higher abundance of Enterobacteriaceae than WT mice and, moreover, that a small subset of these organisms that express the pks pathogenicity island makes its appearance, probably because such organisms have a survival advantage; thus, the inflamed gut of the mice drives changes in the resident microflora that favor the latter's carcinogenic potential.

It is important to note that the E.coli sub-group expressing pks in the studies described above are found quite frequently in normal individuals, particularly those in developed countries; in addition, these organisms are found in neonates and have a tendency to persist in the intestinal microflora for long periods of time. Nevertheless, the rate of carriage of these organisms is far higher than the incidence of intestinal cancer. This is best explained by the fact that, as made clear by the studies described above, they required addition factors to induce neoplasia such as pre-treatment with AOM and the presence of intestinal inflammation. This latter point underscores the fact that the molecular mechanism by which colibactin actually brings about neoplastic transformation in cells initially exposed to another genotoxin agent, such as AOM is not at all understood.

Previous in vitro studies suggest that potentially carcinogenic E. coli secreting colibactin cause genetic changes, but only upon direct contact with epithelial cells (63). One mechanism by which such organisms might exploit the environment in an inflamed gut to enhance their survival and to cause invasive carcinoma is by taking advantage of the increased opportunity to bind to the epithelium under these circumstances. This possibility is favored by the observation that inflammation may increase the expression of adhesion molecules such as CEACAM6, an adhesin that has been shown to bind adherin-invasive E.coli (so-called AIEC organisms) in Crohn's disease patients that is up-regulated by TNF-α or IFN-γ production (64). Alternatively, the inflammation can conceivably lead to decreased mucin expression and thus degrade the epithelial barrier that ordinarily prevents bacterial adhesion.

In studies addressing this latter possibility and also providing additional mechanistic analysis of the relation of tumor development to mucosal immune responses along the lines initially mentioned above in relation to Bacteroides fragilis, Grivennikov et al., analyzed distal colonic tumors occurring in Apc^F/wt mice harboring a Cdx2-Cre transgene that were thus lacking Apc expression on one allele of epithelial cells (49). They found that although these mice did not manifest generalized lamina propria inflammation, their tumors and mesenteric nodes harbored dendritic cells (CD11b⁺/Gr1^low cells) and macrophages expressing IL-23 as well as T cells producing IL-17. Such IL-23 and IL-17 production appeared to be central to tumor cell development since there was reduced tumor cell proliferation in IL-23^−/− mice and in IL-17ra^−/− mice. However, the effect of these cytokines on tumors was probably dependent on the induction of “down-stream” cytokines capable of causing STAT3 activation in epithelial cells such as IL-6 and IL-22 that were also elevated in lesional tissue.

In the same studies, the role of luminal commensal microflora in the induction of the pro-neoplastic IL-23-mediated mechanism became apparent from the fact that tumor growth was associated with increased intestinal permeability and penetration of microbes or microbial products into lesional tissue. In addition, tumor growth and the appearance of cytokines inducing STAT3 activation were diminished in mice rendered germ-free by antibiotic treatment. These findings suggested that microflora penetrating into tumor tissue was inducing IL-23 synthesis via TLR signaling. To test for this possibility tumor growth and cytokine synthesis was determined in irradiated Apc^F/wt mice reconstituted by transplantation of bone marrow cells incapable of developing into cells that express TLR2,4,9 or MyD88, i.e., cells with defective TLR ligand signaling. Indeed, it was found that such reconstituted mice exhibited reduced tumor growth and intra-tumoral IL-17 or IL-6 synthesis. A final observation bearing on the role of the microflora in IL-23 induction and its effect on tumor growth relates to previous studies showing that Apc deficiency leads to hyper-proliferative epithelial cells and reduced goblet cell differentiation, the latter leading to decreased production of mucus (65). This observation was amplified here with data showing that epithelial cells in tumor tissue of Apc^F/wt mice express reduced amounts of mucin associated with diminished tight junction protein production (45). This suggested that the tumor cells were themselves contributing to tumor development via their loss of epithelial cell barrier function and thus their facilitation of bacterial signaling that further encourages tumor growth. It should be noted that in contrast with the AOM/IL-10^−/− tumor model discussed above, in this Apc-deficiency model, the effect of microflora is not limited to bacteria expressing a genotoxic gene such as pkc, but rather is a property of a wide range of bacteria capable of expressing TLR ligands. Nevertheless, in both cases the epithelial abnormality essential for tumor development requires input from the host mucosal environment.

Concluding Remarks

In the above discussion we have emphasized that oncogenesis in epithelial cells subject to continuous signaling by pro-inflammatory factors occurs in a cell already prepared for frank malignant change because of prior oncogenic hits. This is the meaning of the observation that experimental induction of malignancy occurs in mice with epithelial cells exposed to a genotoxic agent such as AOM or in mice with cells harboring a genetic defect involving the Wnt pathway such as those in Apc^min mice. A new and important observation is that particular types of bacteria (E.coli species) are more capable of inducing malignancy than other bacteria. However, in this case malignancy is again induced only in AOM-treated mice so that the oncogenic E.coli are not acting de novo to produce malignant change. In fact, the most likely mechanism of cancer induction by these pro-oncogenic bacteria is related to their capacity to induce epithelial permeablility changes and thereby the entry of non-pathogenic bacteria into the lamina propria that stimulate cytokine production. This last point brings to the fore the idea that inflammation-related gastrointestinal malignancy is in a sense unique because the inflammation recruits microbiome-related stimulation factors that feed into the inflammatory cycle and the STAT3 activation that is the door to further malignant change. Whereas some of the basic factors involved in the pro-neoplastic effect of inflammation have now been delineated, some major questions remain unanswered (Box 1). Insight into these questions should provide inroads into new therapeutic avenues for the treatment of GI disease and cancer.